Development and validation of a predictive model for high-risk immune-related adverse events in gastric cancer patients treated with ICIs

Immune checkpoint inhibitors (ICIs) may cause immune-related adverse events (irAEs), ranging from mild to life-threatening. High-risk irAEs can lead to treatment discontinuation and higher mortality, though ICI-treated patients' death rate is under 5%. Currently, no reliable biomarkers predict irAEs' occurrence or severity. This study investigates the link between accessible biomarkers and high-risk irAEs in gastric cancer patients on ICIs, as well as to develop and assess a predictive model for such events. Data were collected from patients with gastric cancer who received ICIs therapy between May 2020 and March 2025. The incidence and risk factors associated with irAEs were analyzed using the chi-square test or the Mann-Whitney U test. Univariate and multivariate logistic regression analyses were conducted to develop a predictive model. This model was validated through 10-fold cross-validation and assessed using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), calibration curves, and decision curve analysis. A total of 184 gastric cancer patients receiving ICIs therapy were enrolled in this study. The incidence of irAEs of any grade was 21.2%, while the incidence of grade ≥3 irAEs was 12.5%. Multivariate logistic regression analysis identified NLR-1 (<i>p</i> < .001), NLR2-1 (<i>p</i> < .001), PLR-1 (<i>p</i> = .001), tumor thickness (<i>p</i> = .018), CV (<i>p</i> = .001), and intratumoral necrosis (p = .028) as independent predictors of grade ≥3 irAEs. The AUC of the developed model was 0.878, with a sensitivity of 78.26%, specificity of 80.12%, PPV of approximately 80.95%, and NPV of approximately 75.52%. The corrected C-index, derived from bootstrap resampling, was 0.849, and both calibration curves and decision curve analysis confirmed good calibration and clinical utility. These predictors may aid risk stratification and optimized patient management.