Autoimmunity and gastric cancer: pooled risks without pooled biology, methodological cautions before clinical translation – correspondence

Dear Editor, Cheng et al report an ambitious meta-analysis linking a broad range of autoimmune conditions to gastric cancer, with a modest overall elevation in relative risk and higher incidence in men. Their effort is timely and clinically relevant for surgical oncologists who must decide when surveillance is justified and for whom[1,2]. Yet several methodological issues, if addressed, would make the findings more actionable at the bedside[2]. First, the synthesis mixes effect measures of different construct validity (adjusted relative risk (RR)/ hazard ratio (HR) with standardized incidence ratio (SIR)), and in places treats odds ratios as if they were risk ratios under a “rare outcome” assumption. Methodology texts caution against combining heterogeneous estimands, and the OR-to-RR shortcut can mislead when outcomes are not rare or baseline risks differ across regions[3,4]. A clearer analytic pathway is to meta-analyze HR/RR and SIR separately, then, if needed, apply a calibrated cross-walk, rather than presenting a single headline estimate. Second, external validity is sensitive to baseline risk. East Asia, especially Japan, has among the world’s highest gastric cancer rates and a strong male predominance; even a small relative increase can translate into a large absolute risk in such settings. Reporting absolute risk differences (per 100 000 person-years), stratified by region and sex, alongside relative effects would help surgeons gauge screening yield and counsel patients. Third, residual and unmeasured confounding remain plausible. Helicobacter pylori, a Group 1 carcinogen and the dominant causal driver of non-cardia gastric cancer, varies markedly by geography, cohort calendar time, and eradication policy[5]. So do exposures such as long-term proton-pump inhibitor (PPI) use and aspirin/ non-steroid anti-inflammatory drug, which have been associated with higher and lower gastric cancer risks, respectively, in observational data. Most importantly, surveillance intensity differs across autoimmune diseases and health systems, inviting detection bias if endoscopy is more frequent in some groups[6]. Quantifying robustness using E-values, how strong an unmeasured confounder would need to be to fully explain away a given association, would make the conclusions more transparent for clinicians. Fourth, biology is blurred by limited granularity. Many included studies did not distinguish histology (intestinal vs diffuse) or anatomic subsite. Yet autoimmune gastritis/pernicious anemia predominantly tracks with corpus-predominant atrophy, intestinal-type adenocarcinoma, and type I gastric neuroendocrine tumors; other autoimmune phenotypes may not share that pathway. Without histology/subsite stratification, it is hard to translate pooled signals into targeted surveillance recommendations. Finally, “protective” signals (e.g. ulcerative colitis, selected Japanese cohorts with rheumatoid arthritis) should be interpreted cautiously. Differences in H. pylori prevalence/eradication, medication profiles, and endoscopy frequency could readily shift observed risks. Meta-analyses consistently show an inverse association between H. pylori and IBD, complicating causal interpretation of any ulcerative colitis-gastric-cancer “protection.” Triangulation in medication-naïve cohorts, with careful adjustment for eradication history and endoscopic exposure, is advisable before inferring disease-intrinsic protection. We recommend that, in addition to relative risks, the authors systematically present absolute risk differences (per 100 000 person-years) with stratification by age, sex, and region (the current incidence reporting is a good starting point). Bayesian hierarchical meta-regression should be used to incorporate key moderators (region, effect-measure type, calendar period, and proxies for surveillance intensity), with leave-one-out influence diagnostics reported. Sensitivity analyses for unmeasured confounding, such as E-values, should be included. Conditions with a more plausible mechanistic bridge to gastric carcinogenesis (autoimmune gastritis/pernicious anemia and type 1 diabetes) ought to be prioritized for confirmatory analysis, whereas signals from small study sets should be treated as hypothesis-generating. Where feasible, an individual-participant-data meta-analysis should be pursued to enable histologic and anatomic-subsite stratification and to apply uniform lag-time windows. For clinical practice, a pooled estimate of RR ≈ 1.16 alone does not justify a blanket policy of universal gastric screening across all autoimmune diseases. A more defensible near-term approach is targeted endoscopic surveillance focused on autoimmune gastritis and other clearly high-risk phenotypes, integrated with H. pylori testing/eradication and the local baseline incidence, with any broader expansion contingent on more granular evidence. In conclusion, the current evidence does not justify universal gastric surveillance across autoimmune diseases. A prudent interim course is targeted endoscopic monitoring for autoimmune gastritis and clearly high-risk phenotypes, integrated with H. pylori testing/eradication and local baseline risk. Further work should report absolute risks, separate HR/RR from SIR, and address unmeasured confounding and histology/subsite granularity to enable truly actionable guidance for surgical practice.