OBJECTIVE To investigate the influencing factors of linezolid-induced thrombocytopenia(LIT) in children, and to develop and validate a clinically applicable risk prediction model for LIT.

METHODS Pediatric inpatients who received linezolid at the Children's Hospital of Chongqing Medical University between January 2015 and December 2024 were retrospectively enrolled. Univariate analysis and least absolute shrinkage and selection operator(LASSO) regression were employed for variable selection. Multivariable Logistic regression was then performed to identify factors associated with LIT, and restricted cubic spline(RCS) analysis was applied to explore potential nonlinear relationships. Bootstrap resampling was used both to evaluate the stability of variable selection and to construct a nomogram prediction model. Internal validation of the model was also conducted via bootstrap resampling, with predictive performance assessed by the area under the receiver operating characteristic curve(AUC), calibration plots, and other discrimination metrics.

RESULTS A total of 731 eligible children were included, of whom 53(7.3%) developed LIT. LASSO regression combined with multivariable Logistic regression identified that longer linezolid treatment duration(OR=1.0505), lower baseline platelet count(OR=0.9902), higher baseline total bilirubin(OR=1.0174), higher baseline alanine aminotransferase(ALT)(OR=1.0017), and concomitant use of heparin agents(OR=5.3759) were independently associated with LIT. RCS analysis revealed nonlinear relationships for treatment duration and baseline platelet count(P for nonlinearity<0.05); however, the AUC difference between the standard linear model and the RCS-based nonlinear model was not statistically significant(DeLong test: Z=–1.112, P=0.266). Bootstrap resampling for variable stability showed that ALT was selected in only 46.7% of resamples, whereas the other four variables were selected in >80% of resamples. Consequently, the final nomogram was built using the four highly stable core variables. The nomogram achieved an AUC of 0.871(95% CI: 0.827~0.915) with an optimism-corrected AUC of 0.862. The calibration curve demonstrated good agreement between predicted and observed probabilities, with a calibration intercept of 0(95% CI: –0.577~0.584), a slope of 1(95% CI: 0.788~1.268), and a Brier score of 0.053. Decision curve analysis indicated that the model provided net clinical benefit across a reasonable range of threshold probabilities.

CONCLUSION In pediatric patients receiving linezolid, clinicians should closely monitor treatment duration, baseline platelet count, total bilirubin level, and concomitant use of heparin agents. ALT showed low stability in our bootstrap analysis, and there is limited evidence to support its role as an independent predictor; therefore, it should be interpreted with caution in clinical practice. The nomogram model based on the four highly stable core variables demonstrated good discrimination and clinical utility. Its robust and parsimonious structure facilitates easy implementation in clinical settings and may provide a valuable evidence-based reference for early identification of high-risk children and for developing individualized prevention and treatment strategies.