Abstract: Objective To establish a prediction model for risk factors of acute kidney injury (AKI) indued by voriconazole for injection and carry out internal and external validation to ensure clinical safety of medication based on real-world data. Methods The medical records of patients who were treated with voriconazole for injection for a minimum of 3 days in our hospital and aged 18 or older between January 1, 2020 and June 30, 2023 were collected. Among them, patients who were discharged before January 1, 2023 were used as the modeling group, and the rest as the validation group. The basic information of patients and such data as the indicators of experimental examinations, results of clinical diagnosis and combined medications was retrieved. Patients were divided into the AKI group and non-AKI group according to the occurrence of AKI. Multivariate logistic regression was used to analyze the risk factors for AKI induced by voriconazole for injection, and a prediction model was established. Discrimination and calibration were evaluated based on the receiver operating characteristic (Receiver Operating Characteristic, ROC) area under the curve (Area Under Curve, AUC) and H-L test (Hosmerand Lemeshow Test). The applicability of the model was evaluated both internally and externally. Results A total of 625 patients (371 males, 254 females) were included in the study. Among them, there were 489 cases in the modeling group (296 males and 193 females), with 87 cases of AKI, compared with 136 cases in the verification group (75 males, 61 females), with 32 cases of AKI. Multivariate logistic regression analysis of the modeling group showed that bloodstream infections, kidney diseases, cardiovascular diseases, and the use of diuretics were risk factors for AKI, while CrCl (Creatinine Clearance) and Alb(Albumin) were protective factors. The AUC of the model group was 0.750 (95%CI: 0.692~0.808, P<0.001) while the H-L test χ² value was 7.535, P=0.480, compared with 0.821(95%CI: 0.749~0.893, P<0.001) and 13.924, P=0.084 in the verification group was, suggesting that the model had good discrimination. The maximum value of the Youden index of the model was 0.389, and the best value corresponding to the tangent point of the ROC curve was -0.976, the sensitivity was 56.3%, and the specificity was 82.6%. k-fold cross-validation suggested that the accuracy of the model was better (0.823) and the consistency was lower (0.191). Conclusion CrCl, Alb, bloodstream infections, kidney diseases, cardiovascular diseases, and diuretics are independent factors for AKI induced by voriconazole for injection. The prediction model of AKI established based on these 6 variables has some discrimination, which can assist clinical treatment and contribute to safe drug use.

Key words: voriconazole, injection, acute kidney injury, risk factors, real world data, Logistic regression, k-fold cross-validation, prediction model