**Image-Guided Localization Techniques for Surgical Excision of Non-Palpable Breast Lesions: An Overview of Current Literature and Our Experience with Preoperative Skin Tattoo**

**Gianluca Franceschini 1,2 , Elena Jane Mason 1, \* , Cristina Grippo 3 , Sabatino D'Archi 1 , Anna D'Angelo 4 , Lorenzo Scardina 1 , Alejandro Martin Sanchez 1 , Marco Conti 4 , Charlotte Trombadori 4 , Daniela Andreina Terribile 1,2 , Alba Di Leone 1 , Beatrice Carnassale 1 , Paolo Belli 4 , Riccardo Manfredi <sup>4</sup> and Riccardo Masetti 1,2**


**Abstract:** Breast conserving surgery has become the standard of care and is more commonly performed than mastectomy for early stage breast cancer, with recent studies showing equivalent survival and lower morbidity. Accurate preoperative lesion localization is mandatory to obtain adequate oncological and cosmetic results. Image guidance assures the precision requested for this purpose. This review provides a summary of all techniques currently available, ranging from the classic wire positioning to the newer magnetic seed localization. We describe the procedures and equipment necessary for each method, outlining the advantages and disadvantages, with a focus on the cost-effective preoperative skin tattoo technique performed at our centre. Breast surgeons and radiologists have to consider ongoing technological developments in order to assess the best localization method for each individual patient and clinical setting.

**Keywords:** breast cancer; breast-conserving surgery; non-palpable breast lesions; image-guided localization; preoperative breast localization; breast ultrasound

#### **1. Introduction**

Breast cancer (BC) is the most commonly diagnosed cancer and the leading cause of cancer-related death among women [1]. A successful BC treatment is based on a multidisciplinary use of surgery, chemotherapy and radiation therapy, with surgery as the central component of treatment for early-stage breast cancer [2,3]. Breast-conserving surgery (BCS) followed by adjuvant radiotherapy, known as breast conservation therapy (BCT), has become the alternative treatment to mastectomy for early stage breast cancer because of equivalent survival and lower morbidity [4–6].

Local recurrence after BCS is strongly correlated to the surgical margin status, as demonstrated by a large number of follow-up studies [7–11]. The main goal of BCS is to fully remove the tumor with clear margins, while avoiding resection of healthy breast tissue in order to achieve better cosmetic results. Image-guided preoperative localization is

**Citation:** Franceschini, G.; Mason, E.J.; Grippo, C.; D'Archi, S.; D'Angelo, A.; Scardina, L.; Sanchez, A.M.; Conti, M.; Trombadori, C.; Terribile, D.A.; et al. Image-Guided Localization Techniques for Surgical Excision of Non-Palpable Breast Lesions: An Overview of Current Literature and Our Experience with Preoperative Skin Tattoo. *J. Pers. Med.* **2021**, *11*, 99. https://doi.org/10.3390/jpm11020099

Academic Editors: Hisham Fansa and Pim A. de Jong Received: 21 December 2020 Accepted: 1 February 2021 Published: 4 February 2021

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

**Copyright:** © 2021 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/).

mandatory for guiding surgery of non-palpable lesions or surgically relevant extension of palpable lesions to improve both oncological and cosmetic outcomes [12,13]. Over the last decade, methods for preoperative localization of breast lesions for BCS have evolved rapidly due to innovative techniques and discovery of novel agents. However, cooperation and communication between breast surgeons and radiologists still play a crucial role.

Different image guided localization techniques are variably used in different institutions depending on personal choices, skills and available technologies. As a general rule, the method chosen should be the most precise to localize the lesion or marker left after biopsy, thus improving free margin rates and decreasing operative time, and possibly cause little to no discomfort to the patient. Preoperative breast lesions localization techniques currently available are wire localization, carbon marking, radio-guided occult lesion localization (ROLL), radioactive seed localization (RSL), magnetic seed localization and non-radioactive radar localization, intraoperative ultrasound and preoperative skin tattoo localization (Table 1). In this article, we provide an overview of current literature of all commercially available techniques. The aim of this review is to educate practicing radiologists and breast surgeons so they can knowingly select new techniques to improve patient care.

**Table 1.** Comparison of different localization techniques. Abbreviations: ROLL = radio-guided occult lesion localization; RSL = radioactive seed localization; Magseed = magnetic seed localization; IOUS = intraoperative ultrasound; Skin tattoo = preoperative localization with skin tattoo; OR = operating room; US = ultrasound; MRI = magnetic resonance imaging. \* Success is defined as removal of target lesion. \*\* Authors' experience.



**Table 1.** *Cont.*

#### **2. Wire Guided Localization**

Wire localization (WL) was introduced in the 1970s and for many years has served as the only method for preoperative breast localization [36]. Initially, mammography was the only imaging modality used to guide wire placement. Currently, wire localization can be performed under different kinds of image-guidance (mammographic, sonographic and magnetic resonance imaging). WL is the most commonly used method for non-palpable breast lesions, with clear margins reported in a range of 70.8%–87.4% of cases [15]. Different types of wires are available, ranging in length (from 3 to 15 cm), shape (hook, barb or pigtail), materials and numbers of thickened segments [12,13,15,36,37]. Wires are preloaded in a 16–21 G needle introducer: when the tip is just beyond the target, the hook is deployed by fixing the needle firmly with one hand and gently advancing the wire with the other. The needle is then removed over the wire and the thread extending from the tip of the hookwire is secured on the skin surface. Routinely, post-procedural CC and ML mammograms were obtained to confirm accurate placement (Figure 1). The depth of the wire tip from the skin surface is also recorded. In case of extensive disease wires can be placed in multiple numbers, allowing targeted localization in a procedure known as "bracketing wire localization" [38]. WL remains the most widely adopted approach due to the long-term data supporting its effectiveness [39], although success is strongly dependent upon the surgeon's mental reconstruction of the images, perceived intraoperative position of the lesion and wire trajectory [40]. Approximately 2.5% of wire localizations are unsuccessful; factors associated with an increased risk of unsuccessful localization are multiple lesions, small lesions, lesions containing extensive microcalcifications and small surgical specimens [14]. Established advantages of WL are the widespread availability and the moderate price, with one study estimating the cost of a needle at \$22.50 [41]. – – "bracketing wire localization" '

–

–

**Figure 1.** Wire-guided localization. Craniocaudal (**a**) and mediolateral (**b**) oblique mammograms taken after hookwires insertion show optimal wires positioning, with the wires at the biopsy markers site. A specimen radiograph (**c**) contains the hookwire and the residual calcifications (circle).

Moreover, wires emit no ionizing radiation and can be stored safely within the imaging department. This approach also allows localizations of breast lesions under different kinds of image guidance (US, mammography/tomosynthesis or MRI). Although WL is highly

effective, it still yields several disadvantages. The procedure is in itself unpleasant and causes patient discomfort; vasovagal reactions are reported in up to 7–10% of patients, although less frequent for US than for mammography guided procedures [12]. Wire migration within the breast, and more infrequently outside the breast, has also been reported [42,43]. The hookwire can be transected during the surgery, with pieces being retained in the breast post-operatively [44,45]. Finally, this localization approach requires adequate coordination between trained breast radiologists and surgeons because the wire placement has to occur on the day of surgery to avoid displacement. This limitation can lead to inconvenience and delay in the operating room or suboptimal localization. Moreover, wire localization could limit the surgical approach and cause a potential worse cosmetic outcome; the placement route of the wire, chosen by the radiologist, often dictates incision choice for the surgeon who then has to follow the wire's course during dissection.

#### **3. Carbon Marking**

Carbon marking (CM) is an alternative method for non-palpable breast lesion localization first reported by Svane in 1983, consisting of an injection of sterile charcoal powder diluted with saline solution in close proximity to the lesion [46]. The injection can be performed under either sonographic or mammographic guidance, depending on how the target lesion has been biopsied [17]. A dark trail is created from the lesion to the skin, leaving a visible track that guides the surgeon during the operation. As the carbon track is immobile in breast tissue, it cannot dislodge. In contrast, hookwires can migrate when the patient changes position or when traction is applied during surgery. The main advantages of CM are logistics, patient comfort and cost. As CM and biopsies could be concurrent, the patient may be spared an extra invasive procedure. Moreover, surgery may be planned up to 1 month after the carbon injection, making operative planning easier for surgeons and sparing radiologists the pressure to place hookwires immediately before or during an operating session. The success rate using carbon marking is very high, with failure to remove targeted lesion occurring in about 1 in every 100 procedures [16]. However, there are cases in which CM presents technical difficulties. If the lesion is close to the chest wall, particularly in a large breast, or for extensive or multifocal lesions, long and several carbon tracks will be difficult for the surgeon to follow and a hookwire may be preferable. For extensive or multifocal lesions several carbon tracks are difficult to follow, and WL may be preferable [46]. The disadvantages are that the carbon tracks resist slicing, thus the carbon can distort or obscure the lesion. To avoid this, the carbon should be injected only as far as the edge of the lesion. Another possible, although uncommon, complication of CM is the incomplete surgical removal of the injected charcoal, which can cause a late-onset granuloma that may mimic malignant lesions in postoperative controls [47,48]. In terms of missed lesions and clear margin rates, CL shows similar results as WG: the proportion of cases with close or involved margins ranges between 15% (for invasive cancer) and 39% (in situonly lesions) [17,18].

#### **4. Radio-Guided Occult Lesion Localization**

Radio-guided occult lesion localization (ROLL) involves intratumoral injection of a small amount (0.2–0.3 mL) of human serum albumin marked with nuclear radiotracer technetium 99 [49] (Figure 2). This localization technique can be performed either by ultrasonography, stereotactic mammography or MRI.

– **Figure 2.** Radio-guided occult lesion localization (ROLL) technique: (**a**) invasive ductal cancer (arrow) in the left upper outer quadrant in a 77-year-old woman. (**b**) Intratumoral injection (arrow head) of a small amount (0.2–0.3 mL) of human serum albumin marked with nuclear radiotracer technetium 99 in order to perform radio-guided occult lesion localization.

> – – – The radiation dose is of about 7–10 MBq, equivalent to 1–2% of the dose used for a whole-body bone scintigraphy [50]. A handheld gamma ray detection probe is used by the surgeon to locate the lesion, guide the removal and verify the removed specimen and the surgical bed. To allow an adequate detection, surgery has to be performed no later than 24 h after the injection of the radiotracer. ROLL has gained popularity on account of several advantages associated with a reduced excision volume, more accurate centricity of a lesion within the surgical specimen, better cosmetic results and a higher percentage of tumor-free margins, around 92% of cases [20,21]. There are no serious complications related to ROLL, even though experience in the injection is needed to avoid failure of lesion identification, described only in 1–5% of the cases [19]. ROLL can be performed together with sentinel lymph node identification in the same surgical session, in a procedure known as sentinel node and occult lesion localization (SNOLL), that involves the injection of an additional radiotracer (carried by micromolecules instead of macromolecules used for ROLL) [51,52].

#### **5. Radioactive Seed Localization**

– – – Radioactive seed localization (RSL) using Iodine-125 seeds has been proposed in 1999 by Dauway as an attractive alternative to both WL and ROLL. This technique involves targeted placement of a seed, commonly used for brachytherapy, composed of titanium labeled 0.075–0.3 mCi of Iodine-125. Each seed has a half-life (T 1/2) of 59 days and a radioactivity of about 20–30 MBq, a dose equivalent to 3–5% of that used for a whole-body bone scintigraphy [53]. Radioactive seeds can be positioned under different image guidance, ultrasonography, mammography/tomosynthesis or MRI. An 18G needle preloaded or manually loaded with the seed was used, and the tip was occluded by bone wax. Once the needle advanced to the desired location, the seed was deployed through the bone wax by advancing the stilette. At the end of the procedure, regardless of the guidance method, the patient was assessed for radioactivity with a Geiger counter and post-procedural mammograms with two orthogonal images reconfirm proper seed positioning [54]. During surgery a gamma probe set for I-125 guides the surgeon. The different energy peak of technentium-99 and iodine-125 allows one to differentiate the isotope used for sentinel node biopsy. Radioactive seed localization could potentially be performed weeks before the scheduled surgery because of the long half-life (59 days) of I 125; however, according to Nuclear Regulatory Commission guidelines the procedure should be carried out no more than 7 days before surgery in order to minimize radiation exposure [55]. In fact, one of the potential drawbacks of RSL is the presence of radioactivity. Although the activity levels of the seeds are low and considered safe for human exposure, patients are advised to avoid interactions with children and pregnant women to mitigate any potential risk. Moreover, a strict local protocol for quality assurance must be followed in order to guarantee that all implanted seeds are actually removed and recovered by the local Nuclear Medicine

Department. An undeniable benefit of both ROLL and RSL is that the surgeon is no longer impeded by the guidewire when planning breast incision and can use the feedback from the gamma probe to reorientate the surgical approach in real time.

Given oncoplastic breast techniques, this allows greater choice of cosmetically sensitive approaches, such as periareolar, lateral or inframammary fold incisions [56]. Current literature comparing RSL and WL margin status achievements shows variable results, with some studies favoring RSL and more recent studies, including three randomized control trials, suggesting no difference between the two methods [40,57,58]. Due to the real-time intraoperative monitoring of the detected gamma counts from the seed, RSL allows an accurate lesion localization with lower incidence of positive margins and decreased need for repeat surgery than with wire localization. The success rate using RSL is very high: target lesion is effectively removed in nearly 100% of cases and clear margin rates range from 73.5% to 96.7% [22,23].

#### **6. Magnetic Seed Localization**

Magnetic seed is a novel localization technique approved by the FDA in 2016 [24]. This technique shares many similarities with RSL, because it consists in seed placement under sonographic or tomosynthesis guidance, however it does not involve radioactivity. First introduced by Sentimag (Magseed®, London, UK), magnetic seeds are cylindrical markers, measuring approximately 5 mm × 1 mm, made of paramagnetic steel and iron oxide. They can be deployed by an 18 G preloaded needle of different length according to different breast sizes (Figure 3). Following insertion, mammograms in double projection are acquired to confirm correct positioning of the seed. The Sentimag probe employed in the operating room generates an alternating magnetic field that temporarily magnetizes the Magseed, and subsequently measures its magnetic field. The surgical technique is therefore similar to that adopted after ROLL or RSL, involving a live numerical feedback that guides surgical direction and reveals the remaining distance from lesion. A final assessment is conducted by probing the specimen and the surgical cavity, and potentially verified with specimen X-ray confirming excision of seed. While sharing with ROLL and RSL the important benefit of granting maximum liberty in the choice of incision, this technique has the further benefit of avoiding exposure to radiation. It also eases coordination between Radiology and Surgery Departments, because seed placement, initially approved for up to 30 days prior to surgery, has now been extended in Europe and USA for long-term implantation [25]. However, while this seed could be potentially implantable during biopsies and even before neoadjuvant treatment, one major drawback is that it interferes with MR imaging by creating artifacts as wide as 4 cm [12]. Another challenge with this technique is that during surgery all ferromagnetic instruments will interfere with the signal. A dedicated set of non-ferromagnetic surgical instruments is therefore always necessary, and weighs on cost-effectiveness [13]. Studies on the efficacy of this technique in terms of successful excision, clear margins and optimal volume of resection are few and include relatively small populations of patients, however preliminary data is encouraging, with a successful placement rate of 94.42%, a successful localization rate of 99.86% and a percentage of clear margins of 88.75% [24,26,27,59].

**Figure 3.** Magseed positioning in a 49-year-old woman with ductal carcinoma in situ. Ultrasound images of the right upper outer quadrant. Biopsy marker is visible in the lesion (**a**, arrow). Magseed magnetic marker is placed under ultrasound guidance (**b**, arrow shows the needle). Magseed ® marker is clearly seen in the lesion (**c**, arrow).

#### **7. Radiofrequency Identification Tags**

– Radio frequency reflector (RFR) is a non-ionizing electromagnetic wave tagging device for localizing non-palpable breast lesions approved in the United States by FDA in 2014 [28]. The identification tag, as any biopsy clip marker, can be placed by radiologists under mammographic, tomosynthesis or ultrasound guidance. The injection can take place up to 30 days preoperatively. During surgery, the surgeon activates the reflector with the hand piece and follows the signal to guide the excision. The audible and numerical signals change with increasing proximity to the lesion. Once the tissue is removed, the reader console can be used to confirm that all tags have been removed from the tissue cavity. RFRs differ in size and shape from vendor to vendor. One of the first available RFR is SAVI SCOUT (Cianna Medical, Aliso Viejo, CA, USA) and another more recent device is the LOCalizer (Faxitron, Tucson, AZ, USA). The SAVI SCOUT reflector has been rated as MR conditional and be considered safe to image in a static magnetic field of 3 Tesla or less and a maximum spatial gradient magnetic field of 3000 G or less [29]. Whereas metallic interference from nearby surgical instruments can interfere with detection of magnetic seeds, metal does not interfere with detection of radiofrequency signals during surgery [59]. Radiofrequency identification tag is an effective technique: data from the literature report success rates of 97–100% and clear margin rates ranging between 85% and 100% [30,31]. The main advantage of RFR localization over wire localization is the decoupling of the radiology and surgery schedules; moreover, it avoids the risk of complications associated with an external wire component. Compared to RSL, RFR is a non-ionizing system and does not require extensive multidisciplinary coordination or regulatory compliance. Disadvantages of localization with the SAVI SCOUT device include its relatively large size (12 mm), especially for small subcentimetric lesions. The LOCalizer overcomes the size hurdle

since it is smaller. Other limitations include the inability to reposition the reflector once deployed and the maximum lesion detection depth, as studies have reported problems in intraoperative detection of the reflector in women with large breasts and lesions located >6 cm from the overlying skin surface [30].

#### **8. Intraoperative Ultrasound**

Intraoperative ultrasound (IOUS) was first described by Schwartz et al. in 1988 and has gradually spread and evolved with other techniques due to growing experience and technological advances. A sterile-gowned ultrasound probe has to be available in the operating room. The procedure begins at the operating table before incision, once painting and draping procedures have been carried out. The surgeon locates the tumor by ultrasonography and measures its diameter and distance from surrounding hallmarks, such as skin surface, nipple–areolar complex (NAC) and fascia. The surgical approach is then planned in full liberty, and after incision the dissection is carried out by repeatedly reassessing the tumor's position and the distance between the surgical plane and its margins. Once the excision has been completed, specimen ultrasound is performed at the operating table to assess margins, and additional shaving excisions can be acquired if necessary. This technique is highly effective, with identification rates close to 100% [31–35], and studies focusing on margin status have shown that IOUS guided surgeries yield less positive resection margins compared to WGL [32], with free-margin percentages ranging from 81% to 97% [33–35]. Free margin rates are enhanced by IOUS even in resection of palpable lesions [60]. A study by James et al. has instead shown no significant differences in margin status between IOUS and mammographic WGL in patients undergoing surgery for carcinoma in situ, although the authors still recommend performing IOUS as it is more cost-effective [61]. Compared to other techniques, IOUS yields several practical advantages: it does not increase patient presurgery psychological stress, as it is non-invasive compared to techniques involving breast compression or puncture; it grants full liberty to the surgeon in choosing the most convenient oncoplastic surgical approach; it does not aggravate organizational problems and coordination between several departments, as it takes place directly in the operating room and can be carried out completely by the surgeon himself [62]. To this regard, the learning curve of specialists not necessarily familiar with manipulating ultrasounds, such as surgeons, could potentially pose an issue, however a study by Krekel et al. suggests that performance of eight procedures is enough for the surgeon to acquire the expertise necessary to combine ultrasounds to palpation-guided surgery [63]. Drawbacks include technical problems resulting from combining ultrasound with surgery, such as air infiltration beneath the probe that can impede visualization, and refraction issues that can arise when scanning tissue that is irregular in shape [32]. The major, insurmountable issue of this technique is however represented by its inability to localize sonographically invisible tumors. To overcome this problem, some authors have described this technique in combination with hematoma-guided surgery after MRI- or stereotactically-guided biopsies, with mixed results [64].

#### **9. Preoperative Localization with a Skin Tattoo**

Preoperative localization with a skin tattoo is a simple and safe technique amply utilized in our centre, as it is easily performed, extremely well tolerated by patients and effective in terms of successful excision and clear margin rates. This method can be carried out by acquiring either sonographic or mammographic images, depending on the type of lesion, but ultrasounds are employed whenever possible because the procedure is easier. In this case, patients lie in the supine position with their arms extended to mimic the position held during surgery. The tumor is located, and its distance from the skin surface is measured taking care not to apply pressure with the probe, so as to report accurately the depth of the tumor in relation to the skin surface [65]. The distance between the lesion, the nipple and the pectoralis major muscle is also measured, as is the distance between separate lesions in case of multifocal or multicentric disease [44,66]. Radiologists with experience

in this technique visualize the tumors at their largest diameter to achieve the optimal correspondence between the lesion and the skin markers. The tumor's projection on the skin surface is pinpointed with a dermographic skin marker and the drawing is covered to avoid accidental erasure (Figure 4). The whole procedure, performed by an experienced radiologist, takes 5–10 min and provides minimum patient discomfort. Limitations include poor results in case of sonographically invisible lesions, microcalcifications or biopsy markers, but are easily overcome by implementing this technique with a mammographic approach. Stereotactic-guided skin marking is also a non-invasive technique, albeit it provides a little more discomfort to the patient due to breast compression. Mammograms are acquired in double projection and measurements are performed on the images to determine the distance between the lesion and the nipple, the skin surface and the fascia. correspondence between the lesion and the skin markers. The tumor's projection –

). The dermographic skin markers of the tumor's projection on the skin surface ( **Figure 4.** Preoperative skin tattoo. Transverse sonogram showing hypoechoic, round shaped multifocal masses with indistinct margins in the upper outer quadrant of the right breast (**a**,**b**, arrows). The distance between separate lesions is measured (**c**). The dermographic skin markers of the tumor's projection on the skin surface (**d**).

> The radiologist then estimates the projection of the tumor on the skin surface and positions a lead marker in the corresponding spot. In case of bigger lesions, such as extensive microcalcifications, or multifocal disease, multiple lead markers can be employed to determine lesion margins. A second stereotactic pair of images is acquired to confirm the correct localization, and in case of inaccurate positioning, the lead markers can be repositioned more accurately and confirmed by a further mammogram [67] (Figure 5). At the end of the procedure the lead markers are removed, and the skin tattoo is drawn in their place. In the operating room, the mark is exposed and retraced with a specific marker resistant to antiseptic solutions, and painting and draping procedures are carried out carefully without wiping out the ink. Our centre strongly advocates pursuit of the maximum aesthetic result achievable with oncological safety, and because this localization technique employs only a temporary skin tattoo, the surgeon is granted total liberty in choice of incision and oncoplastic technique. The skin flap is dissected in the direction of the tattoo, then the incision is deepened and a lumpectomy is carried out taking into account tumor depth measured during the preoperative localization. In some cases, a non-palpable lesion becomes palpable after dissection of the skin flap, allowing the surgeon to easily complete the excision, however in most cases the excision has to be conducted by reassessing the original position of the skin mark from time to time. Once the excision is completed, metallic clips are placed on the orienting sutures in different numbers, so as to recognize margins in the specimen X-ray. The sample is then placed into a transparent plastic bag and sent to the Radiology Department, and mammograms are acquired in double projection. The tumor is usually visible as a radiopaque nodule, and its position inside the lumpectomy specimen is described as either well centered or close to one or more surgical margins, and reported to the operating surgeon. In dense, glandular specimens

> > 215

the nodule can be difficult to distinguish from the surrounding mammary parenchyma: in these cases, the exam can be completed with a specimen ultrasound [68] (Figure 6). If close margins are detected in either technique the surgeon can acquire further cavity shave margins on the affected border.

) views confirm the appropriate marker (arrow) placement on the microcalcifications' (circle) projection on **Figure 5.** Lead marker positioning during mammographic technique. Metallic marker (**a**). Craniocaudal (**b**) and mediolateral oblique (**c**) views confirm the appropriate marker (arrow) placement on the microcalcifications' (circle) projection on the skin surface. Specimen X-ray contains the microcalcifications (**d**). ) views confirm the appropriate marker (arrow) placement on the microcalcifications' (circle) projection on

**Figure 6.** Radiograph of a dense, glandular specimen with scarcely recognizable nodules (**a**). Subsequent specimen ultrasound demonstrates successful removal of two masses (arrows) (**b**,**c**).

This technique is quick, easily performed by breast radiologists and extremely costeffective. It does not require equipment that is not normally present in any breast surgery department, and is therefore feasible even with scarce resources. Limitations include accurate scheduling to time the procedure before surgery thus avoiding accidental mark erasure, and a certain degree of experience by the surgeon in reassessing the tumor's

position based on the skin mark during dissection. Reports on this technique are widely deficient in the literature, however a preliminary analysis of the data from our highvolume centre examining the outcome of 199 lumpectomies performed for non-palpable breast tumors between August and December 2019 identified a global success rate of 99.5% (198/199) and a clear margins rate of 95.9% (192/199). As these rates did not differ significantly from other localization techniques, this method appears safe and especially ideal in the case of limited resources or spending reviews.

#### **10. Conclusions**

Image-guided preoperative localization of breast lesions is a common procedure that has rapidly evolved throughout the last decades. Continuous technological developments and results from new clinical trials have provided growing insight and new possibilities for breast specialists to select upon various effective techniques. However, to date, no single perfect method exists. Therefore, the optimal approach should be tailored on each patient by taking into account preoperative disease characterization (both radiologic and histologic) and consulting all stakeholders, including surgeons, radiologists and pathologists.

**Author Contributions:** Conceptualization: C.G., E.J.M., S.D., A.D.; data curation, E.J.M., S.D.; writing—original draft preparation, C.G., E.J.M.; writing—review and editing, C.G., E.J.M., G.F., S.D., A.D., D.A.T., P.B.; visualization, A.D., P.B., M.C.; supervision, A.M.S., A.D.L., R.M. (Riccardo Masetti); project administration, L.S., B.C., R.M. (Riccardo Manfredi), C.T. All authors have read and agreed to the published version of the manuscript.

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

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

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

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

#### **References**


### *Article* **GENERATOR Breast DataMart—The Novel Breast Cancer Data Discovery System for Research and Monitoring: Preliminary Results and Future Perspectives**

**Fabio Marazzi 1 , Luca Tagliaferri 1 , Valeria Masiello 1, \* , Francesca Moschella 2 , Giuseppe Ferdinando Colloca 1 , Barbara Corvari 1 , Alejandro Martin Sanchez 2 , Nikola Dino Capocchiano 3 , Roberta Pastorino 4 , Chiara Iacomini 4 , Jacopo Lenkowicz 3 , Carlotta Masciocchi 4 , Stefano Patarnello 4 , Gianluca Franceschini 2,5 , Maria Antonietta Gambacorta 1,3 , Riccardo Masetti 2,5 and Vincenzo Valentini 1,3**


**Abstract:** Background: Artificial Intelligence (AI) is increasingly used for process management in daily life. In the medical field AI is becoming part of computerized systems to manage information and encourage the generation of evidence. Here we present the development of the application of AI to IT systems present in the hospital, for the creation of a DataMart for the management of clinical and research processes in the field of breast cancer. Materials and methods: A multidisciplinary team of radiation oncologists, epidemiologists, medical oncologists, breast surgeons, data scientists, and data management experts worked together to identify relevant data and sources located inside the hospital system. Combinations of open-source data science packages and industry solutions were used to design the target framework. To validate the DataMart directly on real-life cases, the working team defined tumoral pathology and clinical purposes of proof of concepts (PoCs). Results: Data were classified into "Not organized, not 'ontologized' data", "Organized, not 'ontologized' data", and "Organized and 'ontologized' data". Archives of real-world data (RWD) identified were platform based on ontology, hospital data warehouse, PDF documents, and electronic reports. Data extraction was performed by direct connection with structured data or text-mining technology. Two PoCs were performed, by which waiting time interval for radiotherapy and performance index of breast unit were tested and resulted available. Conclusions: GENERATOR Breast DataMart was created for supporting breast cancer pathways of care. An AI-based process automatically extracts data from different sources and uses them for generating trend studies and clinical evidence. Further studies and more proof of concepts are needed to exploit all the potentials of this system.

**Keywords:** breast cancer; DataMart; real world data; predictive model; healthcare

**Citation:** Marazzi, F.; Tagliaferri, L.; Masiello, V.; Moschella, F.; Colloca, G.F.; Corvari, B.; Sanchez, A.M.; Capocchiano, N.D.; Pastorino, R.; Iacomini, C.; et al. GENERATOR Breast DataMart—The Novel Breast Cancer Data Discovery System for Research and Monitoring: Preliminary Results and Future Perspectives. *J. Pers. Med.* **2021**, *11*, 65. https://doi.org/10.3390/ jpm11020065

5

Academic Editor: Enrico Capobianco Received: 30 December 2020 Accepted: 20 January 2021 Published: 22 January 2021

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

**Copyright:** © 2021 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. Background**

In the last few years, breast cancer (BC) curability has been highly improved thanks to implementation of treatments and technologies [1]. In oncology, clinical research is usually supported through prospective clinical trials in which collected records are usually codified by an ontological system [2] and electronic case report forms (CRFs) [3]. However, prospective clinical trials require a considerable number of resources and time to obtain statistically meaningful outcomes. In addition, the results obtained after 5–10 years from a clinical trial often cannot reflect up-to-date technical needs and can be overtaken by new therapeutic choices [4]. Alongside the high-quality data from clinical trials, lowquality but high-quantity data generated by clinical practice are often not used because they are difficult to collect and analyze [5]. Real-world data (RWD) studies represent a possibility for obtaining evidence from clinical practice, because they are considered to be more representative of the patients and trends that are currently being treated. RWD are stored and potentially available inside hospital informatic systems, both in structured and unstructured formats, and carry truly relevant information applicable for different scopes (research, monitoring, alert, etc.).

The use of automated data discovery and Artificial Intelligence (AI) in medical research has also increased exponentially in recent years, and the continuous development of improved computer science and machine learning tools helps raise the efficiency of research by automating various processes that are usually either performed manually or in a suboptimal way [6]. Additionally, even more thriving applications of data discovery and AI in oncological sciences have led to the development of systems capable of substantially improving diagnostic and therapeutic choices [6–8]. Thanks to the automated process of AI, RWD extraction and analysis could become even more feasible without manual work, but, even more, the system could learn from RWD to predict trends and improve processes [9].

The primary aim of this work is to show an integrated, highly replicable approach where the use of modern technologies (e.g., data discovery, transformation, and AI-based technologies) is leveraged in order to extract, validate, and organize RWD data. This approach is applied to the domain of breast cancer and will allow doctors to organize information for patients' treatment history in a time-effective manner, by centralizing such data from different archives distributed in the hospital healthcare systems into a single standardized repository for breast cancer real-world data (called Breast DataMart). This procedure will, in turn, enable focused studies to be much more effective in their aims. In this work, we also highlight how this DataMart can be exploited through machine learning, to obtain models for outcome prediction and development of guardian systems set up to monitor the clinical flow of patients (pts) and provide supporting info for corrective actions, e.g., in time-sensitive treatment schedules. We describe the architectural structure of the Breast DataMart, and two proof-of-concept designs intended to show the potential of the guardian systems and the automated data-extraction procedures.

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

#### *2.1. Domain-Specific Ontology*

At the very start of the project, we focused on the definition of a terminology dictionary aligned to the requirements of the clinician team, in terms of completeness of patients' data, accuracy in the description of clinical workflow, and relationships among entities. This phase of the ontology definition was developed jointly among clinical experts and data scientists, to make sure that the mapping into the target IT framework was accurate and viable.

#### *2.2. Multidisciplinary Team and Rapid Requirement Definition*

The goal of building a framework that can be extensively leveraged across multiple studies and trials has naturally led to organize a team which could offer a comprehensive view of the needs from a clinical research side, tightly connected with the technology experts for the technical design and architectural builds. A multidisciplinary team of experts

was formed by radiation oncologists, epidemiologists, medical oncologists, and breast surgeons, to approach breast cancer RWD from a clinical perspective. To develop a system able to collect, transform, and organize data from different archives within the hospital IT system, data scientists and data management experts worked to capture requirements from clinical teams and translate them into fast prototypes and implementation. Combinations of open-source data science packages and industry solutions (SAS® Viya framework) were used to design the target framework. To validate the DataMart directly on real-life cases, the working team defined tumoral pathology and clinical purposes of proof of concepts (PoCs). The PoCs were immensely helpful for a user-oriented approach to select, classify, and organize data in the DataMart. From the project-management perspective, the working group adopted a rapid development strategy, where the clinical team (i.e., end users) and technology staff were highly integrated in designing, prototyping, and validating intermediate and final outcomes.

#### *2.3. Breast Cancer DataMart Architecture*

The working team chose breast cancer as the initial pathology to be investigated for the DataMart creation process. Besides the expertise of the working team, breast cancer was properly chosen for its high range of possible variables and different archives with relevant information in the hospital IT infrastructure. The approach to extract, transform, and organize information in the target DataMart is based on a multilayer approach: The first layer is based on the hospital IT platform, in which the retrospective data are centralized and structured in accordance with the ontology defined, and prospective data items are collected daily from physicians, analysis laboratories, and electromedical devices, to then be stored and protected with the strictest physical and logical security criteria. The working team also classified sources or "Channel Doors" from which to import data. The second layer is the DataMart structured dataset; the interchange between the first and second layer is handled with a set of IT tools: automated procedures to feed a real-time flow of data stemming from the daily clinical practice; connectors to electromedical devices (e.g., to extract radiomic data); text-mining techniques, which transform unstructured text (e.g., consultancies, exam reports, and diagnoses) into clinically relevant structured data. Raw data from production repositories are extracted in pseudo-anonymized form, to protect patients' privacy. The third layer is the discovery one, where analytics, machine learning, and AI methods are applied to perform the studies (in our case, the PoC). The output of this semi-automated AI layer can be represented in formats which are relevant for the clinical staff through the development of easy-to-use graphical user interfaces or different forms based on the specific study (production of synthetic data, out-come-related risk scoring, etc.).

With this approach, the Breast Cancer DataMart was defined and structured as the shared global archive of all available breast cancer data inside the hospital IT system of Fondazione Policlinico Gemelli IRCCS, which will be continuously updated through the scheduled procedures (Figure 1).

**Figure 1.** GENERATOR Breast DataMart architecture. In this figure, architecture of GENERATOR Breast DataMart is described. On the left, the sources are reported (a description is provided in Table 1). Thanks to Artificial Intelligence (AI) automatism, connection, and procedures, it is feasible to extract these data sources and deposit them inside Breast DataMart. An external server support Breast DataMart. Data extracted are available for further elaboration, such as creation of robots (or BOTs) for implementation of clinical research.


**Table 1.** Data definition and classification, according to their availability.

Finally, to program the DataMart implementation, the working group (WG) divided future processes into 3 distinct phases:

Phase 1: proof of concept;

Phase 2: internal consent with multidisciplinary contribution;

Phase 3: dynamic DataMart with data access for monitored and authorized internal and external requests.

To verify the usability and effectiveness of the DataMart and the overall framework, the team proposed two proofs of concept (PoCs) for testing purposes. The first one was identified as a "waiting time" calculation test from surgery to radiotherapy beginning. The second PoC was set up to calculate and test a series of key performance indicators (KPIs) based on diagnostic and therapeutic performance markers. Each end-product of the two PoCs was defined as a robot (BOT) in accordance with the AI-related features, in terms of AI data governance, automated procedures, and end-user output.

For each PoC, the definition of specific methodology and development pathways were required: DataMart access and usage (including variables selection and definition, archives and channel doors identifications, and data extraction processes), modeling phase (BOT construction), and end-user testing (BOT clinical validation).

#### **3. Results**

Starting from October 2019, the working team organized meetings on a regular basis, in order to keep the workflow initially defined. Meetings were both live and online. All the tasks planned for phase one of the DataMart development were completed.

Data definition: The WG defined data on the basic of their availability. Results are resumed in Table 1.

Archives and channel doors definition: Based on data definition, Multidisciplinary Team then defined where to find data for filling Breast Cancer DataMart and for capture them different "channel door" were identified. "Channel door" identified are reported in Table 2.

Waiting Time Bot: As already mentioned, this PoC addresses the waiting-time calculation from surgery to the start of radiotherapy. We believe this is a relevant testbed for two purposes: as a process BOT, to identify areas to improve and accelerate patients' clinical paths; and as a supporting platform for interventional studies, to track the evolution of the selected cohorts. The team identified pathways for data extraction and elaboration.

ICD9 codes for diagnosis and surgery were selected for identifying pts with breast cancer who underwent surgery. We evaluated 10 main variables to extract by the textmining technology the time lapses in which the RT was performed: Seven were structured variable, such as the date of birth or the kind of surgery, and three were unstructured variables (multidisciplinary board indication to chemotherapy, radiotherapy, or both). Text mining selected patients with multidisciplinary indication to adjuvant chemotherapy and radiotherapy. Waiting-time interval was calculated as the interval between surgery and the first day of treatment.


#### **Table 2.** Archives and channels doors definitions.

From January 2017 till December 2019, a cohort of 2074 patients underwent surgery for breast cancer. Between them, 655 pts were addressed to adjuvant RT alone, 113 to adjuvant chemotherapy alone, and 153 to both. Of this cohort, 1023 underwent RT in our hospital. Mean waiting time was 119 days (31–345). They were divided into three groups, based on waiting-time interval: 154 patients underwent RT within 60 days from the surgery; 407 patients, starting from 60 days after the index breast surgery and up to 90 days; and 462 patients who were treated after 90 days from surgery. Patients who came from other regions, and so, far from our center, experienced a wider delay in the beginning of RT.

The Wating Time BOT showed that it is feasible to extract data from different data sources inside the hospital system, to obtain an output for monitoring real-time pts' waiting time for radiotherapy treatments (Figure 2). Output of this evaluation needs to be implemented and integrated inside the hospital system, to have an alert for managing patients' waiting-time delay. Specific further prospective studies are needed to highlight predictive factors that can influence the timing of RT.

*KPIs Bot*: The goal of this PoC is to create, through data clustering, a group of Key Performance Indicators (KPIs) based on diagnostic and therapeutic performance [11]. Among other potential exploitations, this is a simplified example of how the DataMart can be used for rule-based patients' recruitment. ICD9 codes for diagnosis and surgery were chosen for selecting pts with breast cancer who underwent surgery. In accordance with the aim of the study, we selected nine KPIs to be extracted (Table 3). For each KPI, variables for its definition were selected and divided in structured and not structured. The last one was extracted by text mining. Artificial Intelligence automated pathway of extraction identified 2144 patients. Five different data sources were used for data extraction.

Nine structured (age, ICD9 diagnosis, ICD9 surgery, ICD9 diagnostic exams, data of beginning chemotherapy and/or radiotherapy, data of recovery and dismissal, and data of pathology exam) and four not-structured variables by text-mining elaboration (subtypes, staging, and multidisciplinary board therapeutic indications) for KPIs' calculation were identified. Extraction populated all KPIs, and mean rate of data extraction in text-mining elaboration was 78% and 88.3%, respectively, for staging and subtypes' characterization. KPIs' performance was, respectively, (1) 20.91%, (2) 17.88%, (3) 26.9%, (4) 0.25%, (5) 1.72%, (6) 44.6%, (7) 92.2%, (8) 95%, and (9) 67.3%.

'


#### **Figure 2.** Waiting Time BOT platform.

#### **Table 3.** KPIs description.


**on-**KPIs' extraction was feasible, even if further validation is necessary to implement data extraction and optimize quality of data, to create a simultaneous evaluation of them, integrated inside the hospital system.

#### **4. Discussion**

7 days Artificial Intelligence (AI) is the ability of technology applications to accomplish any cognitive task, at least as humans [12]. Even more than ever, AI is transforming our lives and job-automating processes, and it is becoming an indispensable tool for research and development of technology. Creation of a system based on AI requires the use of neural network replications that are capable of answering some questions, identifying specific patterns of data, and learning from them. For this, it is fundamental to create algorithm connections on which system AI technologies need to run. It has been already reported by Carter et al. that breast cancer care was always supported by AI applications since the 1970s, and now it is even more integrated in diagnostic systems [13], for example, in mammography implementation [12,14]. There are many single experiences of AI application in breast cancer care. An example of this issue is reported by Schaffter

et al., in a study in which AI was applied to build algorithms for interpretation of screening mammography [15]. Another study by Pantanowitz L et al. reported application of AI to pathology activity of quantifying mitotic figures in digital images of invasive breast carcinoma with implementation of accuracy and overall time savings, respectively, in 87.5% and in 27.8% of cases [16].

Moreover, it is demonstrated that, thanks to AI application, it is possible to save time, lower costs, and raise efficacy [12].

In our experience, we created an AI connection to allow not only storage of all data repository of breast cancer patients who were treated in our hospital, but also a system that is capable of being interrogated for different purposes. In fact, data stored in our Breast DataMart were analyzed and used to create two systems: Waiting Time BOT for monitoring waiting time from surgery to radio-therapy, and KPIs' BOT for evaluating different aspects of breast unit performance. In both cases, BOT were comparable with clinical practice and the literature. In fact, waiting time in breast cancer represents a key point of treatments, and its delay can lead to reduced efficacy in terms of breast cancer outcomes [17–20]. In particular, a waiting time of 12 weeks or more from surgery to the start of radiation (for patients who are not candidate to adjuvant chemotherapy) and a waiting time of six weeks or more from completion of chemotherapy to start of radiation (for patients who are candidate to adjuvant chemotherapy) are associated with worse event-free survival after a median follow-up of seven years [17]. Given that radiotherapy should be started as soon as reasonably possible, a monitoring system such as Waiting Time BOT could allow not only to track possible delays in pts' pathway of care, but also to learn to predict factors that can be associated to this delay and can be prevented. On the other hand, the KPIs BOT, which allows users to track the performance of breast tumor pathway of care, is based on a system of indicators published by Altini et al. [11] in 2019. In this system, multidisciplinary evaluation is fundamental, but in the hospital system, services provided by the various departments can be reported on different informatic platforms or archives. This usually requires a data entry or data manager to report the folder manually inside CRFs for data collection [21]. In the literature, systems for tracking breast unit performance are reported, for example, EUSOMA, which is used for quality assurance [22]. However, the GENERATOR Breast DataMart does not want to replace these already established systems, but rather offers the possibility to search for data sources automatically for any type of analysis and can therefore be integrated with them.

Beyond the individual project with AI application, there is a multitude of data that could be analyzed by different prospective for implement patterns of care by the following:


Breast DataMart is a dynamic system based on AI, with the purpose of connecting data patterns from different sources, answer specific questions, and learn from data analyses, to implement outputs. The PoCs we performed in this study demonstrate that it is feasible to achieve this purpose for breast cancer care, using simple pathways. We interrogate the system about waiting-time data, the system returns data of interest, and it learns from them, constructing a "guardian system" to predict waiting time of patients and surgery data. On the other side, we created a second system of data elaboration by KPIs analysis. DataMart system was trained to find and return data of interest for analysis. Final elaboration allows clinicians to have a system integrated in the hospital system for on-line contemporary analysis. DataMart goals are not only to obtain a single PoC, but to have an entire data

repository for breast cancer continually analyzed and processed, with the possibility to perform unlimited queries. Results of these queries can be integrated in the Hospital System for Guardian or Avatar robot system. Future application of Breast DataMart system in breast cancer care is addressed to reduce biases in patterns of care, manage heterogeneity of disease, and create algorithms for implement cost/effectiveness.

Standardized Data Collection (SDC) is a recent methodology for extract and use of real-world data. It is based on the concept that, besides structured data available by clinical trials, we have a multitude of low-quality big data inside electronic and paper folders, from which evidences can be generated, also closer to clinical practice [23–26]. AI introduction leads to the evolution concept that modern oncology not necessarily needs to be built only on SDC, but here AI application allows us to use real-world data to obtain data classificatory, predictive model, or guardian for clinical practice also in a not-time-consuming process. In this way, a system such as Breast DataMart, which we developed, becomes a dynamic application of SDC-captured data, with automated possibility of on-line queries. Moreover, DataMart AI technology, thanks to neural networks applied to building unstructured data and retrieving data through text mining, ensures that otherwise lost data are included in its system. In fact, it is possible to also recover from the hospital system PDF documents and electronic reports. The guarantee that the data contained in them are certified is linked to the officiality of these reports. Finally, since the DataMart is linked to the hospital system, its outputs can be integrated, in turn, into clinical practice, as alert systems, to obtain predictions or simply to describe useful trends to manage cost/effectiveness items.

#### **5. Conclusions**

GENERATOR Breast DataMart was created for supporting breast cancer pathways of care. An AI-based process automatically extracts data from different sources and uses them for generating trend studies and clinical evidence. For testing its use, two proof of concepts on waiting time and KPIs' calculations were built and validated. Further steps will include DataMart population with all data online in the hospital system and to start queries to implement clinical practice. Further studies and more proof of concepts are needed to exploit all the potentials of this system.

**Author Contributions:** Conceptualization, F.M. (Fabio Marazzi), L.T., V.M. and S.P.; Data curation, N.D.C., J.L. and C.M.; Funding acquisition, V.V.; Investigation, N.D.C., R.P., G.F. and V.V.; Methodology, N.D.C., R.P. and C.M.; Project administration, L.T., S.P. and V.V.; Software, C.I. and J.L.; Validation, F.M. (Francesca Moschella), A.M.S., M.A.G., R.M. and V.V.; Visualization, G.F.C. and B.C.; Writing—original draft, V.M. All authors have read and agreed to the published version of the manuscript.

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

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

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

#### **References**


**Petra Tesarova 1, \*, David Pavlista <sup>2</sup> and Antonin Parizek 2**


**Abstract:** The main goal of precision medicine in patients with breast cancer is to tailor the treatment according to the particular genetic makeup and the genetic changes in the cancer cells. Breast cancer occurring during pregnancy (BCP) is a complex and difficult clinical problem. Although it is not very common, both maternal and fetal outcome must be always considered when planning treatment. Pregnancy represents a significant barrier to the implementation of personalized treatment for breast cancer. Tailoring therapy mainly takes into account the stage of pregnancy, the subtype of cancer, the stage of cancer, and the patient's preference. Results of the treatment of breast cancer in pregnancy are as yet not very satisfactory because of often delayed diagnosis, and it usually has an unfavorable outcome. Treatment of patients with pregnancy-associated breast cancer should be centralized. Centralization may result in increased experience in diagnosis and treatment and accumulated data may help us to optimize the treatment approaches, modify general treatment recommendations, and improve the survival and quality of life of the patients.

**Keywords:** breast cancer; pregnancy; chemotherapy; tailoring; personalization

#### **1. Introduction**

The need to protect the fetus from the adverse events associated with the treatment of cancer represents a significant barrier to the implementation of genomic and molecular biological personalization of treatment in a subgroup of pregnant patients with breast cancer. Pregnancy-associated breast cancer (PABC) is defined as breast cancer diagnosed during pregnancy (BCP) or in the first postpartum year or at any time during lactation. BCP is a special situation of concomitant pregnancy and cancer and, due to different subtypes of breast cancer, tumor detection at different stages and diagnosis confirmed at different trimesters of pregnancy does not allow the application of only one standard treatment approach. The reason is also the fact that despite the increasing experience with the treatment of such patients, the published data on PABC are still limited. Prospective studies of breast cancer during pregnancy are almost lacking, and we must rely on data from retrospective case series [1,2].

The development of personalized precision medicine as the ultimate aim of the treatment of PABC is dependent on a better understanding of the pathogenesis of PABC [3].

The advent of big genomic data has shifted our attention from examining single genes to whole exome and transcriptome analysis with the aim of identifying new predictive factors, biomarkers, and therapeutic targets although until now, still only some more frequently mutated genes are tested to achieve better cost-effectiveness, i.e., genes that seem to be associated with better cost-effectiveness, enhanced data analysis, and rapid availability for the immediate clinical decisions [4]. Unfortunately, pregnant patients with breast cancer do not yet benefit from these advances in precision medicine.

**Citation:** Tesarova, P.; Pavlista, D.; Parizek, A. Is It Possible to Personalize the Diagnosis and Treatment of Breast Cancer during Pregnancy? *J. Pers. Med.* **2021**, *11*, 18. https://doi.org/10.3390/ jpm11010018

Received: 1 December 2020 Accepted: 22 December 2020 Published: 28 December 2020

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

**Copyright:** © 2020 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/).

Tailoring the treatment of breast cancer in pregnancy must primarily adapt to the course of pregnancy. Due to the young age, the disease is more often associated with hereditary mutations of risk genes. Cancer is more likely to have a high histological risk profile and is diagnosed at a more advanced stage. Therefore, in clinical practice we are more often faced with the need to treat patients with a very advanced stage of cancer, frequently with the presence of metastases in the skeleton or visceral organs.

#### **2. Epidemiology**

Increasing incidence of PABC is associated with an overall increase of breast cancer in the population and increasing age at conception. PABC is still, however, relatively uncommon (with an incidence of 15 to 35 per 100,000 deliveries, more frequently occurring during the first postpartum year rather than during the pregnancy) although breast cancer is the most common type of cancer in pregnancy [5]. PABC is very rare (one per 1000 pregnancies annually, i.e., 0.07% to 0.1% of all malignant tumors, only) [6].

Pregnancy generally has a lifelong protective effect on breast cancer risk, but it increases the risk of breast cancer for several years after pregnancy with the highest risk at 6 years after delivery and significantly higher risk in older primiparas. There are important differences (in terms of diagnosis, treatment, and outcome) between PABC and breast cancer after pregnancy [2].

#### **3. Pathophysiology**

The pathogenesis of PABC is not fully understood [7]. Pregnancy and lactation are associated with increased levels of estrogens with the impairment of their normal cyclical pattern resulting in resultant molecular and histological changes in the breast gland. Increased estrogen levels may also promote the formation of metastases. Other factors, e.g., immune changes and inflammation [8], also promote carcinogenesis, especially in women with occult disease at conception (more frequent during the involution of the mammary gland) [9]. It should also be stressed that late diagnosis of breast cancer in pregnancy may also contribute to the more frequent presence of metastatic disease.

Pregnancy-associated plasma protein A (PAPP-A) may also play an important role in the development of metastatic PABC (by its collagen-modifying properties) and may help to identify patients at risk of metastatic disease [10].

#### **4. Pathology**

As in non-pregnant women, the most common form of PABC is infiltrating ductal adenocarcinoma. PABC is, however, less differentiated and (as already stressed) diagnosed at more advanced stage. Inflammatory breast cancer is also more frequent in pregnancy than in non-pregnant women [11]. The molecular pattern of PABC is different, namely in terms of more frequent mutations of the mucin gene family, mismatch repair deficiencies, and other non-silent mutations [12].

Estrogen and progesterone receptor expression seems to be decreased in PABC compared to that in non-pregnant patients with breast cancer (25% vs. 55% to 60%) [13] probably with no significant difference in overexpression of human epidermal growth factor receptor 2 (HER2) ([14,15], Table 1). Despite many differentially expressed genes, there seems to be no correlation between genetic changes and histopathological and clinical characteristics of BCP. Further studies in search of putative novel biomarkers that could identify the subpopulation of women in childbearing age at risk of PABC are warranted [16].


**Table 1.** Tailoring treatment according to the type of breast cancer.

HER2, human epidermal growth factor receptor 2.

#### **5. Precision Medicine in Breast Cancer**

Precision medicine involves the identification of molecular signature, biomarkers, and clinical phenotype and the evaluation of their impact in combination with lifestyle and environmental factors on the prevention and treatment of the disease [17]. Cancer biomarkers may be diagnostic, prognostic, predictive, or used to monitor treatment responses. Prognostic biomarkers provide information about a patient's overall cancer outcome, irrespective of therapy [18]. They can identify high-risk patients who may benefit from more aggressive treatment but provide no information on which patients will most likely derive a clinical benefit from any specific therapy. Conversely, modifiable predictive markers responding to the treatment can indicate the probability of a patient gaining a therapeutic benefit from a specific treatment [19].

Breast cancer can be classified based on gene expression and histology including the expression of estrogen receptor (ER), progesterone receptor (PgR), and human epidermal growth factor receptor 2 (HER2) into several subtypes, characterized as luminal, normallike, HER2-overexpressing, and triple negative breast cancer (TNBC) [20]. Gene expression profiling is more in-depth and provides more detailed stratification of breast cancer compared to histology itself. Based on these analyses, breast cancer was shown to be very heterogeneous with substantial variability in biological behavior, pathogenesis, response to treatment, and outcome [21].

Analysis based on microarray gene expression is already available, but its cost prevents its broader use in routine clinical practice with more focused analysis aimed at smaller gene sets (breast cancer index, Endopredict, the Oncotype DX 21-gene recurrence score, the BreastOncPx 14-gene distant metastasis signature, 50-gene signature called PAM50 (Prosigna), and the MammaPrint 70-gene prognosis signature) used for breast cancer stratification may emerge as more cost-effective and help clinicians to pinpoint the use of endocrine treatment and adjuvant chemotherapy [22].

To overcome the need to obtain biopsy samples from primary or metastatic lesions, great attention is paid to the blood-based biomarkers, e.g., circulating tumor cells (CTCs), exosomes and circulating tumor DNA (ctDNA), sometimes called liquid biopsy. CTCs are released from the primary tumor and are related to the propensity of the cancer to form distant metastases [23].

Genomic instability, which is common in cancer, results in genetic and epigenetic heterogeneity, and so the outcomes of patients with the same histologic type of cancer may be different in terms of response to treatment and outcome [24].

Epigenetic modification, e.g., DNA methylation and histone acetylation, is instrumental in the early phase of carcinogenesis. Recently, the role of different types of non-coding RNAs (ncRNAs) regulating gene expression and working as epigenetic modifiers has been uncovered [25].

Evaluation of the expression of both estrogen (ER) and progesterone (PR) receptors is indispensable before the introduction of hormonal treatment, and similarly, evaluation of HER2 amplification is necessary for the prediction of the response to anti-HER2 treatment. Mutation of the gene for the estrogen receptor (ESR1) predicts the risk of resistance to

aromatase inhibitors. Similar markers predicting the response to radiotherapy and different modes of chemotherapy are warranted [26].

Analysis of some of these biomarkers in clinical practice may refine the search for suitable clinical trials with drugs aimed at the identified targets, but pregnant patients, unfortunately, cannot be recruited to the clinical trials. In the treatment of pregnant women, we can use neither standard, breast-cancer-specific immunohistochemical targets, such as hormone receptor or HER2 antigen positivity, nor targets derived from genomic analysis, such as PIK3 (phosphatidylkinase 3) or ESR1 (gene for estrogen receptor 1) mutations, nor those found by pathologists (TILs (tumor infiltrating lymphocytes)). Off-label treatment aimed at molecular targets not typical for breast cancer (KRas, BRAF, EGFR, etc.) cannot be used in the treatment of PABC.

Pregnancy and concomitantly diagnosed breast cancer are currently a major barrier to the use of precision medicine in the treatment of breast cancer. Its inclusion in treatment plans must be postponed until after delivery or modified so that the questions we specifically address in these situations can be answered. Due to the small number of patients and the fetuses, there are currently no (and will hardly be any in the future) clinical studies in this breast cancer subpopulation.

#### **6. Clinical Presentation**

Common signs and symptoms of cancer (lump, thickening, change in the size, shape, inverted nipple, etc.) may be hidden because of the pregnancy-associated physiological changes of the breast gland. This can delay diagnosis and adequate care. Patients with the presence of metastases may develop general symptoms, fatigue, back pain, dyspnea, pain and pressure in the right ribs, etc.

#### **7. Diagnosis**

Physical examination of the breast gland in pregnancy in search for putative cancer is difficult because of pregnancy-associated changes of the breast gland and also the utility of mammography may be limited resulting often in delayed diagnosis of PABC [27]. Any persisting (for more than two weeks) mass should be examined although 80% of the findings in breast biopsies in pregnant women are benign [28]. Mammography is not contraindicated in pregnancy with abdominal shielding (although the decrease of fetal radiation exposure with shielding remains uncertain). The sensitivity of mammography may be decreased due to higher density of the breast gland during pregnancy and lactation, but it still remains useful as a diagnostic tool. Breast ultrasonography can determine whether a breast mass is a simple or complex cyst or a solid tumor without the risk of fetal radiation exposure and may be used to guide the diagnostic biopsy. Gadolinium-enhanced MRI should be (if possible) avoided during pregnancy [29]. Needle core biopsy is the preferred method in any clinically suspicious breast mass and can be safely done during pregnancy, preferably under local anesthesia [30]. Possible infiltration of the lymph nodes by cancer cells should be further evaluated with ultrasound and fine needle aspiration biopsy for cytologic confirmation [31].

#### **8. Staging**

Modifications of the standard staging work-up should be implemented to protect the fetus (Table 2). Chest radiographs to evaluate for lung metastases should be performed with appropriate fetal shielding and limited late in gestation when the gravid uterus is pressing against the diaphragm. Computed tomography (CT) scans should be avoided during pregnancy because of the radiation exposure. Abdominal ultrasound for the evaluation of liver metastases is safe, but in pregnant women, significantly less sensitive than CT or MRI. MRI without gadolinium can be considered only if needed, especially in the first trimester, since there is a limited experience assessing safety during organogenesis [32]. Bone scans must not be used in pregnant patients for the evaluation of bone disease in the absence of signs or symptoms of bone abnormality. As an alternative, skeletal MRI may be

considered (without contrast). Increases in tumor markers CA (cancer antigen ) 15.3 and CEA (Carcinoembryonic antigen) always give rise to the suspicion of metastasis [33]. Locally advanced-stage disease and/or suspicious symptoms should prompt a complete radiographic staging evaluation with modifications and shielding to protect the fetus. Since the therapeutic approach to patients with early or metastatic breast cancer is not usually changed during pregnancy (neither targeted nor hormonal treatment is considered), it is possible to safely leave staging of early breast cancer examinations after delivery, preferably using PET-CT or CT scans [34].


**Table 2.** Tailoring treatment according to the stage of breast cancer.

#### **9. Hereditary Breast Cancer and PABC**

Genetic predisposition to breast cancer is more frequent among pregnant women with cancer. The protective effect of multiparity and breastfeeding may be lost in women who inherit BRCA2 (but not BRCA1) mutations. *BRCA1* (Breast cancer antigen 1) or *BRCA2* (Breast cancer antigen 2) mutations confer the women with a 50–80% lifetime risk of breast cancer and 16–65% lifetime risk of ovarian cancer. These risks far exceed those of breast (13%) and ovarian (1.5%) cancer in the general population [35].

Most cases of breast cancer related to BRCA1 and BRCA2 are diagnosed in young women, and the probability of pregnancy in young women is high. At present, several other genes that increase the risk of breast cancer (*PALB2, CHECK2, CDH1*, etc.) are being identified in genetic screening panels. Genetic counseling is recommended for all patient with PABC [36]. Carriers of BRCA/2 not only have a higher risk of developing PABC but also have probably poorer outcomes with higher probability of developing distant metastases [37].

If a pregnant woman carries a BRCA1/2 mutation, this information may influence the decision on the type of surgery but does not allow the use of PARP (poly-ADP ribose polymerase) inhibitors in pregnancy in case of metastatic spread.

#### **10. Monitoring of the Pregnancy**

The pregnant woman with breast cancer requires careful and continuous monitoring of her pregnancy by her obstetrician and her oncologist. Confirmation of gestational age and expected date of delivery are important, as both are significant factors in treatment planning. For this reason, follow-up should take place at the center with experience in the care of patients with BCP and the gynecologist/obstetrician should be the part of the multidisciplinary team [38]. Breast-feeding should be discontinued immediately after delivery. Since, according to clinical studies, a properly selected cancer treatment does not compromise the cognitive function of the newborn as opposed to its immaturity, it is optimal to complete pregnancy until physiological delivery, if this is possible in terms of the severity of the disease course [39].

#### **11. Prognosis**

Based on smaller studies, maternal outcome may be worse in women with breast cancer diagnosed in pregnancy [40]. The largest cohort study in women treated for PABC, however, demonstrated similar disease-free survival and overall survival comparable to those of the general population [41].

In the registry study that compared over 300 women with breast cancer during pregnancy with almost 870 women who were not pregnant at the time of diagnosis, there was no significant difference in either progression-free survival (PFS, hazard ratio (HR) 1.34, 95% CI 0.93–1.91) or overall survival (OS, HR 1.19, 95% CI 0.73–1.93) [42]. In another smaller study that included 75 women who received standard chemotherapy during the second and third trimesters, women who were pregnant had a significantly improved five-year disease-free survival (72% vs. 57%) and OS (77% vs. 71%) [43].

A 2012 meta-analysis comprising over 3000 cases of gestational breast cancer and 37,100 controls found that gestational breast cancer was associated with a higher risk of death (HR 1.44, 95% CI 1.27–1.63), however, the association appeared to be limited primarily to women diagnosed in the postpartum period (HR 1.84, 95% CI 1.28–2.65) rather than during pregnancy (HR 1.29, 95% CI 0.72–2.24) [44].

#### **12. Treatment of BCP**

Pregnant women with breast cancer should be treated according to the guidelines for non-pregnant patients, with some modifications to protect the fetus (Table 3) [42,45].


**Table 3.** Personalization of breast cancer treatment in pregnancy with regard to its stage (adapted according to [46–48]).


**Table 3.** *Cont.*

#### *12.1. Surgery*

Either breast-conserving surgery or mastectomy are a reasonable option for the pregnant woman with breast cancer. A choice between them is guided by tumor characteristics and the result of the genetic test and patient preferences [49]. Women with breast cancer during pregnancy should undergo an axillary node evaluation. While axillary lymph node dissection is preferred, there are increasing data on the safety and efficacy of sentinel lymph node dissection [50].

The best cosmetic results and the least complications are achieved by surgery on a hormonally unstimulated breast preferably after childbirth after lactation arrest.

#### *12.2. Radiotherapy*

If the breast-conserving surgery is performed, the adjuvant radiotherapy (RT) should be postponed after delivery. The threshold for adverse radiation effects in fetuses is less than 100 mGy. Given the high dosage of fetal radiation, radiation therapy for breast cancer in pregnancy is still considered an absolute contraindication, although this may change in coming years with improving technologies [51].

As methods of stereotactic radiation and improved modalities of delivery are developed, radiation therapy may be an option for more women during pregnancy [46].

#### *12.3. Systemic Antitumor Therapy*

#### • **Pharmacokinetics and Distribution of Drugs in Pregnancy**

Alterations in drug distribution are expected due to the physiologic changes that occur in pregnancy. Pregnancy leads to 40–60% increase in plasma volume even as early as 6 weeks after gestation. Increased fluid volume is associated with decreased plasma albumin, which may interfere with plasma concentration of some protein-bound drugs, e.g., taxanes, but this effect may be counterbalanced by high levels of estrogens, which increase other plasma proteins. Drug clearance by the kidney and liver increases, which may again reduce plasma levels of cytotoxic drugs. Diminished gastric motility may impact the absorption of orally administered drugs. "Third space" of the amniotic sac may play a role as well. The multidrug-resistance p-glycoprotein has been detected in fetal tissues and in the gravid endometrium and may offer some degree of protection to the fetus. However, currently it is not clear how these physiologic changes impact upon active drug concentrations and their resulting efficacy and toxicity. Moreover, pregnant women receive similar body surface-area based chemotherapy doses as non-pregnant women, which are adjusted according to continuing weight gains [52].

#### • **Chemotherapy**

Patients indicated to chemotherapy during pregnancy may only start treatment after the first trimester. Data are available namely for anthracycline-based chemotherapy, often on an every-three-week schedule. Anthracyclines, more specifically doxorubicin, have not been found to significantly affect the cardiac function of children exposed in utero [53]. However, at least four cases of neonatal adverse cardiac effects have been reported after in utero exposure to anthracyclines, and there are several cases of in utero fetal death after exposure to idarubicin or epirubicin. Largely because of these reports, doxorubicin is preferred to idarubicin or epirubicin for the use in pregnancy [54]. Cyclophosphamide also has not been demonstrated to increase neonatal morbidity. In a prospective single-arm study, 87 pregnant breast cancer patients were treated with FAC (5-fluorouracil, adriamycine (doxorubicine), cyclophosphamide) in the adjuvant or neoadjuvant setting [55]. No stillbirths, miscarriages, or perinatal deaths occurred in the cohort of patients who received FAC chemotherapy during their second and/or third trimester. Most of the children did not have any significant neonatal complications. Three children were born with congenital abnormalities: one each with Down syndrome, ureteral reflux, or clubfoot. The rate of congenital abnormalities in the cohort was similar to the national average of 3%.

Taxanes, specifically paclitaxel, have not been found to be teratogenic when administered in the third trimester. Paclitaxel is preferred over docetaxel due to the better transplacental transfer of docetaxel. Taxanes were administered in the second and third trimesters in 38 patients and for the treatment of breast cancer in 27 patients. Despite the limitations and bias inherent in case reports, the use of taxanes appears feasible and safe during the second and third trimesters of pregnancy, with minimal maternal, fetal, or neonatal toxicity [56]. Although taxanes have promising treatment outcomes, we still have information about their safety only from case reports and small case series, and therefore, we must use them with caution [57]. Platinum derivatives may play a role in the treatment of triple negative breast cancer. They are highly protein bound, but the unbound fraction may cross the placenta. Carboplatin may be associated with the derangements of trophoblast invasion and disrupting placental development, which is not complete until 20 weeks of gestation. Although the data regarding the safety of platinum in pregnancy are limited, a systematic review of the use of carboplatin and cisplatin in pregnancy found that no malformation or toxicity was reported in seven carboplatin-exposed neonates [58]. Although only limited case reports are available, anthracycline chemotherapy administered on a dose-dense schedule (i.e., treatment every two weeks) does not appear to increase the risks of maternal or fetal complications compared with treatment administered every three weeks [59]. Chemotherapy should be avoided for three to four weeks before delivery whenever possible to avoid transient neonatal myelosuppression and potential complications, including sepsis and death. Weekly regimens with low hematotoxicity are an exception [60].

#### • **Targeted Treatment**

The use of trastuzumab during pregnancy is relatively contraindicated. Exposure to trastuzumab during pregnancy can result in oligohydramnios, which in some cases may lead to pulmonary hypoplasia, skeletal abnormalities, and neonatal death. Women exposed to trastuzumab during pregnancy require ongoing monitoring of amniotic fluid volume, which is a marker of fetal renal status, throughout the pregnancy [61,62]. In a case report of maternal exposure to lapatinib for 11 weeks during the first and second trimester of pregnancy, there was an uneventful delivery of a healthy female infant, who was developmentally normal at 18 months of age [63].

However, until more information is available, we recommend against the use of lapatinib during pregnancy and lactation. There are currently no significant data on the safety of other anti-HER2 agents such as pertuzumab and ado-trastuzumab emtansine (TDM-1), and therefore, we do not recommend these agents until after delivery. However accidental short-term exposure to these agents during the first trimester does not appear to be associated with increased risk of fetal malformation, which is different compared to the risk from chemotherapy [64].

Currently we have not enough information on the safety of using bevacizumab, PARP inhibitors, and immunotherapy (PD-1 (Programmed death-1) and PDL-1 (Programmed death ligand-1) inhibitors) during pregnancy.

#### • **Endocrine Treatment**

The use of selective estrogen receptor modulators (SERMs) such as tamoxifen during pregnancy should be generally avoided. They have been associated with vaginal bleeding, ambiguous genitalia, miscarriage, congenital malformations (spinal abnormalities, absent ears, craniofacial abnormalities, and cardiac malformation seen in Goldenhar's syndrome), and fetal death [65]. Aromatase inhibitors (AIs) and luteinizing hormone-releasing hormone (LHRH) agonists are both contraindicated in pregnancy. AIs are not used in premenopausal women, but AIs combined with ovarian suppression by LHRH agonists may be used following term delivery.

#### • **Supportive Care**

Antiemetics, including selective serotonin (5-HT) and neurokinin 1 (NK1) antagonists, are used to treat severe nausea and vomiting in pregnant women and are generally considered safe. However, long-term dexamethasone therapy should be avoided, if possible, because of potential maternal and fetal risks. Safe use of G-CSF (Granulocyte-colony stimulating factor) (and recombinant erythropoietin) in human pregnancy has been reported. Although there are no prospective trials evaluating the use of G-CSF or granulocytemacrophage colony-stimulating factor (GM-CSF) in pregnant women, these agents are safe in the treatment of neonatal neutropenia and/or sepsis, but more caution is needed considering the very limited data. Hence, dose-dense chemotherapy is not the optimal strategy in pregnant patients [66].

#### *12.4. Postponement of Treatment*

If a malignant tumor is diagnosed in the first trimester, it is possible to terminate the pregnancy prematurely or postpone treatment until the second trimester. Delay can mean the risk of progression and generalization of the disease depending on the type of cancer and its staging at the time of diagnosis and may worsen prognosis (Table 4) [67]. If the patient has a lower-grade hormone-dependent cancer limited to the breast itself, the risk of delay is lower than in triple-negative cancer with nodal involvement. Delaying chemotherapy by 3–6 months may increase the risk of metastases by 5–10% [68].


**Table 4.** Personalization according to patient preference.

#### *12.5. The Course of Pregnancy, Fetal Monitoring, and Childbirth*

Based on the available evidence, chemotherapy in BC patients may be safe during the second and third trimesters, with cessation of treatment three weeks prior to expected delivery. The most common complications of pregnancy associated with the application of chemotherapy are intrauterine growth retardation, prematurity, low birth weight, and bone marrow toxicity. Prematurity is generally associated with worse neonatal and long-term outcomes and, thus, should be avoided. Fetal condition can be well monitored by regular ultrasound biometrics and Doppler flowmetry. If premature birth is necessary, induction of fetal pulmonary maturity by corticoid administration is indicated. Most women expect vaginal delivery at term, but due to chemotherapy, delivery must be planned and induced, and immediately after delivery, lactation must be stopped.

#### **13. Infant Outcome**

Data suggest that early development among children born to women with cancer appears similar to that of children of the same gestational age, irrespective of in utero exposure to radiation or chemotherapy.

In a study of 129 children born to mothers diagnosed with cancer during pregnancy (over half of whom had breast cancer), cardiac, cognitive, and general development after a median of 22 months was equivalent with controls matched for gestational age [69]. In a subgroup analysis of children exposed to anticancer therapy in utero, similar outcomes were reported for the 96 children exposed to chemotherapy after the first trimester and the 11 children exposed to radiation compared with gestational-age-matched controls. There was a non-significant trend toward higher rates of small for gestational age at birth infants born to women with cancer (22% vs. 15%), particularly if exposed to chemotherapy or radiation. While the median gestational age of the children born to women with cancer was 36 weeks and, thus late preterm, it is unclear whether these children were born early because of early induction given their mothers' diagnosis of cancer.

In the cohort study of 1170 pregnant women with all types of cancer treated at multiple institutions, 39% of whom had breast cancer, 88% of pregnancies resulted in live births [70]. Half of these deliveries were preterm, almost 90% of which were iatrogenic. These studies suggest that low neonatal complication rates are associated with in utero exposure to chemotherapy, but long-term data are limited. Moreover, studies may be limited by the fact that treatment providers may sometimes opt for early delivery induction, even when pregnancy does not affect treatment. One study reported 40% mortality among patients with advanced BCP who received chemotherapy when studied over a 13-year period (1991–2004) [71]. For women with breast cancer during pregnancy, the risk of cancer to the unborn is unknown, although there are no reported cases of childhood cancer arising in children exposed to chemotherapy of their mothers for breast cancer in utero.

#### **14. Termination of Pregnancy**

Early termination of pregnancy does not improve the outcome of BCP. In fact, some series suggest decreased survival in pregnant women who electively terminate their pregnancies compared with that in those who continue the pregnancy. However, these studies are retrospective case reviews and possible bias cannot be excluded; women with more advanced disease or poorer prognostic features possibly were more likely to be counseled to have an abortion [71]. The decision to terminate pregnancy for health reasons is difficult and should always be comprehensively considered in terms of the risk of fetal cancer treatment, the patient's prognosis, and the impact of cancer therapy on the mother's fertility. Although this situation is quite ambiguous, many physicians recommend to the patients with BCP to end pregnancy and so often deprive the patient of their only chance of having a child (Table 4).

#### **15. Metastatic BCP**

During pregnancy we can also diagnose patients with de novo metastatic breast cancer, and some patients with early breast cancer treated in a neo/adjuvant setting later metastasize. The main problem of the care of the metastatic breast cancer in pregnancy is limited treatment options with respect to the fetus. The main goal of therapy is to prolong the patient's life, maintain its quality, not to damage the fetus, and for mother to spend as much time as possible with the child. This situation is extremely physically and psychologically demanding for the patient and affects the whole extended family [72].

#### **16. Tailoring Treatment of Breast Cancer in Pregnancy**

Personalized medicine has changed our approach from a "one size fits all" to the treatment of patients in a more individually tailored way. The goal of clinical research programs with a personalized approach to patients with breast cancer is to evaluate the unique code of RNA and DNA of cancer, enabling individualization of the treatment plan [73].

During pregnancy, tailoring to immunohistochemical markers such as hormone receptors, HER2 or PDL-1 expression, cannot be used at present, due to the risk of fetal harm. Genome testing and the use of next-generation sequencing (NGS) could, in the future, refine the prognosis of cancer and its sensitivity to chemotherapy, as the only acceptable systemic treatment in pregnancy.

From 2010 to 2020, 53 patients with BCP were treated at the Department of Oncology of the First Faculty of Medicine and the General Hospital in Prague. The number and proportion of patients has been influenced by the fact that in our comprehensive cancer center we have a program dedicated to young patients under 35 years of age and pregnant patients with breast cancer are referred to us from almost all over the Czech Republic (Table 5).

**Table 5.** Patients with breast cancer occurring during pregnancy (BCP) were treated at the Department of Oncology of the First Faculty of Medicine and the General Hospital in Prague (2010–2020).


#### **17. Conclusions**

BCP is an example of cancer where individualization of the treatment approach could significantly improve the results of treatment and the hope of patients with concomitant breast cancer and pregnancy to prolong survival. The therapeutic plan must be adapted to the clinical parameters, the degree of pregnancy, the type and stage of the tumor, and the patient's preference. The current options for a personalized treatment approach are not yet widely used in this subgroup of patients, although, in the future it would certainly be possible to focus molecular biology, NGS, and liquid biopsy methods to refine staging, estimate tumor chemosensitivity, and cancer prognosis to assess possible postponement of treatment to the postpartum period. Physicians treating patients with breast cancer in pregnancy have increased responsibility because they are trying to save two lives. While information and data on BCPs are increasing, it is necessary to centralize the treatment of BCP in the hands of experienced oncologists and obstetricians with praxis in this type of high-risk pregnancy and personalized access to each pregnant patient.

**Author Contributions:** P.T. prepared the draft of the paper which was then extensively consulted with both of them and D.P. and A.P. added valuable information from the gynecologist's point view. All authors collaborated on the analysis of available data and their interpretation and contributed significantly to the final version of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** The work on the paper was supported by the research initiative the Ministry of Health of the Czech Republic Progress Q28/LF1 and DRO VFN 64165.

**Conflicts of Interest:** No potential conflict of interest is to be disclosed.

#### **References**


#### *Article*

### **Palbociclib Plus Fulvestrant or Everolimus Plus Exemestane for Pretreated Advanced Breast Cancer with Lobular Histotype in ER**+/**HER2**− **Patients: A Propensity Score-Matched Analysis of a Multicenter Retrospective Patient Series** †

**Armando Orlandi 1, \* , Elena Iattoni 1 , Laura Pizzuti 2 , Agnese Fabbri 3 , Andrea Botticelli 4 , Carmela Di Dio 1 , Antonella Palazzo 1 , Giovanna Garufi 1 , Giulia Indellicati 1 , Daniele Alesini 3 , Luisa Carbognin 5 , Ida Paris 5 , Angela Vaccaro 6 , Luca Moscetti 7 , Alessandra Fabi 2 , Valentina Magri 4 , Giuseppe Naso 4 , Alessandra Cassano 1,8 , Patrizia Vici 2 , Diana Giannarelli 9 , Gianluca Franceschini 8,10 , Paolo Marchetti 4 , Emilio Bria 1,8,**‡ **and Giampaolo Tortora 1,8,**‡


Received: 10 November 2020; Accepted: 16 December 2020; Published: 18 December 2020

**Abstract:** Cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) in combination with endocrine therapy (ET) show meaningful efficacy and tolerability in patients with metastatic breast cancer (MBC), but the optimal sequence of ET has not been established. It is not clear if patients with lobular breast carcinomas (LBC) derive the same benefits when receiving second line CDK4/6i. This retrospective study compared the efficacy of palbociclib plus fulvestrant (PALBO–FUL) with everolimus plus exemestane (EVE–EXE) as second-line ET for hormone-resistant metastatic LBC. From 2013 to 2018, patients with metastatic LBC positivity for estrogen and/or progesterone receptors and HER2/neu

negativity, who had relapsed during adjuvant hormonal therapy or first-line hormonal treatment, were enrolled from six centers in Italy in this retrospective study. A total of 74 out of 376 patients (48 treated with PALBO–FUL and 26 with EVE–EXE) with metastatic LBC were eligible for inclusion. Progression-free survival (PFS) was longer in patients receiving EVE–EXE compared with PALBO–FUL (6.1 vs. 4.5 months, univariate HR 0.58, 95% CI 0.35–0.96; *p* = 0.025). On the propensity score (PS) analysis, PFS was confirmed to be significantly longer for patients treated with EVE–EXE compared to PALBO–FUL (6.0 vs. 4.6 months, *p* = 0.04). This retrospective analysis suggests that EVE–EXE is more effective than PALBO–FUL for second line ET of metastatic LBC, allowing us to speculate on the optimal therapeutic sequence.

**Keywords:** advanced breast cancer; mTOR inhibitor; CDK4/6 inhibitor; endocrine resistance

#### **1. Introduction**

Invasive lobular breast carcinomas (LBCs), which account for up to 15% of all invasive breast cancers (BC), are almost always estrogen-positive (ER, coded by the ESR1 gene) and lacking HER2 amplification and as such are treated with endocrine therapy (ET) [1]. Options for ET have expanded in the last two decades with the availability of new agents, including selective estrogen receptor modulators (SERM), aromatase inhibitors (AIs), and selective estrogen receptor degrader (SERD) [2,3]. However, resistance to therapy and subsequent disease progression continue to be major problems. More than a third of patients with ER-responsive early-stage BC and almost all of those with metastatic disease become refractory to these treatments during the course of their disease [4–6]. New approaches to treatment are clearly required, and to this end, cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) were developed. CDK4/6i palbociclib, ribociclib, and abemaciclib in combination with ET have shown clinically meaningful efficacy and a good tolerability profile in patients with metastatic breast cancer (MBC), in endocrine sensitive and endocrine resistant disease, within the PALOMA, MONALEESA, and MONARCH trials, respectively [7–12]. The MONALEESA-3, MONALEESA-7, and MONARCH-2 trials showed significantly improved overall survival with a combination of a CDK4/6i and ET [10,13].

While subgroup analysis of the PALOMA 2 trial showed that the combination of palbociclib plus letrozole is effective in first-line treatment both in ductal and lobular histotypes, no evidence is currently available on the efficacy of CDK4/6i exclusively in second-line treatment according to histotype (PALOMA 3, MONARCH 2, and MONALEESA 3) [7,12,13]. Recently, a pooled analysis of seven phase III trials (combining the data of the endocrine sensitive and resistant setting) was made to investigate the benefit of adding CDKIs to endocrine therapy in patients whose tumors might have differing degrees of endocrine sensitivity, such as the lobular histotype [14]. This pooled analysis shows that all subsets, including LBC, of patients derived benefits from the addition of a CDKI to endocrine therapy.

For some time in our clinical practice, we have observed that patients with ER-positive metastatic LBC who had relapsed on adjuvant tamoxifen/AIs or had progressed with first-line hormonal therapy tended to show poor responses, and their disease showed faster progression with CDK4/6i [15]. Interestingly, in some of these patients, the subsequent use of a mTOR inhibitor (everolimus) produced greater clinical benefits and prolonged survival. In light of these considerations, we conducted a multicentric, retrospective study to compare the efficacy of the combination of palbociclib plus fulvestrant (PALBO–FUL) with everolimus plus exemestane (EVE–EXE) as second-line ET for hormone-resistant metastatic LBC.

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

This retrospective study enrolled women with metastatic LBC from six Italian oncology centers over a five-year period from 2013 to 2018. Female patients (≥18 years at diagnosis) with metastatic LBC (confirmed by metastasis biopsy) or with a clinical history of disease, compatible with recurrent lobular carcinoma of the previously diagnosed primary breast cancer, positivity for estrogen and/or progesterone receptors, and HER2/neu negativity, who had relapsed during adjuvant hormonal therapy or a first-line hormonal treatment, were eligible for inclusion. Patients were excluded if they relapsed in a period of more than 12 months from the end of adjuvant hormonal therapy or they had not received prior hormonal treatments. Patients received second line therapy with PALBO–FUL or EVE–EXE according to standard approved administration schedules. All patients enrolled in the study provided written informed consent for their data to be used for future medical research. The study was conducted in accordance with Italian legislation on observational studies (Min. Sal. Circular 6 September 2002). Data from the six participating centers were processed and stored at the coordinating center (Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy) in compliance with local privacy regulations.

The primary endpoint was progression-free survival (PFS) defined as the interval between the treatment start date and the disease progression date. Secondary endpoints were objective response rate (ORR, rate of complete objective responses and partial objective responses of the disease to the treatment evaluated using clinical and/or radiological criteria, according to RECIST 1.1 Criteria) and clinical benefit rate (CBR, rate of complete objective responses, partial objective responses, and stable disease in response to the treatment evaluated with clinical and/or radiological criteria).

All continuous data were expressed as mean ±SD, range, and median value; frequencies and percentages were reported for categorical variables. The clinical, biological, and pathological characteristics of tumors at baseline were determined using Fisher's exact test. PFS and overall survival were estimated by the Kaplan–Meier limit product method. The Cox regression model was applied to multivariate survival analysis, and *p* values and hazard ratios (HRs) with 95% CI were obtained. All significant variables in the univariate model were used to build the multivariate model of survival. A propensity score (PS) adjustment for baseline characteristics was conducted for survival analysis. The Statistical Package for the Social Sciences (SPSS) 20.0 software, (Chicago, IL, USA) was used for statistical analysis and integrated with Medcalc software V.9.4.2.0 (Mariakerke, Belgium). In all analyses, the significance level was specified as *p* < 0.05. As the study was explorative, an estimate of the sample size was not calculated.

#### **3. Results**

#### *3.1. Patient Demographics*

Of a total of 376 women screened over the five-year period (2013–2018) in the six centers, 74 were diagnosed with metastatic LBC. Of these, 48 patients received PALBO–FUL and 26 EVE–EXE. Baseline patient characteristics were comparable between the two treatment groups (Table 1). Most patients were post-menopausal (89% and 100% in the PALBO–FUL and EVE–EXE groups, respectively), had non-visceral disease (61 and 68%, respectively), and had less than three sites of metastasis (78 and 79%, respectively). Overall, 43 and 57% of patients in the PALBO–FUL and EVE–EXE groups, respectively, had previously received two lines of endocrine therapy, and 15 and 17% of patients, respectively, had metastatic disease on diagnosis. All patients had received at least one or two lines of endocrine therapy (aromatase inhibitors alone or in combination with tamoxifen, or fulvestrant).


#### **Table 1.** Baseline and treatment characteristics (*n* = 74).

ECOG: Eastern Cooperative Oncology Group.

#### *3.2. E*ffi*cacy and Activity*

Median PFS was significantly longer in patients receiving EVE–EXE than in those receiving PALBO–FUL (6.1 vs. 4.5 months, univariate HR 0.58, 95% CI 0.35–0.96; *p* = 0.025 (Figure 1)). Univariate analysis showed that metastatic stage at diagnosis (HR 2.82, 95% CI 1.43–5.56; *p* = 0.003),

previous chemotherapy exposure (HR 0.41, 95% CI 0.24–0.72, *p* = 0.002), and study treatments (HR 0.58, 95% CI 0.35–0.96, *p* = 0.025), correlated positively with PFS (Table 2). On multivariate analysis, previous chemotherapy exposure was the only factor significantly associated with PFS (HR 0.41, 95% CI 0.24–0.72, *p* = 0.002).

**Figure 1.** Progression-free survival (PFS). N: number; EVE: everolimus; EXE: exemestane; PALBO: palbociclib; FULV: fulvestrant.



PFS was significantly longer in patients receiving EVE–EXE in comparison with PALBO–FUL (6.0 vs. 4.6 months, *p*=0.04) on PS analysis adjusted for prior chemotherapy and synchronous/metachronous metastatic status (Figure 2). Objective response rates in both groups did not significantly differ, with 7 out of 46 patients (ORR 15.2%, 95% CI 4.8–25.6) in the PALBO–FULV group and 9 out of 28 patients (ORR 32.1%, 95% CI 14.8–49.4) in the EVE–EXE group (*p* = 0.0725). Accordingly, no difference in CBR was found between both groups (PALBO–FULV 65.2%, 95% CI 51.4–78.9 and EVE–EXE 67.8%, 95% CI 50.5–85.1, *p* = 1.0) (Figure 3). Only 1 patient experienced a complete response (CR) in the PALBO–FULV group (CR 2%, 95% CI < 1–6.3). Stable disease (SD) and progressive disease (PD) were 35.7% (95% CI 17.9–53.4) and 32.1% (95% CI 14.8–49.4), respectively, in the EVE–EXE group, and 50.0% (95% CI 35.5–64.4) and 34.7% (95% CI 21.0–48.5) in the PALO–FULV group, respectively.

**Figure 2.** Progression-free survival (PFS) after propensity score adjustment. EVE: everolimus; EXE: exemestane; PALBO: palbociclib; FULV: fulvestrant.

**Figure 3.** Objective response rate (ORR) according to RECIST 1.1 and clinical benefit rate (CBR). EVE: everolimus; EXE: exemestane; PALBO: palbociclib; FULV: fulvestrant; CI: confidence interval; *p*-value: chi-square.

#### *3.3. Safety*/*Adverse Events*

In terms of safety and adverse events, both treatments were relatively well tolerated (Table 3). In the PALBO–FUL group, neutropenia (65%) and anemia (41%) were the most commonly reported events, while in the EVE–EXE group, fatigue (64%), stomatitis (35%), and rash (25%) were the most reported adverse events (Table 3). Grades 3 and 4 adverse events (in the main afebrile neutropenia) occurred in 24 patients (52%) in the PALBO–FUL group, and 6 patients (21%) receiving EVE–EXE reported grade 3 adverse events (stomatitis and cutaneous rash and one case of interstitial pneumonitis). Dose reduction was required in 11 (24%) of patients in PALBO–FUL and 12 (43%) in the EVE–EXE group. Treatment discontinuations were all subsequent to disease progression, except in one case—a patient who developed interstitial pneumonitis while receiving EVE–EXE discontinued treatment. No deaths related to study medications were reported.


**Table 3.** Adverse events for any causes observed during the study period.

#### **4. Discussion**

Despite the clinically meaningful efficacy and good tolerability profile of the combination of CDK4/6i and ET in patients with MBC, patients eventually experienced disease progression and the emergence of resistance [16]. Resistance to CDK4/6i plus ET represents the next clinical challenge for the breast cancer community to overcome and requires a deep understanding of the mechanism of CDK4/6i resistance in an endocrine sensitive and resistant setting. Furthermore, there are limited data on the efficacy of these treatments in different BC histological types, in particular in patients with LBC who are often not well represented in clinical trials. While a subgroup analysis of PALOMA 2 trial showed that the combination of palbociclib plus letrozole was effective as a first-line treatment both in ductal and lobular histotypes, and in BOLERO-2, everolimus was shown to be effective both in ductal and lobular histotype hormone-refractory patients [17], no evidence is currently available on the efficacy of CDK4/6i as a second-line treatment based on histotype [18]. Thus, the treatment options for this frequent BC subtype are limited if tumors develop resistance to anti-estrogen treatment regimens.

In our clinical experience, the combination of PALBO–FUL in patients with metastatic LBC whose disease relapsed during adjuvant hormonal therapy or progressed after first-line ET for advance disease did meet the expectation. Most patients showed early disease progression and low clinical benefit [15]. We posed the question, why did patients with LBC show lower than reported responses to CDK4/6i? We know that the development and progression of invasive lobular carcinoma (ILC) are characterized by the loss of E-cadherin–E-cadherin binding in normal cells that prevents beta-catenin inhibition of PTEN, thus permitting the inhibition of AKT [19]. As a consequence of the loss of E-cadherin in LBC, the PI3K/AKT pathway is constitutively activated and represents one of the main pathways of proliferation and growth [20,21]. We hypothesized that in patients with metastatic LBC that becomes resistant to endocrine therapy, the hyperactivation of PI3K/AKT signaling may promote an intrinsic resistance to CDK4/6i through the activation of cyclin E/CDK2—amplification of cyclin E is the only factor that showed a correlation with resistance to a CDK 4/6i (palbociclib) in trials [22,23]. Alternatively, inhibition of the AKT pathway could perhaps represent a superior strategy for these patients.

Everolimus is a sirolimus derivative that inhibits mTOR (a key downstream point of the PI3K pathway) through allosteric binding to mTORC1. The results of the BOLERO-2 trial showed

that dual-blockade with EVE–EXE more than doubled median PFS versus EXE alone in patients with hormone receptor-positive (HR+)/human epidermal growth receptor 2-negative (HER2−) metastatic BC recurring/progressing on prior non-steroidal aromatase inhibitors (NSAIs) (7.8 versus 3.2 months) [24–26]. In addition, results of an Italian observational study suggest that treatment with EXE–EVE is an active and safe therapeutic option for endocrine-sensitive MBC patients in a real-world clinical setting, regardless of treatment lines [27]. These results were confirmed in the BALLET study that enrolled patients more heavily pretreated, with a safety profile consistent with that observed in BOLERO-2 [28]. These results are important because the treatment pattern of MBC is based on the sequence of multiple lines of therapy, and it is therefore vital to determine the possible additive/cumulative effects of different regimens. The combination regimen of EVE and EXE is the only regimen currently registered with an mTOR inhibitor in this setting and represents a valid alternative to the harmful toxicity profile of cytotoxic chemotherapy [29].

In our real-world analysis, median PFS was significantly longer for patients with metastatic LBC receiving EVE–EXE as second-line hormonal treatment compared with PALBO–FUL. Both treatments were well tolerated and only one patient (in the EVE–EXE group) discontinued therapy due to adverse events. Univariate analysis showed that prognosis may be influenced by disease status (de novo metastatic vs. relapsed disease), previous exposure to chemotherapy, and study treatment (PALBO–FUL or EVE–EXE). In particular, patients who had disease relapse and those who received a neo/adjuvant and/or first-line chemotherapy had shorter median PFS, suggesting that de novo metastatic and relapsed disease are characterized by different molecular background which for relapsed cancer is probably the result of clone selection derived from the exposure to previous treatments. The efficacy of chemotherapy in LBC is the subject of much debate, and it is usually reserved for patients with negative prognostic scores, visceral crisis, or when all possible ET lines have been exhausted. Most of our patients received cytotoxic agents (as neoadjuvant/adjuvant), which may have resulted in a detrimental effect, in particular when used in early lines. The PS analysis adjusted for previous chemotherapy exposure and synchronous/metachronous metastatic status confirmed a longer median PFS for patients receiving EVE–EXE (6.0 vs. 4.6 months, *p* = 0.04). Therefore, the activation of the PI3K/AKT pathway in LBC may result in intrinsic resistance to palbociclib after development of refractory disease to prior ET.

The results of this study allow us to speculate that EVE–EXE as a second-line treatment of metastatic LBC may improve therapeutic outcomes. In terms of optimizing sequential therapy, using a CDK4/6i for the first-line treatment of endocrine sensitive tumors is indicated, while mTOR inhibitors could be considered the preferred option when resistance to adjuvant/first-line ET has occurred. Activation of the PI3K/AKT pathway may not solely explain the lack of effectiveness of palbociclib in LBC, and other factors may drive resistance to Palbociclib [30]. Further studies are needed to explore the potential implications of these pathways in the mechanisms of resistance to CDK4/6i. In the era of personalized medicine, improving molecular characterization of cancer to define the best therapeutic program for each patient is paramount.

This retrospective real-world analysis generates the hypothesis of a potential benefit from EVE–EXE in comparison with PALBO–FUL as a second line hormonal-treatment for metastatic luminal breast cancer with lobular histology, and it allows us to speculate on the best therapeutic sequence. However, the limitations of this study, including its retrospective nature and small sample size, need to be addressed.

#### **5. Conclusions**

The results of this retrospective, real-world analysis seem to suggest a potential benefit of EVE–EXE in comparison with PALBO–FUL as a second-line ET of MBC with lobular histology.

**Author Contributions:** Conceptualization, A.O.; methodology, A.O., E.B., G.T.; writing—review and editing, all authors; supervision, A.O., E.B., G.T., and G.F.; All authors have read and agreed to the published version of the manuscript.

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

**Acknowledgments:** E.B. is currently supported by the Associazione Italiana per la Ricerca sul Cancro (AIRC) under Investigator Grant (IG) No. IG20583. G.T. is supported by AIRC, IG18599, AIRC 5 × 1000 21052. E.B. is currently supported by Institutional funds of Università Cattolica del Sacro Cuore (UCSC-project D1-2018/2019). Medical editorial assistance was provided by Edra S.p.A. Financial support for this assistance was provided by Novartis Farma S.p.A. Authors had full control of the content and made the final decision for all aspects of this article.

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

#### **References**


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

© 2020 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 (http://creativecommons.org/licenses/by/4.0/).

*Review*
