Incorporating the metabolic ratio (C-VRC/C-VNO) into a LASSO-logistic model for predicting voriconazole-induced DILI: development and web-based implementation.

OBJECTIVE: Voriconazole (VRC) is widely used as an antifungal agent, yet its potential to cause drug-induced liver injury (DILI) remains a major clinical concern. Therefore, this study aims to identify the relevant risk factors for VRC-induced DILI, construct a predictive model using LASSO-Logistic regression, and present it as a nomogram. Additionally, a web-based prediction tool was developed to facilitate clinical application, ultimately providing a scientific basis for early risk assessment of DILI in patients receiving VRC therapy.

METHODS: Clinical data of inpatients who received VRC for the treatment or prophylaxis of fungal infections at the First People's Hospital of Changzhou from March 2022 to June 2025 were collected retrospectively. The patients were divided into a DILI group and a non-DILI group according to the diagnostic criteria for DILI. Univariate analysis was used to screen for variables with intergroup differences, followed by LASSO regression for further feature optimization. The screened variables were incorporated into a multivariate Logistic regression to construct a nomogram prediction model for DILI. The receiver operating characteristic (ROC) curve, calibration curve and decision curve analysis (DCA) were adopted to evaluate the discrimination, calibration and clinical applicability of the model, with the Bootstrap method for internal validation. Additionally, a web-based interactive prediction calculator was developed based on R Shiny.

RESULTS: A total of 250 patients were enrolled, among whom 47 developed DILI. Univariate analysis revealed statistically significant differences between the DILI group and the non-DILI group in seven variables, namely renal replacement therapy, concomitant bacterial infection, steady-state trough concentration of VRC (C-VRC), C-VRC/daily dose (C-VRC/DD), the ratio of C-VRC to VRC N-oxide concentration (C-VRC/C-VNO, MR), albumin (ALB) and serum creatinine (Scr) (all P < 0.05). LASSO regression further identified five variables associated with VRC-induced DILI: concomitant bacterial infection, C-VRC, MR, ALB and Scr. Multivariate Logistic regression analysis demonstrated that C-VRC, MR, ALB and Scr were independent risk factors for VRC-induced DILI (all P < 0.05). A nomogram prediction model was constructed based on the above factors, with the area under the ROC curve (AUC) of 0.818 (95% confidence interval [CI]: 0.748-0.888). The calibration curve showed a good fit between the predicted values and actual observed values of the model. DCA indicated that the model yielded favorable clinical net benefits within a certain range of threshold probabilities, and the Bootstrap method verified the good stability of the model. The developed web-based calculator enables the calculation of DILI risk probability in individual patients.

CONCLUSION: C-VRC, MR, ALB and Scr are important predictors of VRC-induced DILI. The nomogram prediction model constructed based on LASSO-logistic regression has good predictive performance and clinical applicability, and the developed web-based interactive prediction calculator improves the clinical convenience of the model. This model can be effectively used for the early risk assessment of DILI in patients receiving VRC, providing support for clinical individualized medication and risk prevention and control.