Adjuvant chemoradiotherapy vs chemotherapy after R0 D2 gastrectomy for gastric cancer: Meta-analysis of randomized trials

Abstract

BACKGROUND

Whether postoperative radiotherapy adds meaningful benefit to contemporary multiagent chemotherapy (CT) after an R0 D2 gastrectomy remains uncertain.

AIM

To synthesize randomized evidence comparing postoperative chemoradiotherapy (CRT) with CT alone after R0 D2 gastrectomy and to evaluate disease-free survival (DFS) or recurrence-free survival (RFS), overall survival (OS), recurrence patterns, and treatment tolerability.

METHODS

We searched PubMed/MEDLINE, Europe PMC and ClinicalTrials.gov, verified a targeted Chinese Clinical Trial Registry record, and examined backward and forward citations through August 19, 2026. Eligible randomized trials compared postoperative CRT with CT after R0 D2 gastrectomy. DFS/RFS and OS were synthesized as hazard ratios (HRs) using restricted maximum-likelihood random-effects models and modified Knapp-Hartung confidence intervals (CIs). Two legacy trials contributed secondary reconstructed survival estimates; the published Adjuvant Chemoradiotherapy in Stomach Tumors 2 (ARTIST 2) CI was used directly. Risk of bias and certainty were evaluated using the Cochrane RoB 2 and the Grading of Recommendations Assessment, Development and Evaluation approach, respectively.

RESULTS

Eight independent randomized comparisons were included; seven trials (2256 participants) contributed DFS/RFS estimates. The all-era pooled HR was 0.82 (95%CI: 0.69-0.98; I² = 0%). The oral fluoropyrimidine–platinum doublet subgroup yielded HR = 0.88 (4 trials; 1754 participants; 95%CI: 0.66-1.18), and exclusion of the two legacy secondary reconstructions yielded HR = 0.84 (5 trials; 2105 participants; 95%CI: 0.68-1.03). The OS estimate was imprecise (5 trials; 1580 participants; HR = 0.89, 95%CI: 0.69-1.15). Exploratory analyses of four legacy trials suggested fewer locoregional recurrences [odds ratio (OR) = 0.50, 95%CI: 0.26–0.94], but distant recurrence was uncertain (OR = 0.81, 95%CI: 0.50-1.29). Severe toxicity and treatment completion were not pooled because reporting was insufficiently comparable.

CONCLUSION

Randomized evidence is insufficient to establish a consistent DFS or OS benefit from routine postoperative CRT over effective CT after R0 D2 gastrectomy. The doublet-subgroup CI remains compatible with clinically important benefit and harm; these results do not establish equivalence or absence of effect. A possible local-control benefit is supported mainly by low-certainty legacy evidence.

Key Words: Gastric cancer; D2 lymphadenectomy; Adjuvant chemoradiotherapy; Chemotherapy; Randomized controlled trial; Meta-analysis

Core Tip: This synthesis links eight randomized trials after R0 D2 gastrectomy. The all-era disease-free survival estimate favored chemoradiotherapy, but treatment-backbone and provenance analyses remained imprecise. These findings do not establish equivalence or exclude meaningful benefit. The published Adjuvant Chemoradiotherapy in Stomach Tumors 2 CI is used, and source-tagged numerical inputs and reproducible sensitivity calculations are supplied.

INTRODUCTION

Gastric cancer remains a major global cause of cancer morbidity and mortality, with substantial geographic variation in incidence, stage at diagnosis, and treatment pathways[1,2].

For resectable disease, R0 gastrectomy with an adequate D2 lymphadenectomy is the operative standard in experienced centers. Long-term randomized evidence has shown better locoregional control with D2 than with limited nodal dissection, although surgical quality and perioperative risk materially influence its benefit[3,4].

Systemic therapy has evolved in parallel. Perioperative chemotherapy (CT) is established in many Western pathways, and randomized programs have also defined perioperative or postoperative oxaliplatin-based approaches after D2 surgery[5-7]. Postoperative chemoradiotherapy (CRT) became influential after INT-0116, but most participants in that trial underwent less than a D2 dissection. Its applicability after adequate D2 resection has therefore remained contested[8,9]. Asian randomized programs have further established effective postoperative fluoropyrimidine- and platinum-based CT after D2 gastrectomy, including S-1, capecitabine plus oxaliplatin, and related contemporary regimens[10-14]. Several randomized comparisons subsequently tested postoperative CRT against CT after D2 gastrectomy, including small legacy trials, the Adjuvant Chemoradiotherapy in Stomach Tumors (ARTIST) trial, ARTIST 2, and recent Chinese phase 3 studies[15-23]. The Chemotherapy vs Chemoradiotherapy after Surgery and Preoperative Chemotherapy for Resectable Gastric Cancer (CRITICS) addressed a different perioperative question because randomization occurred before preoperative CT and is not interchangeable with a direct postoperative D2 comparison[24,25].

Earlier meta-analyses suggested improved local control and, in some analyses, disease-free survival (DFS), but they predated the recent randomized comparisons and used heterogeneous treatment regimens and effect sources[26-28]. Korean and American Society for Radiation Oncology (ASTRO) recommendations provide a clinical context for postoperative treatment after adequate surgery[29,30]. Chinese guidelines also describe stage- and treatment-dependent pathways[31,32]. Guideline recommendations are considered separately from the randomized effects estimated in this review.

We therefore synthesized direct randomized comparisons of postoperative CRT vs CT in adults with nonmetastatic gastric adenocarcinoma after R0 D2 gastrectomy. DFS/recurrence-free survival (RFS) was the primary outcome; overall survival (OS), recurrence, tolerability and treatment delivery were secondary outcomes. Trial-report linkage, CT-backbone subgroups and effect-provenance sensitivity analyses were used to examine the applicability and uncertainty of the pooled results.

MATERIALS AND METHODS

Design, reporting standard, and protocol status

This systematic review and meta-analysis was reported with reference to PRISMA 2020 and PRISMA-S[33,34]. The review was not prospectively registered. Because the available documentation did not confirm prospective specification of the CT-backbone subgroup or provenance-based sensitivity analyses, these analyses were treated as exploratory. No equivalence margin or prospective threshold for a clinically important HR was documented. An analysis plan is provided to describe the analytical framework, effect-selection hierarchy, subgroup definitions and sensitivity analyses used in this study; because it was prepared after the initial analyses, it is not presented as evidence of prospective prespecification (Figure 1).

Figure 1 PRISMA 2020 flow diagram. The publication-report and registry streams are shown separately to avoid double-counting registrations and reports. Counts reproduce the screening flow recorded in the original review; the exact search strategies and a record-level eligibility verification ledger are supplied in the audit materials.

Eligibility criteria

Population: Adults with nonmetastatic gastric adenocarcinoma who underwent R0 gastrectomy with D2 lymph-node dissection.

Intervention and comparator: Postoperative CRT vs postoperative CT using a broadly comparable systemic-treatment backbone. CT exposure was not identical in all comparisons: Some CRT strategies replaced CT cycles with radiotherapy and concurrent fluoropyrimidine. The estimated is therefore the effect of the randomized treatment strategy, not necessarily the isolated effect of radiotherapy added to an otherwise identical CT course.

Design: Randomized controlled trials. Multi-arm trials contributed only the direct eligible CRT-versus-CT contrast. Multiple reports from the same trial were linked and counted once.

Outcomes: DFS or RFS (primary), OS, locoregional recurrence, distant recurrence, grade 3 or higher adverse events, treatment completion, and quality of life. Objective response rate and disease-control rate were excluded because they are not meaningful postoperative adjuvant outcomes after complete resection.

Key exclusions were nonrandomized studies; surgery-only comparators; mixed D1/D2 populations without an isolatable D2 cohort; preoperative or perioperative randomization; noncomparable systemic therapy; single-arm studies; duplicate or secondary reports without a new eligible contrast; and protocols without results.

Information sources and search strategy

The search cutoff was August 19, 2026. Sources included PubMed/MEDLINE (127 records), Europe PMC (75 records), ClinicalTrials.gov (24 records), targeted verification of ChiCTR-TRC-12002919, and backward and forward citation searching. Targeted verification of the cited ChiCTR identifier was not a comprehensive search of that registry. Electronic searches were conducted without language restrictions. Full search strings and available screening and eligibility records are provided in Supplementary material of Search strategies, and the audit workbook. Registry searching was used to complement publication searches and to distinguish trial registrations from trial reports[35,36].

Study selection, trial linkage, and data collection

Candidate reports were linked using trial identifiers, recruitment setting, treatment arms and sample size. ARTIST primary and final publications were counted as one trial. ARTIST 2 contributed only SOXRT vs SOX; its S-1 monotherapy arm was not included in this direct contrast. Randomized and outcome-analysis denominators were recorded separately. Numerical inputs were verified against accessible primary reports or explicitly identified secondary sources. Trial linkage, source locations, effect direction and extraction status are documented in the audit workbook.

All hazard ratio (HRs) and odds ratio (OR) were oriented as CRT vs CT, so values below 1 favor CRT. For effects originally reported in the opposite direction, the HR was converted as 1/HR, with the corresponding 95%CI converted as (1/upper CI, 1/Lower CI). For Kwon et al[15] and Kim et al[16], the legacy DFS/OS log-HRs and standard errors were transcribed from Figure 2 of the 2016 meta-analysis[27]; they were not independently reconstructed from participant-level data or newly digitized Kaplan–Meier curves. The respective DFS log-HR/SE pairs were -0.56/0.46 and -0.36/0.31; the OS pairs were -0.11/0.43 and -0.14/0.33. The upstream reconstruction cannot be independently reproduced from these reported summaries alone. In contrast, ARTIST 2 reports HR = 0.971 with 95%CI: 0.663-1.421[21]; these published limits were used in the primary analysis. The earlier approximation and its rounding assumptions were retained only for sensitivity comparison.

Figure 2 Forest plot of disease-free survival/recurrence-free survival using the published Adjuvant Chemoradiotherapy in Stomach Tumors 2 CI (0.663-1.421). Squares reflect inverse-variance weights; horizontal lines show study-level 95%CIs, and the diamond shows the restricted maximum likelihood pooled estimate with modified Knapp–Hartung inference. Hazard ratios below 1 favor chemoradiotherapy. N denotes the participants represented in each extracted comparison; The trial reported by Zhu et al[17] contributes 351 analyzed rather than 404 randomized participants. ARTIST: Adjuvant Chemoradiotherapy in Stomach Tumors; REML: Restricted maximum likelihood; HR: Hazard ratio; CRT: Chemoradiotherapy; CT: Chemotherapy.

Effect measures and statistical synthesis

Time-to-event effects were pooled on the log-HR scale using inverse-variance weights. For a reported two-sided 95%CI, SE[ln(HR)] = [ln(upper) - ln(lower)]/(2 × 1.959964); reversing the treatment direction changes the sign of ln(HR) but not its SE[37]. Reported log-HR/SE pairs were used directly when only secondary reconstructions were available. The optional ARTIST 2 approximation uses z = Φ⁻¹(1 - P/2), SE = |ln(HR)|/z and exp[ln(HR) ± 1.959964 × SE]. This is a Wald-type approximation requiring a two-sided P value for the same contrast; it need not reproduce a CI obtained from a different test or rounded inputs. The published CI was therefore preferred.

Between-study variance (τ²) was estimated by restricted maximum likelihood, with the boundary τ² = 0 explicitly evaluated. With wᵢ = 1/(SEᵢ² + τ²), the pooled log-HR is μ = Σwᵢyᵢ/Σwᵢ, q = Σwᵢ (yᵢ − μ)²/(k - 1), and q* = max (1, q). where q* denotes the variance-scaling factor truncated at a minimum value of 1. The modified Knapp-Hartung CI is exp[μ ± t (0.975, k - 1) × √(q*/Σwᵢ)]. Conventional normal-reference CIs were retained for comparison. Heterogeneity was summarized using Q, τ² and I²[38-40]. Analyses were reproduced using Python 3.13.5, NumPy 2.3.5 and SciPy 1.17.0; the executable script, input files, software versions and outputs accompany the workbook.

The primary synthesis retained all seven trials with usable DFS/RFS effects. Exploratory analyses included: An oral fluoropyrimidine–platinum doublet subgroup (ARTIST, ARTIST 2, Qiao et al[22], and Wang et al[23]); an oxaliplatin-based subgroup (ARTIST 2, Qiao et al[22], and Wang et al[23]); exclusion of Kwon et al[15] and Kim et al[16] separately and together; and exclusion of ARTIST 2 separately and together with Kwon et al[15] and Kim et al[16]. ARTIST used capecitabine/cisplatin, whereas the oxaliplatin-based subgroup is narrower; the four-trial subgroup is not described as uniformly contemporary. Kwon et al[15] used an intravenous fluorouracil/cisplatin backbone, and Xie et al[20] lacked a usable HR despite an oral doublet regimen. Analogous exclusion analyses were conducted for OS where data were available. Exploratory recurrence ORs were recalculated from four legacy trials’ event counts; time windows and outcome definitions were not assumed to be identical. No zero-cell correction was needed for the available recurrence tables. Toxicity and completion were not pooled when definitions or denominators were not commensurate.

Risk of bias, reporting bias, and certainty of evidence

Risk of bias was assessed using the RoB 2 framework for the effect of assignment to intervention[41]. The audit workbook contains 462 signalling-question responses across 21 trial–outcome assessments, together with source locations, access notes and algorithm checks. After source review, the authors reviewed and approved the final outcome-level judgments used in the manuscript and Figure 3. Where primary methods or protocol details were unavailable, the corresponding uncertainty was retained in the judgment rather than reconstructed. Grading of Recommendations Assessment, Development and Evaluation (GRADE) certainty was assessed separately for each evidence body[42], with explicit reasons for downgrading recorded in the audit workbook. Funnel-plot asymmetry and regression tests were not used because fewer than 10 trials contributed to each synthesis[43,44].

Figure 3 Final RoB 2 judgments for the disease-free survival/recurrence-free survival outcome after author review. D1: Randomization process; D2: Deviations from intended interventions; D3: Missing outcome data; D4: Outcome measurement; D5: Selection of the reported result. The corresponding signalling-question rationale and source-access notes are provided in the audit workbook. The trial reported by Xie et al[20] was eligible qualitatively but did not contribute a hazard ratio to the quantitative synthesis. ARTIST: Adjuvant Chemoradiotherapy in Stomach Tumors.

RESULTS

Study selection

The publication search identified 202 database records. After removal of 71 duplicates, 131 unique records were screened; 33 database reports and 2 additional citation or conference reports were assessed in detail. Twenty-six reports were excluded, leaving 9 reports describing 8 independent randomized trials. The registry stream contained 25 records; 4 were linked to included trials, 2 potentially eligible trials had no located results, and CRITICS was excluded from the primary synthesis because randomization occurred before preoperative CT. The two result-unavailable records comprised an adjuvant D2 trial and the ARTEG program; the latter also had a published randomized phase III protocol[45] (Figure 1).

Trial characteristics

The eight independent comparisons enrolled 2453 participants in the direct eligible contrasts as reported; 2400 were represented in published analysis populations because Zhu et al[17] randomized 404 patients but reported the efficacy analysis for 351. Trial size ranged from 61 to 620. All studies were conducted in East Asia. Radiotherapy doses ranged from 45 to 50.4 Gy, and CT evolved from fluoropyrimidine-based legacy regimens to capecitabine/cisplatin, XELOX, or SOX (Table 1).

Table 1 Characteristics of included randomized trials.

Ref.	Direct N	Population	CRT arm	CT arm	Main randomized evidence
Kwon et al[15] (2010)	61	R0/D2; stage IIIA-IV (M0)	FP → 45 Gy/capecitabine → FP	6 FP cycles	5-year DFS 76.7% vs 59.1%; reconstructed DFS HR = 0.57 (0.23-1.41)
Kim et al[16] (2012)	90	R0/D2; stage III-IV (M0)	FL → 45 Gy/FL → FL	5 FL cycles	Overall DFS/OS estimates were imprecise; locoregional recurrence-free survival favored CRT in the report
Zhu et al[17] (2012)	404 randomized; 351 analyzed	D2; stage IB-IV (M0)	FL → 45 Gy IMRT/FL → FL	Matched FL	RFS HR = 0.74 (0.56-0.97); OS HR = 0.81 (0.61-1.06), standardized to CRT vs CT
Lee et al[18]/Park et al[19] (ARTIST)	458	Completely resected; D2	2 XP → 45 Gy/capecitabine→ 2 XP	6 XP cycles	Final DFS HR = 0.74 (0.52-1.05); OS HR = 1.13 (0.78-1.65)
Xie et al[20] (2018)	144 report-eligible	T3-4/N1-3 after D2	2 XELOX → 45 Gy/capecitabine→ 4 XELOX	6 XELOX cycles	3-year DFS 72.8% vs 76.3%; no HR/CI; qualitative only
Park et al[21] (ARTIST 2)	364 direct contrast	Stage II-III, node-positive; D2	SOX + 45 Gy	SOX for 6 months	DFS HR = 0.971; published 95%CI: 0.663-1.421; P = 0.879. Prior P-based approximation retained only for sensitivity audit
Qiao et al[22] (2025)	312	High-risk after D2	4-6 SOX → 45 Gy/S-1; total 6 SOX	8 SOX cycles	3-year DFS HR = 0.82 (0.54-1.25); grade 3/4 acute events 18.5% vs 26.1%
Wang et al[23] (2026)	620	R0/D2; T4 or node-positive; M0	1 SOX → 50.4 Gy/S-1 → 3 SOX	6 SOX cycles	DFS HR = 0.98 (0.73-1.33); OS HR = 0.86 (0.60-1.23)

CI: Confidence interval; CT: Chemotherapy; CRT: Chemoradiotherapy; DFS: Disease-free survival; FL: Fluorouracil plus leucovorin; FP: Fluorouracil plus cisplatin; HR: Hazard ratio; IMRT: Intensity-modulated radiotherapy; OS: Overall survival; RFS: Recurrence-free survival; SOX: S-1 plus oxaliplatin; XP: Capecitabine plus cisplatin; XELOX: Capecitabine plus oxaliplatin.

The ARTIST primary and final reports were linked as one trial. ARTIST 2 contributed only SOXRT (n = 183) vs SOX (n = 181), not the third S-1 arm. The trial reported by Xie et al[20] was retained as an eligible randomized comparison but did not enter the HR synthesis because the conference report provided landmark rates and P values without an HR or CI. The recent Qiao et al[22] and Wang et al[23] trials contributed their overall randomized comparisons; subgroup findings were not substituted for primary effects. The ARTIST 2 published CI was recovered during source checking; replacing the previous reconstruction did not change any headline DFS estimate when rounded to two decimal places.

Risk of bias

Following author review of the source-based RoB 2 register, ARTIST was judged at low risk for DFS/RFS; Kwon et al[15], Kim et al[16], ARTIST 2, Qiao et al[22], and Wang et al[23] had some concerns; and Zhu et al[17] and Xie et al[20] were judged at high risk (Figure 3). For Zhu et al[17], the randomized-versus-analyzed denominator discrepancy remained a material concern because intention-to-treat handling was not sufficiently clear in the available report. Xie et al[20] remained high risk because only a conference report was available, the report and registry denominators differed, and a complete outcome table and extractable HR were unavailable. These judgments were applied to the eligible result rather than inferred from reporting format alone.

DFS/RFS survival

Seven trials with 2256 participants contributed DFS/RFS estimates. Recalculation using the published ARTIST 2 CI yielded HR 0.82 (modified Knapp-Hartung 95%CI: 0.69-0.98; conventional 95%CI: 0.71-0.95; Q = 3.86; I² = 0%; τ² = 0). The pooled point estimate favored CRT. Its interpretation depends on the precision, treatment-backbone applicability and source limitations considered below.

In the oral fluoropyrimidine–platinum doublet subgroup (4 trials; 1754 participants), the pooled HR was 0.88 (95%CI: 0.66-1.18). Excluding the Kwon et al[15] and Kim et al[16] secondary reconstructions yielded HR = 0.84 (5 trials; 2105 participants; 95%CI: 0.68-1.03). The corresponding point estimates remained directionally favorable to CRT; crossing the null reflects uncertainty and does not by itself establish a different treatment effect. The narrower oxaliplatin-based subgroup yielded HR = 0.94 (3 trials; 1296 participants; 95%CI: 0.60-1.47). All three analyses had I² = 0% and τ² = 0. Single-study and combined reconstruction-related exclusions are reported to three decimal places in Supplementary Tables 1 and 2, because some upper confidence limits were close to 1 (Figure 2 and Supplementary Figures 1-3).

OS

Five trials with 1580 participants contributed OS estimates. The pooled HR was 0.89 (modified Knapp–Hartung 95%CI: 0.69-1.15; conventional 95%CI: 0.75-1.07; Q = 2.05; I² = 0%; τ² = 0). The CI was compatible with benefit and harm and did not establish an OS advantage or equivalence. Excluding Kwon et al[15] and Kim et al[16] yielded HR = 0.90 (3 trials; 1429 participants; 95%CI: 0.59-1.37). Several otherwise eligible trials lacked an extractable direct-comparison OS HR (Figure 4 and Supplementary Table 2).

Figure 4 Forest plot of overall survival. Study-level effects are standardized as chemoradiotherapy vs chemotherapy. The diamond shows the restricted maximum likelihood pooled hazard ratio with a modified Knapp-Hartung 95%CI. N reflects the participants represented in the extracted comparison. Legacy Kwon et al[15] and Kim et al[16] log-HR/SE inputs were taken from the 2016 secondary reconstruction. ARTIST: Adjuvant Chemoradiotherapy in Stomach Tumors; CRT: Chemoradiotherapy; CT: Chemotherapy; REML: Restricted maximum likelihood.

Recurrence pattern, safety, adherence, and quality of life

Four legacy trials (960 participants) contributed cumulative recurrence counts transcribed from Figure 4 of the 2016 synthesis[27]. Recalculated exploratory ORs were 0.50 (95%CI: 0.26-0.94) for locoregional recurrence and 0.81 (95%CI: 0.50-1.29) for distant recurrence (I² = 0% for both). These are OR based on cumulative counts, not recurrence HRs or estimates at a uniform follow-up time. The secondary-source counts and unresolved primary definitions are identified in the workbook; the local-control signal should not be generalized directly to current regimens (Table 2).

Table 2 Summary of quantitative syntheses.

Analysis	k /n	Effect	Modified KH 95%CI	I²	GRADE	Interpretation
DFS/RFS, all eras	7/2256	HR = 0.82	0.69-0.98	0%	Low	Point estimate favors CRT; indirect legacy evidence
DFS/RFS, oral fluoropyrimidine–platinum doublets	4/1754	HR = 0.88	0.66-1.18	0%	Moderate	Imprecise; benefit and harm remain possible
DFS/RFS, excluding Kwon et al[15] and Kim et al[16] reconstructions	5/2105	HR = 0.84	0.68-1.03	0%	Low	Conservative CI crosses no effect
Overall survival	5/1580	HR = 0.89	0.69-1.15	0%	Low	Imprecise; no equivalence inference
Locoregional recurrence, legacy	4/960	OR = 0.50	0.26-0.94	0%	Low	Improved local control; uncertain current relevance
Distant recurrence, legacy	4/960	OR = 0.81	0.50-1.29	0%	Low	Uncertain effect

n is the sum of participants represented by the selected outcome comparisons rather than a uniform randomized denominator. Grading of Recommendations Assessment, Development and Evaluation categories are the final author-reviewed certainty ratings for the evidence bodies shown in Table 2; detailed downgrade reasons and sensitivity-analysis membership are supplied in the audit workbook. CI: Confidence interval; CRT: Chemoradiotherapy; GRADE: Grading of Recommendations Assessment, Development and Evaluation; HR: Hazard ratio; OR: Odds ratio; DFS: Disease-free survival; RFS: Recurrence-free survival.

Comparative severe toxicity was not pooled because definitions, time windows and denominators were insufficiently comparable. Qiao et al[22] reported grade 3/4 acute adverse events in 18.5% vs 26.1% of patients (CRT vs CT; P = 0.11). Wang et al[23] reported completion of the allocated strategy in 251/309 CRT participants (81.2%) and 250/311 CT participants (80.4%). Its Figure 1 separately identifies 58 and 61 participants who discontinued the allocated intervention and 10 and 9 who were lost to follow-up, respectively; all 620 were included in the reported ITT analysis[23]. Treatment discontinuation should not be interpreted as complete absence of radiotherapy or as missing survival outcomes in every affected participant. No sufficiently complete, harmonized quality-of-life dataset was available for quantitative synthesis.

Certainty of evidence and reporting bias

After author review, certainty was rated low for the all-era DFS/RFS estimate because of risk-of-bias and indirectness concerns across treatment eras; moderate for the oral fluoropyrimidine–platinum doublet DFS/RFS analysis, downgraded for imprecision; low for OS because of risk of bias and imprecision/incomplete reporting; low for the exploratory legacy locoregional and distant recurrence analyses because of indirectness and imprecision; and very low for comparative severe-toxicity evidence because definitions, time windows and denominators were not sufficiently comparable. No separate certainty rating was assigned to the oxaliplatin-only sensitivity subset or to quality of life because these were not treated as independent confirmatory evidence bodies.

DISCUSSION

This review included eight independent randomized comparisons, seven of which contributed DFS/RFS estimates. The all-era point estimate favored CRT, whereas the oral fluoropyrimidine–platinum subgroup and the analysis excluding two legacy secondary reconstructions were less precise. The point estimates remained directionally similar; statistical significance in one analysis but not another does not demonstrate treatment-effect modification. The OS estimate was also imprecise. Exploratory legacy evidence suggested better locoregional control, but this signal is not equivalent to a demonstrated survival advantage under current treatment conditions.

Earlier syntheses[26-28] reported improved DFS and local control, but they were dominated by the first three or four Asian trials and necessarily preceded ARTIST 2, Qiao et al[22], and Wang et al[23].

A plausible clinical explanation is that the incremental value of local treatment depends on surgical quality, systemic disease control and baseline recurrence risk. These mechanisms were not directly tested by the aggregate analysis. Wang et al[23] reported a DFS HR of 0.98 and an OS HR of 0.86, and ARTIST 2 reported a direct-comparison DFS HR of 0.971[21,23]. Their estimates contribute to uncertainty about a consistent strategy-level benefit, rather than proving that radiotherapy has no biological or clinical effect. Differences in CT exposure between treatment arms further limit attribution of pooled effects to radiotherapy alone.

The Korean guideline gives a conditional recommendation against adjuvant CRT after curative gastrectomy for pathological stage II/III disease, while the ASTRO recommends postoperative CT alone after R0 resection with adequate D2 Lymphadenectomy in the relevant CT-eligible setting[29,30]. Chinese guidance provides additional treatment-pathway context[31,32]. These are guideline recommendations, not evidence that the confidence intervals in this review satisfy an equivalence criterion.

For patients with R0 D2-resected gastric cancer receiving an effective postoperative CT strategy, the present evidence is insufficient to establish a consistent incremental DFS or OS benefit from routine CRT. Importantly, the four-trial doublet DFS CI (0.66-1.18) is compatible with potentially important benefit and harm. Failure to establish a benefit is not proof of no effect, noninferiority or equivalence. Positive margins, inadequate nodal dissection, inability to receive effective systemic therapy, and preoperative or perioperative treatment pathways fall outside the direct comparison synthesized here; recommendations for those settings should not be derived from this pooled estimate.

Signals in node-positive or other high-risk subgroups should not be interpreted as proof of predictive benefit unless a prespecified treatment-by-subgroup interaction is positive and independently validated. The high-risk subgroup result reported by Qiao et al[22] is therefore hypothesis-generating rather than a basis for replacing the overall randomized comparison. Future trials should stratify by operative quality, molecular subtype, nodal burden, and contemporary CT, and should report recurrence pattern, patient-reported outcomes, and treatment delivery uniformly.

Limitations

Strengths of this study include explicit trial-report linkage, separation of randomized and analyzed denominators, standardized effect direction, source-tagged numerical inputs, reproducible small-sample inference, record-level eligibility documentation, and author-reviewed RoB 2 and GRADE decisions. The audit workbook distinguishes primary reported effects, secondary reconstructed effects, and recomputed sensitivity analyses.

Several limitations should be considered. The review was not prospectively registered, and the treatment-backbone and provenance analyses are therefore presented as exploratory rather than prospectively prespecified. Some original database export files and historical screening records were not retained, although the recorded search strategies, screening counts, and available eligibility records are provided in the Supplementary materials. The original search did not directly include EMBASE or CENTRAL, and targeted ChiCTR verification was not a comprehensive registry search. Kwon et al[15] and Kim et al[16] remain dependent on secondary survival reconstructions, although exclusion analyses reduce reliance on these inputs. Some trial methods, protocol details and recurrence definitions were incompletely reported in accessible sources. Treatment regimens, CT exposure, follow-up and outcome definitions differed; I² = 0% does not remove this clinical heterogeneity or indirectness. Toxicity and quality-of-life reporting remained insufficiently comparable for quantitative synthesis. All included trials were East Asian and postoperative, limiting extrapolation to perioperative-treatment settings.

CONCLUSION

Current randomized evidence is insufficient to establish a consistent DFS or OS benefit from routine postoperative CRT over effective CT after R0 D2 gastrectomy. The doublet-subgroup estimate remains imprecise and does not rule out clinically important benefit or harm. These findings do not establish equivalence, noninferiority or absence of a treatment effect. A possible locoregional-control advantage is supported mainly by low-certainty legacy evidence. Treatment decisions should therefore consider the uncertainty and the specific clinical setting rather than treating a nonsignificant estimate as proof that additional radiotherapy is ineffective.