**3. Results**

Between March 2012 and October 2018, a total of 104 NSCLC patients received palliative surgery for spinal metastasis at the authors' institution. Among these patients, 19 were excluded from the analysis for the following reasons: (1) ten due to missing data on EGFR mutation analysis results, (2) five with an unidentified survival period or death, and (3) four who died within two weeks after surgery due to immediate postoperative complications (two pneumonia, one cardiac arrest, and one disseminated intravascular coagulation due to massive bleeding; Figure 1). As a result, 85 patients (58 males and 27 females) with a mean age of 60.9 (range, 32–81) years were analyzed in the current study. The characteristics of the study population are described in Table 2.


**Table 2.** Characteristics of the study cohort.

Seven patients were alive at the last follow-up, with a minimum follow-up period of 12 months, and the remaining 78 died during follow-up. The median postoperative survival period estimated by the Kaplan–Meier estimator was 6.4 months for the entire cohort (*n* = 85; Figure 2). Patients with a positive EGFR mutation had a significantly prolonged survival (*p* = 0.007), and those with synchronous metastasis tended to have longer survival (*p* = 0.101) than their counterparts in the log-rank test (Figure 3). According to the multivariate Cox proportional hazard model, the chronicity of spinal metastasis (hazard ratio (HR) = 1.88 (95% CI: 1.13. 3.12), *p* = 0.015), and EGFR mutation positivity (HR = 2.10 (95% CI: 1.30, 3.38), *p* = 0.002) were significantly associated with postoperative survival (Table 3). All predictors satisfied the proportional hazard assumption.

The Uno's C-index (discrimination ability) of NESMS was improved from 0.59 (95% CI: 0.54–0.65) to 0.62 (95% CI: 0.56–0.69), 0.64 (95% CI: 0.58–0.71), and 0.67 (95% CI: 0.61–0.74) when the chronicity of spinal metastasis, the EGFR mutation positivity, and both factors were added to the NESMS, respectively (Table 4). The improvement was statistically significant when the EGFR mutation positivity alone (adjusted *p* = 0.019) and both factors (adjusted *p* = 0.004) were added to the NESMS. The time-dependent AUC for predicting survival beyond 6 months postoperatively also increased from 0.63 (95% CI: 0.53–0.74) to 0.73 (95% CI: 0.64–0.82) when the two biological factors were added to the NESMS (adjusted *p* = 0.022; Table 5).

**Figure 2.** The Kaplan–Meier estimator graph for the total cohort.

**Figure 3.** Comparison of the Kaplan–Meier curve stratified by the biological factors.


**Table 3.** Results of the multivariable Cox proportional hazards model.

**Table 4.** The changes in the discrimination ability (Uno's C-index) of prognostic models by adding biological factors.


\**p*-value adjusted using the Bonferroni method.


**Table 5.** The changes in the prediction ability (time-dependent area under curve (AUC)) of prognostic models by adding biological factors.

> \* *p*-value adjusted by Bonferroni method.

#### **4. Discussion**

In the late 1990s, gefitinib, an oral EGRF TKI, was introduced as a molecular target therapy for NSCLC patients. A few years later, researchers identified EGFR mutations in NSCLC patients sensitive to gefitinib. Since then, genetic mutation analyses and corresponding molecular target therapies have been game-changers in the managemen<sup>t</sup> of NSCLC, improving the survival of patients with EGFR mutations [11]. Several previous studies have reported the clinical effects of EGFR mutation positivity and TKIs in NSCLC patients with skeletal [17] and spinal metastasis [18]. In the current study, patients with a positive EGFR mutation showed a significantly prolonged postoperative survival period compared to the EGFR mutation-negative group. The EGFR mutation positivity also significantly improved the discrimination (Uno's C-index) and prediction ability (time-dependent AUC at 6 months postoperatively) of a novel prognostic model—the NESMS. These results signify the importance of considering biological profiles in the decision-making process for spinal metastasis.

The timing of diagnosis of spinal metastasis, or the chronicity of spinal metastasis, was considered an additional biological factor in this study, which was significantly associated with postoperative survival. In previous studies, not only postoperative survival but also overall survival, was prolonged in patients with spinal metastasis as the initial manifestation of malignancy (synchronous metastasis) [3,12]. From the standpoint of tumor genetics, these findings can be related to the acquired resistance to first-line (first and second generation) TKIs. Common mechanisms for acquired resistance to TKIs, which usually develop within 12 months after TKI usage [13], are mutations in 20 exons (threonine-to-methionine substitution on codon 790, T790M) and MET oncogene amplification [19,20].

In our series, 7 (18.4%) of the 38 patients in the EGFR mutation-positive group showed a mutation in exon 20 (T790M) later in their disease course, which was not present in the initial molecular analysis. Five of these seven patients had metachronous spinal metastasis, and their exon 20 mutations were found in specimens obtained from the spine surgery. For these patients, third generation TKI (simertinib) or cytotoxic chemotherapy was considered after spinal surgery, and a shorter life expectancy was anticipated. This effect of acquired resistance to a TKI in metachronous metastasis patients was reflected in our finding that the time-dependent AUC 6 months postoperatively was significantly increased when both factors (EGFR mutation and chronicity) were added to the prognostic model (*p* = 0.022) and not when only EGFR mutation positivity was added (*p* = 0.096). As not all patients in our series underwent additional biopsies and molecular analyses during their disease course, the exact number of patients with acquired resistance to TKI in the metachronous metastasis group cannot be derived. Nevertheless, acquired resistance to TKIs can be associated with shortened survival in metachronous metastasis patients, and therefore, the chronicity of spinal metastasis should be considered as a significant biological factor (Figure 4).

**Figure 4.** Comparison of Kaplan–Meier curve stratified by the biological factors. An illustrative case of acquired resistance to tyrosine kinase inhibitor (TKI) in an epidermal growth factor receptor (EGFR) mutation-positive non-small cell lung cancer (NSCLC) patient. (**A**,**B**) A 53 years-old male with lung adenocarcinoma in right upper lobe. EGFR mutation analysis from the lung specimen showed a microdeletion mutation in exon 19. (**C**) After 2 years of systemic treatment with multiple regimens including TKI (gefitinib), the patient was diagnosed with multiple spinal metastasis with spinal cord compression at T7 and T12. (**D**) The patient underwent a palliative decompression and stabilization, and EGFR mutation analysis from a spine specimen revealed a missense mutation of EGFR gene exon 20 (T790M). The patient expired 4 months postoperatively due to disease progression.

We examined the discrimination and prediction ability of the NESMS, a novel and prospectively validated prognostic model, in this study (Table 1). In this system, the primary tumor is stratified according to the modified Bauer score. As all patients in our series had lung adenocarcinoma, the modified Bauer score was 0 for all patients. Therefore, after eliminating the most significant factor from the NESMS, the remaining factors for the decision-making process are ambulatory function and serum albumin. In this setting, if there are two different NSCLC patients with ambulatory status and serum albumin falling into the same category, the decisions for two patients would be the same according to the NESMS, even if the two have significantly different biological profiles (e.g., synchronous metastasis with a positive EGFR mutation versus metachronous metastasis without EGFR mutation). This novel "classification-based" decision-making system, the NESMS, may be useful and straightforward when all spinal metastasis patients with diverse primary cancers are combined; however, its discrimination ability seems to be significantly limited for individual cancers.

We examined the discrimination and prediction ability of the NESMS, a novel and prospectively validated prognostic model, in this study (Table 1). In this system, the primary tumor is stratified according to the modified Bauer score. As all patients in our series had lung adenocarcinoma, the modified Bauer score was 0 for all patients. Therefore, after eliminating the most significant factor from the NESMS, the remaining factors for the decision-making process are ambulatory function and serum albumin. In this setting, if there are two different NSCLC patients with ambulatory status and serum albumin falling into the same category, decisions for two patients would be the same according to the NESMS, even if the two have significantly different biological profiles (e.g., synchronous metastasis with a positive EGFR mutation versus metachronous metastasis without EGFR mutation). This novel "classification-based" decision-making system, the NESMS, may be useful and straightforward when all spinal metastasis patients with diverse primary cancers are combined; however, its discrimination ability seems to be significantly limited for individual cancers.

It is obvious that a prognostic model's performance will improve if more prognostic factors are added to it. However, adding too many factors can make a prognostic model complicated and difficult to use in the clinical setting. Therefore, it is essential to prioritize prognostic factors according to their weights in multivariate logistic or proportional hazard regression analyses. Factors with higher odds or hazard ratios should be incorporated into the system. In our study, a multivariate Cox proportional hazard model (backward stepwise with likelihood ratio test) yielded a higher hazard ratio for EGFR mutation positivity (HR = 2.27 (95% CI: 1.41, 3.66), *p* = 0.001) than ambulatory status (HR = 2.26 (95% CI: 1.29, 3.95), *p* = 0.004) and serum albumin (HR = 1.71 (95% CI: 0.96, 3.02), *p* = 0.068), which are the main components of the NESMS. These results also emphasize the importance and necessity of adding biological factors as modifiers in the decision-making systems for spinal metastasis.

Among the various decision-making systems reported in the literature, there have been efforts to incorporate biological factors into these systems. In 2014, Katagiri et al. introduced a revised version of their prognostic system for spinal metastasis, in which the application of molecular target therapy was considered when stratifying the patient's primary tumor [10]. In their system, lung cancer treated with molecular target therapy was classified as a moderate-growth tumor, while lung cancer without available molecular target therapy was classified as a rapid-growth tumor. Efforts to incorporate biological factors into decisionmaking systems, as shown in the revised Katagiri system, are anticipated to be the future trends in the managemen<sup>t</sup> of spinal metastases.

In this study, we stratified patients by EGFR mutation positivity rather than by the treatment they received (e.g., TKI versus platinum-based chemotherapy), as in a previous study [18]. The most important reason for choosing this categorization is that the EGFR mutation profile, rather than the type of postoperative systemic treatment the patient will receive after surgery, is more available at the time of decision-making for spinal metastasis surgeries. As the purpose of this study was to verify the prognostic value of biological factors and not to compare the treatment outcomes, our categorization seems to be more appropriate. Another reason is the diversity of systemic treatment that a patient with NSCLC receives after surgery, as well as the start point and duration of these treatments. In our series, 41 (48.2%) patients received a combination of molecular target therapy and cytotoxic chemotherapy, whereas only 14 (16.5%) received molecular target therapy alone postoperatively, regardless of EGFR mutation positivity. In addition, the molecular target therapies used in our study patients ranged from first to third generation EGFR TKIs (gefitinib, erlotinib, afatinib, and osimertinib), EGFR monoclonal antibody (cetuximab), anaplastic lymphoma kinase (ALT) inhibitors (crizotinib), mesenchymal-epithelial transition (MET) inhibitors (savolitinib, capmatinib), and PD-1 inhibitors (avelumab, nivolumab, and pembrolizumab). Therefore, it would be impossible and meaningless to stratify patients by postoperative systemic treatment, given the diversity of mechanisms and the treatment effects of these agents.

There are several limitations in the current study. First, because of its retrospective nature, selection bias regarding the inclusion and exclusion criteria cannot be ruled out. Second, there is a possibility that the differences in surgical aggressiveness between individual cases may have influenced the patients' prognosis and survival, such as the case described in Figure 4 [21,22]. However, this possible effect of surgical strategy on patients' outcomes was not considered in the analysis. Third, because this study included only lung adenocarcinoma patients, our results cannot be generalized to spinal metastases of various primary cancers. Finally, and most importantly, because we did not aim to develop a new prognostic model in this study and include all relevant prognostic factors in the analysis, we cannot perform any validations, including calibrations, on our results. We also cannot sugges<sup>t</sup> how to incorporate biological factors into the decision-making systems as a modifier, which is well beyond the current study's scope. Despite these limitations, the results of this study provide valuable information for state-of-the-art care for patients with

spinal metastasis, and sugges<sup>t</sup> future directions for the development of decision-making systems for spinal metastasis.

## **5. Conclusions**

EGFR mutation positivity and the chronicity of spinal metastasis provide additional prognostic value for NSCLC patients with spinal metastasis. These results signify the importance of considering biological profiles in the decision-making process for spinal metastasis.

**Author Contributions:** Conceptualization, H.K., S.Y.C., and B.-S.C.; methodology, H.K., S.Y.C., and J.S.; validation, H.K., S.M., and S.C.P.; formal analysis, S.Y.C., S.C.P., and B.-S.C.; investigation, H.K., S.Y.C., and J.S.; resources, S.M. and S.C.P.; data curation, H.K., S.M., and B.-S.C.; writing—original draft preparation, H.K., S.Y.C., and B.-S.C.; writing—review and editing, all authors; visualization, J.S., S.M., and S.C.P.; supervision, B.-S.C. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Seoul National University Hospital. (IRB No. 2009-060-1155).

**Informed Consent Statement:** Patient consent was waived due to retrospective nature of the study.

**Data Availability Statement:** All relevant raw data from the data presented in the manuscript or the supplementary figures and tables are available from the authors of the study upon request.

**Acknowledgments:** The authors appreciate the statistical consultation provided by the Medical Research Collaborating Center at the Seoul National University College.

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