Prediction of QTc Prolongation in Acute Poisoning with Atypical Antipsychotics Using Machine Learning Techniques: A...

Atypical antipsychotics have experienced a significant increase in use across various disorders, coinciding with a rise in acute intoxication. This retrospective study predicts prolonged QTc interval and the necessity for mechanical ventilation (MV) in patients with acute atypical antipsychotic poisoning using machine learning techniques. This retrospective study included 355 patients with a mean age of 26.1 ± 9.6 years. The overall prevalence of the investigated outcomes was 5.5% for prolonged QTc interval and 7.1% for MV. Eight classifiers were developed, including Logistic Regression, Support Vector Machine, K-Nearest Neighbors, and five tree-based models: Random Forest, XGBoost, LightGBM, CatBoost, and Gradient Boosting Models. Model validation was conducted through external validation using the testing dataset and an internal five-fold cross-validation after optimizing the hyperparameters. As a predictor of prolonged QTc interval, all tree-based models achieved perfect specificity, recall, precision, accuracy, and area under the curve (AUC) of 100% using the training dataset. Similar performance was reported in models predicting the necessity for MV. Upon validation, the tree-based models for predicting prolonged QTc intervals maintained good AUCs, ranging between 0.930 and 0.958 in the training dataset and between 0.927 and 0.949 in the testing dataset. In terms of accuracy, the tree-based models exhibited good values in both external and five-fold cross-validation, with all values above 0.901. The observed declines in recall and precision during the validation of the proposed models underscore the need for future studies to utilize larger validation cohorts, thereby enabling the generalization of the proposed models in relevant clinical settings.