Age-stratified nomogram development and validation for predicting voriconazole trough concentrations above safety threshold in pediatric patients with hematologic malignancy.

PURPOSE: To develop and validate an age-stratified nomogram for predicting the risk of voriconazole (VRC) trough concentrations exceeding the safety threshold in pediatric patients with hematologic malignancies (HM).

METHODS

We retrospectively enrolled pediatric patients with HM who received VRC treatment and underwent therapeutic drug monitoring (TDM) in our hospital, stratifying them by age and randomly assigning them to the training and test cohorts at a ratio of 6:4 and in each stratum. We screened the variables by Least Absolute Shrinkage and Selection Operator (LASSO) regression and established age-stratified prediction models using multivariate logistic regression (Models 1 and 2: ≥ and < 11 years, respectively). We evaluated model performance using the area under the receiver operating characteristic curve (AUC) and assessed the clinical net benefit by decision curve analysis.

RESULTS

LASSO regression identified C-reactive protein, albumin, and neutropenia as well as blood urea nitrogen and direct bilirubin as independent VRC supratherapeutic concentration-influencing factors in children ≥ and < 11 years, respectively. In the test cohort, Model 1 and 2 AUCs were 0.835 and 0.814, respectively. At a high sensitivity threshold (≥ 95%), the models yielded specificities of 43.75% and 29.76%, which could reduce TDM frequency by 33.70% and 23.41%, respectively. Decision curve analysis showed both models yielded greater clinical net benefit than the "treat-all" or "treat-none" strategies.

CONCLUSIONS

The developed age-stratified prediction model could effectively identify pediatric patients with HM at high risk of VRC over-exposure, demonstrating good discrimination and clinical utility, potentially facilitating early risk warning.