Abstract:

The study aimed to construct and validate a logistic regression–based prediction model for tigecycline (TGC)–induced thrombocytopenia, to delineate its principal risk determinants, and to develop a clinically applicable tool for early risk stratification. A retrospective cohort study was performed using data from hospitalized patients who received TGC therapy at a tertiary medical center between January 2020 and January 2024. A total of 64 candidate clinical variables were initially considered and subsequently selected using a hybrid LASSO-Boruta algorithm. Model performance was assessed through 10-fold cross-validation, receiver operating characteristic (ROC) analysis, and calibration curve evaluation. A dynamic nomogram was further developed to facilitate bedside implementation. Of the 919 eligible patients, 224 experienced thrombocytopenia, corresponding to an incidence rate of 24.37%. Nine variables emerged as independent predictors, including age, intensive care unit admission, mechanical ventilation, maintenance dose of TGC, baseline platelet count, red blood cell count, serum creatinine, potassium, and blood urea nitrogen. The model achieved an area under the curve (AUC) of 0.744 in the training cohort and 0.736 in the validation cohort, with a Brier score of 0.149, indicating favorable discriminative ability and calibration. Decision curve analysis (DCA) demonstrated a meaningful clinical net benefit across threshold probabilities ranging from 18% to 63%. Overall, the proposed logistic regression model and its accompanying dynamic nomogram exhibited robust predictive performance and clinical utility, offering a practical and interpretable approach for the early identification of patients at elevated risk of TGC-induced thrombocytopenia.

Key words: Tigecycline, Thrombocytopenia, LASSO-Boruta algorithm, Risk factors, Prediction model