Development and internal evaluation of a clinical prediction model for blood lead re-elevation in lead poisoning patients: a retrospective cohort study.

Background: Blood lead can rise again after chelation therapy for lead poisoning, but few tools identify which patients are at highest risk. We developed and internally evaluated a model to predict this re-elevation from routinely collected clinical data. Methods: We conducted a retrospective cohort study of 163 patients hospitalized for lead poisoning at Chongqing Poison Control Center from 2014 to 2022. Blood lead re-elevation was ascertained from re-admission for lead poisoning after an initial response to chelation therapy. Logistic regression, random forest, and support vector machine models were evaluated. The dataset was randomly divided into training (70%, n = 114) and testing (30%, n = 49) sets. Model performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, calibration, and decision curve analysis. Results: Among 163 patients, 59 (36.2%) experienced blood lead re-elevation. The cohort comprised 44 children (27.0%) and 119 adults (73.0%), and 111 patients (68.1%) were male. Pediatric patients had a higher re-elevation rate than adults (52.3% vs. 30.3%, p = 0.016). The random forest model showed the best internal test-set performance, with an AUC of 0.849, accuracy of 83.7%, precision of 77.8%, and recall of 77.8%. Important predictors included blood lead-to-season ratio, baseline-to-follow-up blood lead ratio, serum calcium-to-blood lead ratio, blood lead-to-hemoglobin ratio, and hemoglobin-to-AST ratio. Conclusion: An internally evaluated model can identify patients at higher risk of blood lead re-elevation after chelation therapy and could guide risk-based follow-up. External validation is needed before clinical use.