**Contents**



## **Preface**

The progress achieved using new techniques and therapies in HN cancers has influenced chemotherapy, surgery and radiotherapy, facilitating considerable improvement in the life expectancy and in some cases even complete healing. The consequent life elongation of affected patients means that they must face the post-surgical consequences of the interventions for the excision of neoplasms and side effects of radiotherapy for a long time. The surgical treatment of these neoplasms, in fact, unfortunately often exposes patients to disabilities that impair function, aesthetics and psychology.

Among HN cancers, OSCCs still have a poor prognosis, and survivors suffer from heavy aesthetic and functional impairments since oncologic therapies, especially surgery and radiotherapy, can damage important structures in the oral cavity. Recently, scientific research in OSCCs has focused mainly on preventive strategies, early diagnosis and reconstructive techniques of the jaws after demolition surgery [1].

In recent years, however, research has also shifted to the evaluation of the life expectancy and quality of life of patients with OSCC, trying to provide new elements to better evaluate the therapeutic strategies available for the treatment of these cancers.

This Special Issue collects in vitro experiments, clinical papers (either observational or experimental) and systematic reviews to promote the spread of precision and tailored medicine. New evidence strongly supports the idea that the use of molecular diagnostic techniques for early diagnosis could improve the life expectancy and that more tailored therapies could favour better quality of live in affected patients [2].

For the abovementioned reasons, the final objective of this editorial and the associated Special Issue is to spread the most recent knowledge on OSCC patient care, with attention to prevention, early diagnosis, therapy and quality of life. The accepted topics include all the branches of oral oncology, and multidisciplinary research is encouraged.

### **References**


**Carlo Lajolo, Gaetano Paludetti, and Romeo Patini** *Editors*

## **Quality of Life after Mandibular Reconstruction Using Free Fibula Flap and Customized Plates: A Case Series and Comparison with the Literature**

**Jorge Pamias-Romero 1,2, Manel Saez-Barba 1,2, Alba de-Pablo-García-Cuenca 1,2, Pablo Vaquero-Martínez 1,2, Joan Masnou-Pratdesaba <sup>3</sup> and Coro Bescós-Atín 1,2,4,\***


**Simple Summary:** The health-related quality of life was evaluated in 23 patients undergoing mandibular reconstruction with free fibula flap and titanium customized plates. A computer-aided design and computer-aided manufacturing technology were used. The University of Washington Quality of Life questionnaire for head and neck cancer patients is a widely used and validated tool, which was self-completed by the patients after 12 months of surgery. In the 12 single question domains, the highest scores were obtained in the domains of taste, shoulder function, anxiety, and pain. The lowest scores corresponded to chewing, appearance, saliva, and mood. The global quality of life was rated as good, very good, or outstanding by 81% of patients. The present results compared favorably with previous studies of mandibular reconstruction using the same questionnaire published in literature.

**Abstract:** A single-center retrospective study was conducted to assess health-related quality of life (HRQoL) in 23 consecutive patients undergoing mandibular reconstruction using the computer-aided design (CAD) and computer-aided manufacturing (CAM) technology, free fibula flap, and titanium patient-specific implants (PSIs). HRQoL was evaluated after at least 12 months of surgery using the University of Washington Quality of Life (UW-QOL) questionnaire for head and neck cancer patients. In the 12 single question domains, the highest mean scores were found for "taste" (92.9), "shoulder" (90.9), "anxiety" (87.5), and "pain" (86.4), whereas the lowest scores were observed for "chewing" (57.1), "appearance" (67.9), and "saliva" (78.1). In the three global questions of the UW-QOL questionnaire, 80% of patients considered that their HRQoL was as good as or even better than it was compared to their HRQoL before cancer, and only 20% reported that their HRQoL had worsened after the presence of the disease. Overall QoL during the past 7 days was rated as good, very good or outstanding by 81% of patients, respectively. No patient reported poor or very poor QoL. In the present study, restoring mandibular continuity with free fibula flap and patient-specific titanium implants designed with the CAD-CAM technology improved HRQoL.

**Keywords:** quality of life; mandibular reconstruction; free fibula flap; patient-specific implant; plates; University of Washington Quality of Life Questionnaire

**Citation:** Pamias-Romero, J.; Saez-Barba, M.; de-Pablo-García-Cuenca, A.; Vaquero-Martínez, P.; Masnou-Pratdesaba, J.; Bescós-Atín, C. Quality of Life after Mandibular Reconstruction Using Free Fibula Flap and Customized Plates: A Case Series and Comparison with the Literature. *Cancers* **2023**, *15*, 2582. https://doi.org/10.3390/ cancers15092582

Academic Editors: Remco De Bree, Carlo Lajolo, Gaetano Paludetti and Romeo Patini

Received: 27 March 2023 Revised: 18 April 2023 Accepted: 28 April 2023 Published: 30 April 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Refinements in surgical techniques have led to significant improvement in oncological, functional, and aesthetic outcomes in oral cancer. Currently, one of the main goals of mandibular defect reconstruction is to provide patients with the best possible healthrelated quality of life (HRQoL) [1]. Assessment of results of treatment is a key aspect for the accurate selection of patients and the choice of the most appropriate reconstruction technique [2,3].

The microvascular or free fibula flap, originally described by Hidalgo et al. [4] in 1989, is considered the "gold standard" flap for the reconstruction of mandibular defects. More recently, the use of computer-aided design (CAD) and computer-aided manufacturing (CAM) (CAD-CAM) technology [5] promoted a paradigm shift in the diagnostic and therapeutic approach of defects in the maxillofacial region. Further introduction of 3D-printed titanium using direct metal laser sintering (DMLS) as an additive manufacturing technique allowed the development of custom-made plates or patient-specific implants (PSIs), improving accuracy and efficiency in mandibular reconstruction procedures. PSIs could provide the missing link in the digital flow process for mandibular reconstruction, and in doing so they would avoid potential shortcomings that are inherent to pre-modelled reconstruction plates and improve final precision [6,7]. Thus, computer-generated PSIs would be the next logical step in the digital planning and design flow rather than an independent device, as they represent the metallic cast that accurately reflects the surface of the reconstructed bone compounds and keeps geometry stable [8].

The evidence of PSI printed titanium implants for reconstruction of mandibular continuity defects is scarce. In a systematic review of the literature of 31 clinical studies with 139 patients, benefits identified included finite element analysis of the digital design, dimensional accuracy, shorter duration of surgery, augmenting dental/masticatory function, and capacity for dental implant rehabilitation, although the evidence predominantly was low level and at moderate-to-high risk of bias [9]. The published articles provided valuable evidence of the use of 3D-printed titanium PSIs with reported benefits seemingly outweighing their limitations and of the important role to be played by such implants in mandibular reconstruction for improving patient outcomes. However, in none of the studies included in the review was HRQoL evaluated. Improvements in different domains of HRQoL and patient satisfaction after free fibula flap reconstruction of segmental mandibulectomy have been rarely reported [10–15], but as far as we are aware, no studies have specifically assessed HRQoL outcomes in the setting of mandibular reconstruction using free fibula flaps combined with 3D-customized titanium plates.

Therefore, the aim of this study was to assess the impact of using free fibula flaps associated with CAD-CAM technology and PSI titanium plates on HRQoL in patients with mandibular pathology undergoing reconstruction for continuity defects.

### **2. Materials and Methods**

### *2.1. Study Design and Participants*

This was a retrospective study of all consecutive patients undergoing mandibular reconstruction with free fibula flaps using CAD-CAM technology and titanium PSI for the repair of mandibular defects of malignant or benign etiology operated on at the Service of Oral and Maxillofacial Surgery of Hospital Universitari Vall d'Hebron in Barcelona (Spain) between October 2015 and July 2019. Inclusion criteria were as follows: adult patients scheduled for primary or secondary mandibular reconstruction due to benign or malignant pathology, whether diseases had been treated previously or not; use of CAD-CAM technology including virtual planning, mandibular resection, fibula cutting guides for modelling, and PSI; use of free fibula flap for the reconstruction of the mandibular bone defect; and follow-up for at least 1 year after surgery. Patients were excluded if one or several components of the CAD-CAM technology were lacking (such as virtual planning, mandibular resection guides, fibula cutting guides for modelling), PSI was not used, or the free fibula flap failed.

The study protocol was approved by the Clinical Research Ethics Committee of Hospital Universitari Vall d'Hebron (codes PR(AG)93/2016, approval date 1 March 2016) (Barcelona, Spain). Written informed consent was obtained from all participants.

### *2.2. Protocol for Mandibular Reconstruction with Free Fibula Flap*

Briefly, the presurgical stage included the following steps: (a) virtual planning (image processing, segmentation, resection, cutting, and reconstruction planning); (b) CAD (mandibular resection guides, fibula cutting guides, custom-made reconstruction plates, and custom-made prostheses); and (c) manufacturing stage (polyamide models from StereoLitography (STL) file format for resection and cutting guides for the mandible and fibula, STL model for the mandible, 3D printing and manufacturing titanium plates [PSI], and custom-made polyetheretherketone [PEEK] prosthesis) (Figure 1). Custom-made plates were manufactured using direct metal laser sintering using an EOSINT M270 system (EOS GmbH, Electro Optical Systems Company, Munich, Germany).

**Figure 1.** Patient-specific implant (PSI): (**a**) screw hole with thread; it contains information on the screw angle; (**b**) an enveloping design to help place the plate in the optimal position; and (**c**) patient's information code (left). PEEK prosthesis. Positioning of the PSI in the remaining healthy bone (right).

The surgical procedure (Figure 2) included the following steps: (1) mandibular resection using resection guides; (2) modelling of the fibula flap using cutting guides and placement of immediate implants (if required); (3) plate binding in the donor zone before sectioning the vascular pedicle; (4) positioning and binding of the flap in the mandibular defect; (5) microsurgical anastomosis; (6) positioning and binding of the PEEK prosthesis with miniplates and screws (if required); and (7) final repositioning of soft tissues and wound closure. All plates were customized for each patient. In all cases, PSI modelling was performed in the limb while the flap remained vascularized.

**Figure 2.** Details of the surgical procedure: (**a**) mandibular resection; (**b**) fibula flap modelling; (**c**) plate binding in the donor zone; and (**d**) positioning and binding of the flap in the mandibular defect.

Anatomical models, surgical guides, and custom-made plates were designed using the specific design software "D-matic Medical ® 10.0 by Materialise". Biomodels were manufactured directly using a rapid prototyping machine that used tridimensional solid support technology (Stratasys, Eden Prairie, MN, USA). Plates were manufactured using direct sintering with metal laser using an EOSINT M270 system (Electro-Optical Systems, GmbH, Munich, Germany).

### *2.3. Evaluation and Follow-up*

Patients were visited postoperatively by the same investigator (J.P.-R.) during their stay in the hospital, after 1 week of hospital discharge, and at 1, 3, 6, and 12 months thereafter. Postoperative complications were evaluated using the Clavien–Dindo classification [16]. Complications related to PSI (presence or absence of intraoral or extraoral exposure) and the PEEK prosthesis (stability) were evaluated clinically. Prosthesis failure was determined when the prosthesis was extra-orally exposed and had to be removed. Other variables were evaluated by orthopantomography and computed tomography (CT) scan performed at least 6 months after mandibular reconstruction, including merging of fibula fragments (between different fibula fragments and between the fibula fragments and the remaining mandible), stability of screws, plate adjustment (defined as the presence of close contact between the PSI, the fibula, and the mandible), and presence or absence of PSI fracture.

Esthetical evaluation included photographs of the patients before and after surgery. Additionally, pre- and post-surgical panoramic radiographs and 3D-cone-beam computed tomography (CBCT) scans were acquired, and image superposition was used to assess the correlation between virtual planning and the results obtained.

### *2.4. Health-Related Quality of Life*

At least 12 months after surgery, patients were contacted by phone and were appointed for a face-to-face visit to assess HRQoL. After signing the informed consent, they completed the University of Washington Quality of Life Questionnaire (UW-QOL v4) for head and neck cancer patients [17,18]. A Spanish validation version of the UW-QOL instrument was used [19]. The UW-QOL is a self-administered questionnaire specifically for head and neck cancer patients that measures health and quality of life (QoL) over the past 7 days. The questionnaire includes 12 single question domains (pain, appearance, activity, recreation, swallowing, chewing, speech, shoulder function, taste, saliva, mood, and anxiety) and 3 global questions, one about how patients feel relative to before they developed their cancer, one about their HRQoL, and one about their overall QoL. A free-text box is also included, so that the patient may write down any other comment he or she wishes to make on QoL that had not come forth in the previous questions. Domains are scaled from 0 (worst possible response) to 100 (best possible response). Domain scores include the mean (SD), the percentage of patients selecting the best possible response (100), and the percentage of patients choosing each domain. The domains can also be ranked by order.

### *2.5. Study Outcomes*

The primary outcome was HRQoL assessed by means of the UW-QOL questionnaire at least 12 months after mandibular reconstruction using free fibula flap and titanium PSI based on the CAD-CAM technology. Secondary outcomes were complications related to the PSI and the PEEK prosthesis.

### *2.6. Statistical Analysis*

Categorical variables are expressed as frequencies and percentages, and continuous variables are expressed as mean and standard deviation (SD) or median and interquartile range (IQR) (25th–75th percentile) or range (maximum–minimum). The chi-square test or the Fisher's exact test were used for the comparison of categorical variables, and the Student's *t*-test or the Mann–Whitney *U* test were used for the comparison of quantitative variables according to conditions of application. Statistical significance was set at *p* < 0.05.

### **3. Results**

### *3.1. Clinical and Surgical Characteristics*

The study population consisted of 23 patients (56.5% men) with a mean age of 52.8 (14.2) years. Fifteen patients (65.2%) had malignant tumors and locally advanced disease. Four patients had received neoadjuvant radiotherapy or combined radiochemotherapy.

Central defects according to the classification of Boyd et al. [20] were the most common (56.5%). PSIs were inserted in the occlusal zone in 15 patients and in the basal zone in the remaining 8. The skin flap was used as an internal intraoral layer in 20 patients, as an external skin layer in 2, and both as internal and external layers in 1. One patient required bilateral nasolabial flaps because of a defect that involved a large amount of soft tissue. Arterial anastomosis was most frequently performed with the facial artery and venous anastomosis with the thyrolinguofacial trunk. Osseointegrated dental implants were placed immediately in 2 patients and in a second step in 3.

The mean (SD) ischemia time was 122 (4) minutes, and the mean duration of surgery was 10.2 (1.4) hours. Immediate postoperative complications were recorded in 11 patients, which were classified as grade I in 7 and grade IIIb in 4 (2 cases of cervical bleeding and 2 of compartment syndromes in the donor limb). These 4 patients were reoperated under general or local anesthesia. In all cases, complications were solved. The mean length of hospital stay was 23 days (range 10–55 days), without significant differences between patients without and with complications (17 [4.4] vs. 26.3 [12.9] days, *p* = 0.062).

The microvascular fibula flap survived in 100% of the patients. Postoperatively, 12 patients received chemotherapy and/or radiotherapy adjuvant treatment. Table 1 shows the main clinical characteristics of patients and surgery-related data.


**Table 1.** Clinical and surgical data of the 23 patients included in the study.


SD: standard deviation; TNM: tumor node metastasis; T: tumor; N: node; RT: radiotherapy; CT: chemotherapy; type C: defect consisting of the entire central segment containing four incisors and two canines; LCL: lateral defect-to-bilateral angle defect; CL, LC: lateral angle-to-bilateral canines; CH: lateral segment defect including the condyle and central defect; type H: lateral defect of any length, including the condyle but not significantly crossing the midline; type L: defect of the same type without the condyle.

Image superposition studies showed a high correlation (greater than 92% in most patients) between preoperative virtual surgical plan and the results obtained.

The mean length of follow-up was 26 months (range 12–50 months). Twenty-two patients (95.6%) were alive at 12 months after surgery. One patient developed a recurrence of their oral cancer and the other patient died due to cancer progression.

### *3.2. Health-Related Quality of Life*

Twenty-one patients (91.3%) completed the UW-QOL questionnaire, after a median of 27 months (IQR 19–41 months) after primary surgery. Two patients did not complete the questionnaire; one patient had an advanced stage of the oral cancer due to recurrence, and the other patient had died.

Table 2 shows the results obtained in the 12 single question domains of the UW-QOL questionnaire. The highest mean scores were found for "taste" (92.9 [13.1]), "shoulder" (90.9 [18.4]), "anxiety" (87.5 [24.5]), and "pain" (86.4 [12.8]). In contrast, the lowest mean scores were observed in the domains of "chewing" (57.1 [39.6]), "appearance" (67.9 [19.6]), and "saliva" (78.1 [27.3]).


**Table 2.** Results obtained in the 12 single question domains of the UW-QOL questionnaire.


\* This asks about which three domain issues were the most important during the past 7 days, and results expressed as the percentage of patients choosing each domain.

The highest percentages of patients selecting the best possible response (100) were 76% for "shoulder" and "taste", 70% for "anxiety", 67% for "swallowing", and 52% for "recreation" and "saliva". The lowest percentages corresponded to 10% for "appearance", 38% for "chewing", and 43% for "activity".

In relation to importance of domain, "chewing", "appearance", and "speech" were selected by 62%, 48%, and 43% of patients, respectively. "Recreation" and "shoulder" were chosen by only 5% of patients, respectively. The rank order of domains was consistent with the importance already assigned to the different domains.

In the three global questions of the UW-QOL questionnaire (Table 3), 80% of patients considered that their HRQoL was as good as or even better than it was compared with their HRQoL before cancer, and only 20% reported that their HRQoL had worsened after the presence of the disease. Additionally, HRQoL and overall QoL during the past 7 days were rated as good, very good, or outstanding by 81% of patients, respectively. No patient reported poor or very poor QoL.


**Table 3.** Responses to three global questions of the UW-QOL questionnaire.

Key to ratings: A: (0) much worse, (25) somewhat worse, (50) about the same, (75) somewhat better, (100) much better. B: (0) very poor, (20) poor, (40) fair, (60) good, (80) very good, (100) outstanding. C: (0) very poor, (20) poor, (40) fair, (60) good, (80) very good, (100) outstanding. \* Best scores: A = % of scoring 50, 75, or 100; B and C = % scoring 60, 80, or 100.

Thirteen patients (61.9%) provided an answer in the free-text box of the questionnaire. Four patients explicitly stated their satisfaction with the outcomes of surgery, but 9 patients would like to undergo dental rehabilitation for improving chewing and aesthetic functions. Other complaints were the possibility of a secondary reconstruction to improve appearance (3 cases), reduction in the extension of mouth opening (1 case), decreased saliva output and taste alterations (1 case), paresthesia (1 case), and delayed wound healing and/or paresthesia in the graft area of the lower limb.

### *3.3. PSI-Related Complications*

At 6 months after surgery, 22 out of 23 patients (95.6%) underwent clinical and radiological assessment. One patient moved to another city and was lost to follow-up. The fibula fragments were properly consolidated in all 22 patients. In 19 patients (86.4%), PSIrelated complications did not occur, whereas complications were recorded in the remaining

3 patients (13.6%). Extraoral and intraoral exposure of the PSI was clinically documented in 2 patients, and in both cases, the plate was removed, but the segments of the microvascularized fibula flap were found to be well consolidated. In the remaining patient, there was a lack of consolidation between the fibula and the remaining mandible, with screw instability and plate mobility. In this patient, removal of both the plate and the remaining segment of the mandibular ramus were performed.

### *3.4. PEEK Prosthesis-Related Complications*

A PEEK prosthesis for the reconstruction of the mandibular inferior border was performed in 14 patients (60.9%) (immediate reconstruction in 13 cases and at a later stage in 1). In 6 patients (42.8%), the prosthesis became exposed and had to be removed. Five of these 6 patients had received radiotherapy (RT) in the neoadjuvant or adjuvant setting. Removal of the PEEK prosthesis was significantly more common in patients treated with RT than in those who had not received RT (83.3% vs. 16.7%, *p* = 0.031).

### **4. Discussion**

This study shows that in patients undergoing extensive mandibular resection leading to wide mandibular continuity defects, the use of a surgical procedure based on CAD-CAM technology with free fibula flap and titanium PSI was associated with high scores in the UW-QOL questionnaire at least 12 months after surgery. In the 12 single question domains, mean scores were higher than 80 (with 100 being the highest possible response) in 9 domains (75%), with only 3 domains scoring below 80%. In the three global questions of the UW-QOL instrument, HRQoL before diagnosis of malignancy and overall QoL in the previous 7 days, high scores were achieved, as 80% and 81% of patients selected the options of much better and good, very good, or outstanding, respectively.

Assessment of QoL is a clinically relevant outcome in monitoring the treatment success and the sequelae of illness in patients with oral cancer. Subjective measures of health status can be evaluated by generic or disease-specific instruments, but due to the complex anatomy of the oral cavity, it is desirable to use specific HRQoL measures. These measures are more sensitive in assessing the impact of oral conditions on daily life activities. The relatively large number of questionnaires that are specific for diseases of the oral cavity (e.g., 14-item Oral Health Impact Profile [OHIP-14], Oral Impacts on Daily Performances [OIDP], Oral Health-Related Quality of Life [OHRQoL], European Organization for Research and Treatment of Cancer Head and Neck cancer questionnaire [EORTC-H&N35]) [21], underscores the fact that there is no gold standard tool. The UW-QOL instrument is one of the most used and validated questionnaires for patients with head and neck cancer and has shown good psychometric properties that have been specifically developed for this pathology [17,18]. Furthermore, the incorporation of importance-rating domains makes UW-QOL unique among head and neck cancer instruments [22,23]. The Spanish version of this questionnaire was validated by Nazar et al. [19] in 2010. In fact, the following characteristics of the UW-QOL questionnaire stand out: (1) it provides a specific "appearance" item related to disfigurement; (2) it allows for the evaluation of appearance problems through "recreation", "anxiety", and "mood" domains; and (3) it is quick and simple for patients to complete (it may take 5 minutes) and is easy to process.

Despite the advantages of the UW-QOL questionnaire, few studies have used this instrument for assessing HRQoL after mandibular reconstruction using free fibula flaps. In 2019, Petrovic et al. [24] conducted a systematic review of the literature and found only 6 studies in which QoL outcomes following mandible reconstruction using free fibula flap had been evaluated using the UW-QOL questionnaire. All these studies were retrospective case series. Apart from these 6 publications, we did not find any subsequent publication of the use of this questionnaire after free fibula flap reconstruction of the mandible. Therefore, the present results are compared with data reported in these 6 studies [14,15,25–28]. As shown in Table 4, mean scores obtained in our study were higher than those reported by

others, except for "appearance". Overall, "chewing" was the domain with the lowest mean values in all studies followed by "appearance", "anxiety", "speech", and "swallowing".


**Table 4.** Mean scores of the 12 single question domains of the UW-QOL questionnaire.

SD: standard deviation.

In relation to the domains in which the best score (of 100) was obtained, data were reported in four studies, with "pain", "shoulder function", "activity", and "recreation" as those with the most favorable evaluation (Table 5).


**Table 5.** Best scores obtained in the 12 single question domains of the UW-QOL questionnaire.

Data as % best score (of 100) for each domain; NR: not reported.

A remarkable finding was that the "chewing" domain had the lowest score both in our study and in the 6 studies analyzed. Additionally, this domain showed a rate of importance of 62% in the present study as compared with 76.8% in the remaining studies. On the other hand, when considering the rank order assigned to the different domains, "chewing" ranked first in all studies but one (Table 6).

**Table 6.** Importance of domain and rank order assigned to the 12 single question domains of the UW-QOL questionnaire.


Data as percentage of patients choosing which three domains were the most important during the past 7 days. Rank order of domains in parenthesis; NR: not reported.

Chewing has been shown to score worse after segmental mandibulectomy and reconstruction using composite free tissue transfer [29]. In these patients, rehabilitation with implant-supported prosthesis appears to improve QoL outcomes [30–32]. In a pilot study of 10 patients of early loaded implant-supported fixed dental prosthesis following mandibular reconstruction, patient satisfaction improved significantly after dental rehabilitation as compared to mandibular reconstruction alone [33]. Dental implants were placed in only 5 patients in our series, but 9 of the 13 patients (69.2%) reported the desire to undergo dental rehabilitation for improving chewing and aesthetic functions in the free-text box. Prosthetic rehabilitation, however, should be indicated on a case-by-case basis [31]. This decision should be based on several considerations including the medical history, prognosis, comorbidities and, particularly, the patient's desires and expectations. In addition, special attention should be paid to the surgical planning of implants, soft tissue management, and prosthodontics in order to avoid complications and achieve stable long-term results. We also believe that tests of swallowing function could help identify patients with a preserved swallowing function, which are in fact those who would benefit most from this kind of rehabilitation.

"Appearance" in the preceding 7 days was another domain selected as one of the most important by 48% of our patients, which is consistent with percentages between 49% and 67% reported in other studies [14,15,25,28]. Although "appearance" was considered an important factor, 71.4% of our patients stated in the questionnaire that their appearance had suffered slight or no changes, 19% a moderate change, and only 9.5% (2 patients) reported feeling disfigured. However, appearance did not seem to be a reason for social isolation, as "recreation" was rated as only 5% in the importance of domain and in the 7th position of the rank order. As for the overall QoL during the past 7 days, 81% reported that it was good, very good, or outstanding, and only 4 patients (20%) considered that QoL was fair. Poor or very poor ratings were not observed.

In relation to secondary outcomes, only 3 patients presented PSI-related complications, with a rate of 13.6%, which is consistent with 12.2% reported in the systematic review of Goodson et al. [9]. Plate removal was required by only 2 patients because of exposure, but no deficiencies in the consolidation process between the fibula fragment and the mandible were found.

A PEEK prosthesis was used in 14 patients for the correction of mandibular asymmetry after free fibula flap reconstruction [34]. In 6 patients (42.8%), the prosthesis was exposed and had to be removed. It should be noted that 5 of these 6 patients had received RT for the treatment of their oncological disease. Patients in whom the PEEK prosthesis was not exposed to RT did not present complications, with satisfactory aesthetic results and stability of the mandibular contour.

Esthetical evaluation was performed using pre- and post-surgical photographs, panoramic radiographs, and 3D-CBCT scans showing a high correlation between virtual surgical plan and the results obtained. Other techniques, such as cephalometric analysis and photogrammetry, were not used as the study was focused on the assessment of QoL as a primary subjective domain.

Limitations of the study include the single-center characteristics, retrospective design, and a small study population. Additionally, patients with malignant and benign conditions were included, which may have different risk factors related to QoL, particularly the use of radiation therapy and chemotherapy. However, the aim of the study was to assess the impact of the reconstructive process of the mandible (CAD-CAM, free fibula flap, and customized titanium plates) on QoL rather than the pathology itself, and in this respect, the population was homogeneous. Patients included in other series reported in the literature with which a comparison was made (Table 6) also included patients with ameloblastoma, osteoradionecrosis, and oral squamous cell carcinoma. Preoperative data of HRQoL using the same UW-QOL questionnaire was not obtained, so a within-group comparison of QoL before and after surgery was not feasible. Although only 4 patients received neoadjuvant or adjuvant RT and/or chemotherapy, the impact of this oncological treatment (e.g., impairment of salivary gland, trismus, mucositis, mouth opening limitation, etc.) was not evaluated. Other risk factors, such as oral health status, smoking, or age were not evaluated either. However, the use of a validated HRQoL instrument, such as the UW-QOL questionnaire, after a period of at least 12 months after surgery is a strength of this study. Moreover, a detailed comparison of the present findings with other studies published in the literature in which the UW-QOL questionnaire was completed by patients undergoing similar mandibular reconstruction procedures with free fibula flap is an interesting and distinctive aspect of the study.

### **5. Conclusions**

Restoring mandibular continuity with free fibula flap and patient-specific titanium implants designed with the CAD-CAM technology improved HRQoL. High scores in most specific domains of the UW-QOL questionnaire were obtained at 12 months after surgery, except for "chewing" which had the lowest score. The global QoL was considered good, very good, or outstanding by 81% of the patients. Further studies with a larger study population are necessary to confirm the present findings.

**Author Contributions:** Conceptualization, J.P.-R. and C.B.-A.; methodology, J.P.-R.; software, J.P.-R.; validation, J.P.-R. and C.B.-A.; formal analysis, J.P.-R.; investigation, J.P.-R., M.S.-B., A.d.-P.-G.-C., P.V-M., J.M.-P. and C.B.-A.; data curation, J.P.-R. and C.B.-A.; writing—original draft preparation, C.B.-A.; writing—review and editing, J.P.-R., M.S.-B., A.d.-P.-G.-C. and P.V.-M.; supervision, J.P.-R. The authors decline the use of artificial intelligence, language models, machine learning, or similar technologies to create content or assist with writing or editing of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinki and approved by the Clinical Research Ethics Committee of Hospital Universitari Vall d'Hebron (codes PR(AG)93/2016, approval date 1 March 2016), Barcelona, Spain.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Study data are available from the corresponding author upon request.

**Acknowledgments:** The authors thank Marta Pulido, for editing the manuscript and editorial assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Article* **Comparison of T1/2 Tongue Carcinoma with or without Radial Forearm Flap Reconstruction Regarding Post-Therapeutic Function, Survival, and Gender**

**Katharina El-Shabrawi 1,2,\*, Katharina Storck 2, Jochen Weitz 3,4, Klaus-Dietrich Wolff <sup>3</sup> and Andreas Knopf 1,2**


**Simple Summary:** Surgical therapy for tongue carcinoma is challenging due to the various important functions of the tongue. In order to compensate for loss of tongue tissue and function, flap reconstruction has been firmly established. Interestingly, a large number of early-stage tongue cancer receive flap reconstruction despite minor tissue loss. This study aims to investigate functional and survival differences as well as epidemiologic characteristics in tongue carcinoma patients with or without flap reconstruction. Our retrospective and prospective analyses show no significant survival or functional differences between the groups with or without flap reconstruction. Still, we were able to demonstrate that the possibility of flap reconstruction leads to a more generous tumor resection, less frequent presence of close margin, and subsequently less frequent use of toxic adjuvant therapy regimens. Moreover, for the first time, a significantly higher female ratio could be depicted in the reconstruction group (*p* = 0.02). These findings suggest that, apart from oncologic and functional factors, proportional aspects should be taken into consideration for future decisions on the optimal reconstruction method.

**Abstract:** Background: Flap reconstruction is commonly used in advanced tongue carcinoma in order to compensate for the loss of tongue tissue and function. Surprisingly, a large number of reconstructed early-stage tongue cancer can be found. Survival or functional benefits in these cases remain unclear. Methods: A retrospective data analysis of 384 surgically treated tongue carcinoma patients was conducted aiming to find epidemiologic and survival differences between patients with (*n* = 158) or without flap reconstruction (*n* = 226). A prospective functional analysis was performed on 55 early-stage tongue cancer patients, 33 without and 22 with radial-forearm flap reconstruction, focusing on post-therapeutic swallowing function as the primary endpoint, speech as the secondary endpoint, xerostomia, quality of life, and mouth opening. Results: Consistent with the current literature, we demonstrated the significantly more frequent use of flap grafts in advanced tongue carcinomas. For the first time, we depicted a higher female ratio in the reconstructed group (*p* = 0.02). There were no significant differences in survival or functional outcomes between the groups. The none-reconstructed group showed more frequent use of adjuvant C/RT despite presenting fewer N+ stages. Conclusions: The higher female ratio in the reconstruction group is plausible due to the anatomically smaller oral cavity and relatively larger carcinoma in women. A higher presence of close margins in the none-reconstruction group may explain the more frequent use of adjuvant C/RT. Since we found no survival or functional differences between the groups, we propose a critical approach toward flap reconstruction in T1/2 tongue carcinoma. At the same time, proportional aspects and adequate resection margins should be taken into account.

**Citation:** El-Shabrawi, K.; Storck, K.; Weitz, J.; Wolff, K.-D.; Knopf, A. Comparison of T1/2 Tongue Carcinoma with or without Radial Forearm Flap Reconstruction Regarding Post-Therapeutic Function, Survival, and Gender. *Cancers* **2023**, *15*, 1885. https:// doi.org/10.3390/cancers15061885

Academic Editors: Romeo Patini, Carlo Lajolo and Gaetano Paludetti

Received: 6 February 2023 Revised: 17 March 2023 Accepted: 19 March 2023 Published: 21 March 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Keywords:** tongue cancer; primary closure; flap reconstruction; functional outcome; survival outcome; gender distribution

### **1. Introduction**

Tongue cancer represents one of the most common subtypes of oral cancer, with the majority of it being squamous cell cancer [1]. Annually, there are estimated 354,900 new cases and 177,400 deaths associated with oral cancer worldwide [2]. In addition, an increasing incidence of tongue cancer has been reported over the last few years [3]. The two major risk factors for developing tongue cancer are tobacco and alcohol abuse. Despite its growing significance in the development of oropharyngeal cancer, human papilloma virus (HPV) does not show major relevance in the etiology of tongue cancer [4]. Apart from radiation and chemotherapy, primary surgical resection holds the greatest importance in the therapy of tongue cancer. In a randomized, prospective trial, Iyer et al. showed the significant advantage of surgery compared to primary chemo/radiotherapy (C/RT) regarding survival in oral cancer patients [5]. While surgical treatment remains the first-choice therapy for tongue cancer, a higher focus on its functional outcome must be established. The anatomical reduction in tongue tissue evolves into a functional loss and impairs essential abilities such as speaking, swallowing, and eating, as well as the overall quality of life [6,7]. Furthermore, deteriorated tongue functionality can lead to a decline in survival through complications such as aspiration. For this reason, especially for wide tumor resections in advanced tumor stages, flap reconstruction has been firmly established in order to compensate for the loss of tongue volume and function [6]. However, for early-stage tongue cancer (T1/2), there is an inconsistent opinion on the necessity of flap reconstruction [8,9]. In the literature, there is poor data concerning the functional or survival benefits of performing flap reconstruction after smaller resections of tongue tissue. However, a large number of flap-reconstructed early-stage tongue cancers can be observed. Oncologically, the possibility of flap reconstruction during tumor resection might allow an extension of safety margins and thus better survival outcomes. Consequently, the higher rate of tumor-free margins could result in less use of adjuvant therapy in nodal-negative patients. Considering postoperative functionality, flap reconstruction could improve tongue mobility and thus have a positive effect on speech and swallowing function. These aspects prompted us to investigate not only functional but also survival outcomes after partial glossectomy in patients with or without flap reconstruction. We performed a retrospective analysis on a cohort of 384 surgically treated tongue cancer patients regarding therapeutic modalities and survival as well as epidemiological characteristics. Based on this retrospectively analyzed cohort, we recruited patients with surgically treated T1 and T2 tongue carcinomas for further prospective analyses of their functional outcomes. To receive conclusive results, we defined clear inclusion criteria and recruited only patients in tumor stages T1 and T2 with carcinomas of only the mobile tongue to exclude the possible involvement of the floor of the mouth. To form comparable cohorts, only patients who received reconstruction with or without radial-forearm flaps were included. Postoperative functionality was assessed in swallowing as the primary study endpoint, speech as the secondary endpoint, sicca symptoms, mouth-opening, and overall quality of life. The null hypothesis was that flap reconstruction improves functional outcomes.

### **2. Materials and Methods**

### *2.1. Retrospective Analysis*

To get a rough breakdown of patients' epidemiology, therapeutic regimes, and particularly the type of tongue reconstruction, a total of 384 patients with surgically treated tongue carcinoma were analyzed retrospectively. Patients from both the department of otorhinolaryngology and the clinic for oro-maxillofacial surgery were included. Epidemiological data, e.g., age, gender, therapy, reconstruction method, as well as survival data (death/loss

to follow-up), were analyzed according to the UICC manual, 7th edition. Patients with a known history of head and neck cancer were excluded from the study. Retrospective data collection and analysis were confirmed by the local ethical committee.

### *2.2. Patient Selection*

A homogeneous sub-cohort of T1/2 cancer patients was chosen for further functional assessment. To avoid surgical bias, only patients with T1/2 cancer of the mobile tongue and without the involvement of the floor of the mouth were included. Patients were divided into two groups: patients with and without radial forearm flap reconstruction (RFF).

From the 384 retrospectively identified patients, there were 204 T1/2 patients without flap reconstruction and 111 T1/2 patients who received any flap reconstruction. To receive reliable results regarding swallowing function, we excluded all patients with secondary tumors in the head and neck area. To build comparable groups, we additionally excluded all patients who received other flaps than RFF. Finally, 143 T1/2 patients without flap reconstruction or radial forearm flap reconstruction remained. RFF was selected because it is considered the first choice for small defects of the tongue [10]. A total number of 55 patients who had undergone partial glossectomy agreed to participate in the study, the other 88 patients could not be recruited due to residential distance, missing contact information, personal reasons, their medical condition, or were already deceased (Figure 1). Informed consent was obtained from all 55 patients.

**Figure 1.** Patient selection.

Reconstruction with an RFF was performed on 22 patients. Thirty-three patients underwent surgery with primary closure or healing by secondary intention. The date of surgical intervention and the date of functional analysis were defined and calculated as delta therapy analysis in months. Additionally, clinical parameters on age, sex, height, and weight, TNM-staging (referring to the UICC 7th edition), grading, and treatment modalities, as well as personal risk factors such as nicotine consumption in pack years and alcohol consumption in quantity and quality, were collected. Furthermore, histopathological data on maximum tumor diameter, maximum depth of penetration, minimal tumor-free margin, as well as tumor-free margins at first pass or by follow-up resection were gathered retrospectively. At least two experienced pathologists histologically reviewed all tumor samples.

The local ethical committee approved the study (289/16S).

### *2.3. Swallowing Assessment*

To assess the extent of dysphagia, patients completed the 100 mL water swallowing test (WST) [11], which portrays the primary endpoint of the study. The WST previously proved to be a valuable tool to assess post-treatment swallowing performance in head and neck cancer patients [12,13]. In addition, swallowing dysfunction was determined by the M.D. Anderson Dysphagia Inventory (MDADI) [14], which contained 20 dysphagiarelated questions and was self-completed by the patients. A score between 0 and 100 could be achieved, with lower scores indicating higher levels of dysphagia. Moreover, a score for changes in eating habits was developed based on the Toxicity Criteria of the Radiation Therapy Oncology Group (RTOG) and the European Organization for Research and Treatment of Cancer (EORTC) [15]. Patients were categorized into five different groups by the examiner according to their dietary changes (Table 1).


**Table 1.** Dietary changes.

### *2.4. Speech Assessment*

Speech problems were evaluated using the Speech Handicap Index (SHI) [16], which was self-completed by the patients. The questionnaire consisted of 30 items dealing with daily impairments in social interactions. A score between 0 (no problems) and 120 (high grade of speech problems) could be obtained.

### *2.5. Xerostomia Assessment*

Mouth dryness was assessed in patients performing the Saxon Test [17]. To quantify saliva production, patients insalivated a 5 × 5 cm sterile sponge for 2 min. The sponge was weighed before and after salivation. Additionally, all patients completed the Visual Analogue Scale xerostomia questionnaire (VAS) for salivary dysfunction [18], which contained eight items dealing with problems caused by mouth dryness. For each item, a symptom severity scale between 0 and 100 was calculated; higher levels indicating more symptoms. The test included the following questions:


### *2.6. Mouth-Opening*

Due to surgery and radiotherapy in tongue cancer patients, problems with mouth opening are frequently reported [19]. We assessed mouth opening in all patients as the maximal distance from the upper alveolar ridge to the lower alveolar ridge using a measuring compass. Measurements were taken in millimeters. Furthermore, we determined the Mallampati score in all patients.

### *2.7. Quality of Life Assessment*

In order to assess the physical and psychosocial quality of life after cancer surgery, patients completed the Head and Neck module of the EORTC Quality of Life Questionnaire (QLQ-H&N35) [20]. The questionnaire contained 35 items and portrayed frequently reported health and lifestyle changes in patients diagnosed with head and neck cancer face, including the categories of pain, swallowing, teeth problems, problems with opening mouth, mouth dryness, sticky saliva, loss of senses, coughing, speech problems, feeling ill, social eating, social contact, sexuality, use of pain killers, use of oral supplements or a feeding tube, as well as weight loss and weight gain. A symptom score between 0 (= no symptoms) and 100 was calculated for each category.

### *2.8. Statistical Analysis*

Statistical analysis of the data obtained was performed with the Statistical Package for the Social Sciences (SPSS Inc., Chicago, IL, USA). Continuous variables were represented by the arithmetic mean and the corresponding standard deviation. Categorical variables were represented by absolute and relative frequencies. Group comparisons between reconstructed and none-reconstructed patients were performed for all retrospectively gathered data as well as for all prospectively assessed functional results, with the WST being the primary study endpoint. Group comparisons of categorical variables were performed using the chi-square test or, for small data sets, the Fisher exact test. For continuous variables, the unpaired Student's t-test was applied. Survival rates were defined from the date of surgical intervention and were calculated and illustrated using the Kaplan–Meier method. Differences between survival rates were assessed using the log-rank test. A *p*-value of <0.05 was defined as significant in all statistical tests.

### **3. Results**

### *3.1. Retrospective Analysis*

From a total of 384 surgically treated T1–4 tongue carcinoma patients, 226 patients received primary closure and 158 underwent flap reconstruction. One hundred and four patients received reconstruction using a radial-forearm flap. Regarding T-status, the nonereconstruction and reconstruction groups showed a distribution of 51% vs. 32% of T1, 39% vs. 38% of T2, 6% vs. 22% of T3, and 4% vs. 8% of T4 cases. Both groups were compared, and there were significantly more cases of advanced T-status in the reconstruction group (*p* < 0.0001). Subsequently, we depicted significantly more cases of advanced N-status and positive R-status in the reconstruction group (*p* = 0.03; *p* = 0.02). The TNM status is summarized in Table 2. Adjuvant C/RT was significantly more often performed in the reconstruction group (*p* < 0.0001). Interestingly, we found a significantly higher female ratio (26% vs. 38%) in the reconstruction group (*p* = 0.02). Survival analysis was performed on T1 and T2 cases. Analyses showed a mean overall survival (OS) of 116 months in the none-reconstruction group and 118 months in the reconstruction group. There was no

significant advantage regarding OS from performing flap reconstruction (*p* = 0.47, Figure 2).

**Table 2.** Retrospective data.


**Figure 2.** Overall survival (OS) for T1/2 carcinoma comparing primary closure and flap reconstruction.

To identify whether the significantly higher female ratio in the reconstruction group was due to an also advanced T-status in women, we analyzed the distribution of T status among men and women in the reconstructed patients. Analyses showed a significantly higher proportion of T3–4 status in men (34% vs. 21%; *p* = 0.005) among all reconstructed patients (Table 3).


**Table 3.** T status distribution among men and women in reconstructed patients (*n* = 158).

In order to specify the results for early tumor stages, we performed an additional data analysis for T1 and T2 stages only. Here, there was no statistically significant difference for N, M, and R stages between the reconstructed and non-reconstructed tongue carcinomas. Regarding therapy regimens, adjuvant therapy was used more frequently in the reconstruction group, which can be attributed to the correspondingly higher presence of N+ in this group (Table 4).


**Table 4.** Calculation of N, M, R stages, therapy, and reconstruction method for T1/2 cases only.

### *3.2. Epidemiology of the Prospective Functionally Analyzed Cohort*

To form comparable groups for the assessment of functional outcomes in early-stage tongue cancer, we excluded all patients who received different flaps than the RFF. Finally, a total of 55 patients diagnosed with tongue cancer in tumor stages T1 and T2 participated in this study, comprising 33 patients who underwent partial glossectomy without reconstruction and 22 patients who underwent partial glossectomy with RFF reconstruction. At this point, it should be taken into account that biases with regard to patient selection are possible due to the small number of cases. Patients' treatment decision was made after the tumor board consensus and recommendation of head and neck surgeons. The mean duration between surgery and functional analysis was 78 months and 66 months, respectively, without statistically significant differences between the groups (*p* = 0.49, Table 5). The mean age of patients at initial diagnosis was 49 years and 51 years. A higher female ratio (15% vs. 36%) was found in the RFF-reconstruction group, although it could not be regarded as significant (*p* = 0.09). The height-to-weight ratio did not differ significantly between the groups (*p* = 0.06). Only one patient in the RFF-reconstruction group required a gastral tube; none required a permanent tracheostomy.


**Table 5.** Epidemiologic data of the functionally analyzed cohort.


A distribution of T1 (67–68%) and T2 (32–33%) was evenly reported in both groups, as well as a majority of N0-status (82% and 77%). When compared to the RFF-reconstruction group, the none-reconstruction group showed a significantly advanced grading, with 89% of patients being diagnosed with G2 or G3 (*p* = 0.016). In both groups, the maximum tumor diameter ranged from 16 to 17 mm and the maximum depth of penetration from 8 to 9 mm. The minimally achieved tumor-free margin amounted to 4 mm in both groups. R0 resection on the main sample could be achieved in 67% of cases in the none-reconstruction group and in 77% of cases in the RFF-reconstruction group (*p* = 0.45). Other patients underwent R0 resection during the same surgery by follow-up resection. In this study, there was only one patient in the RFF-reconstruction group who was diagnosed with R1 resection in the final histology. The patient refused further surgical procedures and underwent adjuvant treatment.

All patients received a neck dissection. In both groups, the majority of patients did not undergo an additional adjuvant therapy, referring to nodal negativity and R0 resection. Additional radiotherapy was performed on 33% and 14%, and additional chemoradiotherapy was performed on 6% and 14%, respectively (*p* = 0.81). Subsequently, seven patients without RFF underwent adjuvant treatment escalation due to close margins, while only one patient in the RFF group underwent adjuvant treatment escalation referring to R1 status. However, this tendency did not achieve statistical significance (*p* = 0.09).

There were slightly fewer patients with prior or active use of nicotine in the nonereconstruction group, comprising 55% and 46% of none-smokers (Table 6). The average smoking time ranged from 17py in the none-reconstruction group to 15py in the RFF group. A non-significant higher consumption of alcohol could be depicted in the nonereconstruction group, showing an average daily alcohol consumption of 615 mL or 397 mL respectively (*p* = 0.46). Drinking habits, considering active, prior, or none alcohol consumption, did not differ significantly between the groups. In both groups, the patients' majority continued alcohol consumption after surgery (42% and 73%). In both groups, the most common liquid consumed was beer.


**Table 6.** Noxae.

### *3.3. Swallowing*

The primary endpoint of this study was swallowing function determined through the 100 mL water swallowing test (WST). The mean amount of water swallowed per second in patients with primary closure was 17.89 mL/s. Patients receiving an RFF reconstruction swallowed 15.96 mL/s. In this objective assessment of patients' swallowing function, there were no statistically significant differences between both groups (*p* = 0.39, Table 7). Additionally, no significant differences in nasal reflux (18% vs. 5%; *p* = 0.1) or in the RTOG dysphagia score (0.44 vs. 0.86; *p* = 0.09) were found. As a subjective assessment of patients swallowing function, the MD Anderson dysphagia inventory showed a remaining swallowing function range of 60–65/100 in both groups (Figure 3). No significant differences between the groups were reported in this test (*p* = 0.28).


**Table 7.** Functional analyses.

**Figure 3.** Results of the Speech Handicap Index (SHI) and the MD Anderson Dysphagia Inventory (MDADI) compared. A low symptom score in the SHI represents fewer problems with speech. A low score in the MDADI indicates greater difficulties in swallowing function.

### *3.4. Speech*

The Speech Handicap Index represented the secondary endpoint of this study and depicted patients' subjective speech issues. The mean total SHI score ranged from 25 points in the none-reconstruction group to 37 points in the RFF-reconstruction group, showing fewer speaking problems in the none-reconstruction group (Figure 3). There were no significant differences between the groups (*p* = 0.14).

### *3.5. Xerostomia*

Saliva production in the Saxon test ranged from 2.07 g/2 min in the none-reconstruction group to 1.87 g/2 min in the RFF-reconstruction group. The Sicca VAS score revealed symptom scores from 18% (Q5) to 30% (Q8) in the none-reconstruction group and 22% (Q4) to 29% (Q1) in the RFF-reconstruction group (Figure 4). In both the objective and subjective tests, no significant differences were found between the groups (*p* = 0.53; *p* = 0.71).

**Figure 4.** Results of the Sicca VAS score.

### *3.6. Mouth-Opening*

The maxilla-mandible (gingiva-to-gingiva) distance ranged from 60.34 mm in the nonereconstruction group to 61.14 mm in the RFF-reconstruction group. The none-reconstruction group showed a Mallampati score of 2.53, and the RFF-reconstruction group had a score of 2.36, respectively. In both tests, there were no statistically significant differences between the groups (*p* = 0.81; *p* = 0.62).

### *3.7. Quality of Life*

The most severe symptom reported in the QLQ-HN35 was mouth dryness in both groups, followed by coughing in the none-reconstruction group and social eating in the RFF-reconstruction group. On any of the symptom scales, there were no statistically significant differences found between the groups. Figure 5 illustrates all 18 symptom scales for both groups.

**Figure 5.** Results of the EORTC QLQ-HN35 for primary closure and RFF reconstruction compared.

### **4. Discussion**

Flap reconstruction is firmly established in the advanced stages of tongue carcinomas, aiming to restore volume and preserve tongue mobility [6,21]. Our retrospective analyses of the large cohort of 384 tongue cancer patients showed significantly more cases of advanced T- and N-status as well as positive R-status in the reconstruction group when compared to the none-reconstruction group. In addition, escalation of therapy via adjuvant C/RT was depicted more often in the reconstruction group, which is coherent with advanced T-, N-, and positive R-status. These findings are consistent with the current literature [22–24]. Interestingly, our retrospective analyses also depicted a large amount of reconstructed early-stage tongue cancer. In fact, we were able to denote that the majority of reconstructed patients were diagnosed with either T1 (32%) or T2 (38%). This observation suggests either a functional or a survival benefit from performing flap reconstruction and prompted us to further investigate these parameters.

From a surgical point of view, survival can be improved by extending the distance between the tumor and the resection margin. Several studies have shown that the margin size and achieving R0 on the main sample correlate with significant improvements in recurrence-free intervals (RFI) and thus OS [25,26]. Consequently, the possibility of flap reconstruction might surgically lead to a more generous tumor resection and improve the chances of R0 on the main sample. Through this consideration, reconstruction could contribute to a survival benefit. Analyses of the histopathological data of our functionally examined patients did not show a statistically significant difference concerning R0 on the main sample between the groups. This indicates that sufficient resection can be achieved even without flap reconstruction. These results were confirmed by the analysis of overall survival in the large retrospective group. There were no statistically significant differences regarding OS in the T1 and T2 stages between the none-reconstruction and reconstruction groups (116 vs. 118 months, *p* = 0.47). In contrast, we were able to depict the more frequent use of adjuvant C/RT in the none-reconstructed group when compared to the RFF group (39% vs. 28%), despite the predominant N0 status in both groups (82% and 77%). Indications for adjuvant radiotherapy in T1/2 oral cancer include positive resection margins (R1) or the presence of lymph node metastases (N+) [27]. Lymph node metastases with extracapsular extension (ECE) are an indication of adjuvant chemo-radiotherapy [28]. In the case of close margins, adjuvant radiotherapy is also frequently applied [27]. Since the none-reconstructed group showed fewer R0 resections on the main sample, there are higher chances of close margins in these cases. The higher presence of close margins consequently led to the more frequent use of adjuvant C/RT escalation in the none-reconstruction group, comprising 21% and 5%, respectively. In turn, these results suggest that, due to the possibility of flap reconstruction, a more generous tumor resection was possible in the RFF group, with lower chances of close margins and therefore a less frequent need for adjuvant C/RT and its respective concomitant toxicities. However, this tendency failed to achieve statistical significance in our study (*p* = 0.09). Lu et al. compared margin size and recurrence rate in a large cohort of 347 early-stage tongue carcinoma patients and found that the utilization of flap reconstruction achieved a significantly larger pathologic free margin and had significantly lower recurrence rates [29].

The surgical objective after partial tongue resection should not only include the best oncological outcome but also consider functional aspects. Essential functions such as airway protection, swallowing, and speaking should be safely practicable and maintain an optimal quality of life for the patient. Regarding functionality in advanced tumor stages, Canis et al. compared microvascular flap reconstruction and primary closure in T3 tongue carcinomas and showed significant functional impairments in patients who did not receive flap reconstruction [30]. Our study assessed functionality in T1 and T2 tongue carcinomas in both subjective and objective test batteries. Our results show that there is no significant functional advantage in any field examined. Regarding swallowing, the none-reconstruction group even showed a slightly better mean of water-drinking time, a lower symptom score in the QLQ-HN35, a lower RTOG dysphagia score, and better results in the MD Anderson dysphagia inventory, despite not being statistically significant. Only the presence of nasal reflux was observed to be less frequent in the RFF reconstruction group. A similar trend could be observed in the subjective speech questionnaires: both the QLQ-HN35 symptom score for speech problems and the SHI portrayed slightly more problems with speech in the RFF-reconstruction group, although this could not be regarded as statistically significant. These findings demonstrate that the flap not only serves to provide bulk and improve mobility but can also interfere with processes such as articulation or swallowing and necessitates training. Similar results were obtained in a study by Ji et al. showing significantly worse functional outcomes in articulation, tongue mobility, and speech intelligibility in flap-reconstructed tongue carcinoma patients [6]. A study by Kaur et al. placed the main emphasis on subjective patient satisfaction and found that higher levels were achieved for the primary closure of smaller defects and for flap reconstruction of larger tongue defects [31]. Interestingly, the most important factor regarding the QLQ-HN35 with the highest symptom scores for both groups was mouth dryness, which was also demonstrated to be relevant in the Sicca VAS score. This subjectively present mouth dryness was objectively confirmed in the Saxon test, demonstrating a mean saliva production of <2.75 g/2 min in both groups, which was defined as pathological [17]. A major factor that can cause dry mouth is the use of adjuvant C/RT.

The time range between the end of therapy and functional assessment was 78 months and 66 months, respectively; therefore, this study mainly depicts long-term functional outcomes. In this period of time, other factors might have impacted speech and swallowing function, which must be considered when applying the results for clinical purposes. Still, a comparison considering long-term functional outcomes was possible since the period

between therapy and functional assessment did not differ significantly between the groups (*p* = 0.49).

When analyzing our retrospective cohort, we saw a statistically significant higher distribution of women receiving flap reconstruction (*p* = 0.02). To identify whether this was due to an also advanced T-status in women, we analyzed the T-status distribution among reconstructed male and female patients. Surprisingly, analyses showed significantly smaller T status in reconstructed women when compared to men (*p* = 0.005). In conclusion, this shows that women received flap reconstruction despite having a smaller T-status. Anatomically, women tend to have smaller oral cavities and less tongue tissue, which indicates that T1 or T2 tongue carcinomas in women are proportionally larger than in men. Therefore, the more frequent usage of flap reconstruction is conclusive. Generally, these findings suggest that not only the tumor size should be taken into consideration but also its proportion to the remaining tongue tissue and protection of the mandible. A uniform classification system for tongue reconstruction does not yet exist. However, possible solutions have been pointed out in various studies so far. Mannelli et al. proposed a strategic approach for surgery of different types of tongue defects where a specific reconstruction algorithm is available for each type of defined tongue defect [23]. Ansarin et al. proposed a similar classification system that is based on the anatomical and functional components of the tongue and the spread routes of tongue cancer [32]. A proportional concept with consideration of the actual size of the tongue and oral cavity is missing in both studies. Given our results, the distance between the tumor and the tongue midline, for instance, could serve as an indicator of the relative volume loss and consequently be used as a criterion for flap reconstruction. Our research revealed no other previous study describing differences in tongue reconstruction in women.

To further determine whether N or R status exerted an influence on the decision to perform flap reconstruction, we specifically analyzed their distribution in early tumor stages of tongue carcinoma in the retrospective cohort. Again, there was no significant difference between the groups. This underlines the fact that the decision to perform flap reconstruction depends individually on the local expertise of the surgeon.

Apart from fitting the proportionally perfect flap, performing microvascular flap reconstruction also involves the management and training of non-innervated tissue, which is crucial for abilities such as speaking and eating. Furthermore, RFF reconstruction involves the loss of the radial artery at the donor site and possible complications such as necrosis of the flap due to vascular anastomosis insufficiency [33,34]. A major advantage when performing primary closure of the tongue is the shorter operative time and thus lower risk of delirium, more likely avoidance of tracheostomy, and no or short intensive care duration.

To our knowledge, this is the first study analyzing objective and subjective parameters in surgically treated early-stage tongue cancer, specifically comparing none-reconstruction to RFF-reconstruction, which is one of the strengths of this study. TNM stages, age, and gender were also evenly distributed between the functionally studied groups, which supports the results from the group comparison. At the same time, the small case number of the functionally examined cohort must be taken into account, as must the possible bias in patient selection. Moreover, the long time range between the end of therapy and functional assessment must be considered when applying the results for clinical purposes, as other factors might have impacted speech and swallowing functions during this period. In contrast, the retrospective analysis of 384 surgically treated tongue carcinoma patients provides a broad overview regarding epidemiological differences and treatment regimens. By including patients from the ENT and maxillofacial surgery departments, we were able to form a very heterogeneous collective, which can be well applied to actual clinical practice. In addition to overall survival, it would have been interesting to measure disease-specific survival as well. This could not be conducted because it was not evident from our existing data if death was caused by a tumor diagnosis. In summary, our findings indicate that a more restrained approach to the usage of flap reconstruction in smaller carcinomas of the

tongue is favorable and that the loss of tongue tissue proportionally to the remaining tongue volume should be taken into consideration for an optimal functional outcome. For surgical therapy of T3/4 tongue carcinoma, flap reconstruction is undoubtedly recommended.

### **5. Conclusions**

Our study demonstrated that there are no statistically significant differences regarding functional and survival outcomes between flap reconstruction and none-reconstruction in early-stage tongue carcinomas. This suggests that the implications of reconstructing T1 and T2 tongue carcinomas should be deeply evaluated beforehand, taking possible complications and necessary training into consideration. At the same time, we showed that the possibility of flap reconstruction leads to a more generous surgical resection, less frequent presence of close margin, and subsequently less frequent use of toxic adjuvant therapy regimens. Furthermore, we demonstrated for the first time that women were significantly more likely to be reconstructed by flap surgery, even when presenting a smaller T-status. This indicates that reconstruction cannot be determined by the tumor size alone but requires a proportional approach based on existing anatomical circumstances. Future research is needed to identify and develop clear guidelines for the usage of flap reconstruction in early-stage carcinomas of the tongue.

**Author Contributions:** Conceptualization, K.E.-S. and A.K.; methodology, K.E.-S., A.K., K.S., J.W. and K.-D.W.; software, K.E.-S., A.K., J.W. and K.S.; validation, K.E.-S., A.K., K.S., J.W. and K.-D.W.; formal analysis, K.E.-S., A.K., K.S., J.W. and K.-D.W.; resources, K.E.-S., A.K., K.S., J.W. and K.-D.W.; data curation, K.E.-S., A.K., K.S., J.W. and K.-D.W.; writing—original draft preparation, K.E.-S., A.K., K.S., J.W. and K.-D.W.; writing—review and editing, K.E.-S., A.K., K.S., J.W. and K.-D.W.; visualization, K.E.-S. and A.K.; supervision, K.E.-S., A.K., K.S., J.W. and K.-D.W.; project administration, K.E.-S., A.K. and K.S.; funding acquisition, A.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinki, and ap-proved by the Institutional Ethics Committee (Ethikkommission Klinikum rechts der Isar, Technische Universität München (289/16S)).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data can be shared up on request.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Article* **Patterns of Postoperative Trismus Following Mandibulectomy and Fibula Free Flap Reconstruction**

**Rex H. Lee 1, Cara Evans 1, Joey Laus 1, Cristina Sanchez 2, Katherine C. Wai 1, P. Daniel Knott 1, Rahul Seth 1, Ivan H. El-Sayed 1, Jonathan R. George 1, William R. Ryan 1, Chase M. Heaton 1, Andrea M. Park <sup>1</sup> and Patrick K. Ha 1,\***


**Simple Summary:** Trismus is a serious sequela of head and neck cancer (HNC) treatment that can profoundly affect quality of life. While the relationship between radiotherapy and trismus in HNC has been established, the surgical risk factors for trismus in HNC patients are largely unclear. This study reports the prevalence of postoperative trismus in a large cohort of patients who underwent mandibulectomy and fibula free flap reconstruction. Patients with a posterior mandibulotomy that involved or removed the ramus had significantly higher rates of persistent trismus >6 months after surgery, which was also demonstrated in a multivariable logistic regression. These findings may inform future surgical planning and potentially optimize functional outcomes in patients undergoing significant mandibular resection.

**Abstract:** The factors that contribute to postoperative trismus after mandibulectomy and fibula free flap reconstruction (FFFR) are undefined. We retrospectively assessed postoperative trismus (defined as a maximum interincisal opening ≤35 mm) in 106 patients undergoing mandibulectomy with FFFR, employing logistic regression to identify risk factors associated with this sequela. The surgical indication was primary ablation in 64%, salvage for recurrence in 24%, and osteonecrosis in 12%. Forty-five percent of patients had existing preoperative trismus, and 58% of patients received adjuvant radiation/chemoradiation following surgery. The overall rates of postoperative trismus were 76% in the early postoperative period (≤3 months after surgery) and 67% in the late postoperative period (>6 months after surgery). Late postoperative trismus occurred more frequently in patients with ramus-involving vs. ramus-preserving posterior mandibulotomies (82% vs. 46%, *p* = 0.004). A ramusinvolving mandibulotomy was the only variable significantly associated with trismus >6 months postoperatively on multivariable logistic regression (OR, 7.94; 95% CI, 1.85–33.97; *p* = 0.005). This work demonstrates that trismus is common after mandibulectomy and FFFR, and suggests that posterior mandibulotomies that involve or remove the ramus may predispose to a higher risk of persistent postoperative trismus.

**Keywords:** trismus; mouth opening; mandibulectomy; fibula free flap; postoperative; ramus; MIO; interincisal opening; head and neck; survivorship

### **1. Introduction**

Trismus, or restricted mouth opening, is an increasingly recognized condition among patients with head and neck cancer (HNC). The impact of trismus on quality of life can be devastating, including marked limitations in communication, inadequate nutrition, and chronic pain, which predispose to social isolation and depression [1–3]. Difficulty performing adequate dental care with decreased mouth opening also frequently leads to

**Citation:** Lee, R.H.; Evans, C.; Laus, J.; Sanchez, C.; Wai, K.C.; Knott, P.D.; Seth, R.; El-Sayed, I.H.; George, J.R.; Ryan, W.R.; et al. Patterns of Postoperative Trismus Following Mandibulectomy and Fibula Free Flap Reconstruction. *Cancers* **2023**, *15*, 536. https://doi.org/10.3390/ cancers15020536

Academic Editors: Carlo Lajolo, Gaetano Paludetti, Romeo Patini and Orlando Guntinas-Lichius

Received: 15 October 2022 Revised: 18 December 2022 Accepted: 12 January 2023 Published: 16 January 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

poor oral hygiene and caries, which is especially concerning in irradiated patients given the risk of osteoradionecrosis if dental extraction is required [4]. The prevalence of trismus in HNC patients varies widely by study, from less than 10% to greater than 50% [5–8]. This broad range suggests that trismus development is influenced by many overlapping demographic and treatment parameters, which are challenging to individually delineate in varied patient cohorts. Furthermore, there is inconsistency in the definition of trismus between HNC care providers, such as the use of subjective jaw mobility assessments rather than quantitative measurements and disagreement over the numerical cutoffs for mouth opening that constitute trismus [6]. These limitations in the literature create barriers to comparing the scope of this problem across patients and institutions.

While radiotherapy (RT) has a well-recognized and dose-dependent relationship with trismus in HNC patients, the influence of surgical interventions on trismus is less clear [8–10]. Previous work has demonstrated that certain intraoperative actions may decrease trismus risk, such as prophylactic coronoidectomy, division of the ipsilateral masseter and/or medial pterygoid muscles, and immediate rather than delayed reconstruction of surgical defects [8,11–13]. It is essential to identify the surgical factors that predispose to trismus, and understand how these operative features are affected by preoperative clinical characteristics, so that surgical resections may be modified to optimize functional outcomes.

One surgical cohort that may be especially susceptible to postoperative trismus are patients undergoing mandibulectomy with fibula free flap reconstruction (FFFR). These patients represent a unique challenge in the early postoperative period, as physicians may be reticent to initiate early jaw stretching exercises out of concern for stressing the newly placed bone graft. There is little data on the prevalence and severity of trismus in patients with significant mandibular resection (i.e., segmental mandibulectomy resulting in a bony continuity defect). A previous study in HNC patients receiving free flap reconstruction of lateral segmental mandibular defects found that 94% of patients demonstrated "little to no trismus" postoperatively, with a mean follow-up time of 17 months after surgery [14]. In stark contrast, another group reported that approximately 30% of patients experienced trismus following flap reconstruction of posterior mandibular defects at the most recent follow-up (average of 42 months) [15].

Multiple key questions remain unanswered with respect to trismus after mandibulectomy and FFFR. The lack of quantitative mouth opening measurements over time, including change from preoperative baseline to postoperative follow-ups, is a major limitation in determining the natural course and trajectory of this complication. Furthermore, key modifying characteristics that predispose to postoperative trismus following mandibulectomy and FFFR have not been explored, including patients' preoperative mouth opening, indication for surgery, and anatomy of the bony resection. Identifying these factors may have actionable implications for clinical practice. In this study, we sought to define the scope of trismus in patients after mandibulectomy and FFFR, including factors predictive of worse trismus outcomes.

### **2. Materials and Methods**

### *2.1. Patient Selection*

The study subjects were identified from a database composed of 277 patients who underwent fibula free flap reconstruction (FFFR) from August 2011 to March 2022. A retrospective chart review of all patients in the database was first performed to determine the patients' baseline (preoperative) mouth opening status. The patients with a documented preoperative quantitative measurement, most commonly maximum interincisal opening (MIO), were advanced to the next step of the workflow. Aside from MIO, the charts were also queried for multiple related terms indicative of this metric, including "maximum jaw opening" (MJO) and "mouth opening", as well as units of measurement ("mm" and "cm"). For patients without any quantitative indication of preoperative mouth opening, the otolaryngology provider notes were then searched for a qualitative description of whether the patient had "trismus" or "no trismus" prior to surgery. If available, the degree of mouth opening described in our institution's preoperative anesthesia note was also used to confirm the trismus status in these patients, with "poor" mouth opening indicating trismus and "good" or "excellent" indicating lack of trismus. An anesthesia mouth opening descriptor of "fair" was considered ambiguous, and such patients were not advanced further if this was the only indication of preoperative trismus status. The patients with neither a preoperative quantitative measurement (referred to collectively as "MIO" from this point forward) nor qualitative description of mouth opening were not included in subsequent steps.

Next, the patient charts that met the above criteria for indication of preoperative mouth opening were searched for quantitative postoperative mouth opening measurements using the same search strategy described above. All documented postoperative MIO measurements were recorded, often from multiple sources, including progress notes from speech–language pathologists and dental oncology colleagues, who work closely with the Head and Neck Surgery program at our institution. All patients with preoperative mouth opening status (either MIO or qualitative) and at least one documented postoperative MIO (*n* = 131) were advanced to the next step of the workflow. For patients without any documented postoperative MIO, the date of the last follow-up with our department was recorded. Those patients without any postoperative MIO and a last follow-up of more than three years prior to the chart review were excluded from further analysis.

The 131 patients meeting the aforementioned preoperative and postoperative measurement criteria then underwent chart abstraction for demographic, clinical, and treatmentrelated variables. Only the patients who underwent FFFR for segmental mandibulectomy defects (including hemimandibulectomy) were analyzed; 22 patients who had received a maxillectomy prior to FFFR were excluded, as well as 3 patients who received reconstruction only for a remote ablative surgery. This led to a final study size of 106, 52 of whom had both preoperative and postoperative MIO and 54 who had postoperative MIO and only qualitative preoperative mouth opening descriptors.

### *2.2. Defining Trismus and Postoperative Analysis Intervals*

We defined trismus as an MIO ≤ 35 mm and the lack of trismus as an MIO > 35 mm, based on multiple prior studies assessing functional and quality of life outcomes [16,17]. The presence or absence of trismus was analyzed at two main intervals: an early postoperative period (≤3 months after surgery) and a late postoperative period (>6 months after surgery). The ≤3 month timepoint was chosen to maximize the number of patients with an early postoperative measurement, as the majority of patients (76/106) had one or two MIOs within this time interval. The late timepoint was intended to encompass persistent postoperative trismus with a higher likelihood of representing a chronic condition; a period of >6 months postoperatively captured at least one MIO in 58/106 patients. In contrast to these early and late timepoints, relatively fewer patients (38/106) had MIOs in the intermediate timepoint between 3 and 6 months after surgery, which also plausibly represents a transition period between acute and chronic trismus in which patients may experience resolution. For patients with multiple MIO measurements within the early or late time periods, the most recent MIO was used. We defined ΔMIO as the difference between a patient's preoperative MIO and the MIO measured at the most recent follow-up.

### *2.3. Categorizing Mandibulotomy Anatomy*

For all patients, the sites of the anterior and posterior mandibulotomies were recorded and broadly divided into "ramus-involving" and "ramus-preserving" cuts. "Ramusinvolving" cuts were defined as mandibulotomies that removed any part of the ascending ramus; these cuts were either entirely superior and posterior to the mandibular angle, or they began at the angle and traversed superiorly/posteriorly to involve a portion of the ascending ramus (e.g., spanning from the angle to the sigmoid notch or coronoid process). In contrast, "ramus-preserving" mandibulotomies involved only the angle region itself or were positioned anterior to the angle.

### *2.4. Statistical Analysis*

The rates of postoperative trismus at early and late timepoints were compared using the Pearson chi-square test. The proportion of patients with postoperative trismus were compared by surgical indication, presence vs. absence of preoperative trismus, receipt of adjuvant RT/CRT vs. no adjuvant therapy, and ramus-involving vs. ramus-preserving posterior mandibulotomy. We used the Bonferroni correction method to account for multiple comparisons of the preoperative and postoperative trismus rates, with a corrected significance threshold of α < 0.005 (αoriginal of 0.05/11 total comparisons).

Logistic regression was performed with the binary outcome of yes/no trismus >6 months postoperatively, using the most recent MIO available. Univariate logistic regressions were conducted for age, preoperative trismus status, surgical indication, adjuvant therapy, and posterior mandibulotomy location. A multivariable analysis was also conducted with the same variables. Tumor T stage was not included as a variable in the final model, as staging information was available only for patients undergoing primary ablation and not surgery for recurrent disease or osteonecrosis.

The distribution of ΔMIO between groups was visualized with a waterfall plot for all patients with a measurement at >6 months, separated by the same variables used in the regression analyses. To compare the magnitude of ΔMIO change between these variables, the mean ΔMIO between groups were compared via two-tailed student's *t*-test. Statistical analyses were performed using Stata software version 17.0 (StataCorp LLC; College Station, TX, USA), and the figures were generated via GraphPad Prism 9.3.1.

### **3. Results**

### *3.1. Demographic and Treatment Characteristics of the Patient Cohort*

A total of 106 patients who underwent mandibulectomy and FFFR met the study criteria. The preoperative demographic and clinical characteristics are presented in Table 1. The cohort was 58% male and 42% female, with an average age of 62.1 years (SD 14.3). The indication for surgery was primary ablation with curative intent in 64% (68/106), salvage resection for recurrent disease in 24% (25/106), and mandibular osteonecrosis in 12% (13/106, including 11 patients with osteoradionecrosis and 2 patients with bisphosphonaterelated osteonecrosis of the jaw). Overall, 27% (29/106) of the cohort had a history of prior head and neck irradiation, either for a tumor that subsequently recurred or for a distinct primary tumor prior to developing a second malignancy. Of the patients undergoing surgery for primary ablation with tumor staging available (63/68), the T stage was 11% T1 (7/63), 8% T2 (5/63), 2% T3 (1/63), and 79% T4 (50/63); the N stage was N0 in 49% (31/63) and N+ in 51% (32/63).

Prior to surgery, 45% (48/106) of patients had preoperative trismus, while 55% (58/106) did not have preoperative trismus (Table 1). The posterior mandibulotomy was ramusinvolving in 56% (59/106) and ramus-preserving in 44% (47/106). Following mandibulectomy and FFFR, 36% (38/106) received adjuvant radiation (RT), 23% (24/106) received adjuvant chemoradiation (CRT), and 42% (44/106) were not administered adjuvant therapy (Table 1). The proportion of patients administered adjuvant therapy was higher in those without preoperative trismus compared to patients with preoperative trismus (68% vs. 32%, *p* = 0.001) (Table 2). Of those who underwent adjuvant RT/CRT with precise start and end dates available, the median time from surgery to completion of the adjuvant therapy was 112 days (Q1 = 99 days, Q3 = 131 days). Among all patients, the median time from surgery to the most recent follow-up with our department was 687 days (Q1 = 241 days, Q3 = 1274 days).


**Table 1.** Baseline demographic characteristics and treatment variables for the study cohort.

SD, standard deviation; RT, radiation; CRT, chemoradiation; NA, not available; NR, not relevant (for indications of salvage or osteonecrosis).

**Table 2.** Proportion of patients experiencing trismus at three timepoints: preoperative baseline, early postoperative period (≤3 months after surgery), and late postoperative period (>6 months after surgery).


Preop., preoperative; Postop., postoperative; RT, radiation; CRT, chemoradiation. Asterisks (\*) designate significance with α < 0.005 (Bonferroni correction for multiple comparisons).

### *3.2. Trismus Prevalence at Early and Late Postoperative Timepoints*

In the early postoperative period (≤3 months after surgery), a total of 76 patients had at least one MIO measurement, and the majority experienced trismus (58/76, 76%). Eighty-nine percent of the patients with preoperative trismus demonstrated early postoperative trismus, and 65% of the patients without preoperative trismus demonstrated early postoperative trismus (*p* = 0.014). There were no significant differences in the rates of early postoperative trismus by surgical indication, receipt of adjuvant RT/CRT, or between patients with ramus-involving and ramus-preserving posterior mandibulotomies (Table 2).

In the late postoperative period (>6 months after surgery), a total of 58 patients had at least one MIO measurement. Overall, 67% of patients (39/58) had trismus at this time point. The rates of late postoperative trismus for patients with preoperative trismus and patients without preoperative trismus were 80% and 58%, respectively (*p* = 0.072). Trismus prevalence was again not significantly different when compared by surgical indication or receipt of adjuvant RT/CRT. However, patients with ramus-involving mandibulotomies had significantly higher rates of long-term trismus when compared to ramus-preserving mandibulotomies (82% vs. 46%, *p* = 0.004) (Table 2).

### *3.3. Exploring Variables Associated with Persistent Postoperative Trismus*

To delineate the patient-level and treatment-level factors associated with persistent trismus, we employed logistic regression for patients with at least one MIO measurement >6 months postoperatively. For the univariate analysis, the location of the posterior mandibulotomy was the only variable significantly associated with the presence of postoperative trismus at >6 months, with ramus-involving mandibulotomies demonstrating an odds ratio (OR) of 5.52 (95% CI, 1.67–18.17) compared to ramus-preserving mandibulotomies (*p* = 0.005). The presence of existing preoperative trismus approached significance (OR, 2.95; 95% CI, 0.89–9.77; *p* = 0.077). Patients who underwent surgery for osteonecrosis could not be included in the regression model, as all eight osteonecrosis patients with MIO measurements taken >6 months had postoperative trismus (i.e., an indication of osteonecrosis perfectly predicted the regression outcome). This necessitated the reduction of the surgical indication variable from three categories to two (primary ablation vs. salvage for recurrence). Surgical indication, age, and receipt of adjuvant therapy were not significantly associated with persistent postoperative trismus in the univariate analysis (Table 3).

**Table 3.** Univariate and multivariable logistic regressions for the presence of persistent late postoperative trismus (>6 months postoperatively).


RT, radiation; CRT, chemoradiation.

In the multivariable logistic regression with the same variables, posterior mandibulotomy location remained the only variable significantly associated with trismus >6 months postoperatively, with an OR of 7.94 for ramus-involving vs. ramus-preserving cuts (95% CI, 1.85–33.97; *p* = 0.005). As with the univariate analysis, age, surgical indication, and adjuvant therapy were not significantly associated with persistent postoperative trismus in the multivariable regression (Table 3).

### *3.4. Comparing* Δ*MIO at the Late Postoperative Timepoint*

Of the 58 total patients with postoperative MIO measurements taken >6 months postoperatively, 27 patients also had a preoperative MIO available. The overall distribution of ΔMIOs from the preoperative visit to the most recent visit's MIO measurement >6 months postoperatively for these 27 patients is shown in Figure 1A, separately stratified by preoperative trismus status, receipt of adjuvant therapy, surgical indication, and ramus involvement of the posterior mandibulotomy. The mean ΔMIOs between these groups at the same timepoint are displayed in Figure 1B. On average, patients without preoperative trismus had a decline in ΔMIO, while those with existing preoperative trismus had a mean positive change in ΔMIO (−7.07 vs. +1.83 mm, *p* = 0.038). The mean ΔMIO for patients who received adjuvant RT/CRT was −5.11 mm and +1.63 mm in those without adjuvant therapy (*p* = 0.159). By indication, the mean ΔMIOs for primary ablation, salvage, and osteonecrosis were −4.72, +0.50, and −0.67 mm, respectively (*p* = 0.362 for a comparison

between primary ablation and salvage surgery). The mean ΔMIO for the ramus-involving vs. ramus preserving mandibulectomies were −3.18 and −3.00 mm, respectively (*p* = 0.970).

**Figure 1.** (**A**) Waterfall plot of ΔMIO for all patients with preoperative MIO and MIO measurement taken >6 months postoperatively (*n* = 27), separated by preoperative trismus status, receipt of adjuvant therapy, surgical indication, and ramus involvement of the posterior mandibulotomy; (**B**) mean ΔMIO > 6 months postoperatively for the same 27 patients when compared by preoperative trismus status, receipt of adjuvant therapy, surgical indication, and ramus involvement of the posterior mandibulotomy.

### **4. Discussion**

Trismus is a complex, multifactorial condition that represents a formidable challenge in patients with HNC. In surgically treated patients, data on trismus outcomes are limited, especially when compared to the preponderance of work assessing the relationship between trismus and RT. Many previous studies utilize heterogenous HNC patient cohorts receiving a variety of treatment approaches, which makes it difficult to disentangle the individual demographic and treatment-related characteristics that contribute to this sequela. In this study, we focused specifically on one unique surgical population—those undergoing mandibulectomy and FFFR.

The first step towards conducting meaningful trismus studies that are generalizable across patients and institutions is specifying an appropriate definition of trismus. The current evidence suggests that a mouth opening of ≤35 mm is a suitable and clinically meaningful demarcation of trismus that predicts health-related quality of life [16,17]. However, even in the presence of an appropriate trismus metric, obtaining consistent post-treatment jaw opening measurements is a significant challenge. By far, the largest obstacle in this study was a lack of regular MIO measurements taken at both preoperative and postoperative visits. From a database of 277 fibula free flap patients, fewer than half (131/277) had postoperative MIO measurements and either a quantitative or qualitative indication of preoperative mouth opening. Furthermore, of the 106 patients who met the final study criteria, only 52 had a quantitative mouth opening measurement (MIO) taken preoperatively.

Because most patients in this cohort did not have both baseline preoperative MIOs and regular postoperative MIOs, our ability to compare the magnitude of change in mouth opening over time was limited (only 27 total patients with follow-up >6 months after surgery had both preoperative and postoperative MIOs). Nonetheless, the comparison between the mean ΔMIO at >6 months was significant between the patients with preoperative trismus compared to the patients without preoperative trismus (ΔMIOavg of +1.83 vs. −7.07 mm, respectively; *p* = 0.038). Indeed, while half (6/12) of the patients with preoperative trismus experienced increased MIO at >6 months, only 13% (2/15) of patients without preoperative trismus demonstrated an increase in MIO at this late time point. This suggests that while patients who are trismus-free preoperatively will generally experience a decline in postoperative MIO, patients with existing preoperative trismus

may exhibit a marginal improvement in postoperative MIO. However, it is important to recognize that the majority of patients with preoperative trismus continued to have chronic trismus (i.e., MIO ≤ 35 mm) postoperatively. In addition, because MIO measurements were taken at different intervals for different patients, it was not feasible to infer a general timeline of the trajectory of MIO decline following mandibulectomy and FFFR. Future work that incorporates consistent mouth opening measurements at defined and frequent postoperative intervals may allow for the construction of a generalizable timeline for the development of (and recovery from) trismus in this population.

A key limitation of this study was our inability to assess the impact of jaw stretching interventions on the risk of trismus development in this cohort of mandibulectomy and FFFR patients. At our institution, home jaw stretching and neck stretching exercises are routinely taught and implemented with postsurgical patients by our team of speech–language pathologists (SLPs). However, details on the adherence to these regimens and the time periods over which stretching is practiced are often unable to be accurately extrapolated from retrospective chart review. The limitations of the documentation also precluded the analysis of the benefit of active device-based intervention (using products such as the TheraBite® or OraStretch®) in this study. Given the high out-of-pocket cost of these devices coupled with inconsistent clearance by insurance providers, it is often unclear if patients even received the stretching device recommended to them, let alone the adherence to the stretching exercises with the device itself. In patients with mandibulectomy and FFFR, concern for imparting excessive stress on the newly placed vascularized bone graft may generate reticence over initiating early jaw stretching exercises postoperatively. There have been reports of serious complications while using the TheraBite® in the post-treatment setting, including a mandibular fracture in an HNC patient with undiagnosed mandibular osteoradionecrosis (ORN) [18], and fracture of titanium mandibular reconstruction plates in the setting of mandibular recurrence following mandibulectomy and FFFR [19]. However, it is notable that both complications occurred in patients with bone that was ostensibly already structurally compromised (due to ORN or recurrent cancer). The true risk of trismus devices in mandibulectomy and FFFR patients with a healthy flap and expected postoperative healing has not been studied. Additional work will help to clarify the earliest time period following surgery that is safe for device-assisted jaw stretching, and the optimal timing for trismus interventions in these patients.

Our work suggests that the location of the posterior mandibulotomy may affect the risk of postoperative trismus after mandibulectomy and FFFR. At >6 months, patients with ramus-involving posterior mandibulotomies experienced trismus at nearly twice the rate of patients with ramus-preserving cuts (82% vs. 46%), the most robust difference in magnitude among any variables compared in this study. Multivariable logistic regression revealed a nearly eight-fold greater odds of persistent postoperative trismus for patients with ramusinvolving posterior mandibulotomies, when adjusted for age, preoperative trismus status, surgical indication, and receipt of adjuvant therapy. However, we could not demonstrate a significant difference in ΔMIO between these groups, most likely due to the very small number of patients with paired preoperative and postoperative MIO measurements, as discussed above. There are multiple possible explanations for the observation of marked differences in late postoperative trismus rates by mandibulotomy location. Acute or chronic inflammation of structures surrounding the mandibular ramus, condyle, or coronoid can either directly limit movement around the temporomandibular joint and/or induce pain resulting in a reflex trismus (as is often also seen in non-neoplastic conditions such as peritonsillar abscesses or lateral pharyngeal space infections) [20]. Violation of the region surrounding the ramus during surgery may also result in the fibrotic shortening of the pterygoids or pterygomandibular ligament during healing, thereby further contracting mouth opening. While the extent of disease is the largest factor dictating whether the ramus can be spared during resection, our data identify patients at a particularly high risk of late postoperative trismus, who may especially benefit from vigilant surveillance of mouth opening and early trismus interventions. It is also critical to note that a large portion of

HNC patients undergoing mandibulectomy and FFFR will demonstrate adverse features on surgical pathology, necessitating adjuvant RT/CRT for optimal oncologic control. Radiation itself has a well-known, dose-dependent relationship with trismus secondary to fibrosis of the masticatory apparatus, especially with large doses to the masseter and medial pterygoid muscles [21–23]. It will be important to determine how radiation interacts with postoperative anatomy in mandibulectomy and FFFR patients, and how irradiation of the retained masticatory musculature contributes to trismus severity specifically following surgical disruption of the ramus.

### **5. Conclusions**

In summary, this study described the scope of postoperative trismus following mandibulectomy and FFFR, including potential contributing factors to this sequela. We demonstrated that most patients (76%) experienced trismus in the early (≤3 months) postoperative period and persistent trismus remained common in the late (>6 months) postoperative period (occurring in 67% of patients in this study). Using logistic regression, we found that a posterior mandibulotomy involving or removing the ramus was associated with a substantially higher odds of persistent trismus after surgery. To our knowledge, this is the first study that implicates mandibulotomy location as a surgical risk factor for postoperative trismus. Larger-scale studies are critical for identifying patients at the highest risk of trismus after mandibulectomy with FFFR, and delineating precise temporal changes in mouth opening may inform key postoperative intervals for active trismus intervention.

**Author Contributions:** Conceptualization, R.H.L., C.E., C.S., P.D.K., R.S., I.H.E.-S., J.R.G., W.R.R., C.M.H., A.M.P. and P.K.H.; Methodology, R.H.L., C.E., J.L., C.S., K.C.W. and P.K.H.; Formal analysis, R.H.L. and K.C.W.; Investigation, R.H.L.; Writing—original draft, R.H.L. and P.K.H.; Writing—review & editing, R.H.L., C.E., J.L., C.S., K.C.W., P.D.K., R.S., I.H.E.-S., J.R.G., W.R.R., C.M.H., A.M.P. and P.K.H.; Visualization, R.H.L.; Supervision, A.M.P. and P.K.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of the University of California, San Francisco (UCSF).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** De-identified data presented in this study are available upon reasonable request from the corresponding author.

**Conflicts of Interest:** W.R.R. is a member of the Scientific Advisory Boards for Olympus and Rakuten Medical and was briefly a consultant for Intuitive Surgical in 2021. P.K.H. has received educational funding from Stryker and Medtronic and is a consultant for Privo Technologies and Checkpoint Surgical.

### **Abbreviations**


### **References**

1. Lydiatt, W. Trismus: A sequela of head and neck cancer and its treatment. *JCO Oncol. Pract.* **2020**, *16*, 654–655. [CrossRef] [PubMed]


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Article* **Exploring Precise Medication Strategies for OSCC Based on Single-Cell Transcriptome Analysis from a Dynamic Perspective**

**Qingkang Meng 1,†, Feng Wu 2,†, Guoqi Li 1, Fei Xu 1, Lei Liu 1, Denan Zhang 1, Yangxu Lu 1, Hongbo Xie <sup>1</sup> and Xiujie Chen 1,\***


**Simple Summary:** At the time of diagnosis, most oral squamous cell carcinoma (OSCC) patients are in the middle or advanced stages, and advanced patients usually have poor prognosis after traditional therapy. One of the primary causes has been demonstrated to be heterogeneity. However, most of the current studies on tumor heterogeneity are static, while the development of cancer is dynamic. Thus, understanding the tumor development process from a dynamic perspective is deeply necessary. Here, we combined static and dynamic analysis based on single-cell RNA-Seq data to comprehensively dissect the complex heterogeneity and evolutionary process of OSCC. We pioneered the concept of pseudo-time score, which is closely related to patient's prognosis. Finally, we identified candidate drugs and proposed precision medication strategies to control OSCC in two respects: treatment and blocking. Our findings offer new insights for clinical practice and could help improve the treatment of advanced OSCC.

**Abstract:** At present, most patients with oral squamous cell carcinoma (OSCC) are in the middle or advanced stages at the time of diagnosis. Advanced OSCC patients have a poor prognosis after traditional therapy, and the complex heterogeneity of OSCC has been proven to be one of the main reasons. Single-cell sequencing technology provides a powerful tool for dissecting the heterogeneity of cancer. However, most of the current studies at the single-cell level are static, while the development of cancer is a dynamic process. Thus, understanding the development of cancer from a dynamic perspective and formulating corresponding therapeutic measures for achieving precise treatment are highly necessary, and this is also one of the main study directions in the field of oncology. In this study, we combined the static and dynamic analysis methods based on single-cell RNA-Seq data to comprehensively dissect the complex heterogeneity and evolutionary process of OSCC. Subsequently, for clinical practice, we revealed the association between cancer heterogeneity and the prognosis of patients. More importantly, we pioneered the concept of pseudo-time score of patients, and we quantified the levels of heterogeneity based on the dynamic development process to evaluate the relationship between the score and the survival status at the same stage, finding that it is closely related to the prognostic status. The pseudo-time score of patients could not only reflect the tumor status of patients but also be used as an indicator of the effects of drugs on the patients so that the medication strategy can be adjusted on time. Finally, we identified candidate drugs and proposed precision medication strategies to control the condition of OSCC in two respects: treatment and blocking.

**Keywords:** single-cell RNA-Seq; oral squamous cell carcinoma; cell trajectory inference; precise medication; drug discovery

**Citation:** Meng, Q.; Wu, F.; Li, G.; Xu, F.; Liu, L.; Zhang, D.; Lu, Y.; Xie, H.; Chen, X. Exploring Precise Medication Strategies for OSCC Based on Single-Cell Transcriptome Analysis from a Dynamic Perspective. *Cancers* **2022**, *14*, 4801. https:// doi.org/10.3390/cancers14194801

Academic Editors: Carlo Lajolo, Gaetano Paludetti, Romeo Patini and Lorenzo Lo Muzio

Received: 22 July 2022 Accepted: 28 September 2022 Published: 30 September 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Oral squamous cell carcinoma (OSCC) is the most common type of head and neck squamous cell carcinoma (HNSC), accounting for approximately 95% of cases [1]. At present, the clinical classification and treatment of OSCC are mainly based on the TNM staging system of the American Joint Committee on Cancer (AJCC) and the International Union for Cancer Control (UICC). In general, patients in the early stages (i.e., stages I and II)—approximately 30–40% of the patients diagnosed—have small tumors without significant lymph node involvement. Surgery and radiation therapy can provide effective tumor control and improve long-term survival in approximately 70–90% of early stage patients [2]. For patients with advanced stages (i.e., stages III and IV), which are characterized by varying degrees of surrounding tissue invasion, lymph node involvement, and metastatic spread, it is difficult to eliminate or kill tumors completely via surgery or radiotherapy. Although systematic treatment with drugs (e.g., platinum, taxanes, antifolates, cetuximab, etc.) can be conducted for the remission of recurrence and metastasis [2], more than 65% of these patients have a poor prognosis due to the significant heterogeneity, which appears not only between individuals but also within the same individual or even the same tissue—that is, intratumoral heterogeneity [3]. Therefore, conquering advanced OSCC has become an urgent problem to be solved in the field of HNSC treatment. Although many studies have been carried out successively, and some progress has been made in the diagnosis and treatment of OSCC, most advanced patients will still experience recurrence or metastasis (or both) [2]. Further study of the heterogeneity of advanced OSCC and, accordingly, exploration of improved therapeutic strategies to perform more precise treatment and increase the cure rate is highly necessary.

In recent years, the rapid development of single-cell sequencing technology has provided a powerful tool for basic cancer research. Relative to traditional bulk sequencing methods, single-cell sequencing takes a single cell as the basic unit and, therefore, provides much better insight into the heterogeneity between cells. The methods based on single-cell sequencing bring new opportunities to dissect the intratumoral heterogeneity of tumors at high resolution. However, in addition to this, there is asynchrony in the development process of tumor cells in organisms—that is, the cycle states of different cells are not the same [4]. Bulk-based sequencing is performed on the whole tissue, in which cells with different cycle states are contained, thus also masking temporal heterogeneity between cells. Hence, at present, a large number of methods have emerged for single-cell trajectory inference, which can infer the position of each cell on the pseudo-time axis through algorithms based on expression profiles, so as to reproduce the trajectory of cell differentiation over time. The traditional clinical stage can reflect the time status of the development of a tumor; therefore, single-cell pseudo-time trajectory inference combined with clinical stages can more clearly analyze the dynamic process of tumor development, so as to more accurately discover the biological factors that promote tumor evolution and, finally, to promote the development of clinical treatment.

Therefore, our study emphasized the dynamic development process of OSCC. Unsupervised clustering based on single-cell transcriptomics was first performed to analyze the heterogeneity of OSCC. Pseudo-time trajectory inference was subsequently performed based on the unsupervised clustering results to dissect the complex development trajectory of OSCC from the early to the late stages. Next, to validate the cell trajectory, an analysis of receptor–ligand-based cell communication was conducted. At present, the single-cell field is under the transition from descriptive biology to predictive biology [5]; therefore, in order to promote the theory in clinical applications, we used bulk RNA-Seq data from the TCGA HNSC cohort to analyze the effects of different cell cluster compositions on patients' survival time. More importantly, we proposed pseudo-time score that can combine heterogeneity with the pseudo-time occupied by cell clusters and can quantify the status of patients during cancer development. We also used the external cohort from the ICGC [6] database to validate the pseudo-time score. Finally, based on the heterogeneity and complex development trajectory of OSCC, we discovered potential targeted therapeutic drugs that

can not only personalize the treatment according to the individual's cell cluster composition but also block the development and deterioration of OSCC as much as possible in order to jointly improve the current situation of clinical treatment of advanced OSCC.

### **2. Results**

### *2.1. High Heterogeneity of OSCC*

We first performed quality control on the data by examining the number of cells and the gene expression levels of all patients; we removed five patients, including 24 cells, which are represented in red in Figure 1A,B. Next, we extracted 12,000 highly variable genes from a total of 23,686 genes to perform feature selection and to improve the accuracy of downstream analysis. Principal component analysis (PCA), which is a widely used dimensionality-reduction method, was used to process the expression profiles of highly variant genes. The optimal number of dimensions was determined by the elbow method, as shown in Figure 1C. Finally, the top 20 principal components were selected for unsupervised clustering (Figure S1).

**Figure 1.** Data preprocessing and unsupervised clustering: (**A**) Violin plot of gene expression counts. The *x*-axis represents the patient ID, while the *y*-axis represents the sum values of gene expression counts. The dots represent the cells. Patient IDs that failed quality control are marked in red. (**B**) Violin plot of expressed gene numbers. The *x*-axis represents the patient ID, while the *y*-axis represents the number of expressed genes. The dots represent the cells. Patient IDs that failed quality control are marked in red. (**C**,**D**) The results of cell clustering. The dots represent the cells. The *x*- and *y*-axes represent the two dimensionalities of UMAP, respectively. Cells are colored by cluster label in panel C and by clinical stage in panel D.

Based on the top 20 principal components, we performed unsupervised clustering on 2176 cancer cells, which were divided into 10 clusters (Figure 1C,D). We found that, with the progression of the clinical stages, the number of cell clusters also gradually increased, indicating that there is greater heterogeneity in advanced OSCC (Figure 1D). This could explain the difficulty of completely curing advanced OSCC with current clinical treatments.

### *2.2. Different Development Fates of OSCC Cells*

To reveal the developmental process of OSCC from the early to the advanced stages, we performed cell trajectory inference analysis based on pseudo-time for all cells using Monocle (Figure 2A,C). It is worth noting that the development of OSCC cells did not follow a single trajectory and was divided into two paths by a branch point. In terms of cell clusters (Figure 2A), clusters 9 and 5 appeared before the branch point, Path I developed along with the order of clusters 4, 10, 7, and 3, and Path II developed along with the order of clusters 6 (2), 8, and 1. From the perspective of clinical stages (Figure 2B), the temporal order was followed by stages I, II, III, and IV, consistent with the actual clinical development progress. Cells from stages I and II mainly appeared before the branch point, which appeared at stage III. Cells developed along the two paths and eventually deteriorated into stage IV. However, the two paths had different deterioration trends. Some cells from the branch point directly developed into stage IV (Path II), while the other cells remained in stage III (Path I) for a long time before eventually developing to stage IV. Therefore, identifying the key genes driving these two paths, with the goal of finding appropriate drugs to prevent the progression of cancer, would make significant progress in improving the current treatment of advanced OSCC.

**Figure 2.** Cell trajectory inference: (**A**–**C**) Cell development trajectory. The dots represent the cells. The *x*- and *y*-axes represent the two dimensionalities of UMAP, respectively. Cells are colored by cluster label in panel A, by clinical stage in panel B, and by pseudo-time value in panel C. Based on the pseudo-time, the cell evolution direction could be determined, which was mainly divided into two paths: Path I (yellow) and Path II (red). (**D**) Results of cell communication analysis. The axis represents the four cell clusters near the branch point. The color and the number labeled by this heatmap were determined by the receptor–ligand pair number between two cell clusters.

### *2.3. Characteristic Analysis of Intercellular Communication*

In the whole body or a specific tissue, the complex interactive relationships between cells constitute the cell communication network, which can reflect the tightness of the connection between different cells. Based on this, we hypothesized that the stronger the intercellular communication, the closer the connection, and the closer it is in the time series.

Therefore, to verify the accuracy of the cell trajectory inference, we performed receptor– ligand-based cell communication analysis. Throughout the trajectory, directly connected cells should tend to interact more tightly, while indirectly connected cells should have a weaker interaction because they require the transition of the intermediate cells—that is, there should be a tighter interaction between adjacent cell clusters in the trajectory. Here, we used CellPhoneDB to perform receptor–ligand-based analysis on cells of four clusters (4, 5, 6, and 9) near the branch point that separates the two paths in order to reveal their intercellular interaction relationships. The level of intercellular interactions was measured by the number of receptor–ligand pairs. As shown in Figure 2D, from the overall point of view, the tightness of the connection between cell clusters 5-4, 5-6, and 5-9 was much higher than that between the other cluster pairs, and as expected, these cell clusters were located adjacent to one another in the pseudo-time trajectory. In addition, we found that the interaction strength between clusters 4 and 5 was the highest. These two cell clusters not only were located close to one another in the pseudo-time trajectory but also belonged to the same stage (III), so the intercellular communication connection between them should also have been closed. The number of receptor–ligand pairs between clusters 5 and 9 was the second highest, at 33. These cells were from stages I, II, and III, but they were all located before the branch point, so the level of communication between them was higher than that of others after the branch point.

In conclusion, the analysis of intercellular communication based on receptor–ligand pairs was consistent with the results of cell trajectory based on pseudo-time and, thus, could further validate the extremely complex landscape during the development of OSCC cells from the early to the late stages.

### *2.4. Biological Factors Driving the Two Paths at the Branch Point in the Cell Development Trajectory*

In order to further explore the driving factors of the branch point that separated Paths I and II, we used the BEAM method to perform pseudo-time-based differential gene expression analysis on four cell clusters (4, 5, 6, and 9) near the branch point. As a result, we identified 267 genes that showed significant fluctuation in their expression at the branch point (Figure 3; Table S2). Compared with traditional differential gene expression analysis, BEAM combines pseudo-time to reflect the continuous changes in gene expression. Through hierarchical clustering of these genes, all genes were divided into two gene sets. We found that the majority of these genes showed mutually exclusive expression characteristics in two directions—that is, high expression characteristics in Path I but low expression characteristics in Path II, or vice versa—indicating that these genes regulate two mutually exclusive cell fates. In addition to this, genes in a gene set of hierarchical clustering usually have co-expression characteristics that may co-regulate some biological functions.

**Figure 3.** Key genes driving the occurrence of the branch point: The heatmap of gene expression levels during the progression of OSCC. Gene clusters were generated by hierarchical clustering.

Therefore, in order to reveal the biological functions regulated by these genes, we performed Gene Ontology (GO) enrichment analysis on two gene sets (Figure 4). The results showed that genes highly expressed in Path I (gene cluster 2) were significantly enriched mainly in biological processes, such as the cell migration and apoptosis signal regulation pathways (Figure 4B). The important role of cell migration and apoptosis in the development and progression of OSCC has been confirmed [1]. Marker genes regulating the cell cycle, apoptosis, and migration have differential expression in OSCC patients or are significantly associated with prognosis [1], such as survivin and heat shock proteins (HSPs) associated with apoptosis. Survivin is a member of the inhibitor of apoptosis proteins (IAP) family that inhibits capase 3, 7, and 9, and its expression is higher in OSCC patients than in epithelial dysplasia patients. High expression of heat shock protein 27 (HSP27) is associated with a better prognosis. The expression of urokinase plasminogen activator receptor (UPAR)—a marker gene associated with cell migration—was negatively correlated with prognosis [1]. Therefore, the genes highly expressed in Path I suggest that OSCC cells may have active metastatic characteristics in clinical stage III. The genes highly expressed in Path II (gene cluster 1) were mainly enriched in biological functions related to MHC class II in BPs (biological processes), CCs (cellular components), and MFs (molecular functions) (Figure 4A). MHC is a collective term for a group of genes encoding major histocompatibility antigens in animals, also known as HLA in humans, which is also involved in the immune process of the body as an antigen. There is literature confirming the upregulation of class II molecules of the major histocompatibility complex (MHC) by keratinocytes in oral squamous cell carcinoma [7]. However, the significance of its high expression in advanced OSCC is currently unclear. Another study confirmed that the keratinocyte line expressing MHC II has the characteristics of the absence of CD80 and CD86 in head and neck cancer, which may be a way for tumors to evade immune surveillance [8].

**Figure 4.** GO enrichment analysis of key genes: The results of GO enrichment analysis of gene clusters 1 and 2 are shown in panels (**A**,**B**), respectively. The *x*-axis represents GO terms, which are displayed in three colors according to the three types (BPs, CCs, and MFs), while the *y*-axis represents -log10(p.adjust). Only the 10 most significant GO terms of each type are shown here.

### *2.5. Relationship between Prognosis and Heterogeneity of Advanced OSCC*

Our findings show that advanced OSCC has more significant heterogeneity than early OSCC, and it is difficult to characterize this heterogeneity via traditional bulk research. Therefore, to further dissect the relationship between this heterogeneity and prognosis, we integrated clinical information and expression profiles from the TCGA database to explore the prognosis of stage III and IV patients with greater heterogeneity, which would also indirectly verify the accuracy of our predicted cell clusters.

Since the data from the TCGA database are at the bulk level, we first mapped corresponding cell clusters to bulk expression profiles to infer the cell cluster composition of each patient. Based on the custom background gene sets (Table S3) derived from differential expression analysis, CIBERSORT was performed for stage III and IV patients from TCGA (Table S4). Next, in order to reveal the impact of cell cluster composition on the prognosis of patients, we performed unsupervised hierarchical clustering based on the results of CIBERSORT. The patients in stages III and IV were divided into two groups. Each patient group had similar cell cluster composition, and there were significant differences between the patient groups (Figure 5A,D). For patients in stage III, including four clusters, group 1 mainly contained cell cluster 7, while group 2 mainly contained cell clusters 5 and 10. For patients in stage IV, composed of six clusters, group 1 mainly contained cell cluster 6, while group 2 mainly contained cell cluster 2.

Survival analysis was performed for the two groups of patients in stages III and IV. The results showed differences in survival time between the two groups (Figure 5B,E), indicating that different cell compositions could impact prognosis. For patients in stage III, the survival time of patient group 1 was shorter, while for stage IV, the survival time of patient group 1 was longer.

**Figure 5.** Association between heterogeneity and clinical prognosis. (**A**,**D**) The results of CIBERSORT. Patients were divided into two groups by hierarchical clustering. Patients in stage III are shown in

panel A, while patients in stage IV are shown in panel D. (**B**,**E**) Survival analysis. The survival curve is shown above, with the *x*-axis representing the survival time (days), the *y*-axis representing the survival probability, and the censored values indicated by "+" in the curve. The color of the curve is consistent with panel A (**D**) and represents two patient groups. Below is the risk table showing the number (or percentage) of survivors at each time point. The results of patients in stage III are shown in panel B, while those of patients in stage IV are shown in panel E. (**C**,**F**) Survival analysis based on pseudo-time score for grouping. The color of the curve represents the two patient groups, which are distinguished by their pseudo-time score. The result of patients in stage III is shown in panel C, while those of patients in stage IV are shown in panel F.

In order to explore the reason(s) that this heterogeneity has an impact on the survival time of patients, we constructed pseudo-time score by combining cell trajectory inference to quantify the temporal status of each patient group. The pseudo-time score considers both the cell composition of patients and the development order of each cell cluster in the same clinical stage. The higher the score, the closer the patient is to the advanced stage. The results showed that, in stage III, Spatient group 1 = 16.35 and Spatient group 2 = 15.19. As for stage IV, Spatient group 1 = 18.59 and Spatient group 2 = 20.44. According to the survival analysis, the groups with smaller pseudo-time score tended to have a better prognosis. Therefore, the difference in prognosis status between patient groups lies in the different cell compositions, which contribute to the different temporal status during the development of cancer. Our results also show that even patients at the same clinical stage would have many differences in cell composition and prognosis.

To further analyze the relationship between the pseudo-time score and patients' prognosis, we regrouped patients according to their pseudo-time score, and the results showed that the pseudo-time score could be used as the marker to distinguish survival time in both stage III and stage IV patients, with *p*-values of 0.047 and 0.043, respectively, as measured by the log-rank test. The patient groups with the higher score had the worse prognosis (Figure 5C,F).

To confirm the above results, the same methods were used to process the dataset from ICGC, and the cell cluster composition inference of patients was performed using CIBERSORT based on the same background gene sets. Then, these patients were divided into two groups based on their pseudo-time score, and survival analysis also illustrated that the two groups of patients showed significant differences in survival time (*p* = 0.049, log-rank test) (Figure S2).

In summary, the pseudo-time score that we constructed has a significant correlation with prognosis, can be used as a prognostic marker, and is robust across multiple datasets.

### *2.6. Identification of Candidate Drugs Based on PPI Networks*

Transcriptomic analysis at the single-cell level with high resolution is a way to improve the efficacy of medication by dissecting the diverse cell clusters in the tumor and by selecting the targeted therapy strategy. In addition, it is also essential to find key genes with synchronous expression changes during the development of OSCC and to use these genes as potential targets to discover blockers to slow or even block the deterioration of OSCC in order to prolong the treatment time in clinical practice and to maximize the lifespan of patients. Therefore, we performed drug discovery in the following two respects: (1) searching for drugs to block OSCC progression and (2) searching for cell-cluster-specific drugs to achieve targeted therapy.

For the first respect, 267 key genes at the branch point revealed by the BEAM analysis were first used for protein–protein interaction (PPI) network construction using the STRING [9] database. Cytoscape software was used for network visualization (Figure S3). There were 218 nodes and 649 edges in the PPI network. We used the cytoHubba [10] plugin built in Cytoscape to identify the hub genes, and all 12 topological analysis methods were taken into account to improve the robustness. Then, the expression trends of these

hub genes with pseudo-time were manually checked, and only genes showing a significant trend of upregulation in Path II and downregulation in Path I were retained as marker hub genes. As a result, 15 marker hub genes were identified (Figure 6). Except for ALDH3A1, CD40, CXCL11, HLA-DRA, and HLA-DRB have been validated by the literature to have positive relationships between expression and prognosis, so we removed them from candidate drug targets, while the others presented malignant characteristics (Table S5). Finally, using the 10 remaining genes as targets, 195 drugs were extracted from integrated drug– target relationships, including first-line antitumor drugs, such as cisplatin, fluorouracil, methotrexate, gemcitabine, p-phenylenediamine, etc. (Table S6, Sheet 1). Among them, 90 drugs were validated based on CCLE experimental data (Table S6, Sheet 2), and 77 of the remaining 105 drugs were validated based on the literature (Table S6, Sheets 2 and 3). The drugs targeting multiple targets were paid more attention to. Cyclosporine and valproic acid (VPA) targeted all 10 proteins. In recent years, VPA has been found to be a histone deacetylase inhibitor (HDACi). Many experimental studies have shown that VPA can inhibit the growth and proliferation of tumor cells by inducing cell cycle arrest, apoptosis, and differentiation and by inhibiting tumor angiogenesis and metastasis [11,12]. It is known that the combination of cisplatin (CDDP) and cetuximab (CX) is one of the standard first-line treatments for OSCC. However, this therapeutic regimen is often associated with resistance, suggesting that new combinatorial strategies need to be improved. Federica Iannelli et al. demonstrated that the introduction of VPA to the conventional treatment for recurrent/metastatic HNSCC represents an innovative and feasible antitumor strategy that warrants further clinical evaluation [13]. Another study showed VPA acting as a histone deacetylase inhibitor (HDI) in OSCC cells and normal human keratinocytes (HKs), potentiating the cytotoxic effect of cisplatin in OSCC cell lines and decreasing the viability of OSCC cells as compared to HKs [14]. Taken together, these results provide initial evidence that VPA might be a valuable drug in the development of better therapeutic regimens for HNSCC.

We performed drug sensitivity predictions at single-cell resolution for these drugs, and the drug sensitivity of cell clusters was represented by the mean value of drug sensitivity of the cells in each cluster. According to the pseudo-time trajectory, all cell clusters were divided at the branch point into Paths I and II. Fifty-three drugs were predicted to have higher sensitivity in Path II, suggesting that these drugs are more effective at inhibiting malignant developmental processes (Table S6, Sheet 4). Of these, we found that fulvestrant simultaneously exhibited the highest drug sensitivity in cell clusters 4, 5, and 6, which appeared near the branch point and, thus, likely represented an earlier exacerbation progression (Figure 7A). A study has shown that estrogen can participate in the progression of precancerous lesions of HNSC by inhibiting apoptosis and by promoting the proliferation of advanced HNSC cells [15]. Antiestrogen may be beneficial as a chemopreventive agent for HNSC [15], and fulvestrant is an antiestrogen drug, so our results were consistent with those of the previous study.

For the second respect, marker genes were first calculated using the SC3 [16] method for all cell clusters. With manual examination, a total of 459 genes remained after statistical filtering (Table S7, Sheet 1). These genes showed significant upregulation in specific cell clusters, so they could be used as essential cell-cluster-specific marker genes for drug discovery. As described above, the PPI networks of each cell cluster were constructed (Figure S4), and then, 12 topological methods from cytoHubba were taken into account for identification of the hub genes. Then, 76 hub genes were filtered (Table S7, Sheet 2), and 478 drugs were ultimately discovered (Table S8). Some of these drugs overlap with the blocking drugs found in our study and are all first-line anticancer drugs, such as valproic acid, cisplatin, fluorouracil, methotrexate, temozolomide, etc. Based on CCLE experimental data, 195 drugs were validated as being effective in HNSC cell lines (Table S9, Sheet 1), and 168 of the remaining 283 drugs were validated through the literature (Table S9, Sheets 1 and 2). Many previous studies have shown that the combined use of certain drugs can enhance their effects; for example, the combination of curcumin and copper can enhance the inhibitory effect on the migration and activity of OSCC cells [17]. Our findings show that curcumin and copper can be used for blocking the development of OSCC and targeting cell cluster 1, indicating that the two have the potential for combination, and are consistent with the results of the previous study. Next, we analyzed the sensitivity of drugs and extracted drugs that had higher sensitivity in their targeting of cell clusters. Finally, there were 102 drugs, including many drugs that have been proven to be effective for HNSC treatment (Table S9, Sheet 3). Cell cluster 1 is at the end of the pseudo-time trajectory and, thus, represents highly advanced OSCC cells. Among the selected drugs, there were 71 drugs targeting cell cluster 1, including common anticancer drugs, such as paclitaxel, gemcitabine, carboplatin, decitabine, etc. (Figure 7B). We performed literature validation for all of these drugs, and 91 of the 102 had been reported in previous studies for cancer treatment or combined medication. Therefore, the candidate drugs discovered in our study could be of great significance to changing the current situation of treatment for advanced OSCC.

**Figure 6.** Analysis of 15 target genes' expression trends: This figure shows the expression trends of genes progressing with pseudo-time. The *x*-axis represents pseudo-time, while the *y*-axis represents the gene expression level. Dots represent cells, which are colored according to clinical stages. The curves were fitted by gene expression level, and the colors represent two different development paths in the cell trajectory.

**Figure 7.** Sensitivity of candidate drugs: The *x*-axis represents cell clusters, and the *y*-axis represents the logFC (drug sensitivity data from the CCLE are represented using relative log fold-change values of cell lines' viability to DMSO). Cell clusters with the highest drug sensitivity mentioned in our study are marked in red. (**A**) Sensitivity of fulvestrant. (**B**) Sensitivity of gemcitabine.

### **3. Methods**

### *3.1. Data and Preprocessing*

The single-cell RNA sequencing data of OSCC used in this study were obtained from the GEO database (GSE103322), with a total of 5902 cells from 15 patients [18] including all clinical stages. These data were preprocessed using the method described in the article published by Sidhart V. Puram et al. [18], and all cells were accurately classified into cancer and non-cancer types. The gene expression did not conform to a normal distribution in five patients due to the low number of cells, which may have had a bad impact on the accuracy of the downstream analysis; therefore, these five patients were removed. There were 2200 cancer cells in total. The detailed information of all patients is shown in Table S1. Because our study focused on cancer cells, non-cancer cells were removed.

Bulk sequencing data were obtained from the HNSC cohort of the TCGA database and contained a total of 501 patients.

The dataset used for pseudo-time score validation was obtained from the ICGC database (ORCA-IN, Sequence-Based Gene Expression).

### *3.2. Unsupervised Clustering of Cells*

We used the R package Seurat v4.0 [19]—a toolkit developed specifically for single-cell data. First, the R function "VlnPlot" was used to evaluate the gene expression levels of all cells to ensure that the gene expression level was distributed in an approximately Gaussian manner in each patient, so as to reduce the impact of individual differences on downstream analysis. Then, the expression profiles were log-normalized using the R function "NormalizeData", and 12,000 high variant genes were identified from 23,686 genes using the R function "FindVariableFeatures". Genes with high variation can better reflect the biological similarities and differences between cells, which is conducive to improving the accuracy of downstream unsupervised clustering. Next, we used the R function "ScaleData" to scale the expression of highly variant genes in order to balance the weight of genes in the downstream analysis. Because single-cell data are usually sparser compared with bulk sequencing and their redundancy is higher, principal component analysis (PCA) is an important and essential step in single-cell transcriptome analysis. PCA can effectively remove data noise, extract important information, and improve the accuracy and speed of downstream analysis. PCA was performed using the R function "RunPCA" on the expression profiles of highly variable genes, and then, the R function "ElbowPlot"—which ranks the principal components based on the percentage of variance explained by each component—was used to determine the optimal dimension number of the data. The optimal number of principal components would appear near the "elbow" (i.e., the inflection point). Unsupervised clustering based on the top 20 principal components was implemented for all cells using the R function "FindClusters".

### *3.3. Trajectory Inference of Cell Development*

We used the R package Monocle [20]—a powerful tool for single-cell RNA-Seq data processing and cell trajectory inference—for analysis. Monocle uses algorithms to learn the gene expression changes that each cell experiences during a state transition, to mine the overall trajectory, and then to place each cell at the appropriate location in the trajectory. In addition, Monocle can combine with UMAP (Uniform Manifold Approximation and Projection) [21] so as to make the cell trajectory more intuitive. Therefore, all cells were first embedded by UMAP based on unsupervised clustering results.

The R functions "learn\_graph" and "order\_cells" were used for the inference of cell trajectory, both of which used default parameter settings. Subsequently, the R function "plot\_cells" was used to visualize the results of cell trajectory inference with the pseudo-time.

### *3.4. Gene Expression Analysis at the Branch Point of the Trajectory*

The BEAM (branch expression analysis modeling) [22] algorithm provided by Monocle can analyze the pseudo-time-based gene expression changes at the branch point of the trajectory, revealing the important genes that drive the occurrence of cell trajectory division. The threshold was set as *<sup>q</sup>*-value < 1 × <sup>10</sup><sup>−</sup>8. This stricter threshold was designed to identify essential genes more accurately. In addition, only the genes expressed in more than 20% of cells were retained in order to screen widely expressed genes in cells.

In order to reveal the biological functions of these genes, they were first divided into two gene sets by hierarchical clustering. The genes in the same gene set would have co-expression characteristics and may regulate similar biological processes. Thus, Gene Ontology (GO) [23] enrichment analysis was performed for each gene set separately using the R package clusterProfiler [24] with the threshold *p*.adjust < 0.05 and *q*-value < 0.05.

### *3.5. Cell Communication Analysis*

In order to verify the results of cell trajectory inference and to further determine our conclusions, receptor–ligand-based cell communication analysis was performed using the Python package CellPhoneDB [25] for cell clusters adjacent to the branch point. The iteration parameter was set to 2000. The number of receptor–ligand interaction pairs was visualized using a heatmap, and a dot plot was used to show pairs with statistical *p*-values < 0.05 and mean expression > 0.

### *3.6. Survival Analysis*

In order to reveal the impact of heterogeneity on the survival status of patients, we downloaded the RNA-Seq data from the HNSC cohort with clinical information of patients from the TCGA [26] database. CIBERSORT [27] is a linear support-vector regression-based deconvolution algorithm that enables researchers to perform sample annotation based on a set of background genes. Hence, the selection of the background gene set would have a great impact on the accuracy of the downstream analysis. Since the type of expression profile derived from the TCGA database is at the bulk level, we used the differentially expressed genes that can represent the characteristics of each cluster separately as a background to infer the proportion of cell clusters of each bulk sample using the R package CIBERSORT. For stage III, the threshold for gene background screening was *p*-value < 0.01 and log2fc > 1.5. Since patients in stage IV have greater heterogeneity, the threshold was set as *<sup>p</sup>*-value < 1 × <sup>10</sup>−<sup>4</sup> and log2fc > 1.5. This stricter threshold can help identify the genes that represent the characteristics of each cell cluster more effectively and can improve the accuracy of the proportion inference. The results of CIBERSORT were subsequently filtered by setting the threshold as *p* < 0.05 and correlation > 0.3. The Ward.D algorithm [28] was used to perform hierarchical clustering to divide patients in stage III and stage IV into two groups. The survival [29] and survminer [30] R packages were used for survival analysis, plotting of the Kaplan–Meier curve, and the log-rank statistical test.

### *3.7. Pseudo-Time Score*

Here, we hypothesized that patients in the same stage would also have relatively early or advanced cancer cells due to the temporal heterogeneity and that cancer cells in the advanced stage would have more malignant features, leading to a worse prognosis. Therefore, in order to prove our hypothesis, based on the previous trajectory inference results, the pseudo-time score was constructed to quantify the temporal status of each patient. The formula was S = ∑Pi × Ti, where Pi denotes the proportion of cell cluster i of the patient, while Ti denotes the pseudo-time value of cell cluster i. Pi was calculated using CIBERSORT. Ti is the average pseudo-time value of each cell from the cell cluster i, which can be calculated by Monocle. Specifically, for patients in stage III, S = ∑Pi × Ti (i = 4, 5, 7, or 10), while for patients in stage IV, S = ∑Pi × Ti (i = 1, 2, 3, 6, 8, or 10). As for the patient groups, the pseudo-time score is the average value of S of each patient from each group.

In order to further reveal the relationship between pseudo-time score and prognosis, all patients were automatically grouped based on their pseudo-time score using survminer to perform survival analysis. Similarly, statistical significance was tested using the log-rank method.

The dataset from the ICGC database was used for further validation. This dataset contains 40 patients, including a stage II patient, 3 stage III patients, and 36 stage IV patients. The cell cluster composition of each patient was inferred using CIBERSORT, based on the same background gene sets as the TCGA data above, and then, the pseudo-time score of each patient was calculated and the patients were grouped for survival analysis as described above.

### *3.8. Drug Discovery*

There may often be many cell clusters in a patient under the single-cell resolution. For such a complex system of multi-cell clusters, we should discover specific drugs for each cell cluster, so as to select multitarget drugs or drug combinations according to the heterogeneity of patients for the elimination of all cell clusters of patients—rather than just the dominant cell cluster, which can often cause drug resistance. Therefore, we first screened specific drugs for each cell cluster. Secondly, discovering drugs to block or delay the development of OSCC would also be an idea to effectively improve the cure rate.

The discovery of drug targets is the first step. The marker genes of each cell cluster were identified using the R package SC3 [16], with the threshold set as *p*-value < 0.01 and AUROC (the area under the receiver operating characteristic) >0.8. Then, all marker genes were manually checked to ensure that they had significantly high expression characteristics in specific cell clusters. In biomolecular networks, hub nodes often have crucial biological significance, so we used hub genes in the network as targets to find candidate drugs. First, we used STRING [9] and selected the default threshold to construct protein–protein interaction (PPI) networks based on the key genes that drive the occurrence of trajectory branching from BEAM analysis and cell-cluster-specific marker genes from SC3. The visualization of PPI networks and further topological analysis were based on Cytoscape [31] software. To identify hub genes more accurately, we deeply mined the PPI networks based on 12 topological analysis methods built into the cytoHubba [10] plugin. The hub genes were defined as the intersection genes of the top 50% of each of the 12 analysis methods and were used as targets for drug discovery.

Here, we collected 216,428 drug–target relationships including 21,650 human targets and 2470 approved drugs from seven commonly used data sources: the DrugBank database (v5.1.9) [32], the Therapeutic Target Database (TTD) [33], the BindingDB database [34], the PharmGKB database [35], the Drug–Gene Interaction Database (DGIdb, v4.2) [36], the IUPHAR/BPS Guide to PHARMACOLOGY [37], and the Comparative Toxicogenomics Database (CTD) [38]. Finally, using the aforementioned genes as targets, we obtained the preliminary candidate drugs.

### *3.9. Validation of Candidate Drugs*

To validate the candidate drugs identified in our study, we combined experimental data, computational methods, and previous literature to assess the effectiveness of the candidate drugs.

The drug sensitivity experimental data and cell line expression profiles were first obtained from the Cancer Cell Line Encyclopedia (CCLE) [39] database, and all OSCC cell lines were extracted. In order to validate drug effectiveness from experiments indirectly, we computed the mean drug sensitivity of all OSCC cell lines for each drug contained in the CCLE. For drugs that were not included or showed no efficacy in the CCLE, additional literature validation was performed.

In order to reveal the drug specificity during the development of OSCC and of cell clusters, we used the R package oncoPredict [40]—a powerful tool for predicting drug responses based on background data (here, we used drug sensitivity and expression data from the CCLE)—to calculate the drug sensitivity of each cell. Then, the drug sensitivity of each cell cluster was defined as the mean drug sensitivity of cells from this cell cluster. We extracted blocking drugs and cell-specific drugs with high specificity. Drug sensitivity data from the CCLE were represented using the relative log fold-change (logFC) values of cell lines' viability to DMSO. Therefore, for blocking drugs, the following conditions were used for screening: (1) The mean logFC in all cells was lower than 0. (2) The mean logFC of cell clusters located in Path II was lower than that in Path I. For cell-cluster-specific drugs, we screened by the following criteria: (1) The mean logFC in all cells was lower than 0. (2) The mean logFC of targeting cell clusters was lower than that of non-targeting cell clusters.

### **4. Discussion**

The effective curing of advanced OSCC has been a clinical challenge because of its high heterogeneity and metastatic characteristics. Current precision treatment strategies at the single-cell level only focus on static heterogeneity and do not consider dynamic characteristics due to tumor cells' development and evolution. Therefore, our study applied traditional single-cell transcriptome analysis and dynamic cell trajectory inference theory to further explore heterogeneity and precise treatment strategies for OSCC.

During the cell clustering analysis, we found that, with the development of tumor cells, their heterogeneity became greater and greater. This complex heterogeneity reflects the characteristics of multiple evolutionary modes of tumor cells in the same stage that have different sensitivities to chemotherapeutic drugs, which is consistent with the fact that advanced OSCC is difficult to cure. Therefore, we innovatively quantified the drug sensitivity of specific cell clusters and selected drugs with high sensitivities in those specific cell clusters.

The cell trajectory inference with pseudo-time reflects the temporal heterogeneity of OSCC, which forms a complex exacerbation process composed of two paths through a branch point and gives the temporal characteristics of each cell. We integrated this temporal characteristic and cell cluster composition of patients to establish the concept of patients' time score, with significant implications for prognosis—that is, the lower the time score, the better the prognosis. Based on this, it can be seen that even patients in the same clinical stage have different temporal status.

In addition, the patient's response to drug treatment is also dynamic, so RNA-Seq can be performed at different time points during the treatment to focus on the disease progression based on our proposed pipeline. If the time score decreases gradually with the treatment, the prognosis of the patients will be better; otherwise, the medication strategy needs to be adjusted.

Finally, based on the key genes driving the differentiation of cell development trajectory and the cell-cluster-specific marker genes, we used biomolecular network theory and topological analysis to mine hub genes with extremely important biological significance, which were used as targets to find candidate drugs. As a result, there were a total of 167 drugs targeting key genes at the branch point in cell trajectory, which could be used to delay or block the further progression of OSCC. There were a total of 363 cell-cluster-specific drugs, which could be used for targeting medication based on patients' cell composition. These drugs can be combined to treat patients with multiple cell clusters, but some drugs (such as artenimol) can also target multiple cell clusters at the same time. We prefer the latter, so as to maximize the efficiency of the medication while reducing the side effects. Candidate drugs were validated by both literature and computational methods combined with experimental data.

In conclusion, our research provides a new pipeline to dissect the complex heterogeneity of cancer from a dynamic point of view. More importantly, our study proposes the concept of patients' pseudo-time score, which has great clinical value. A series of precision medicine insights for the clinical treatment practice of OSCC were presented—that is, the selection of appropriate drugs for tumor control, which aims to not only block or delay the progression of OSCC but also comprehensively kill various cancer cells. This strategy may also be adapted for other cancers in the future.

However, due to data limitations, no more in-depth research has been carried out. In addition, due to the absence of clinical trials, this study has some limitations. As single-cell spatial sequencing and cell trajectory inference technologies mature, we have reason to believe that the challenges in completely curing OSCC or other cancers will eventually be overcome.

### **5. Conclusions**

Our study confirms the complex heterogeneity of OSCC both statically and dynamically. Advanced-stage patients have greater heterogeneity than early stage patients. The development of OSCC is not a simple pathway, and cell trajectory inference confirms its complex dynamic landscape, which contains multiple developmental pathways. According to this, the proposed pseudo-time score is closely related to the patient's prognosis and has good prognostic prediction potential, as validated by external datasets. We searched for candidate drugs for the treatment and blocking of OSCC, most of which were validated in the literature and via computational methods based on experimental data. In conclusion, our study comprehensively parses the developmental characteristics of OSCC, offering new insights into the basic research and clinical treatment of OSCC.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers14194801/s1, Figure S1. Principle component selection. The x-axis represents the number of PC and the y-axis represents the standard deviation. The optimal number of principal components appears at the inflection point and was marked by a red arrow. Figure S2: Survival analysis for validation data. The survival curve is shown above, with the x-axis representing the survival time (days), the y-axis representing the survival probability, and the censored values are indicated by "+" in the curve. Below is the risk table showing the number (or percentage) of survivors at each time point. Figure S3. PPI network of key genes of the branch point. The nodes represent genes and the edges represent interaction relationships. Node colors represent gene sets, consistent with Figure 3. Figure S4. PPI networks of cell cluster-specific genes. The nodes represent genes and the edges represent interaction relationships. Node colors represent the auroc of genes generated by SC3 and the legend is shown below. Table S1: The detailed information of patients. Table S2: Key genes which promote the occurrence of the branch in cell trajectory. Table S3: Background genes of stage III for cibersort. Table S4: CIBERSORT result of patients in stage III. Table S5: literature validation for targets. Table S6: In this table, the drug-target relationship is represented as 0-1, and 1 represents that the gene is the target of the drug. Table S7: Marker genes from SC3. Table S8: drug-target-cluster relationships. Table S9: This table shows the literature validation results of CCLE intersection drugs. Only drugs that have not shown efficacy (logFC > 0) on OSCC from the experimental data of CCLE were verified, and drugs that already have been proved effective (logFC < 0) were marked in green color.

**Author Contributions:** Conceptualization, Q.M. and X.C.; Data curation, Y.L.; Formal analysis, Q.M., F.W., G.L., F.X., Y.L. and H.X.; Funding acquisition, X.C.; Methodology, L.L., G.L., F.X. and D.Z.; Project administration, Q.M. and X.C.; Software, D.Z.; Writing—original draft, Q.M.; Writing—review and editing, L.L., F.W., F.X., D.Z., H.X. and X.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Natural Science Foundation of China (grant No. 61671191).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The expression data are available through the Gene Expression Omnibus with accession number GSE103322.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article*

## **Oral** *Candida* **spp. Colonisation Is a Risk Factor for Severe Oral Mucositis in Patients Undergoing Radiotherapy for Head & Neck Cancer: Results from a Multidisciplinary Mono-Institutional Prospective Observational Study**

**Cosimo Rupe 1,\*, Gioele Gioco 1, Giovanni Almadori 2, Jacopo Galli 2, Francesco Micciché 3, Michela Olivieri 3, Massimo Cordaro <sup>1</sup> and Carlo Lajolo <sup>1</sup>**


**Simple Summary:** This study aims to find a correlation between *Candida* spp. oral colonisation prior to radiotherapy and (i) the development of severe oral mucositis (OM) (grade 3/4) and (ii) early development of severe OM (EOM). *Candida* spp. in the oral cavity appears to be a predictive factor of EOM. Preventive treatment could aid in reducing incidence of EOM. Further clinical trials are required to confirm our findings.

**Abstract:** Background: This study aims to find a correlation between *Candida* spp. oral colonisation prior to radiotherapy (RT) and (i) the development of severe oral mucositis (OM) (grade 3/4) and (ii) early development of severe OM (EOM). Methods: The protocol was registered on ClinicalTrials.gov (ID: NCT04009161) and approved by the ethical committee of the 'Fondazione Policlinico Universitario Gemelli IRCCS' (22858/18). An oral swab was obtained before RT to assess the presence of *Candida* spp. Severe OM occurring before a dose of 40 Gy was defined as EOM. Results: No patient developed G4 OM, and only 36/152 patients (23.7%) developed G3 OM. Tumour site and lymphocytopenia were risk factors for severe OM (OR for tumour site: 1.29, 95% CI: 1–1.67, *p* = 0.05; OR for lymphocytopenia: 8.2, 95% CI: 1.2–55.8, *p* = 0.03). We found a correlation between *Candida* spp. and EOM (OR: 5.13; 95% CI: 1.23–21.4 *p* = 0.04). Patients with oral colonisation of *Candida* spp. developed severe OM at a mean dose of 38.3 Gy (range: 28–58; SD: 7.6), while negative patients did so at a mean dose of 45.6 Gy (range: 30–66; SD: 11.1). Conclusions: *Candida* spp. in the oral cavity appears to be a predictive factor of EOM.

**Keywords:** oral mucositis; radiotherapy; head and neck cancer; oral *Candida* spp.; oral candidiasis; chemotherapy; radiochemotherapy

### **1. Introduction**

More than 900,000 new cases of head–neck cancer (HNC) are diagnosed worldwide, with 40,000 new cases and 7890 deaths reported annually in the United States. HNC can arise in multiple anatomic subsites (i.e., the oral cavity, oropharynx, hypopharynx, nasopharynx, larynx, and salivary glands) [1].

HNC treatment is challenging and requires a multidisciplinary approach with a team of specialists, including head and neck surgeons, radiation oncologists, medical oncologists, nutritionists, nuclear physicians, and oral oncologists [2,3].

**Citation:** Rupe, C.; Gioco, G.; Almadori, G.; Galli, J.; Micciché, F.; Olivieri, M.; Cordaro, M.; Lajolo, C. Oral *Candida* spp. Colonisation Is a Risk Factor for Severe Oral Mucositis in Patients Undergoing Radiotherapy for Head & Neck Cancer: Results from a Multidisciplinary Mono-Institutional Prospective Observational Study. *Cancers* **2022**, *14*, 4746. https:// doi.org/10.3390/cancers14194746

Academic Editor: David Wong

Received: 5 September 2022 Accepted: 27 September 2022 Published: 29 September 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Approximately 60% of patients with HNC require radiotherapy (RT), with or without induction chemotherapy [4], and a substantial proportion of patients suffer significant treatment-related adverse effects [5], including acute adverse effects (i.e., mucositis and dermatitis) that occur during treatment and late adverse effects (i.e., dysgeusia, osteoradionecrosis, and trismus) that occur in the weeks following the end of therapy [6–8].

Oral and oropharyngeal mucositis (OM) caused by RT and combined systemic therapies appears to be a significant side effect that presents numerous clinical signs and symptoms [9]. It affects the patient's quality of life (QoL) and is associated with symptoms such as pain, bleeding, dysphagia, local infections, increased susceptibility to secondary and systemic infections, impaired food intake, and weight loss [10,11]. The incidence of OM in patients treated with RT is estimated to be approximately 80%, becoming nearly ubiquitous in patients undergoing radiochemotherapy (RTCT) [12].

Severe OM (grade 3/4), according to the Radiation Therapy Oncology Group (RTOG), appears in approximately 43% of patients undergoing combination treatment [13]. It may cause inadequate food intake; this further results in the development of severe nutritional deficiencies and a need for parenteral nutrition. In addition, approximately 15% of patients require an interruption of RT or dose reshaping of concomitant systemic therapy, thus influencing the effectiveness of the treatment [14]. However, the incidence and severity of this condition varies depending on several factors (i.e., cancer subsite, radiation dose, volume of the irradiated mucosa, daily fractionation, association with CT, habit history, oral health prior to initiation of treatment, and neutrophil recovery period) [15].

Nevertheless, while dosimetric parameters are best known to correlate with the time of onset and severity of side effects, the available literature does not provide clear evidence about the clinical parameters that can predict OM development or may indicate worsening of OM [16].

Oral candidiasis is a common fungal disease caused by overgrowth of *Candida* spp. in the mouth. Acute pseudomembranous candidiasis and acute erythematous candidiasis are the most frequent clinical patterns of oral candidiasis, requiring complex treatments (i.e., adequate oral hygiene, topical agents, and systemic medications) that often lead to chronic candidiasis in patients with HNC [17,18]. Oral candidiasis, especially in its acuteerythematous manifestation, may enhance OM-related symptoms and result in worsening of the clinical condition of patients. Thus, treatment of oral candidiasis is recommended when RT in the head and neck region is scheduled. Nevertheless, *Candida* spp. can be found in 50% of the population as a component of the oral microbiota, and it can become a pathogen even after the initiation of RT [19]. Furthermore, alterations in the mucosal layer structure caused by OM often allow bacteria and fungi to penetrate damaged tissue and cause infections, increasing the risk of oral candidiasis development [20].

The primary objective of this observational prospective cohort study was to understand whether the presence of *Candida* spp. in the oral cavity, evaluated using an oral swab taken prior to initiation of RT, is a risk factor for the development of severe OM during RT. The secondary objectives were (i) to understand whether oral colonisation of *Candida* spp. is a risk factor for the early development of severe OM (EOM), defined as an OM developed at 40 Gy of the cumulative radiation dose, (ii) to understand whether other clinical parameters (radiation dose, dose received by the oral cavity and oropharynx, smoking history, white blood cell count (WBC), chemotherapy (CT), and cancer subsite) are risk factors for the development of severe OM or (iii) EOM, and (iv) to evaluate the overall incidence of OM in the studied cohort.

### **2. Materials and Methods**

### *2.1. Setting*

The protocol was registered on ClinicalTrials.gov (ID: NCT04009161) and was approved by the ethical committee of the 'Fondazione Policlinico Universitario A. Gemelli IRCCS in Rome' (22858/18). The study was conducted in accordance with the Declaration of Helsinki, and all patients signed an informed consent form. Patients with HNC

seeking treatment at the Oral Medicine, Head and Neck Department, with a scheduled external beam RT at Gemelli Advanced Radiation Therapy (ART), Fondazione Policlinico Universitario A. Gemelli-IRCSS, between March 2017 and August 2021, were consecutively recruited for this study. All included patients visited the hospital prior to initiation of RT. This paper was written in accordance with the STROBE guidelines (Table S1).

### *2.2. Participants*

The inclusion criteria were HNC diagnosis, indication for RT (either adjuvant or neoadjuvant), and treatment with a curative intent. The exclusion criteria were as follows: indication for palliative treatment, presence of clinically detectable signs of oral candidiasis, patients who received neoadjuvant CT before RT, and metastatic disease.

### *2.3. Variables—Anamnesis*

Before the clinical examination, anagraphic and anamnestic data (age, sex, and comorbidities) were recorded, particularly focusing on the oncologic history of the patient (tumour site, histological type of cancer, stage of the tumour, and previous oncologic treatments) and on the exposure to risk factors for the oncologic disease (i.e., smoking).

### *2.4. Variables—Oral Examination*

Subsequently, clinical evaluation of the oral mucosal conditions was performed, focusing on the presence of signs of oral candidiasis.

Furthermore, oral colonisation by *Candida* spp. was recorded using a sterile swab (eSwab®®, Copan's Liquid Amies Elution Swab, Copan Italia SPA, Brescia, Italy); it was rubbed on the following mucosal surfaces of the oral cavity: hard palate, tongue, upper and lower vestibule, and ending at the commissures of the mouth. Post sample collection, sterile swabs were placed in tubes containing 1 mL of transport medium. The tubes were then stored at 4 ◦C until further processing. Processing involved streak inoculation of the swab onto Sabouraud dextrose agar (SDA) plates, followed by incubation at 37 ◦C for 48 h, according to the manufacturer's instructions.

The unstimulated salivary flow rate was assessed using the spitting method. Patients were instructed to collect saliva for 5 min in a graded tube. The stimulated salivary flow was determined in a similar manner. Saliva secretion was stimulated by applying a solution of 2% citric acid to the sides of the tongue at intervals of 30 s. An unstimulated salivary flow (USF) of over 0.4 mL/min was considered normal [21]. Furthermore, a blood count was performed before the beginning of RT, and the following variables were recorded: number of leukocytes, neutrophils, and lymphocytes. Leukopenia was defined as a leukocyte count < 4 × 109/L, neutropenia as <1.5 × 109/L neutrophils, and lymphocytopenia as <1 × 109/L lymphocytes.

### *2.5. Variables—RT and OM*

RT was delivered using the volumetric-modulated arc radiotherapy (VMAT) technique with a linear accelerator, and treatment was administered in five daily fractions per week for 6–7 weeks. The definition of volume is in accordance with international RT guidelines [22,23].

The treatment plan was optimised to ensure adequate coverage of the target (D95% of the treatment volume received > 95% of the prescribed dose) and to respect the constraints of the various organs at risk, identified during contouring.

Treatment included a daily image-guided radiation therapy (IGRT) check with Cone-Beam CT and, if necessary, the radiation dose was re-planned between 30 Gy and 40 Gy, in case of tumour shrinkage or anatomical changes. Patients were instructed to receive supportive therapy according to the centre's procedures and international guidelines [24]. Each patient underwent at least one weekly examination during RT, in which the diagnosis of OM took place: if present, OM was recorded according to the National Cancer Institute Common Toxicity Criteria for Adverse Events (CTCae, version 4.0) [25]. Each patient was assigned a single OM grade, corresponding to the most severe OM grade recorded during

RT and during the immediate RT follow-up. When OM signs disappeared, the patients terminated their study involvement. For patients who developed severe OM, the dose at which OM developed was also recorded.

According to Mallick et al., since the onset of G3-G4 OM occurs between 50 Gy and 60 Gy [26], we assumed that the onset of severe toxicity at 40 Gy should be considered as early acute toxicity. Onset of OM at a dose of 40 Gy or less was defined as an 'early onset mucositis (EOM)'.

Severe OM was managed according to the centre's procedures and international guidelines: in case of G3 OM, the patients were treated by analgesic drugs to reduce the pain [24].

### *2.6. Statistical Analysis*

The sample size was calculated, with a 90% confidence level and 80% power, by comparing two proportions: considering 45% as the expected incidence of severe OM in the presence of *Candida* spp. in the oral cavity and 25% as the expected incidence in the absence of oral colonisation of *Candida* spp. The required sample size was 136 patients, with 68 in each group (positive or negative for oral cavity swabs). Considering a dropout rate of 10%, the final sample size was 150 patients.

The following variables were recorded as baseline patient characteristics (sex, age, histological type and stage of the tumour, site of the tumour, risk factors such as smoking history, previous oncological surgery, salivary flow, presence of *Candida* spp. oral colonisation), basal treatment characteristics (scheduled CT, Total Radiation Dose, daily fraction, dose received by the oral cavity and oropharynx), and treatment-related toxicity parameters (presence and grade of OM).

Qualitative variables were described using absolute and percentage frequencies, while the Kolmogorov–Smirnov test was performed to evaluate the normal distribution of quantitative variables. Quantitative variables were summarised either as mean and standard deviation (SD) if normally distributed, or as median and percentiles otherwise.

OM was reclassified into three categories: absence of OM, grade I or II OM, and grade 3 or 4 OM.

Correlation analysis between OM onset and clinical characteristics of the patients was performed. The Mann–Whitney U test and Kruskall–Wallis test were performed to compare the continuous variables with non-parametric distribution, while the parametric variables were analysed through an ANOVA test; Pearson's χ<sup>2</sup> test and Fisher's exact test were used to compare the discontinuous variables. Statistical analysis was stratified according to the following variables: development of severe OM and early onset of severe OM.

Univariate analysis was performed to determine the risk factors associated with the onset of OM, and risk factors were introduced in a stepwise logistic regression analysis to identify independent predictors of OM. The same statistical analysis was used to determine risk factors for EOM. All statistical analyses were performed using IBM SPSS Statistics software (IBM Corp. Released 2017. IBM SPSS Statistics for Apple, Version 25.0 (IBM Corp., Armonk, NY, USA).

### **3. Results**

One hundred and sixty-three patients were enrolled in the study; 11 patients were excluded per the inclusion criteria: 5 patients suffered from oral candidiasis at baseline, so they were treated but excluded from the final sample, whereas 6 patients had received a planning for a palliative treatment. The final sample included 152 patients (49 female and 103 male), with a mean age of 60.3 years (range: 22–86). One hundred and fifteen (75.7%) patients had locally advanced oncologic disease (stage III–IV), 93 patients (61.2%) received treatment with curative intent, and 59 patients (38.8%) received adjuvant treatment. Oral cavity swabs were positive in 68 of the 152 patients (44.7%), and the remaining (84/152, 55.3%) swabs showed negative results prior to the initiation of RT. The mean total RT dose was 67.6 Gy (50–72), and the dose was fractionated in 2 Gy/die in majority of the patients (136/152, 89.5%). One hundred and twenty patients out of 152 (78.9% of the total sample)

developed OM. Severe OM occurred in 36 patients (23.7% of the total sample). None of the patients developed G4 OM. Termination of RT before reaching the target dose was not required in any patient, and all patients completed their scheduled treatment; three patients had to discontinue RT for a few days. However, they still finished their RT course. Patient and treatment characteristics and related toxicity parameters are shown in Table 1.

**Table 1.** Basal patients' characteristics, basal treatment characteristics of the studied population and treatment related toxicity parameters. SD: standard deviation; SCC: squamous cell carcinoma, RT: radiotherapy.



The study flow chart is presented in Figure 1.

**Figure 1.** Strobe flow chart of the study.

Results of the statistical analysis stratified according to the development of severe OM are shown in Table 2.

In the univariate analysis, the clinical parameters associated with severe OM onset were the tumour site and RT-CT treatment. Patients with different tumour sites showed a different incidence of severe OM (χ<sup>2</sup> test, *p* < 0.05), and nasopharyngeal cancers were associated with the highest incidence of OM (9/19 patients, 47.4%), while laryngeal cancers had the lowest incidence of OM (1/28, 3.6%). Patients who developed severe OM (25/36 patients; 69.4%) were more frequently treated with combined RT-CT (χ<sup>2</sup> test, *p* = 0.05), while patients who did not develop severe OM often did not receive combined treatment (61/116; 52.6%). The prevalence of oral colonisation of *Candida* spp. was higher in patients with severe OM (20/36; 55.6%) than in other patients in the cohort (48/116; 41.4%), but the correlation was not statistically significant. Furthermore, severe OM correlated with leukopenia, neutropenia, and lymphocytopenia (χ<sup>2</sup> test, *p* < 0.05). However, when inserted in a multiple logistic regression model, only tumour site and lymphocytopenia were statistically significant risk factors for severe OM development (OR for tumour site: 1.29, 95% CI: 1–1.67, *p* = 0.05; OR for lymphocytopenia: 8.2, 95% CI: 1.2–55.8, *p* = 0.03).

Table 3 summarises the characteristics of patients who developed EOM. Oral colonisation by *Candida* spp. was correlated with an early onset of severe OM in the univariate analysis (χ<sup>2</sup> test, *p* < 0.05), showing that the presence of *Candida* spp. in the oral cavity is a risk factor for the development of EOM (OR: 5.13, 95% CI: 1.23–21.4 *p* = 0.04). Patients with oral colonisation by *Candida* spp. developed severe OM at a mean dose of 38.3 Gy (range: 28–58), while patients with negative oral swabs developed severe OM at a mean dose of 45.6 Gy (range: 30–66). From a clinical point of view, these patients experienced severe OM approximately four days before those with a negative swab.


**Table 2.** Clinical variables of the studied population, according to the development of severe mucositis.

<sup>a</sup> Correlation between leucocytes and severe OM—Mann–Whitney test—*p* < 0.05. <sup>b</sup> Correlation between neutrophils and severe OM—Mann–Whitney test—*p* < 0.05. <sup>c</sup> Correlation between lymphocytes and severe OM— Mann–Whitney test—*p* < 0.05. <sup>d</sup> Correlation between Leukopenia and severe OM—χ<sup>2</sup> Test—*p* < 0.05. <sup>e</sup> Correlation between neutropenia and severe OM—χ<sup>2</sup> Test—*p* < 0.05. <sup>f</sup> Correlation between lymphocytopenia and severe OM—χ<sup>2</sup> Test—*p* < 0.05. <sup>g</sup> Correlation between tumour site and severe OM—χ<sup>2</sup> Test—*p* < 0.05. <sup>h</sup> Correlation between chemotherapy and severe OM—χ<sup>2</sup> Test—*p* = 0.05. <sup>i</sup> Multiple Logistic Regression showed how tumour site and lymphocytopenia are risk factors for severe OM: tumour site—OR: 1.29, 95% CI: 1–1.67, *p* = 0.05. Lymphocytopenia-OR: 8.2, 95% CI: 1.2–55.8, *p* = 0.03.


**Table 3.** Clinical variables of the studied population, according to the development of severe mucositis. WBC: white blood cell count.

<sup>a</sup> Correlation between oral candida and early onset severe OM—χ<sup>2</sup> Test—*p* < 0.05; (OR: 5.13, 95% CI: 1.23–21.4 *p* = 0.04).

### **4. Discussion**

RT has played a fundamental role in the multidisciplinary management of patients with HNC over the last few decades. Nevertheless, some adverse events such as OM are a major concern for clinicians. OM is an acute inflammation that may initially manifest as redness of the mucous membrane, which eventually evolves into ulceration and formation of a pseudomembrane, leading to a temporary disruption of mucosal integrity [27]. OM pathogenesis, according to the most widely accepted biological model, appears to be a complex multistep process. This theory, proposed by Sonis et al. [28], describes an initiation phase, followed by upregulation and activation, signal amplification, ulceration, and ultimately a healing phase. It is crucial to understand that OM does not simply result from the epithelial injury caused during RT or CT; the epithelium, the underlying connective tissue, and the type of injury are the main characteristics of this mechanism, although other factors (the oral environment, immunological conditions, and performance status) also play a role in OM pathogenesis. Despite the wide interest in this topic, evidence regarding the risk factors for OM is limited, representing a challenge for healthcare professionals involved in the supportive care field [29]; in particular, mycetes harbouring in the oral cavity could play a pathogenic role in the onset and perpetration of OM. Few studies have investigated the possible role of *Candida* spp. in the onset of OM.

The main objective of this study was to identify the clinical impact of oral *Candida* spp. colonisation on OM onset and its possible effects on OM severity. Although the prevalence of *Candida* spp. in the oral cavity was higher in patients with severe OM (20/36; 55.6%) than in other patients of the cohort (48/116; 41.4%), this correlation was not statistically significant (χ<sup>2</sup> test, *p* = 0.097). Nevertheless, in our population, *Candida* spp. was identified as the only risk factor for EOM (OR: 5.13, 95% CI: 1.23–21.4 *p* = 0.04; Table 3).

The early onset of severe OM is a critical issue in the HNC patients' management, since it may further reduce the QoL of patients [30] and may increase the need for intensive supportive care and hospitalisation [31] and the need for treatment interruption, impairing the control of the disease [32]. Several studies have demonstrated a correlation between EOM and different CT regimens, RT dose per fraction, and treatment timing; however, [33] no studies have demonstrated a correlation between EOM and *Candida* spp. colonisation. Although the reason why *Candida* spp. colonisation resulted as the only risk factor for EOM is unclear, several hypotheses can be put forward. First, *Candida* spp. may accelerate the cascade of events leading to OM through their virulence factors, which can directly damage the epithelial layer by the production of cytolytic enzymes (proteinases, hemolysins, siderophores, and phospholipases) [34]. In addition, as highlighted by in vitro studies [35], *Candida* spp. blastospores may stimulate peripheral blood mononuclear cells to secrete tumour necrosis α (TNF-α) and the type 1 cytokine interferon-γ (IFN-γ), which contribute to OM pathogenesis [28]. However, it has been demonstrated that *Candida* spp. may influence bacterial growth, especially in the oral cavity, through different interaction mechanisms (adhesion and quorum sensing) [36]. This condition may cause a worsening of OM; in fact, it has been hypothesized that dynamic changes in the oral microbial community composition may be involved in OM pathogenesis [37]. HNC patients, often dentally compromised even before the initiation of RT [38], show microbial and inflammatory profiles different from healthy individuals [39], which probably contributes to the high incidence of OM in this group of patients.

Furthermore, an irradiated oral and oropharyngeal mucosa presents an ideal environment that favours opportunistic oral candidiasis; and ulcerations in the mucosal layer structure caused by OM may allow fungal penetration into the damaged tissue [20], enhancing the interactions between *Candida* spp. and the connective tissues. The ongoing inflammatory process, resulting in an anaerobic condition, could promote interactions between *Candida* spp. and pathogenic bacteria of the oral cavity [36]. Although not definitively proven, it is likely that RT may slightly change the oral microbiota composition, favouring the onset of oral candidiasis [40]. Another issue is that irradiated patients may often experience an impairment of the local and systemic immune systems, as highlighted by our findings (leukopenia was present in 12.5% of our samples) and previously published papers [41]. Furthermore, *Candida* spp. hyphae induce a higher production of IL-10, an immunosuppressive cytokine, indirectly worsening the local immune response [35]. Another factor that can be involved in this process is the reduction in salivary flow in irradiated patients. Although hyposalivation is considered a late-onset RT adverse event, it has been

demonstrated that salivary flow may be reduced to 50–70% of the baseline after 10–16 Gy of RT [20,42]. Radiation induces a decrease in amylase activity, bicarbonate levels, and pH and a significant increase in the viscosity [43] of saliva, mainly due to the disruption of the mucin network [44], which could promote the transformation of *Candida* spp. into a pathogen. A recent observational study demonstrated that *Candida* spp. had higher biofilm formation capability in a population of irradiated patients than in healthy individuals [45]. Nevertheless, our findings do not demonstrate a direct correlation between hyposalivation and EOM; thus, further studies are required to evaluate this hypothesis.

Previous studies have investigated the role of *Candida* spp. in irradiated HNC patients, with contradictory results. Singh et al. [19] argued that OM is a risk factor for oral candidiasis, suggesting that *Candida* spp. may overinfect pre-existing lesions. Suryawanshi et al. [46] concluded that *Candida* spp. colonisation does not influence the severity of OM, whereas oropharyngeal candidiasis may play a role in increasing the duration and discomfort of OM. In contrast, Rao et al. [47] found that biweekly prophylactic administration of fluconazole, an antimycotic, during RT-CT for HNC reduced the incidence of OM. Although their study was retrospective and lacked a control group or a pre-RT *Candida* spp. colonisation assessment, our findings may confirm their interesting results. Further clinical trials, designed to investigate this specific outcome, are needed to confirm this hypothesis; performing an oral swab before starting RT should be considered a routine procedure during pre-RT dental evaluation, given its low cost and the potential impact of *Candida* spp. on OM. Future studies should evaluate the preventive eradication of *Candida* spp. and assess whether this therapy has a positive effect in reducing the incidence of EOM.

Other important results retrieved by this study include the incidence of OM in a cohort of patients with HNC and the role of other possible risk factors for OM. Among the studied population, while the overall incidence of OM was 78.9%, in accordance with the 80% estimate reported by Trotti et al. [13], only 36 of 152 (23.7%) patients developed severe OM, which is a lower rate than that reported in previous studies [30]. While severe OM occurred in 36 out of the 152 patents, no patient developed G4 or G5 OM. The low incidence of severe OM might have been due to the use of the VMAT technique, which helps reduce the duration of RT sessions and spares healthy tissues [48].

Based on the results of the multiple logistic regression, two main clinical parameters were significantly associated with a higher incidence of severe OM: the site of the tumour (OR: 1.29, 95% CI: 1–1.67, *p* = 0.05) and lymphocytopenia (OR: 8.2, 95% CI: 1.2–55.8, *p* = 0.03).

As expected, the larynx had the lowest incidence of severe OM (1/28 patients, 3.6%), followed by the hypopharynx (1/9 patients, 11%). In contrast, the nasopharynx was the site that showed the highest incidence (9/19 patients, 47.7%). This finding confirms the results of the vast majority of the literature [41,49], and can be due to the introduction of more targeted radiation therapy techniques and the use of increasingly smaller treatment volumes (PTV) in clinical practice, which helped to spare the oral and oropharyngeal mucosa, maintaining clinical efficacy on the tumoural tissues.

Although a few studies have investigated the role of lymphocytopenia in the onset of OM [41] and found a correlation, the evidence is still limited [29]. It is reasonable to think that lymphocytopenia may have a direct or indirect role in modulating the inflammatory process leading to OM (i.e., dysregulation of the inflammatory processes and enhancement of the risk of bacterial colonisation of the epithelium, which stimulates cytokine production); however, the effective role of lymphocytes in the pathogenesis of OM needs to be clarified by other studies. Notably, Munneke et al., in a randomised clinical trial, highlighted how activated innate lymphoid cells are associated with reduced susceptibility to OM in a cohort of patients undergoing allogeneic haematopoietic stem cell transplantation (HSCT) [50], thus indirectly confirming our findings.

Previous studies have revealed how concomitant CT can predict OM in patients with HNC [10,16,51]. Our results, in accordance with the previous results, showed a higher incidence (χ<sup>2</sup> test, *p* < 0.05) of severe OM (29.1%) in RT-CT-treated patients than in patients treated with RT alone (16.7%). However, when included in a logistic regression model, this correlation was not statistically significant. Similarly, CT was not a statistically significant risk factor for EOM. It is likely that the number of patients treated with concomitant CT (86 patients) within our sample was not sufficient to draw statistically significant results.

This study has several strengths. Our results revealed how severe OM developed significantly earlier in patients with HNC with oral *Candida* spp. colonisation prior to initiation of RT. To our knowledge, this is the first study to prospectively investigate this relationship in patients with HNC undergoing RT, and its novelty lies in the fact that it highlights how oral *Candida* spp. can also play a crucial role in the onset of severe OM, apart from the simple overinfection of already existing lesions.

Furthermore, since this was a monocentric study, the RT treatment plan was always decided by the same radiotherapist (F.M.) with the same device, and the whole sample was homogeneous by mean total dose of radiation, which allowed the identification of risk factors different from the RT dose or RT technique.

This study also had some limitations. Its monocentric nature may have limited the reliability of our results on a larger scale, especially regarding the effects of different types of RT and other devices. Furthermore, because this study aimed to evaluate the onset of OM as a primary outcome, the duration of severe OM was not recorded. These data, however, need to be recorded in future studies and will contribute to a deeper description and understanding of the clinical impact of OM in terms of the number of visits during therapy, supportive therapy, economic burden, and perceived QoL changes. Another possible limitation of this study may be the lack of evaluation of the oral microbiota of included patients; it was not possible to carry out this analysis given its high economic impact. Nevertheless, given its significance in promoting the occurrence of oral candidiasis, further studies should evaluate its impact on opportunistic infection-related diseases.

### **5. Conclusions**

In conclusion, our findings show that oral colonisation by *Candida* spp. is a predictive factor for EOM. Performing an oral swab test before initiation of RT should be considered as a routine procedure during pre-RT dental evaluation, given its low cost and the impact of *Candida* spp. on OM. Therefore, physicians, dentists, and otolaryngologists should be aware that OM prevention strategies should be implemented before the initiation of RT in patients with positive oral cavity swabs. Future studies should evaluate the preventive eradication of *Candida* spp. and assess whether this therapy has a positive effect in reducing the incidence of EOM.

**Supplementary Materials:** The following supporting information can be downloaded at https: //www.mdpi.com/article/10.3390/cancers14194746/s1: Table S1: STROBE Guidelines.

**Author Contributions:** Conceptualization, C.R., C.L. and F.M.; methodology, C.R. and C.L; software, C.R. and C.L.; validation, M.C., J.G. and G.A.; formal analysis, C.R. and C.L.; investigation, M.O. and G.G.; data curation, C.R. and C.L.; writing—original draft preparation, C.R.; writing—review and editing, C.L. and F.M.; visualization, G.G.; supervision, M.C., J.G. and G.A.; project administration, C.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** This study was conducted in accordance with the Declaration of Helsinki, approved by the Ethics Committee of the Università Cattolica del Sacro Cuore (Ref. 22858/18) and registered at ClinicalTrials.gov (ID: NCT04009161).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding authors. The data are not publicly available because of privacy concerns.

**Acknowledgments:** The authors would like to acknowledge the help received from Sunstar Europe S.A., Route de Pallatex 11, P.O. Box 32, 1163 Etoy, Switzerland. (prot. CA-19–0888).

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Trimodal Therapy in Esophageal Squamous Cell Carcinoma: Role of Adjuvant Therapy Following Neoadjuvant Chemoradiation and Surgery**

**Xiaokun Li †, Siyuan Luan †, Yushang Yang †, Jianfeng Zhou, Qixin Shang, Pinhao Fang, Xin Xiao, Hanlu Zhang and Yong Yuan \***

> Department of Thoracic Surgery, West China Hospital, Sichuan University, Chengdu 610000, China; drlixiaokun@163.com (X.L.); luansiyuan30@163.com (S.L.); yangysh07@163.com (Y.Y.); 2020224020107@stu.scu.edu.cn (J.Z.); qixinshang0405@gmail.com (Q.S.); 15877929137@163.com (P.F.); drxiaoxin@foxmail.com (X.X.); drzhanghanlu@wchscu.cn (H.Z.)

**\*** Correspondence: yongyuan@scu.edu.cn

† These authors contributed equally to this work.

**Simple Summary:** In this work, we aimed to explore the effectiveness of adjuvant therapy after trimodal therapy (neoadjuvant chemoradiotherapy and esophagectomy) in patients with thoracic esophageal squamous cell carcinoma (ESCC). Overall survival (OS) and disease-free survival (DFS) were both compared for adjuvant and non-adjuvant groups. Propensity score matching was used to eliminate the confounding factors between the two groups. Meanwhile, subgroup analysis based on a neoadjuvant-treated node stage (ypN) was performed to precisely stratify the patients and to guide the clinical decision-making at the point of care. As of now, there is no guideline or recommendation on the treatment of ESCC patients with adjuvant therapy after neoadjuvant chemoradiotherapy followed by surgery. The results of our study indicate that adjuvant therapy after trimodal therapy could shorten OS and DFS in patients with ESCC. Meanwhile, adjuvant therapy is an independently unfavorably prognostic factor for DFS. Therefore, adjuvant therapy is not recommended for ESCC patients after trimodal therapy, especially patients without nodal metastases after neoadjuvant therapy. To our knowledge, this is the first retrospective study using subgroup analysis to examine the effect of adjuvant therapy in ESCC patients after trimodal therapy by comparing overall survival and disease-free survival. The results of our study add useful evidence to recent guidelines.

**Abstract: Background:** The aim of this study was to determine the role of adjuvant therapy after neoadjuvant chemoradiotherapy and esophagectomy for esophageal squamous cell carcinoma (ESCC). **Methods:** The study retrospectively reviewed 447 ESCC patients who underwent neoadjuvant chemoradiotherapy and esophagectomy. Patients were divided into an adjuvant therapy group and no adjuvant therapy group. Propensity score matching was used to adjust the confounding factors. **Results:** 447 patients with clinical positive lymph nodes and no distant metastasis treated with neoadjuvant chemoradiotherapy and esophagectomy were eligible for analysis. After propensity score matching, there were 120 patients remaining in each group. Patients receiving adjuvant therapy had a significantly shorter post-resection overall survival (OS) and disease-free survival (DFS) when compared to patients not receiving adjuvant therapy (log-rank, OS: *p* = 0.046, DFS: *p* < 0.001). Receiving adjuvant therapy is not an independently prognostic factor for OS (hazard ratio (HR): 1.270, HR: 0.846–1.906, *p* = 0.249) but a significantly unfavorable independent prognostic factor for DFS (HR: 2.061, HR: 1.436–2.958, *p* < 0.001). **Conclusions:** The results of our study indicate that adjuvant therapy after neoadjuvant chemoradiotherapy and surgery could reduce the OS and DFS in patients with ESCC. Therefore, adjuvant therapy is not recommended for ESCC patients after neoadjuvant chemoradiotherapy and esophagectomy, especially patients without nodal metastases after neoadjuvant therapy.

**Keywords:** esophageal cancer; neoadjuvant chemoradiotherapy; esophagectomy; adjuvant therapy

**Citation:** Li, X.; Luan, S.; Yang, Y.; Zhou, J.; Shang, Q.; Fang, P.; Xiao, X.; Zhang, H.; Yuan, Y. Trimodal Therapy in Esophageal Squamous Cell Carcinoma: Role of Adjuvant Therapy Following Neoadjuvant Chemoradiation and Surgery. *Cancers* **2022**, *14*, 3721. https://doi.org/ 10.3390/cancers14153721

Academic Editor: David Wong

Received: 16 June 2022 Accepted: 27 July 2022 Published: 30 July 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Esophageal cancer (EC) is the sixth leading cause of cancer deaths worldwide and the second deadliest gastrointestinal cancer after gastric carcinoma [1]. The literature reports that approximately 200,000 people die of EC annually worldwide, and most cases of EC are diagnosed at advanced stages [1]. Esophageal squamous cell carcinoma (ESCC) represents the predominant subtype of EC, with most cases occurring in eastern Asia. The morbidity rate varies extremely across areas and countries [2,3].

Although a tremendous improvement of therapeutic modalities has been recently observed, patients' quality of life remains poor, and the five-year survival rate rarely exceeds 40% [3]. Currently, surgery remains the major treatment for patients with early stage resectable ESCC, whereas neoadjuvant therapy (chemotherapy, radiotherapy, or their combination prior to surgery) followed by esophagectomy is the standard of care for those with locally advanced disease (cT1-2N+ or cT3-4aN1-3). It has been proven that patients with locally advanced esophageal cancer can benefit from trimodal therapy (neoadjuvant concurrent chemoradiation followed by surgery), when compared to surgery alone [2]. However, additional adjuvant therapy (chemotherapy and/or radiotherapy after surgery) may be necessary for patients that do not fully respond to neoadjuvant therapy, characterized by pathologically confirmed residual disease and lymph node metastasis. Nevertheless, the use of adjuvant therapy remains controversial for these patients because the therapeutic efficacy may be insufficient to control the residual disease. In addition, patients are at an additional risk of adverse events. Currently, there is no guideline recommendation to treat ESCC patients with adjuvant therapy after they receive neoadjuvant chemoradiotherapy and esophagectomy [2]. Due to a restricted number of clinical studies concerning this topic, the indication for adjuvant therapy after trimodal therapy is highly dependent on the patient and the institution [4]. Although there are several large-scale studies investigating the utility of adjuvant therapy after neoadjuvant therapy and surgery in western populations, the majority of the cases included in these cohorts are esophageal adenocarcinoma and the information regarding treatment regimens is missing [5–7]. Therefore, no clear evidence could guide the utilization of adjuvant therapy after trimodal therapy in patients with ESCC, especially in the east Asian region.

To add evidence to this important clinical question, we conducted a single-center and retrospective cohort study to investigate the role of adjuvant therapy following neoadjuvant chemoradiotherapy and surgery in patients with thoracic ESCC. Meanwhile, subgroup analysis based on neoadjuvant treated node stage (ypN) was performed to further explore the impact of adjuvant therapy on ESCC patients.

### **2. Materials and Methods**

### *2.1. Patients*

There were 447 ESCC patients undergoing neoadjuvant chemoradiotherapy and esophagectomy retrospectively reviewed at the West China Hospital from January 2014 to July 2020. The study was approved by the human participants' committee of the West China Hospital of Sichuan University. Surgeons informed the patients concerning the risks of the neoadjuvant/adjuvant therapy and esophagectomy. The written consent of the study's participants and permission to use resected specimens were obtained preoperatively. This study was approved by the Institutional Review Board of West China Hospital, Sichuan University in April 2021 (2022-636).

The inclusion criteria are listed as follows: (1) patients were pathologically diagnosed with ESCC before treatment, (2) patients received neoadjuvant chemoradiotherapy and esophagectomy, (3) patients were staged according to the American Joint Committee on Cancer (AJCC) 8th edition (the patients from 2014 to 2016 were staged according to AJCC 7th edition and then re-staged for the purpose of the study) [8], (4) patients were diagnosed as clinical lymph node metastasis positive (cN+) based on imaging evidence and no distant metastasis (cM0) before any treatments, (5) detailed data of the pathological information and adjuvant therapy were collected, and (6) patients were assessed as negative surgical margin pathologically after radical esophagectomy with complete tumor resection (R0 resection).

Patients were excluded if they had missing pathological information data, had unknown adjuvant treatment status, died prior to eligibility (≤60 days) for adjuvant therapy, had pathologic M1 disease, or had a documented recurrence of cancer prior to administration of adjuvant therapy. Only patients with ESCC were included. The CONSORT diagram (Figure 1) shows the inclusion and exclusion criteria of our study.


**Figure 1.** CONSORT diagram.

Patients were divided into adjuvant and non-adjuvant therapy groups for the log-rank test and Cox regression analysis. Demographic characteristics, comorbidities, operative data, postoperative complications, and pathological information were collected for all patients. Patients were followed up every 3 months for the first 2 years, and every 6 months thereafter. Neck and abdominal ultrasound, chest computerized tomography (CT), gastroscopy, and blood tests were performed on the basis of patient's symptoms during follow-up. The patient status (including death and survival), the tumor status (including tumor recurrence and metastasis), and the patient loss of follow-up were all documented. Our follow-ups were implemented via telephone or outpatient department visit. The last follow-up was conducted on 1 January 2022.

### *2.2. Neoadjuvant Therapy*

The selection of neoadjuvant therapy depended on the preoperative clinical stage of the ESCC patients. For patients with cN1-3 and/or cT4a-b, neoadjuvant chemoradiotherapy was routinely administered. The chemotherapeutic drugs were selected according to National Comprehensive Cancer Network (NCCN) Guidelines for esophageal and esophagogastric junction cancers and previous publications [2,9,10]. Neoadjuvant chemoradiotherapy included two cycles of chemotherapy with sequential or concurrent radiotherapy. The neoadjuvant chemoradiotherapy treatment cycle was 21 days (treatment during weeks 1 and 4). Paclitaxel (China Shiyao Pharmaceutical Group Co., Ltd., Shijiazhuang, China) in a dose of 175 mg/m2 (day 1) or carboplatin (Qilu Pharmaceutical Group Co., Ltd., Jinan, China) in a dose of area under the concentration–time curve 5 (day 1), with a combination of cisplatin (Qilu Pharmaceutical Group Co., Ltd., Jinan, China) in the amount of 75 mg/m2/24 h (days 1–2 or days 1–3), was given intravenously. Patients received concurrent radiation up to a total dose of 40–50.4 gray (Gy), delivered in 1.8–2.0 Gy fractions, beginning on day 1 of the first chemotherapy cycle (week 1) and ending at the completion of the second chemotherapy cycle (week 4). Sequential radiation to the same doses was arranged after the end of the second chemotherapy cycle. Intensity-modulated radiotherapy technique was used to perform radiotherapy in all patients. We referred to the Ryan scoring system to score tumor regression grades (TRGs) [11]. TRGs 0–3 are defined as follows: TRG 0: complete response (no viable cancer cells), TRG 1: near complete response (rare small groups of cancer cells), TRG 2: partial response (residual cancer with evident tumor regression), and TRG 3: poor or no response (extensive residual cancer with no evident tumor regression). Three pathologists reexamined the results of the pathological sections, and the final TRG had to be agreed upon by two or more pathologists. The strategy of neoadjuvant chemoradiotherapy is showed in Table S1.

### *2.3. Surgical Procedure and Pathology*

McKeown esophagectomy with cervical anastomoses or Ivor Lewis esophagectomy with thoracic anastomoses combined with radical lymph node dissection was performed in a standardized manner. The gastric conduit was used to reconstruct the upper digestive tract during esophagectomy. The lymph nodes were then separated by surgeons from the dissected peri-esophagus and esophagus tissues. Specimens were sent to the pathology department for further analysis where representative sections of the tumor and periesophageal tissues were taken for sufficient pathologic evaluation and staging.

### *2.4. Adjuvant Therapy*

In our institution, each patient was evaluated by a multidisciplinary team by whom adjuvant therapy selection was determined. The final decision was left up to the patients' preference. The adjuvant chemotherapeutic regimens were selected according to the NCCN Guidelines for esophageal and esophagogastric junction cancers and previous publications [2,9,12]. Generally, the chemotherapy regimens included 5-fluorouracil and cisplatin, repeated twice every 3 weeks. 5-fluorouracil in a dose of 800 mg/m2 was given by continuous infusion on days 1 through 5. Cisplatin in a dose of 80 mg/m<sup>2</sup> was administered by intravenous drip infusion for 2 h on day 1. An intensity-modulated radiotherapy technique was used to administer radiotherapy with a total dose of 45 to 50.4 Gy (1.8–2.0 Gy/d). Combined chemoradiotherapy included giving radiotherapy from the first day of the first chemotherapy cycle. Two cycle of Tislelizumab (200 mg, D1), Sintilimab (200 mg, D1), or Pembrolizumab (200 mg, D1) administered by intravenous injection combined with radiotherapy was implemented for patients undergoing adjuvant immune radiotherapy. The immunotherapy was repeated twice every 3 weeks. Typically, adjuvant therapy is administered 4 to 6 weeks after esophagectomy based on NCCN Guideline [2,12]. The strategy for adjuvant therapy is showed in Table S2.

### *2.5. Statistical Analysis*

Pearson's chi-square test was used to compare categorical variables expressed as frequencies. Kaplan-Meier curves were used to analyze overall survival (OS) and diseasefree survival (DFS), and the log-rank test was employed to determine statistical significance between the adjuvant and non-adjuvant therapy groups. A Cox regression model was used to determine variables independently associated with OS and DFS for patients undergoing neoadjuvant therapy and esophagectomy. Variables were selected for multivariate Cox regression model entry if *p* < 0.05 on univariate analysis. In addition, hazard ratios with 95% confidence intervals (CI) were reported, and we assessed whether the treatment effect differed in certain subgroups by testing the treatment-by-subgroup interaction effect with the use of Cox models via univariate analysis. All tests were two-sided and *p* < 0.05 was considered to be statistically significant. All statistical analyses were implemented with R (version 3.5.3). SPSS version 27.0 software (SPSS Inc. Chicago, IL, USA) was used to perform propensity score matching. The confounding factors including gender, age, smoke history, tumor length, neoadjuvant treated tumor, node, and metastases (ypTNM) stage, neoadjuvant treated tumor (ypT) status, ypN status, tumor differentiation, lymphovascular invasion, peripheral nerve invasion, and tumor regression grade were employed to develop the propensity score matching. The nearest-neighbor method with a caliper width of 0.02 was used to match the selected cases from two groups at a ratio of 1:1.

### **3. Results**

### *3.1. Patient Characteristics*

After application of the inclusion and exclusion criteria, 447 patients with cN+ and cM0 following neoadjuvant chemoradiotherapy and radical esophagectomy were available for analysis. Demographic characteristics, comorbidities, operative data, postoperative complications, and pathological information of the included patients are displayed in Table 1. The median tumor length was 3 cm, which was used as the cut-off value. A complete response (TRG 0) was reported in 150 (33.6%) patients, a near complete response (TRG 1) in 73 (16.3%) patients, a partial response (TRG 2) in 170 (38.0%) patients, and a poor or non-response (TRG 3) in 68 (13.4%) patients. Adjuvant therapy was performed in 141 (31.5%) patients. Of these, 49 (34.8%) received adjuvant chemotherapy, 15 (10.6%) received adjuvant radiotherapy alone, 40 (28.4%) received chemoradiotherapy, and 37 (26.2%) received immuno-radiotherapy. A total of 306 (68.5%) patients received no adjuvant therapy. Patients receiving adjuvant therapy were more likely to have a younger age, a history of smoking, an upper tumor site, a poorer tumor stage, more positive lymph nodes, advanced stage, increased lymphovascular and peripheral nerve invasion, and poorer response to neoadjuvant therapy. Due to the heterogeneity between the two groups, propensity score matching was used to balance the baseline characteristics between the adjuvant group and the non-adjuvant group. After propensity score matching, there were 120 patients remaining in each group and the patients were adjusted for all the potential confounding factors (Table 1). After propensity score matching, there were 38 (31.7%) patients receiving adjuvant chemotherapy, 13 (10.8%) receiving adjuvant radiotherapy alone, 35 (29.2%) receiving chemoradiotherapy, and 34 (28.3%) receiving immuno-radiotherapy.




Pearson's chi-squared test was used to compare categorical variables expressed as frequencies. An independentsample Student's *t*-test was used to compare continuous variables. COPD, chronic obstructive pulmonary disease; ypTNM, neoadjuvant-treated TNM; ypT, neoadjuvant-treated tumor stage; ypN, neoadjuvant-treated node stage; TRG, tumor regression grade.

### *3.2. Survival Analysis*

The median follow-up time was 13.4 months (interquartile range 6.7–24.47 months) for the overall cohort, 13.38 months (6.8–24.39 months) for those who received adjuvant therapy, and 13.43 months (6.3–25.3 months) for those who did not. After propensity score matching, patients that received adjuvant therapy had a shorter post-resection OS compared to patients not receiving adjuvant therapy (log-rank, OS: *p* = 0.046 (Figure 2a)). Meanwhile, patients receiving adjuvant therapy also had a shorter post-resection DFS compared with patients not receiving adjuvant therapy (log-rank, DFS: *p* < 0.001 (Figure 2a)).

Subgroup survival analysis was performed stratified by the ypN stage (Figure 3). Among the patients with ypN1–3, there was no significant difference in OS between the adjuvant and non-adjuvant groups (*p* = 0.500) (Figure 3a). Meanwhile, no significant difference was found in DFS for patients with ypN1–3 (*p* = 0.400) (Figure 3b). When comparing the OS between the two groups in patients with ypN0, the adjuvant therapy group had a significantly shorter OS when compared with non-adjuvant therapy group (*p* = 0.001) (Figure 3c). Meanwhile, for ypN0 patients the adjuvant therapy group also had a significantly shorter DFS compared to the non-adjuvant therapy group (*p* < 0.001) (Figure 3d).

**Figure 2.** After propensity score matching, Kaplan-Meier curves were used to analyze overall survival (OS) and disease-free survival (DFS), and the log-rank test was employed to determine statistical significance between the two groups. (**a**) Comparison of OS between patients receiving and not receiving adjuvant therapy. Patients receiving adjuvant therapy had a shorter post-resection OS compared with patients not receiving adjuvant therapy (log-rank, OS: *p* = 0.046) (**b**) Comparison of DFS between patients receiving and not receiving adjuvant therapy. Patients receiving adjuvant therapy also had a shorter post-resection DFS compared with patients not receiving adjuvant therapy (log-rank, DFS: *p* < 0.001).

**Figure 3.** After propensity score matching, Kaplan-Meier curves were used to analyze overall survival (OS) and disease-free survival (DFS), and the log-rank test was employed to determine statistical significance between groups. Subgroup survival analysis were performed stratified by the neoadjuvant treated node (ypN) stage (**a**) In the patients with ypN1–3, there was no significant difference in OS between two groups (*p* = 0.500) (**b**) In the patients with ypN1–3, there was no significant difference in DFS between two groups (*p* = 0.400) (**c**) In the patients with ypN0, the adjuvant therapy group yielded a significantly shorter OS compared with non-adjuvant therapy group (*p* = 0.001) (**d**) In the patients with ypN0, the adjuvant therapy group yielded a significantly shorter DFS compared with non-adjuvant therapy group (*p* < 0.001).

### *3.3. Cox Regression Analysis*

There were 13 variables included in the univariate Cox regression model (Table S3). Eight variables were selected for multivariate Cox regression model entry due to *p* < 0.05 on univariate analysis (Table 2). The results of Cox regression analysis on OS shows that only the ypTNM stage was an independent prognostic factor for OS in patients undergoing neoadjuvant therapy and esophagectomy. However, receiving adjuvant therapy was not an independent prognostic factor for OS (hazard ratio (HR): 1.270, 95% CI: 0.846–1.906, *p* = 0.249). The results of the Cox regression analysis on DFS show that ypTNM stage and adjuvant therapy were independent prognostic factors for DFS patients undergoing neoadjuvant therapy and esophagectomy. Meanwhile, receiving adjuvant therapy was a significantly unfavorably independent prognostic factor for DFS (HR: 2.061, 95% CI: 1.436–2.958, *p* < 0.001).

**Table 2.** Cox regression model for variables independently associated with adjuvant therapy status for patients with positive nodal disease after neoadjuvant chemoradiotherapy and radical esophagectomy. Ten variables were selected for multivariate Cox regression model entry due to *p* < 0.05 in univariate analysis.


Cox regression model was used to determine variables independently associated with OS and DFS for patients undergoing neoadjuvant therapy and esophagectomy. ypTNM, neoadjuvant treated TNM; TRG, tumor regression grade.

### *3.4. Subgroup Analysis by Forest Plot*

Figure 4 shows the hazard ratios with 95% CIs for the OS outcome in prespecified subgroups. According to the results of the overall analysis, adjuvant therapy was not a prognostic factor for OS (HR: 1.613, 95% CI: 0.999–2.604, *p* = 0.051). However, adjuvant therapy was an unfavorable prognostic factor for DFS (HR: 2.353, 95% CI: 1.535–3.607, *p* < 0.001). In subgroup analysis, for patients with ypN0, adjuvant therapy was an unfavorable factor for OS (HR: 4.274, 95% CI: 1.714–10.654, *p* = 0.002) and DFS (HR: 5.425, 95% CI: 2.490–11.820, *p* < 0.001). Nevertheless, for patients with ypN1–3, adjuvant therapy was not a prognostic factor for OS (HR: 0.818, 95% CI: 0.452–1.480, *p* = 0.506) or DFS (HR: 1.252, 95% CI: 0.734–2.137, *p* = 0.410). Table 3 contains brief information on the outcomes of prior high-quality publications and the present study.

**Figure 4.** Subgroup analysis with Cox regression model. (**a**) Hazard ratios with 95% CI for the overall survival in prespecified subgroups. For patients with ypN0, adjuvant therapy is an unfavorable factor for OS (HR: 4.274, 95% CI: 1.714–10.654, *p* = 0.002). For patients with ypN1–3, adjuvant therapy is not a prognostic factor for OS (HR: 0.818, 95% CI: 0.452–1.480, *p* = 0.506). (**b**) Hazard ratios with 95% CI for the disease-free survival in prespecified subgroups. For patients with ypN0, adjuvant therapy is an unfavorable factor for DFS (HR: 5.425, 95% CI: 2.490–11.820, *p* < 0.001). For patients with ypN1–3, adjuvant therapy is not a prognostic factor for DFS (HR: 1.252, 95% CI: 0.734–2.137, *p* = 0.410).

**Table 3.** Previous publications evaluating the therapeutic value of adjuvant therapy following neoadjuvant therapy and esophagectomy.


NCDB, National Cancer Database; EAC, esophageal adenocarcinoma; ESCC, esophageal squamous cell carcinoma; ypN, neoadjuvant treated node status.

### **4. Discussion**

In this retrospective study, we evaluated the effectiveness of adjuvant therapy on ESCC patients treated with neoadjuvant therapy and surgery. Concurrent neoadjuvant chemoradiotherapy followed by surgery has been considered as a preferred treatment strategy for patients diagnosed as ESCC in China [16,17]. However, a guideline regarding the use of adjuvant therapy after trimodal therapy in patients with ESCC is still missing. According to NCCN guidelines, the use of adjuvant therapy is recommended for all patients with esophageal adenocarcinoma after trimodal therapy, regardless of the existence of positive lymph nodes and pathologic response [2]. However, on account of different epidemiological characteristics it remains unclear if ESCC patients can benefit from adjuvant

therapy. Therefore, we conducted this retrospective study to explore the effect of adjuvant therapy after neoadjuvant chemoradiotherapy and surgery in ESCC patients. Meanwhile, subgroup analysis was performed to precisely stratify the patients undergoing neoadjuvant therapy followed by esophagectomy and to provide clinical evidence that can be utilized to guide the multimodal care of ESCC patients.

Burt et al. [7] first conducted a large-scale retrospective study based on data from the National Cancer Database (NCDB) to investigate the role of adjuvant therapy after trimodal therapy in patients diagnosed as EC. Their study indicated that EC patients with residual nodal disease after treatment with neoadjuvant chemoradiation could benefit from adjuvant chemotherapy. However, this benefit cannot be found in patients with no residual nodal disease. Whereafter, Samson et al. [6] reported a retrospective cohort study based on NCDB data only including patients with pathologic node-positive EC after neoadjuvant chemotherapy. Their study came to the same conclusion as the study reported by Burt et al. [7]. In the same year, Mokdad et al. [5] explored the effect of adjuvant chemotherapy after trimodal therapy in patients with gastroesophageal adenocarcinoma. They concluded that patients with gastroesophageal adenocarcinoma could obtain a survival benefit from adjuvant chemotherapy after trimodal therapy regardless of pathologic node status. In 2019, Drake et al. [13] investigated the effect of adjuvant chemotherapy after neoadjuvant therapy and esophagectomy. Their study only included esophageal adenocarcinoma patients with nodal metastases and concluded with the same findings as the study reported by Mokdad et al. [5]. Thereafter, Semenkovich et al. [14] conducted a multicenter retrospective cohort study including both ESCC and esophageal adenocarcinoma patients with nodal metastases, which showed that the patients with pathologic node-positive EC could benefit from adjuvant therapy after neoadjuvant therapy and surgery. Huang et al. [15] conducted a retrospective cohort study including 228 ESCC patients to investigate the effect of adjuvant therapy after neoadjuvant chemotherapy and surgery in 2019. The results of their study showed no significant difference in OS or DFS between the adjuvant therapy group and the non-adjuvant therapy group after propensity score matching. However, subgroup analysis based on status of nodal metastases were not implemented.

In our study, we only included patients with thoracic ESCC. Meanwhile, propensity score matching was used to eliminate the confounding factors, which makes the results more reliable. The results indicated that patients undergoing adjuvant therapy after trimodal therapy yielded significantly shorter OS and DFS when compared to patients not receiving adjuvant therapy. The results were consistent in patients with pathologic node-negative ESCC. However, for patients with nodal metastases after neoadjuvant chemoradiotherapy, no significant difference was seen between adjuvant therapy groups and no adjuvant therapy group. The results were opposite to the study reported by Matsuura et al. [18]. They conducted a retrospective study enrolling 113 thoracic ESCC patients with three or more pathologic positive lymph nodes. The included patients received neoadjuvant chemotherapy followed by radical surgery. The clinical efficacy of adjuvant chemotherapy was then evaluated. Their study concluded that adjuvant therapy may offer a significantly additional benefit to the prognosis of EC patients who have many positive lymph nodes even after neoadjuvant chemotherapy and surgery. The potential reason for their different results could be that their study only included patients with three or more pathologic positive lymph nodes (ypN2–3). However, the evidence was not irrefutable due to the small study population.

Theoretically, adjuvant therapy is expected to instigate a favorable effect and prolong survival for patients. However, in our institution adjuvant therapy could not prolong survival for patients with ESCC after trimodal therapy, even for patients with pathologic positive lymph nodes (ypN1–3). On the contrary, for ESCC patients with pathologic negative lymph nodes (ypN0), adjuvant therapy could be an unfavorable prognostic factor. There are several potential explanations for these results. Patients who were already treated with preoperative chemoradiotherapy could be insensitive to repeated systemic

treatment after surgery [19,20]. Meanwhile, as a systemic treatment, all types of adjuvant therapy may cause adverse systemic effects on patients [21]. Especially for ESCC patients, prolonged fasting or reduce of meal could lead to poor nutritional status, which makes them frailer after receiving adjuvant therapy [22,23]. Moreover, the unfavorable impact of adjuvant therapy on the immune system could further weaken the patient's resistance to the tumor, leading to tumor recurrence after adjuvant therapy [24]. Therefore, adjuvant therapy is not recommended for ESCC patients after trimodal therapy regardless of status of nodal metastases.

There are several limitations present in this study. The retrospective nature of this study design could reduce the reliability of the results. Therefore, propensity score matching was used in this study to eliminate the selection bias of included patients. Meanwhile, the sample size is small because of the single-center setting. More participants will be employed in our future study.

### **5. Conclusions**

The results of our study indicate that adjuvant therapy after neoadjuvant chemoradiotherapy and surgery could reduce the OS and DFS in patients with ESCC. Meanwhile, adjuvant therapy is an independently unfavorably prognostic factor for DFS. Therefore, adjuvant therapy is not recommended for ESCC patients after neoadjuvant chemoradiotherapy and esophagectomy, especially patients with node-negative after neoadjuvant therapy. A large-scale well-designed prospective study will be needed to confirm these results.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/cancers14153721/s1, Table S1: Treatment protocol of neoadjuvant therapy; Table S2: Treatment protocol of adjuvant therapy. Table S3. Univariate Cox regression model was used to determine variables associated with overall survival and disease-free survival for patients undergoing neoadjuvant therapy and esophagectomy.

**Author Contributions:** Conceptualization, X.L.; methodology, S.L.; software, X.L.; validation, X.L., S.L. and Y.Y. (Yushang Yang); formal analysis, J.Z.; investigation, X.L., P.F.; resources, Y.Y. (Yong Yuan), H.Z.; data curation, Q.S.; writing—original draft preparation, X.L., X.X.; writing—review and editing, X.L., Y.Y. (Yong Yuan); visualization, X.L.; supervision, X.L.; project administration, X.L.; funding acquisition, Y.Y. (Yong Yuan). All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Natural Science Foundation of China (81970481), the Sichuan Science and Technology Program (2022YFS0048), key projects of Sichuan Provincial Department of Science and Technology (22ZDY1959), and the 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University (2020HXFH047, ZYJC18010 and 20HXJS005).

**Institutional Review Board Statement:** This study was approved by the Institutional Review Board of West China Hospital, Sichuan University at April 2021 (2022-636).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


**Population-Based Cohort Study**
