Combined predictive value of dose, MODS, and NLR for lethal risk of diquat poisoning: a retrospective cohort study

Objective Accurate risk assessment for fatal diquat poisoning is critical to improving clinical outcomes. Therefore, this study aimed to evaluate the combined value of the multiple organ dysfunction syndrome (MODS) and neutrophil-to-lymphocyte ratio (NLR) in predicting in-hospital all-cause mortality from acute diquat poisoning. Methods This work retrospectively enrolled diquat self-poisoning patients in Guangzhou 12th People’s Hospital between 2017 and 2023. Demographic, exposure, and 24-h laboratory data were extracted. A time-dependent Cox model evaluated MODS as a time-varying predictor. LASSO and logistic regression identified independent predictors and established the prediction equation; restricted cubic splines assessed dose-response. Bootstrap internal validation evaluated discrimination and calibration; however, variable selection was not repeated within each resample, and therefore the optimism-corrected performance metrics should be interpreted with caution. SHAP quantified feature contributions. Results NLR was significantly elevated in non-survivors compared to survivors (median 31.37 vs. 11.19, P < 0.001). In the time-dependent Cox analysis, MODS remained independently associated with mortality (hazard ratio [HR] = 35.24, 95% CI: 13.57–91.50, P < 0.001). In the exploratory logistic regression, MODS (adjusted odds ratio [aOR] = 34.63, 95% CI: 7.47–245.65, P < 0.001), diquat dose (aOR = 1.34 per 10 mL, 95% CI: 1.22–1.63, P < 0.001), and NLR (aOR = 1.07, 95% CI: 1.02–1.13, P = 0.01) were independently associated with in-hospital all-cause mortality. The combined model achieved an AUC of 0.976 (Bootstrap 95% CI: 0.951–0.998), with an optimism-corrected AUC of 0.970, an apparent calibration intercept of 0, an apparent calibration slope of 1, and a Brier score of 0.0553 (95% CI: 0.0140–0.0775). SHAP values ranked feature importance as diquat dose (0.179) > MODS (0.145) > NLR (0.069). Sensitivity analyses confirmed the robustness of these findings. Conclusion MODS and NLR were independently associated with in-hospital all-cause mortality in acute diquat poisoning. When applied after the first 24–72 h of admission, the combined model shows promise as a potential exploratory tool for risk stratification; however, external validation in larger multicenter cohorts is required before any consideration of bedside application.