The association of survival with adjuvant chemotherapy in patients with ypTNM stage I gastric cancer after neoadjuvant chemotherapy.

Introduction
As an aggressive malignancy with a poor prognosis, gastric cancer (GC) is one of the leading causes of cancer-related death worldwide [1]. With the increasing use of upper gastrointestinal endoscopic screening and improvements in diagnostic capabilities, more patients are being diagnosed at earlier stages. As a result, early-stage GC has become more prevalent in East Asian countries, such as Japan and South Korea [2]. Although the proportion of early-stage GC in China has been gradually increasing, advanced gastric cancer (AGC) still represents a significant burden due to factors such as lower rates of endoscopic screening and dietary habits, with approximately 80% of GC cases being diagnosed at an advanced stage [3].
The role of neoadjuvant chemotherapy (NAC) in improving both short- and long-term outcomes for patients with locally advanced gastric cancer (LAGC) has been well-established through multiple randomized controlled trials (RCTs) [4–6]. In these trials, all patients who received NAC also underwent postoperative adjuvant chemotherapy (AC), a treatment strategy that is currently the standard of care and is included in clinical guidelines [7]. Recent developments in immune checkpoint inhibitors (ICIs) and targeted therapies have further optimized perioperative treatment regimens, leading to significant tumor downstaging [8]. Consequently, an increasing number of patients are being reclassified as ypTNM stage I, even achieving pathological complete response (pCR) [9]. Current guidelines generally recommend that patients with pathological stage IA GC do not require postoperative AC [7, 10]. Moreover, a multicenter study has suggested that postoperative AC does not significantly improve survival outcomes for patients with stage IB GC [11]. However, the necessity of postoperative AC for patients with ypTNM stage I GC after NAC remains a subject of ongoing debate.
Given the limited research and the controversial conclusions surrounding this issue, we conducted this multicenter study based on the SEER database, utilizing various propensity score (PS) methods such as inverse probability of treatment weighting (IPTW) to balance covariates. The aim of this study was to explore the correlation between postoperative AC and long-term survival in patients with ypTNM stage I GC following NAC. Additionally, we sought to identify prognostic factors and develop nomograms to provide evidence-based recommendations for clinical decision-making regarding the necessity of AC in this patient population.

Methods

Patient population
Data on GC patients were extracted from the SEER database (SEER 17 registries, November 2023 data submission, 2000–2021) of the National Cancer Institute, using SEER*Stat software (v8.4.3). Importantly, the SEER database strictly upholds patient confidentiality; therefore, data analysis for did not require obtaining informed consent from participants. This study has been approved by the Ethic Committee of the Chinese PLA General Hospital (No: S2024-498-01). This work has been reported in line with the strengthening the reporting of cohort, cross-sectional and case–control studies in surgery (STROCSS) criteria [12].
A cohort of patients with resected ypTNM stage I GC, diagnosed between 2010 and 2021, who received systemic chemotherapy either preoperatively or both preoperatively and postoperatively, was enrolled in the study. According to the exclusion criterias, the following groups of patients were excluded: patients with pathologically confirmed diagnoses of neuroendocrine carcinoma (ICD-O-3: 8246/3), stromal sarcoma (8935/3), or gastrointestinal stromal tumor (GIST) (8936/3), as well as those with missing tumor size data or inconsistent or unspecific information regarding radiotherapy. After applying these inclusion and exclusion criteria, the final dataset consisted of the entire cohort of eligible patients (Fig. 1).

Study variables
Information on age at diagnosis, sex, tumor size, primary site, tumor differentiation, histologic type, ypTNM stage, ypT, ypN and treatment of GC patients were collected from SEER database. Age was assessed in two categories (≤ 60, > 60 years) [13]. Tumor size was categorized as ≤ 53 mm and > 53 mm, determined by using “maxstat” function in R package “survminer” according to overall survival (OS) (Figure S1). Primary site was divided into proximal (ICD-O-3 topography code C16.0, C16.1) and other part based on SEER site recode variables. Prior research has demonstrated marked prognostic heterogeneity between signet ring cell gastric cancer and other histological subtypes [14]. Histologic type was characterized as signet-ring carcinoma yes (ICD-O-3 histologic codes 8940) and ypTNM stage categories were redefined based on American Joint Committee on Cancer (AJCC) Staging Manual, 8th edition.

Outcomes and follow-up
The primary endpoint was OS, defined as the time in months from the date of diagnosis to the date of death, the date last known to be alive, or the study cut-off date. Cancer-specific survival (CSS) was defined as the time from diagnosis to death due to GC. Median follow-up time was calculated with the reverse Kaplan–Meier method. The median follow-up periods were 54 months (95% confidence intervals, 44–72 months).

Nomogram development
Combined with clinical significance and multivariable regression analysis, including age, sex, tumor diameter, tumor differentiation, ypTNM and primary site in the final prediction model. The prediction model was implemented into nomograms. The survival prediction model was developed based on patients with AC. It was validated by measuring both discrimination and calibration. Discrimination was evaluated using the concordance index (C-index). Calibration was evaluated with a calibration curve. Decision curve analysis (DCA) was conducted to evaluate the clinical utility of the nomogram.

Statistical analysis
We used IPTW to reduce imbalance in measured confounders between treatment groups, which has already been proved as an effective approach in multiple clinical retrospective studies [15–18]. A multivariable logistic regression model was used to estimate the PS of individuals receiving AC. Covariate balance was assessed using standardized mean difference (SMD), with values ≤ 0.1 indicating acceptable balance.
The chi-square test was applied to analyze the baseline clinical characteristics between the Non-AC and AC groups. IPTW-weighted Kaplan-Meier method was employed to generate OS and CSS curves, and the standard or IPTW-weighted log-rank tests were conducted to evaluate differences in survival outcomes. The Cox proportional hazards models were utilized to assess whether the use of AC improved OS and CSS. The prognostic factors were identified using univariate and multivariate Cox proportional hazards regression models. All statistical analyses were conducted using R version 4.2.3, with two-sided p-values of < 0.05 considered statistically significant.

Results

Demographic and clinicopathological characteristics
This study included 661 patients with ypTNM stage I GC, of whom 209 had T1N0M0, 367 had T2N0M0, and 85 had T1N1M0 disease. After conducting a correlation analysis for two groups before IPTW, it was discovered that tumor primary site (P < 0.001) was significantly associated with AC.
A PSM model was used to reduce confounders of the two treatment strategies. The distribution of all included variables was adequately balanced after IPTW adjusted by all potential confounders (Table 1). Finally, the clinicopathological characteristics were comparable, with no significant difference after IPTW. Figure S2 shows the covariate balance and distribution of PS after matching.

Relationship between systemic adjuvant chemotherapy with survival
The 5-year OS rate in all enrolled patients was 74.1% (95% CI, 70.2%–78.2%), and the 5-year CSS rate was 81.3% (95% CI, 77.8%–85.0%). The median OS and CSS were not reached during the follow up. HR estimates from both unweighted and IPTW Cox regression analyses are presented in Table 2; Fig. 2. Overall, the results indicated a statistically significant OS benefit associated with receiving AC before and after IPTW adjustment (weighted HR = 0.63; 95% CI, 0.43–0.93).

After IPTW adjustment, multivariate analysis revealed that AC was significantly associated with improved OS (P = 0.0155; HR 0.63; 95% CI, 0.43–0.92). Additionally, age ≤ 60 years and tumor diameter ≤ 53 mm were identified as independent protective factors for OS. In contrast, male and proximal GC emerged as independent risk factors for OS (Fig. 3). Tumor diameter and primary site were also identified as independent prognostic factors for both OS and CSS (Table S1, S2). Multiple additional sensitivity analyses, including 1:2 PSM, multivariate analysis adjusted for PS and landmark analysis showed similar result (Figure S3, Table S3).

Subgroup analysis
Based on the multivariate Cox regression results, we performed subgroup analyses on independent prognostic factors. The findings indicated that the use of AC in patients with proximal GC did not significantly affect outcomes. In contrast, AC use in patients with non-proximal GC significantly improved prognosis, as demonstrated by OS (P = 0.0082, HR = 0.53; 95% CI, 0.33–0.85) and CSS (P = 0.041, HR = 0.55; 95% CI, 0.31–0.98) (Fig. 4a, b).

Further analysis of other clinicopathological subgroups revealed that patients older than 60 years who received AC experienced a significant improvement in OS (P = 0.0372, HR = 0.62; 95% CI, 0.40–0.97) (Fig. 4c, d). However, no significant differences were observed with AC use in the remaining subgroup analyses (Figure S4).

Development of the prediction model to guide adjuvant chemotherapy decision making
To estimate individualized risk, multivariable Cox regressions were conducted for patients who underwent AC. The first nomogram (Fig. 5a) estimates the predicted 3-year and 5-year OS with AC, and the second nomogram (Fig. 5b) estimates 3-year and 5-year CSS with AC. A total score was calculated based on each patient clinical characteristic score, and their corresponding 3 and 5-year survival rates were obtained.

The C-indexes for OS were 0.75 (95% CI, 0.66–0.83) in the group treated with AC and 0.79 (95% CI, 0.69–0.89) in CSS. Area under time-dependent receiver operating characteristic curve (time‐dependent AUC) of the nomograms for 5-year OS and CSS prediction were 0.767 (95% CI, 0.671–0.864) and 0.797 (95% CI, 0.691–0.903), respectively (Fig. 5c, d). The calibration curve showed good agreement between the predicted and observed clinical outcomes. In addition, DCA confirmed the nomogram’s clinical utility in guiding AC decisions, and the corresponding results were displayed in Figure S5.

Discussion
Neoadjuvant therapy is associated with a higher rate of tumor downstaging in GC. However, there remains ongoing debate regarding the necessity of postoperative AC for patients with GC who achieve ypTNM stage I after receving NAC. While several studies have investigated perioperative chemotherapy for LAGC [5, 6, 19], limited research exists on the decision-making process and long-term survival outcomes for the specific group of patients with tumor downstaging to ypTNM stage I. Our study, which is the first to utilize the SEER multicenter database combined with the IPTW method to balance covariates, demonstrates that postoperative AC improves the prognosis for these patients. Additionally, we developed a prognostic model for patients with ypTNM stage I following NAC and postoperative AC, which provides valuable clinical insights for identifying treatment-sensitive populations.
Our study found that the addition of AC after NAC significantly improves OS in patients with ypTNM stage I GC (HR = 0.64, 95% CI, 0.43–0.94). Previous studies have suggested that AC does not improve prognosis in patients with pTNM stage I GC [11, 20], and the 6th Japanese Gastric Cancer Guidelines recommend postoperative observation rather than AC for these patients [7]. However, research by Xie et al. demonstrated that postoperative AC in stage IB GC patients is associated with a better CSS, suggesting that early-stage GC patients may benefit from AC [21]. Notably, these guidelines and clinical researches primarily addressed early-stage GC without considering patients who received NAC prior to surgery and were downstaged to early-stage disease. In a study on patients who received NAC and were downstaged to ypT1−2N0, with clinical T stages of T3−4 before chemotherapy, postoperative AC significantly improved prognosis (HR = 0.13, 95% CI, 0.05–0.35) [22], which was consistent with our findings. However, this study has a limitation with small sample size, whereas our study, utilizing a larger and multicenter database, offers more robust and convincing evidence. Therefore, our results demonstrate that in patients whose tumors are sensitive to NAC and acquire ypTNM stage I, the continuation of AC can consolidate treatment effects and improve prognosis.
Furthermore, through IPTW and multivariable Cox regression analysis, we identified several independent prognostic factors for ypTNM stage I GC, including primary site, tumor diameter, age and sex. Notably, the primary site was found to be associated with prognosis; tumors located in the proximal gastric had worse outcomes compared to those in other parts of the stomach, which is consistent with multiple previous studies [23, 24]. Building on this result, our subgroup analysis revealed that ypTNM stage I GC patients with tumors located in non-proximal stomach benefited more from AC, while in proximal GC, the use of AC did not show a significant difference in improving prognosis. Subgroup analysis from the PRODIGY clinical trial showed significant survival benefits for perioperative chemotherapy in the subgroup of tumors located in non-proximal stomach (HR = 0.70, 95% CI, 0.27–0.91) [6], aligning with our findings. This difference in therapeutic response may be attributed to the distinct morphology, clinical behavior, and therapeutic responses of proximal GC compared to tumors originating in other parts of the gastric, suggesting that proximal GC may present a relatively independent characteristic [25, 26]. Additionally, statistical data indicate that with the increasing control of Helicobacter pylori infection, the incidence of GC is declining, whereas the incidence of proximal GC, driven by factors such as chronic gastroesophageal reflux disease (GERD) and obesity, is on the rise [27]. The differences in treatment strategies for these distinct etiologies warrant further investigation. Larger tumors may reflect a more aggressive biological phenotype, with deeper invasion, increased lymphovascular involvement, and a higher likelihood of both regional and potential distant microscopic metastasis. These characteristics can contribute to poorer survival outcomes in patients with larger tumors [28].
Age is another independent prognostic factor in ypTNM stage I GC. Compared to patients under 60 years old, those over 60 have a poorer prognosis. However, subgroup analysis revealed that receiving AC improves prognosis in patients over 60 years old. Among elderly patients with LAGC, an increasing proportion are receiving NAC, which has led to improved outcomes [29, 30]. Additionally, while the 5-year OS rate is significantly lower in elderly GC patients compared to younger patients, there is no significant difference in 5-year CSS between the two groups. This result suggests that non-tumor-related factors, rather than the cancer itself, may play a more important role in affecting long-term survival in elderly ypTNM stage I GC patients. Therefore, it is essential to consider the management of comorbidities and non-cancer-related diseases in this population [31].
Another contribution of this study is the development of a prognostic model to predict long-term survival in patients with ypTNM stage I GC following NAC. We incorporated clinicopathological factors into regression models to establish and validate nomograms. While prior prognostic models for early-stage GC based on clinicopathological factors already exist, our model demonstrates comparable predictive performance, as assessed by various indicators, including the C-index and time-dependent AUC [32–34]. However, unlike previous models, ours specifically focuses on GC patients who were ypTNM stage I after NAC and subsequently received AC. This model enables the prediction of individualized benefits from AC. The nomograms incorporate risk factors that can be easily obtained from routine hospitalization records. By utilizing variable scoring, clinicians can accurately predict patient outcomes and assess the effectiveness of treatments.
Our study has several limitations. Although the SEER database provides a large population-based sample, it lacks some key variables, such as CEA and CA72-4, which have been reported to correlate with GC prognosis [35–37]. In addition, detailed treatment information, including specific chemotherapy regimens, dose intensity, and treatment duration, could not be obtained. Moreover, SEER database does not provide preoperative clinical TNM (cTNM) staging data, making it impossible to assess the degree of downstaging following NAC. Furthermore, key pathological features such vascular or lymphatic invasion, and pathological treatment response were not recorded. These missing variables may influence both treatment decisions and survival outcomes. Although PSM and sensitivity analyses were applied to reduce bias, residual confounding inherent to the retrospective design cannot be completely excluded.
Another limitation is that the SEER database does not provide information regarding the quality of surgical treatment, such as surgical methods, extent of lymphadenectomy, or perioperative complications; therefore, the potential impact of surgical quality on survival outcomes could not be evaluated. Moreover, due to sample size limitations, this study did not explore the long-term prognosis and the relationship with AC in patients who achieved pCR, a group with significant therapeutic response. With the advent of immunotherapy, an increasing number of patients are achieving higher rates of pCR, future large-scale studies are needed to assess the prognostic implications for this population. Finally, given the small sample size for the prognostic model, we did not perform a training-test split to ensure predictive capacity [38]. External validation using larger multicenter real-world datasets is warranted to confirm the generalizability of our findings.

Conclusion
In conclusion, to the best of our knowledge, our study is the first to suggest that the addition of AC after NAC may provide a survival benefit for patients with ypTNM stage I GC. However, the administration of AC might not confer a statistically significant survival advantage for ypTNM stage I GC located in the proximal stomach. Furthermore, we developed a novel prognostic model that offers a new approach for predicting long-term survival in ypTNM stage I GC patients.

Supplementary Information