A Regression Model to Predict Linezolid Induced Thrombocytopenia in Neonatal Sepsis Patients: A Ten-Year Retrospective Cohort Study.

Background: Linezolid-induced thrombocytopenia (LIT) is the main factor limiting the clinical application of linezolid (LZD). The incidence and risk factors of LIT in neonatal patients were possibly different from other populations based on pathophysiological characteristics. The purpose of this study was to establish a regression model for predicting LIT in neonatal sepsis patients. Methods: We retrospectively included 518 patients and divided them into the LIT group and the non-LIT group. A logistic regression analysis was used to analyze the factors related to LIT, and a regression model was established. A receiver operating characteristic (ROC) curve was drawn to evaluate the model's predictive value. We prospectively collected 39 patients' data to validate the model and evaluate the effect of LZD pharmacokinetics on LIT. Results: Among the 518 patients, 103 patients (19.9%) developed LIT. The Kaplan-Meier plot revealed that the overall median time from the initiation of LZD treatment to the onset of LIT in preterm infants was much shorter when compared with term infants [10 (6, 12) vs. 13 (9.75, 16.5), p = 0.004]. Multiple logistic regression analysis indicated that the independent risk factors of LIT were lower weight at medication, younger gestational ages, late-onset sepsis, necrotizing enterocolitis, mechanical ventilation, longer durations of LZD treatment, and lower baseline of platelet level. We established the above seven-variable prediction regression model and calculated the predictive probability. The ROC curve showed that the predicted probability of combined body weight, gestational age, duration of LZD treatment, and baseline of platelet had better sensitivity (84.4%), specificity (74.2%), and maximum AUC (AUC = 0.873). LIT occurred in 9 out of 39 patients (23.1%), and the accuracies of positive and negative predictions of LIT were 88.9 and 76.7%, respectively. Compared with the non-LIT patients, the LIT patients had higher trough concentration [11.49 (6.86, 15.13) vs. 5.51 (2.80, 11.61) mg/L; p = 0.028] but lower apparent volume of distribution (Vd) [0.778 (0.687, 1.421) vs. 1.322 (1.099, 1.610) L; p = 0.010]. Conclusion: The incidence of LIT was high in neonatal sepsis patients, especially in preterm infants. LIT occurred earlier in preterm infants than in term infants. The regression model of seven variables had a high predictive value for predicting LIT. LIT was correlated with higher trough concentration and lower Vd.