Identification of prognostic factors and development of a prognostic model in patients with acute diquat (200 g/L) poisoning

INTRODUCTION: The increased use of diquat (200 g/L) liquid formulations has resulted in a substantial escalation in the number of diquat ingestions. Hence, it is valuable to identify prognostic factors and establish predictive models to assess clinical outcomes. METHODS: = 75) based on the admitting hospital. Risk factors were identified using multivariate logistic regression, random forest, and least absolute shrinkage and selection operator regression techniques. Generalized linear models were developed and converted into nomograms. The performance of the models was evaluated using receiver operating characteristic curves, calibration curves, and decision curve analyses. RESULTS: The study found four clinically relevant predictors of death: aspartate aminotransferase activity, lactate concentration, cystatin C concentration and neutrophil count. The model's discrimination was excellent, with an area under the curve of 0.899 (95% CI: 0.848-0.950) for the training set and 0.952 (95% CI: 0.885-1.000) for the test set as well as good calibration curves, indicating that the predicted probabilities matched the actual observed outcomes, and excellent decision curve analysis, indicating clinical usefulness. DISCUSSION: This study has some limitations. First, its retrospective nature inevitably introduced selection bias. Second, despite employing a multi-center design, the participating sites were geographically close. Thirdly, the overall sample size, as well as that of the non-survival cohort, was relatively small. CONCLUSION: We validated predictive models that can stratify the risk of early death in diquat poisoning. Future large-scale, prospective studies across diverse regions are warranted to validate the generalizability of the model.