Linezolid induced thrombocytopenia in critically ill patients: Risk factors and development of a machine learning-based prediction model

Abstract

Introduction: Linezolid is an antimicrobial with broad activity against Gram-positive bacteria. Thrombocytopenia is one of its most common side effects often leading to severe complications. The aim of this study is to identify factors related with development of this condition in critically ill patients and to develop and evaluate a predictive machine learning-based model considering easy-to-obtain clinical variables.

Methods: Data was obtained from the Medical Information Mart for Intensive Care III. Patients who received linezolid for over three days were considered, excluding those under 18 years and/or lacking laboratory data. Thrombocytopenia was considered as a platelet decrease of at least 50% from baseline.

Results: Three hundred and twenty patients met inclusion criteria of which 63 developed thrombocytopenia and presented significant greater duration of treatment, aspartate-aminotransferase, bilirubin and international normalized ratio; and lower renal clearance and platelet count at baseline. Thrombocytopenia development was associated with a worse outcome (30 days mortality [OR: 2.77; CI95%: 1.87-5.89; P < .001], 60 days mortality [OR: 3.56; CI95%: 2.18-7.26; P < .001]). Thrombocytopenia was also correlated with higher length of hospital stays (35.56 [20.40-52.99] vs 22.69 [10.05-38.61]; P < .001). Median time until this anomaly was of 23 days (CI95%:19.0-NE). Two multivariate models were performed. Accuracy, sensitivity, specificity and AUROC obtained in the best of them were of 0.75, 0.78, 0.62 and 0.80, respectively.

Conclusion: Linezolid associated thrombocytopenia entails greater mortality rates and hospital stays. Although the proposed predictive model has to be subsequently validated in a real clinical setting, its application could identify patients at risk and establish screening and surveillance strategies.

Keywords: Critical care; Drug safety; Linezolid; MIMIC-III; Machine learning; Predictive model; Thrombocytopenia.