Neoadjuvant FLOT versus SOX chemotherapy in locally advanced gastric cancer: secondary outcomes of a single-centre, open-label, randomised, exploratory phase 2 trial.

Introduction
For millions of patients diagnosed with locally advanced gastric cancer (LAGC) worldwide, selecting an optimal neoadjuvant chemotherapy regimen represents one of the most consequential treatment decisions in modern oncology. Gastric cancer remains the fourth most common malignancy and second leading cause of cancer-related mortality globally, with particularly high incidence rates in East Asian countries.1
The perioperative chemotherapy landscape transformed dramatically following landmark trials demonstrating clear survival benefits over surgery alone.2, 3, 4, 5 The German FLOT4 trial revolutionised treatment by establishing FLOT (5-fluorouracil, leucovorin, oxaliplatin, and docetaxel) as superior to ECF/ECX regimens, with significant improvements in both pathological response and overall survival.6,7 This led to FLOT adoption as the preferred regimen in Western guidelines.8,9
However, a parallel evolution occurred in East Asian countries, where different patient populations and healthcare systems favoured oral fluoropyrimidine-based approaches. The SOX regimen (S-1 plus oxaliplatin) emerged as a compelling alternative, with large-scale Chinese trials (RESOLVE and RESONANCE) suggesting comparable efficacy to triplet regimens while maintaining favourable tolerability.10,11 SOX offers practical advantages: reduced infusion time, fewer cycles, and potentially lower rates of severe neutropenia.12
The treatment landscape continues evolving with immunotherapy integration. The phase 3 DRAGON IV trial demonstrated that adding camrelizumab and rivoceranib to SOX significantly improved pathologic complete response.13 However, current guidelines show limited evidence for neoadjuvant immunotherapy outside specific biomarker-selected populations (dMMR/MSI-H). The Chinese Society of Clinical Oncology (CSCO) 2023 guidelines note that for neoadjuvant immunotherapy, ‘there is no sufficient evidence for standard recommendations,' and recommend prioritizing clinical trial participation.14 Similarly, the Japanese Gastric Cancer Association (JGCA) 2021 guidelines state there is ‘no clear recommendation for neoadjuvant chemotherapy' for resectable advanced gastric cancer.15
Real-world data shows immunotherapy usage in only 14.56% of neoadjuvant treatments,16 highlighting that chemotherapy backbone selection remains the predominant clinical decision for most patients worldwide.
Despite widespread clinical use of both regimens, a critical knowledge gap persists: direct comparative data regarding long-term survival outcomes remain limited. Most published studies have been single-arm phase 2 trials or retrospective analyses, with few prospective head-to-head comparisons available.
We previously reported comparable pathological response rates between FLOT and SOX in the Dragon III trial (20.0% versus 32.4% complete/subtotal tumour regression, p = 0.29).17 However, pathological response may not always translate directly into survival benefit, requiring validation with long-term survival data.
The primary objective was to compare long-term survival outcomes between neoadjuvant FLOT and SOX regimens in patients with LAGC using extended follow-up data from the Dragon III trial. Secondary objectives included analysing disease recurrence sites, identifying prognostic factors, and exploring relationships between pathological response and clinical outcomes.

Methods

Study design
The Dragon III trial was an investigator-initiated, phase 2, open-label, randomised controlled trial comparing neoadjuvant FLOT and SOX regimens for patients with locally advanced gastric cancer. The trial design, eligibility criteria, and treatment protocols have been published previously.17 The original Dragon III trial had pathological response (tumor regression grading) as the primary endpoint, which was fully reported in our initial publication in Nature Communications (2020).17 This manuscript reports the pre-specified secondary endpoints of overall survival (OS) and disease-free survival (DFS) with extended follow-up through May 8, 2025. The trial was prospectively registered at ClinicalTrials.gov (NCT03636893) before patient enrolment.
The study was conducted at Ruijin Hospital, Shanghai Jiao Tong University School of Medicine Between Aug 22, 2018, and Nov 14, 2019. The Institutional Review Board of Ruijin Hospital approved the study protocol, including long-term follow-up plans (Approval No KY202074). All patients provided written informed consent for participation and long-term follow-up.
The single-centre design was deliberately chosen for this phase 2 exploratory trial to enable complete standardisation of surgical techniques (all procedures performed by experienced gastric cancer specialists following Japanese guidelines), chemotherapy protocols, and follow-up procedures, while facilitating exceptional follow-up completeness crucial for survival endpoint validity. Ruijin Hospital is a high-volume, specialized gastric cancer centre, enabling standardised care delivery and reducing inter-institutional variability that could confound survival outcomes. This study is reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines for randomised controlled trials.

Participants
Inclusion criteria were designed to select patients with resectable locally advanced disease. Patients aged 18–80 years with histologically confirmed adenocarcinoma of the stomach or oesophagogastric junction were eligible if they had clinical stage cT3-cT4b, cN1-cN3, cM0 disease. Key inclusion criteria included adequate organ function, ECOG performance status ≤2, and written informed consent. Sex was determined based on biological factors recorded in medical records at enrollment. The detailed inclusion and exclusion criteria have been published elsewhere.17

Randomisation and masking
Patients were randomised 1:1 to receive either neoadjuvant FLOT (n = 40) or SOX (n = 34) using simple randomisation without stratification. The randomisation procedure followed rigorous methodology as detailed in our previously published article.17 A computer-generated randomisation sequence was created by an independent statistician at the Clinical Research Centre for 100 patients with 1:1 allocation using SPSS random number generators with a reproducible fixed seed. The sequence employed sealed envelope allocation concealment, with envelopes opened sequentially according to patient admission sequence. After confirming eligibility and obtaining informed consent, participants were assigned to treatment groups by the independent statistician. Treatment assignments were communicated to clinical teams via telephone or text message. Neither patients nor investigators were blinded to treatment allocation due to the different administration protocols of the regimens.
The apparent imbalance (40 FLOT versus 34 SOX) resulted from the pre-planned trial stopping point after achieving our target per-protocol population (n = 55), not from flawed randomisation methodology. This approach is standard in exploratory phase 2 trials where enrolment stops based on evaluable patients rather than total randomised patients. The slight imbalance occurred by chance during the randomization process itself.

Procedures
Both regimens followed established dosing and scheduling. The FLOT regimen consisted of four cycles of 5-Fluorouracil 2600 mg/m2, Leucovorin 200 mg/m2, Oxaliplatin 85 mg/m2, and Docetaxel 50 mg/m2, every 2 weeks. The SOX regimen consisted of three cycles of oral Tegafur/Gimeracil/Oteracil (S-1) 80 mg/m2 twice daily on days 1–14 and oxaliplatin 130 mg/m2 on day 1, repeated every 3 weeks. Surgery was performed 2–4 weeks after completion of neoadjuvant chemotherapy.
Surgical procedures were standardised at a high-volume centre. All surgical procedures were performed by experienced surgeons. Standard D2 gastrectomy was performed according to Japanese Gastric Cancer Treatment Guidelines. Adjuvant chemotherapy regimens were clearly defined in the protocol: patients received continuation of their assigned preoperative regimen (FLOT group continued FLOT; SOX group continued SOX) according to institutional guidelines. This approach maintains randomisation integrity throughout the treatment course and follows international perioperative treatment guidelines as recommended by landmark trials including FLOT4. Adjuvant therapy commenced 4–8 weeks post-surgery when clinically appropriate.
Comprehensive long-term follow-up ensured complete data capture. Following the completion of primary treatment, patients entered long-term follow-up for survival endpoints. Follow-up visits were scheduled every 3 months for the first 2 years post-surgery, every 6 months for years 3–5, and annually thereafter. Vital status was verified through multiple sources: direct clinical follow-up, telephone contact, cross-referencing with local death registry records, and review of medical records from other healthcare facilities when applicable.

Outcomes
The secondary endpoints were rigorously defined. The primary endpoint (pathological response) was previously reported in Nature Communications.17 Overall survival (OS) was defined as the time from randomisation to death from any cause. Patients alive at the time of analysis were censored at the date of last follow-up contact. Disease-free survival (DFS) was defined as the time from randomisation to the first occurrence of local recurrence, regional recurrence, distant metastases, or death from any cause. Disease recurrence was assessed using CT imaging with contrast, upper endoscopy for local recurrence, histological confirmation when feasible, and multidisciplinary team review of imaging findings.
Pathological response was assessed using validated criteria. Treatment response was assessed using the Becker tumour regression grading system: TRG 1a (complete regression), TRG 1b (subtotal regression, <10% residual tumour), TRG 2 (partial regression, 10–50% residual tumour), and TRG 3 (minimal regression, >50% residual tumour).18 Two independent pathologists, blinded to treatment allocation, evaluated the specimens.

Statistical analysis
The original study protocol (dated January 10, 2017) and statistical analysis plan outlined therein were submitted with this manuscript per journal requirements. The long-term survival analysis reported herein was pre-specified as a secondary endpoint in the original protocol, with overall survival defined as “time from randomisation to death from any cause” and disease-free survival defined as “time from randomisation to relapse of the disease.” All survival analyses were performed according to the pre-specified statistical plan outlined in the original protocol, with no deviations from the planned analytical approach. We acknowledge that the original Statistical Analysis Plan contained limited methodological detail, which represents a limitation in the trial documentation standards.
The Dragon III trial was designed as an exploratory phase 2 study without formal power calculations for survival endpoints. The trial recruitment was completed after the surgical treatment of the 55th patient who met criteria for per-protocol analysis, consistent with the original protocol's exploratory design and empirical sample size estimation. This stopping point was pre-planned to obtain preliminary data for efficacy assessment and to inform future phase III trial design. All analyses were conducted according to the pre-specified statistical plan without knowledge of survival outcomes at the time of study completion.
Post-hoc power analysis revealed that, with 41 observed deaths and a median follow-up of 65.7 months, the study had approximately 80% power to detect a hazard ratio of 0.55 or 1.8 at α = 0.05, using a two-sided log-rank test.
Missing baseline and intraoperative data were not imputed. All survival analyses followed intention-to-treat principles including all 74 randomised patients regardless of treatment completion. All analyses followed intention-to-treat principles with robust methodology. All survival analyses were performed according to the intention-to-treat principle, including all 74 randomised patients regardless of treatment completion or protocol adherence. Overall survival and disease-free survival were estimated using the Kaplan–Meier method. Survival differences between treatment groups were assessed using the log-rank test. Hazard ratios with 95% confidence intervals were estimated using Cox proportional hazards regression. Proportional hazards assumptions were verified using Schoenfeld residuals test and visual inspection of log–log survival plots. All statistical tests mentioned in results are described in this methods section.
Comprehensive subgroup and multivariable analyses were performed. Post-hoc subgroup analyses were performed for tumour regression grade, type of gastrectomy, postoperative pathological stage, and other baseline characteristics. Cox regression with stepwise selection was used to identify independent prognostic factors. All analyses were performed using SPSS version 30.0. Statistical analyses were performed using SPSS version 30.0. Figures were created using R version 4.5.0. R packages used included survival (version 3.8–3), ggplot2 (version 3.5.2), survminer (version 0.5.0), ggpubr (version 0.6.0), readr (version 2.1.5), and dplyr (version 1.1.4). Two-sided p-values <0.05 were considered statistically significant. The analysis of both overall survival and disease-free survival without multiplicity adjustment represents a limitation. p-values should be interpreted as exploratory findings in the context of this phase II study designed to inform future phase III trial design. Two independent statisticians verified all analyses.
Subgroup analyses for type of gastrectomy, postoperative pathological stage, and other baseline characteristics were performed post-hoc and were not prespecified in the original protocol. These should be interpreted as exploratory findings requiring validation in future studies.
No major deviations from the prespecified statistical analysis plan were made. All survival analyses followed the original protocol specifications.

Role of the funding source
This research received no specific grant from any funding agency. BKS and ZGZ had full access to all study data and take responsibility for data integrity and accuracy of analysis. BKS had final responsibility for the decision to submit for publication.

Results

Patient enrolment and treatment completion
Between Aug 22, 2018, and Nov 14, 2019, 74 patients with locally advanced gastric adenocarcinoma were randomised to neoadjuvant FLOT (n = 40) or SOX (n = 34) (Fig. 1). Patient flow through the trial has been previously published.17 Pathological response results (tumour regression grading TRG1a and TRG1b) were fully reported in our initial publication.17 The current manuscript focuses on the pre-specified secondary survival endpoints.
Treatment completion rates were comparable between groups. In the FLOT group, nine patients did not proceed to surgery: one withdrew consent, four refused surgery, three required early surgery due to acute bleeding, and one experienced a serious adverse event (acute cerebral infarction). In the SOX group, 10 patients did not undergo planned surgery: two withdrew consent, three refused surgery, one died from grade IV haematological toxicity followed by multiple organ failure, three violated protocol (two requested alternative chemotherapy and one was diagnosed with peritoneal metastases after allocation), and one experienced deep venous thrombosis detected during surgical evaluation. This resulted in 55 patients (31 FLOT, 24 SOX) completing all planned chemotherapy and undergoing surgical resection, as detailed in Fig. 1. When including patients who underwent early emergency surgery (3 in FLOT group for acute bleeding complications), surgical completion rates were 85% (34/40) for FLOT and 71% (24/34) for SOX, with an overall rate of 78% (58/74). The 55 patients represent those who completed both planned chemotherapy and elective surgical resection according to protocol timing. Analysis of dropout reasons reveals that patient choice (withdrawal/refusal) accounted for 10 of 16 cases (63%) who did not proceed to surgery, while treatment-related factors caused only 2 cases (12%). This pattern primarily reflects patient autonomy and cultural factors in treatment decision-making rather than treatment intolerance. The study population included 50 males (67.6%) and 24 females (32.4%). Sex distribution was similar between treatment groups (FLOT: 29 males, 11 females; SOX: 21 males, 13 females, p = 0.43). Baseline characteristics showed no significant differences between treatment groups (Table 1).

Follow-up and adjuvant treatment
Complete long-term follow-up was achieved for all patients. With data cutoff on May 8, 2025, median follow-up was 65.7 months (range: 0.97–80.5) for the intention-to-treat population. Complete vital status was obtained for all patients: 63.7 months (range: 8.4–80.5) in the FLOT group and 66.7 months (range: 0.97–80.5) in the SOX group. No patients were lost to follow-up (Supplementary Table S1).
Post-surgical treatment was well-balanced between groups. Among the 55 patients who underwent surgical resection, the majority (52.7%) initiated adjuvant therapy within 5–8 weeks post-surgery. Most patients (85.5%) completed their planned adjuvant regimen, and 72.7% received at least three cycles. No significant differences in adjuvant therapy parameters were observed between treatment groups (all p > 0.05), confirming the comparability of post-surgical treatment between the FLOT and SOX arms (Supplementary Tables S2 and S3).
Both regimens demonstrated favourable safety profiles. Safety data for neoadjuvant chemotherapy were reported in our initial publication.17 Briefly, there was no significant difference in chemotherapy-related haematological or non-haematological adverse effects between groups. Nine events of grade 3–4 haematological toxicity were observed in the FLOT group versus five in the SOX group. Both regimens demonstrated favourable toxicity profiles, with notably lower rates of leucopeniaand neutropenia than reported in the original FLOT4 trial. This was potentially influenced by more frequent granulocyte-colony stimulating factor (GCSF) use in the FLOT group (94.9%) compared to the SOX group (43.3%), though GCSF was only used for treatment rather than prophylaxis per protocol. All 55 patients in the per-protocol population completed the planned chemotherapy with full doses of their respective regimens.

Survival outcomes: comparable long-term results
Both regimens achieved exceptional overall survival with no significant difference. Survival data are summarised in Table 2. At data cutoff, 41 deaths had occurred (21 [52.5%] in the FLOT group and 20 [58.8%] in the SOX group). No significant difference in overall survival was observed between treatment groups, with median OS of 61.5 months (95% CI: not reached) for FLOT versus 67.8 months (95% CI: 25.7–109.9) for SOX (hazard ratio [HR] 1.101, 95% CI: 0.595–2.036; p = 0.76) (Fig. 2). The 3-year overall survival rates were 62.5% (95% CI: 47.4–77.6) for FLOT and 55.9% (95% CI: 39.2–72.6) for SOX.
Disease-free survival similarly demonstrated comparable outcomes. Disease-free survival showed no statistically significant difference between groups (HR 1.060, 95% CI: 0.597–1.884; p = 0.84), with median DFS of 23.0 months (95% CI: 8.0–38.0) for FLOT versus 25.5 months (95% CI: 12.0–38.9) for SOX (Fig. 3). While 1-year DFS rates appeared numerically higher in the SOX group (83.4% versus 67.3%), this difference diminished at later time points, with 3-year DFS rates of 43.6% for SOX and 31.6% for FLOT. Cox proportional hazards assumptions were verified through Schoenfeld residuals testing and visual inspection of log–log survival plots (Supplementary Figure S1), confirming the validity of Cox regression analyses for both overall survival and disease-free survival comparisons.

Recurrence patterns: notable differences in metastatic sites
Recurrence rates were similar between regimens. Recurrence sites are summarized in Table 3. Among 55 patients who underwent surgical resection, recurrence occurred in 27 (49.1%) patients during follow-up, with a numerically higher rate in the FLOT group (17/31, 54.8%) compared to the SOX group (10/24, 41.7%).
Most common recurrence sites were consistent between groups. The most common sites of recurrence in both treatment groups were retroperitoneal lymph nodes (FLOT: 22.6%; SOX: 20.8%) and peritoneum (FLOT: 16.1%; SOX: 16.7%). Notably, lung metastases were observed exclusively in the FLOT group (12.9% versus 0%), while brain metastases were more common in the SOX group (8.3% versus 3.2%).

Prognostic factors: clinicopathological features outweigh regimen choice
Multiple clinicopathological factors significantly impacted survival outcomes. Comprehensive subgroup analyses for overall survival are summarised in the Supplementary Table S4 and illustrated in Fig. 4. Multiple factors were significantly associated with survival outcomes, with pathological response proving more important than regimen selection.
Pathological response was strongly prognostic. Pathological response, as measured by tumour regression grade, showed significant impact on survival (p = 0.020), with complete/subtotal regression (TRG 1a + 1b) associated with markedly longer median OS (80.5 months) compared to partial/minimal regression (47.6 months).
Surgical factors were equally critical. Gastrectomy extent emerged as another significant prognostic factor (p = 0.0030), with partial gastrectomy demonstrating superior outcomes compared to total gastrectomy (median OS not reached versus 37.3 months; HR 3.619, 95% CI: 1.360–9.628).
Traditional prognostic markers maintained their significance. Pathological staging also influenced survival (p = 0.030), with gradually decreasing survival from Stage I (median not reached) to Stage II (80.5 months) to Stage III (33.5 months). Both vessel invasion (p = 0.020) and nerve invasion (p = 0.040) were significant negative prognostic indicators. Lauren classification demonstrated prognostic value (p = 0.020), with intestinal type showing more favourable outcomes compared to other histological subtypes. Across all prognostic factors, patients who dropped out before surgery consistently showed the poorest outcomes (median OS 29.6 months).

Independent prognostic factors: surgery emerges as key driver
Gastrectomy type was the only independent survival predictor. Cox regression with stepwise selection identified gastrectomy type as the only independent predictor of overall survival (p = 0.0020) (Table 4, Fig. 4). Compared to partial gastrectomy: total gastrectomy HR 3.619 (95% CI: 1.360–9.628; p = 0.010), unknown gastrectomy HR 5.134 (95% CI: 1.844–14.291; p = 0.0020). Importantly, treatment group, tumour regression grade, Lauren classification, and postoperative stage were eliminated during stepwise selection. However, the prognostic significance of gastrectomy type requires careful interpretation as this finding likely reflects the well-established impact of tumour location rather than representing a truly modifiable surgical factor.

Discussion
This first prospective head-to-head comparison provides first prospective evidence of comparable long-term survival between neoadjuvant FLOT and SOX regimens, informing evidence-based treatment selection for locally advanced gastric cancer. With both regimens achieving exceptional median overall survival exceeding 5 years (61.5 versus 67.8 months, p = 0.76), clinicians worldwide now have evidence-based freedom to select regimens based on patient and institutional factors rather than survival concerns.
Our survival results align with the FLOT4 trial's median overall survival of 50 months,7 though direct comparisons require caution given differences in patient populations and staging systems. Despite lower pathological response rates with FLOT in our Asian population (20.0% versus 37% in FLOT4), long-term survival outcomes remained excellent, suggesting pathological response may not fully capture clinical benefit in all patients.6 This finding is consistent with previous studies suggesting that while tumour regression grade is prognostic, the relationship between pathological response and survival may vary across populations and regimens.19
Our findings are particularly relevant for Asian populations, where treatment completion patterns may differ from Western trials. While our surgical completion rate (85% FLOT, 71% SOX) was lower than highly controlled trials like FLOT4 (97%) and PRODIGY (93%), this difference primarily reflects real-world clinical practice factors including patient autonomy and family involvement in treatment decisions. In the context of Chinese clinical practice, various neoadjuvant regimens are utilized for locally advanced gastric cancer based on institutional experience, patient factors, and evolving treatment guidelines.14 Our findings demonstrating equivalent survival between FLOT and SOX provide important comparative data within this broader therapeutic landscape, supporting evidence-based selection among available regimens rather than defaulting to institutional preference alone Recent large Asian trials (RESOLVE, RESONANCE) also reported attrition between enrolment and analysis populations, and our adjuvant therapy completion rate (85.5%) compares very favourably to the RESONANCE trial's 19.2% completion rate in their adjuvant-only arm, demonstrating excellent treatment tolerability in our population.
The study's most clinically relevant finding is that several clinicopathological factors were stronger survival predictors than neoadjuvant regimen choice. This paradigm shift redirects clinical focus from “which regimen is better” to “how do we optimise response and surgical management.
Pathological response emerged as a critical predictor. Tumour regression grade, gastrectomy type, vessel invasion, nerve invasion, and Lauren classification20 all demonstrated prognostic significance in our analysis. Patients achieving complete/subtotal tumour regression had markedly better survival (80.5 versus 47.6 versus 29.6 months for TRG 1a + 1b versus 2 + 3 versus dropout, p = 0.020), reinforcing the importance of pathological response as a prognostic factor. This aligns with foundational work by Becker et al. and Mandard et al., who established tumour regression grading systems that remain essential for prognostic assessment in upper gastrointestinal cancers.18,21
Surgical factors proved equally critical. Gastrectomy type emerged as the strongest independent survival predictor (HR 3.619 for total versus partial gastrectomy, p = 0.010). While this may reflect tumour location and disease extent, it highlights surgical factors' impact on outcomes. Superior outcomes with partial gastrectomy may relate to preserved function, reduced morbidity, and selection of patients with more favourable characteristics. This finding is consistent with contemporary surgical series demonstrating that extent of resection influences both immediate and long-term outcomes in patients with gastric cancer.
Traditional prognostic factors maintained their relevance. The observed associations between vessel invasion, nerve invasion, and survival further strengthen our prognostic model. Patients with negative vessel invasion demonstrated significantly better survival (80.5 versus 31.2 months, p = 0.020) compared to those with positive findings. Similarly, negative nerve invasion was associated with superior outcomes (not reached versus 47.6 months, p = 0.040). Lauren's classification20 also demonstrated prognostic significance, with intestinal-type histology associated with better survival than other histological types (not reached versus 47.6 months, p = 0.020).
Our analysis identified gastrectomy type as the strongest independent predictor of survival in multivariable analysis. However, this finding requires nuanced interpretation due to potential confounding by tumour location. The choice between total and partial gastrectomy is primarily dictated by tumour anatomical location rather than representing a modifiable surgical decision. Proximal gastric tumours, which necessitate total gastrectomy, are well-established to have worse prognosis than distal tumours due to several factors: (1) molecular differences with higher prevalence of diffuse-type tumours, (2) surgical complexity with more challenging access and potential for positive margins, (3) greater likelihood of advanced local disease at presentation, and (4) differences in lymphatic drainage patterns affecting metastatic spread. Therefore, our finding should be interpreted as reinforcing the importance of tumour location as a prognostic factor rather than suggesting that surgical approach itself is modifiable. This insight emphasizes the need for potentially more aggressive perioperative approaches in patients with proximal tumours, regardless of the chosen chemotherapy regimen.
The comparable outcomes between FLOT and SOX support findings from recent Asian trials, including RESOLVE and RESONANCE, which demonstrated promising efficacy for SOX regimens.10,11 These results, along with findings from other major Asian gastric cancer trials,22 reinforce the importance of optimising treatment selection for Asian populations. Our study provides the first prospective head-to-head survival comparison between these regimens, supporting either approach based on institutional and patient factors. The demonstrated equivalence has important implications for treatment selection in Asian populations, where SOX offers practical advantages including oral administration and reduced treatment cycles.
Disease-free survival showed no significant difference between groups (23.0 versus 25.5 months, p = 0.84). Sites of recurrence differed notably, with no lung metastases in the SOX group versus 12.9% in the FLOT group, warranting validation in larger studies. These differences may reflect distinct drug-specific effects on sites of metastases or could be attributed to the relatively small sample size.
The high adjuvant therapy completion rate (85.5%) and similar treatment adherence indicate both regimens were well-tolerated and feasible in the perioperative setting. This is particularly important given that completion of planned perioperative therapy is associated with improved outcomes in patients with gastric cancer.7
Our findings have important implications for the evolving therapeutic landscape where immunotherapy combinations are emerging. The demonstrated equivalence between FLOT and SOX provides crucial evidence for chemotherapy backbone selection in future combination strategies with immune checkpoint inhibitors and targeted agents. Both regimens now serve as validated platforms for integration with emerging therapies, as evidenced by the recent success of the DRAGON IV trial combining immunotherapy with SOX.13
The survival equivalence we observed creates opportunities for personalized treatment selection based on patient factors, institutional experience, and practical considerations, while establishing a foundation for future precision medicine approaches. Both FLOT and SOX can now be confidently combined with emerging agents without concerns about compromising the chemotherapy backbone efficacy.
Given equivalent survival outcomes, regimen selection should be individualised. SOX offers practical advantages including reduced cycles (3 versus 4), fewer hospital visits, and potentially lower neutropenia rates.17 Patient factors (age, performance status, treatment preference) and institutional considerations (experience, protocols, cost) should guide decision-making.
Our findings must be interpreted within the evolving landscape of gastric cancer treatment. Recent advances in targeted therapy and immunotherapy are transforming treatment paradigms.23,24 The integration of trastuzumab deruxtecan for HER2-positive disease and immune checkpoint inhibitors represents significant progress.24,25 The DRAGON IV trial's demonstration of improved pathological response with immunotherapy plus chemotherapy suggests these combinations represent the next treatment frontier.13 Future research should explore whether the survival equivalence we observed between FLOT and SOX is maintained when combined with immunotherapy and targeted agents.
Study limitations include multiple important considerations. As a phase 2 exploratory study, our findings provide important preliminary evidence but should be interpreted cautiously and require validation in adequately powered phase 3 trials. The study was not formally powered for survival endpoints, limiting the strength of survival comparisons. While single-centre design limits immediate generalizability, it enabled standardised care delivery and exceptional follow-up completeness (100% with zero patients lost over 65.7 months median follow-up), which is crucial for survival endpoint validity and rarely achieved in multicentre trials. The identification of gastrectomy type as an independent prognostic factor is likely confounded by tumour location, which was not included in our multivariable analysis. Future studies should incorporate tumour location stratification to better understand the relative contributions of anatomical factors versus surgical approach. The exclusively Asian population, while limiting global generalizability, provides crucial data for this high-incidence region where treatment patterns may differ from Western populations. Our study did not stratify patients by microsatellite instability status or other biomarkers that may influence treatment selection in the era of precision medicine. While our results suggest comparable outcomes between regimens (HR near 1.0), this study was not designed to formally test equivalence. Definitive equivalence would require a specifically designed non-inferiority or equivalence trial with appropriate pre-specified equivalence margins and formal sample size calculations.
This study was conducted in a homogeneous Chinese population at a single centre in Shanghai, which may limit generalizability to other ethnic populations. Sex distribution analysis showed no significant difference between treatment groups (p = 0.43). The influence of sex on treatment outcomes was not formally analysed due to the exploratory nature and sample size limitations of this phase 2 study. Treatment patterns, patient preferences, and clinical outcomes may vary across different racial and ethnic groups due to genetic, cultural, and socioeconomic factors that were not assessed in this study. For immediate clinical application, these results support individualised treatment selection.
This 5-year analysis provides the first prospective evidence of comparable survival outcomes between FLOT and SOX neoadjuvant regimens, both achieving excellent median overall survival exceeding 5 years. As preliminary evidence from a phase 2 exploratory study, these findings require validation in adequately powered phase 3 trials but provide important guidance for evidence-based treatment selection in current clinical practice.
The demonstrated comparable outcomes enable clinicians to select regimens based on patient factors, institutional expertise, and practical considerations without compromising survival expectations. These findings are particularly relevant for current clinical practice where immunotherapy access remains limited globally, while also establishing a foundation for rational chemotherapy backbone selection in future combination strategies.
Our results inform treatment decisions for the majority of patients with gastric cancer (pMMR/MSS, ∼95%) who require chemotherapy optimization, while enabling biomarker-stratified trial design for the minority (dMMR/MSI-H, ∼5%) who may benefit from immunotherapy approaches. The identification of clinicopathological factors as stronger predictors than regimen choice, while acknowledging potential confounding by tumour location, provides insights for patient counseling and treatment planning.
Both regimens serve as validated platforms for integration with emerging therapies, supporting the development of personalized perioperative care for patients with gastric cancer. These preliminary results address an important evidence gap in current guidelines while providing essential groundwork for the evolving therapeutic landscape, pending confirmation in larger definitive trials.

Contributors
BKS conceived and designed the study, collected the patient data, performed the data analysis, and drafted the manuscript. ZY, BZ, FY, HZ, JL, TM, MS, ZZ, YZ, WL, CY contributed to data collection and analysis. CL and ZGZ designed the study and edited the final manuscript. All authors met the criteria for authorship, according to the guidelines of the International Committee of Medical Journal Editors. All authors read and approved the final manuscript. BKS and ZGZ had access to and verified the underlying study data.

Data sharing statement
Individual patient data supporting the conclusions will be made available to qualified investigators upon reasonable request, subject to appropriate data-sharing agreements, and ethics approval. Requests for data should be directed to the corresponding author.

Declaration of interests
All authors declare no competing interests.