A Comparison of Clinicopathologic Outcomes Across Neoadjuvant and Adjuvant Treatment Modalities in Resectable Gastric Cancer

This comparative effectiveness study assesses the association of various neoadjuvant and adjuvant treatments for resectable gastric cancer with pathologic complete response, surgical margin status, and overall survival.


Introduction
Treatment of resectable gastric cancer (RGC) uses a multimodal approach, and significantly improving outcomes for patients remains challenging given the aggressive nature of the disease.
Randomized clinical trials have found improved progression-free survival and overall survival (OS) when comparing perioperative chemotherapy vs surgical treatment alone, 1-3 adjuvant chemotherapy vs surgical treatment alone, 4,5 and adjuvant chemoradiotherapy vs surgical treatment alone 6 ; however, large head-to-head analyses comparing each of these treatment regimens and combinations have not been performed, to our knowledge. Although perioperative chemotherapy with 5-fluorouricil, leucovorin, oxaliplatin, and docetaxel (FLOT) has become the standard of adjunctive therapy for RGC, 7 it remains unclear whether alternative combinations of neoadjuvant and adjuvant chemotherapy with or without radiation may be better than perioperative chemotherapy alone. [8][9][10] Owing to the lack of data regarding the optimal treatment strategy for RGC, there is controversy among major guidelines regarding use and timing of each treatment modality and there is currently no global standard of care. [11][12][13][14] Accordingly, we evaluated the association of various combinations of neoadjuvant chemotherapy, adjuvant chemotherapy, and radiation with outcomes in the treatment of gastric cancer that was cT2-T4b, any N, and M0. Through a modality-by-modality approach, we compared clinical and pathologic factors for each treatment combination across 3 end points, including pathologic complete response (pCR), surgical margin status (SMS), and OS, to investigate the optimal treatment strategy for RGC.

Methods
Given that all patient identification variables were removed, Cedars-Sinai Medical Center determined that institutional review board review and informed consent were not needed. The International Society for Pharmacoeconomics and Outcomes Research (ISPOR) reporting guideline was followed by formulating a prespecified research question and analysis and reporting the results in a standardized method. 15

Patient Selection
Patients with RGC diagnosed from 2004 to 2015 were found through a deidentified National Cancer Database (NCDB) file (eFigure in the Supplement). Among 183 204 initial patients with gastric cancer, we screened for patients with cancer that was stage cT2-T4b, any N, and M0 and who underwent definitive surgical treatment. Patients who had nonadenocarcinoma histology, M1 or unknown M status, or absent or unknown carbohydrate antigen 19-9 and carcinoembryonic antigen levels were excluded. From the remaining 23 021 patients with RGC, we excluded 17 575 patients because they had no vital status for OS follow-up, unknown surgical treatment status, unknown local therapy status, or unknown chemotherapy administration status. Among the remaining 5446 patients with RGC, we excluded 2382 patients who were missing data to assign to a treatment group (defined in a later section) to reach a final study cohort of 3064 patients with RGC. Owing to missing data for each primary outcome measure, several analyses used smaller sample sizes than the total number of patients with RGC. For analyses of pCR, aCT and aCRT were excluded.
Race and ethnicity classification was determined by the NCDB. Race and ethnicity were assessed to evaluate whether this patient demographic was associated with treatment modality, similar to other comparisons using other demographics, such as age, sex, and income ( Table 1). The other race and ethnicity group was based on NCDB classification, and specific groups included and reasons why race and ethnicity were combined were not provided.

Treatment Groups and Primary End Points
We stratified our RGC cohort into 9 treatment groups or treatment timeline groups to assess their association with primary end points of pCR, SMS (ie, positive margins), and OS. The following distinct treatment timeline groups were included: neoadjuvant chemoradiation only (nCRT), neoadjuvant chemotherapy only (nCT), adjuvant chemotherapy only (aCT), adjuvant chemoradiation only (aCRT), neoadjuvant chemotherapy and adjuvant radiation (nCTaRT), chemotherapy with timing unknown (CTTU), chemoradiation therapy with timing unknown (CRTTU), radiation therapy with timing unknown (RTTU), and no perioperative therapy (NT). For neoadjuvant therapy-only groups (ie, nCRT and nCT) and adjuvant therapy-only groups (ie, aCT and aCRT), all patients who received chemotherapy with or without radiation therapy prior to or after definitive surgical treatment, respectively, were included irrespective of sequence of modalities (ie, sequenced or concurrent chemotherapy and radiation therapy) and time from surgical procedure. For the nCTaRT subgroup, chemotherapy could have occurred any time prior to surgical treatment (ie, neoadjuvant treatment) and radiation therapy any time after surgical treatment (ie, adjuvant treatment). The subgroups CTTU, CRTTU, and RTTU included patients for whom the NCDB did not clearly specify when that specific treatment modality was given during the treatment course for RGC but stated that the treatment modality was given at some point along with definitive surgical treatment. For example, CTTU comprised patients who received some form of chemotherapy with unspecified timing; therefore, this subgroup may have included patients receiving nCT, aCT, or both (ie, perioperative treatment). The NT subgroup defined patients for whom RGC was treated with a surgical procedure only.
The primary end points were pCR, SMS, and OS, which was calculated from diagnosis to the date of death or censored at last follow-up. The main estimator variable was treatment group.

Statistical Analysis
Analyses were performed using R statistical software version 4.0.5 (R Project for Statistical Computing) with 2-sided tests at a significance level of P = .05. The missing data pattern for variables with missing values was examined using the Little method. 16 To decrease risk of bias from missing data, missing values were imputed with respect to other variables as estimators using fully conditional specification implemented by the multivariate imputation by chained equations algorithm using the Mice package under the missing-at-random assumption, with 40 imputed data sets corresponding to a maximum loss of efficiency of 5%. [17][18][19] We analyzed 40 complete data sets separately, and the results were combined using the Rubin formula. 20 Data were analyzed from September 2019 through February 2020.
Baseline characteristics were compared between treatment groups with Kruskal-Wallis test for age and χ 2 test or Fisher exact test for categorical variables. Logistic regression models were employed to estimate the association of treatment group with pCR and SMS with and without adjustment for baseline characteristics. Univariate and multivariable analyses of OS were performed using Cox proportional hazards models. 21 The proportional hazards assumption was assessed with scaled Schoenfeld residuals. 22 Median follow-up was calculated using the reverse Kaplan-Meier method. 23 Survival functions were estimated by the Kaplan-Meier method and compared using log-rank tests. 24 Post hoc pairwise comparisons between treatment timeline groups in their associations with each end point were further performed, and P values were adjusted for multiple tests using the Holm procedure. 25     1.54-3.91; P = .006) were associated with increased odds of SMS compared with nCT; no other comparisons reached statistical significance (Table 3).
In multivariable analysis among 3061 patients (owing to missing data for OS), treatment group was associated with OS, but baseline sociodemographic, clinical, or pathologic variables were not ( Figure;

Discussion
In this comparative effectiveness research study of RGC, we found that nCRT and nCT were   found an increased rate of overall survival for nCRT, but this increase was not statistically significant (HR, 0.65; 95% CI, 0.42-1.01; P = .055). 29 However, the POET trial was limited to patients with cancers of the lower esophagus and gastric cardia; therefore, it is unclear how this can be extrapolated to all gastric cancers. 29 Retrospective studies comparing nCRT with nCT for gastric cancers, specifically, found similar data regarding improved pCR rates of nCRT compared with nCT, but there were no statistically significant improvements in OS. 30 variables had the highest ORs for pCR, it is not surprising that nCRT, nCT, and RTTU also had the lowest ORs for SMS. We found that nCTaRT and aCT were associated with increased odds of having a surgical margin, which is consistent with positive margins being an indication for adjuvant therapy. 34 Additionally, we found that CTTU was shown to be the best estimator associated with OS, with the lowest HR for mortality, and had the longest median survival. This finding is not inconsistent with those of the aforementioned studies, which found improved OS with nCT and perioperative chemotherapy, and several phase III trials have found an overall survival benefit for adjuvant chemotherapy compared with surgical treatment alone 4,5 and adjuvant CRT compared with surgical treatment alone. 6 The Multicenter Randomized Phase III Trial of Neoadjuvant Chemotherapy Followed by Surgery and Chemotherapy or by Surgery and Chemoradiotherapy in Resectable Gastric Cancer (CRITICS), 35 which compared preoperative chemotherapy followed by D2 surgical treatment and aCT or aCRT, also found that aCT improved OS compared with aCRT. Our finding that the receipt of CTTU (which includes neoadjuvant, adjuvant, and perioperative chemotherapy) was associated with the greatest OR for survival suggests that chemotherapy should remain as the core pillar in the multimodal approach of RGC treatment.

Limitations and Strengths
This study has several limitations. The major limitation of this study is selection bias given that treatment regimens were selected for patients based on age, performance status, and tumor characteristics. Adjuvant treatment strategies are also influenced by preoperative treatments and surgical outcomes. Therefore, selected treatment regimens inherently introduce bias because more intense approaches are selected for patients who are medically fit, which may be associated with improved outcomes. In addition, pairwise comparison of OS by treatment group was limited by a small sample size of 23 patients for RTTU. That small sample size was also a limitation in survival time comparisons; however, we recognize that RT alone is not a standard paradigm in the perioperative treatment of RGC with definitive surgical treatment planned. Additionally, the study included only patients who underwent surgical treatment, which may overestimate the association of neoadjuvant treatment with outcomes because individuals who progressed on neoadjuvant treatment were excluded. Another limitation of the study is the retrospective nature of the NCDB, which includes missing variables and incomplete information about treatment regimens and did not allow for standardization of treatment regimens. Despite these limitations, the strengths of this study are the large number of patients included and the comprehensive statistical analysis, which adjusted for multiple variables.

Conclusions
To our knowledge, this study is the first to compare the effectiveness of the various treatment modalities for RGC by using a modality-by-modality approach. Although we were unable to tease out whether the neoadjuvant or adjuvant portion of patients receiving chemotherapy with timing unknown was associated with the greatest increase in OR for OS, both groups may be associated with OS outcomes given that perioperative FLOT is now widely recognized as a standard in RGC.
This retrospective study, which used a modality-by-modality approach to compare treatments for RGC, found that nCT and nCRT were associated with pCR. We also found that nCRT had increased ORs for pCR compared with nCT. These findings suggest that nCRT may be more effective than nCT in treating tumor burden and may be associated with improved OS. These results support the ongoing TOPGEAR trial, which is investigating the safety and effectiveness of nCRT vs nCT and whether these results lead to differences in OS. Additionally, our study suggests the importance of chemotherapy (including neoadjuvant and adjuvant therapy) as the standard of treatment in the multimodal approach for RGC based on the association of chemotherapy with increased OS.