Invasive Fungal Infections (IFD) typically occur in individuals with severely compromised immune systems, impaired host barrier functions, or exposure to high-risk environments (Taccone et al., 2015;Perreault et al., 2019;Boutin et al., 2024). Voriconazole, a broad-spectrum triazole antifungal, is extensively employed in treat severe IFD (Patterson et al., 2016;Yang et al., 2022). However, as the clinical use of voriconazole has increased, adverse effects such as drug-induced liver injury (DILI), neurotoxicity, visual disturbances, and gastrointestinal reactions have become more prevalent (Jin et al., 2016). Among these, DILI is the most common and severe adverse reaction, and it is a major reason for treatment discontinuation. DILI can be categorized into non-idiosyncratic and idiosyncratic types. Non-idiosyncratic liver injury is dose-or duration-dependent, with a short latency period, and is attributed to direct hepatotoxicity from drugs or their metabolites (Tiwari et al., 2025). In contrast, voriconazole-associated DILI is primarily idiosyncratic and linked to individual genetic and metabolic variations, as well as immune responses, with no clear dose dependency. Conventional animal toxicology studies often fail to accurately predict clinical toxicity risks in such cases (Tiwari et al., 2025;Du et al., 2023). Recent studies have reported a DILI incidence of 12.9%-32.45% during voriconazole therapy (Perreault et al., 2019;Shen et al., 2022;Hanai et al., 2023;Wang et al., 2022;Zhou et al., 2022), highlighting it as a critical safety concern in clinical practice. Early identification and timely intervention can significantly mitigate these risks; however, there remains a lack of specific predictive tools for voriconazole-related liver injury.Voriconazole-induced DILI has been linked to mitochondrial dysfunction, oxidative stress, and disrupted bile acid homeostasis (Watkins, 2019;Wu et al., 2020). Critically, the specific targets responsible for the induction of liver injury remain unidentified. Therefore, it is crucial to elucidate the risk factors associated with voriconazole-induced DILI and to predict its occurrence following voriconazole administration. Previous studies have employed machine learning algorithms to develop predictive models. These models utilized clinical features to forecast voriconazole trough plasma concentrations, aiding in liver injury risk assessment (Cheng et al., 2023). Furthermore, comparisons among various machine learning models revealed that the logistic regression model exhibited superior performance in predicting voriconazolerelated hepatotoxicity (Ma et al., 2023). Despite the development of these models, limitations persist, including inadequate interpretability, a strong dependence on clinical features, and a lack of intuitive visualization of the risk assessment process in clinical applications.This study delineates key predictors of voriconazole-associated DILI and to develop a nomogram-based risk prediction model utilizing the identified independent risk factors. This approach addresses the limitations of existing research by enhancing model interpretability and clinical utility, thereby facilitating individualized dosing regimens and improving medication safety.This study retrospectively analyzed the electronic health records of patients who used voriconazole for therapeutic and prophylactic indications of invasive fungal diseases at The First Affiliated Hospital of Jinan University from June 2020 to June 2024.Inclusion criteria: age ≥18 years, who received voriconazole for the treatment or prophylaxis of IFD, underwent liver function biochemical tests during voriconazole therapy, and had all baseline liver function indices (prior to voriconazole initiation) within the upper limit of normal (ULN).Exclusion criteria: age <18 years; presence of liver injury caused by viral hepatitis, autoimmune hepatitis, hepatic failure, alcoholrelated liver diseases, metabolic-associated and fatty liver diseases, drug and toxin-induced liver injury, biliary tract diseases, or hepatocellular carcinoma; absence of baseline liver function tests performed before voriconazole initiation; and baseline liver indices (alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (γ-GGT), total bilirubin (T-BiL), and alkaline phosphatase (ALP)) exceeding 1×ULN).The electronic health records of enrolled patients were retrospectively extracted from the hospital's integrated information system. The data collection encompassed demographic characteristics, therapeutic regimens, serial laboratory parameters (specifically liver function profiles), documented adverse drug reactions (ADRs), and clinical outcomes.Severity (Meunier and Larrey, 2023).The causal relationship between liver injury and voriconazole use was evaluated using the Roussel Uclaf Causality Assessment Method (RUCAM) (Danan and Teschke, 2015). Patients with a RUCAM score of ≥6 were classified as having highly probable or probable voriconazole-induced drug-induced liver injury. Those with a RUCAM score of <6 were excluded from further risk factor analyses due to insufficient evidence to reliably attribute liver injury to voriconazole therapy.Diagnosis of Liver Injury Clinical Types: The clinical type of liver injury is classified using the R-value, which is calculated as the ratio of ALT/(ALT ULN) to ALP/(ALP ULN). The categories are defined as follows: Hepatocellular Injury (R ≥ 5): Characterized by a predominant elevation of ALT or AST. Cholestatic Injury (R ≤ 2): Defined by elevated ALP and/or γ-GGT. Mixed Pattern (2 < R < 5): Exhibits combined hepatocellular and cholestatic biochemical features (Aithal et al., 2011;Colombo, 2020;Andrade et al., 2019;Freites-Martinez et al., 2021). Classification by Disease Course: Acute Liver Injury: Resolves within 6 months of onset. Chronic Liver Injury: Persists for more than 6 months with unresolved biochemical or histological abnormalities.Patient data were systematically entered and organized using Microsoft Excel 2020. After thorough verification, the dataset was imported into SPSS 27.0 (IBM Corp.) for comprehensive statistical analysis. Normality for all continuous variables was tested with the Kolmogorov-Smirnov test. For non-normally distributed data, between-group comparisons were performed using the Mann-Whitney U test, with data reported as median (interquartile range). Normally distributed continuous variables are presented as mean ± standard deviation and compared via Student's t-test. Categorical variables underwent χ 2 test or Fisher's exact test when the expected frequencies were less than.Least absolute shrinkage and selection operator (LASSO) regression was applied to identify significant risk factors for voriconazole-associated drug-induced liver injury by shrinking the coefficients of less relevant predictors to zero. To avoid omitting potential risk factors, candidate variables selected by both univariate analysis (P ≤ 0.1) and LASSO regression were included in a final multivariable logistic regression model to define independent risk factors (Hanai et al., 2023;Ranganathan et al., 2017).To ensure a robust and unbiased model evaluation, random stratified sampling was employed to partition the dataset into training and testing sets in a 7:3 ratio. The training set was utilized for model development, while the testing set served for model performance evaluation. The dataset was stratified based on the binary outcome of liver injury occurrence (yes/no). This approach ensured a comparable prevalence of positive events (liver injury = 1) across the two resulting subsets. Within each stratum, a simple random sampling procedure was implemented using a fixed random seed set.seed (42) to guarantee the reproducibility of the data split. The class imbalance between positive cases (hepatotoxicity) and negative cases was addressed using the Synthetic Minority Oversampling Technique. A nomogram model was developed using R software (version 4.4.2) to visualize the risk prediction algorithm. The model's performance was evaluated by its area under the curve (AUC), with internal validation via 1,000 bootstrap resamples to correct for overfitting. Its clinical utility was further quantified using decision curve analysis (DCA).A total of 768 patients who were exposed to voriconazole between June 2020 and June 2024 were retrospectively analyzed after applying the inclusion and exclusion criteria outlined in Figure 1. The cohort comprised 487 males (63.4%) and 281 females (36.6%), with 39 patients (5.1%) reported a history of alcohol use and 91 patients (11.8%) had a smoking history.The majority of patients were from the hematology department (n = 387, 50.4%), followed by the respiratory department (n = 120, 15.6%) and the ICU (n = 108, 14.1%). The identified fungal pathogens included Aspergillus spp. (n = 152, 19.8%), Candida spp. (n = 122, 15.9%), and other fungi (n = 63, 8.2%), such as fungal spores, Lichtheimia corymbifera, Trichosporon asahii, and Pneumocystis jirovecii.Infections predominantly involved the respiratory system (n = 365, 47.5%), followed by the urinary system (n = 38, 4.9%). Following voriconazole therapy, 508 patients (66.1%) demonstrated clinical improvement. Detailed results are presented in Supplementary Table S1.A total of 213 patients (27.7%) experienced ADRs during voriconazole therapy, with a median onset time of 4 days. The most common ADR was DILI (n = 95, 12.4%), followed by neurotoxicity (n = 25, 3.3%), visual disturbances (n = 9, 1.2%), and other reactions (n = 24, 3.1%), which included rash, flushing, transient fever, tinnitus, diarrhea, elevated serum creatinine, and palpitations. Among 95 patients with voriconazole-induced liver injury, the severity was graded as 1 in 44 (46.3%), 2 in 30 (31.6%), 3 in 18 (18.9%), and 4 in 3 (3.2%) patients. According to clinical phenotype, the injury was classified as hepatocellular in 33 (34.7%), cholestatic in 42 (44.2%), and mixed in 20 (21.1%) patients.Management strategies involved switching from oral voriconazole tablets to intravenous formulations for patients experiencing intractable nausea or vomiting, administering hepatoprotective agents (e.g., glutathione, polyene phosphatidylcholine) to those with abnormal liver function, adding mecobalamin for mild neurotoxicity, and discontinuing voriconazole in cases of severe ADRs. These interventions led to symptom resolution or significant improvement in 89.2% of affected patients (n = 190/213). Detailed results are presented in Supplementary Table S1.The study cohort of 768 patients was randomly divided into a training set (n = 537, 70%) and a test set (n = 231, 30%) using a 7: 3 allocation ratio. Clinical data from the training set underwent univariate analysis and LASSO regression analysis, followed by logistic regression to identify independent risk factors for voriconazole-induced liver injury. The results of these analyses were summarized in Table 1;Figure 2. In the training set (n = 537), patients were stratified into two groups: DILI (n = 62) and Non-DILI (n = 475). Analysis of the training set indicated that the occurrence of voriconazole-induced DILI was marginally associated with septic shock (P = 0.071). Significant associations were found with the concomitant use of sulfamethoxazole (P = 0.034), glucocorticoids (P = 0.049), caspofungin (P = 0.024), ezetimibe (P = 0.045), and terbutaline (P = 0.028). Additionally, borderline trends were noted for rabeprazole (P = 0.063), omeprazole (P = 0.088), bromhexine (P = 0.087), salbutamol (P = 0.088), and montelukast (P = 0.093). Laboratory parameters associated with DILI risk included an elevated white blood cell count (P = 0.089), procalcitonin (PCT) (P = 0.048), and total cholesterol (TC) (P < 0.05). These key indices were subsequently entered into a LASSO regression model, and variables with non-zero coefficients were retained. At λ.1se = 0.0367, the selected variables included sulfamethoxazole, caspofungin, glucocorticoids, β-blockers, bromhexine, montelukast sodium, terbutaline, ezetimibe, PCT, white blood cell count, and total cholesterol, which were then incorporated into a logistic regression to establish a new prediction model (Supplementary Table S2).The final model identified the concomitant use of glucocorticoids, ezetimibe, caspofungin, and elevated TC levels as independent risk factors for voriconazole-induced DILI. Detailed regression coefficients, odds ratios, and corresponding P-values are presented in Table 2.Based on the high risk factors identified through logistic regression, a nomogram for predicting voriconazole-induced DILI was constructed using R 4.4.2 (Figure 3). Each risk factorezetimibe use, glucocorticoid use, caspofungin use, and TC levels-was assigned a score ranging from 0 to 100 points, proportional to its contribution to the outcome. Individual scores were calculated based on the value of each predictor, and the total score was converted into a probability of DILI occurrence through a predefined functional relationship. Higher total scores correlated with an increased risk of DILI.The ROC curve for the training set (Figure 4A) demonstrated an AUC of 0.728 (95% CI: 0.660-0.797), with a sensitivity of 0.661, Frontiers in Pharmacology frontiersin.org Frontiers in Pharmacology frontiersin.org specificity of 0.722, and an optimal cut off value for TC of 4.485 mmol/L to stratify the risk of DILI.The calibration of the nomogram was evaluated to assess the accuracy of the predicted probabilities for clinical outcomes. The calibration curve illustrated the concordance between the predicted and observed event rates. Following internal validation with 1,000 bootstrap resamples, the calibration curve of the training set demonstrated close alignment with the ideal line (45 °reference), indicating good calibration performance with minimal deviation between the predicted and observed results (Figure 4B). The model's clinical value was assessed via DCA, as illustrated in Figure 4C. The black line, representing the "None" strategy, indicates the net benefit when no patients are diagnosed with DILI, while the light gray line, corresponding to the "All" strategy, reflects the net benefit assuming all patients are diagnosed with DILI. Within the threshold probability range of 20%-50% (0.2-0.5), the nomogram, depicted by the red curve, demonstrated a higher net benefit compared to both the "All" and "None" strategies. This finding suggests that the nomogram is optimal for guiding clinical decisions in patients with moderate-risk thresholds. At lower threshold probabilities, the model may also assist in identifying individuals who require intervention, thereby enhancing the overall net clinical benefit.Following the successful development of the prediction model, the nomogram underwent internal validation using the test set. ROC curve analysis (Figure 4D) revealed an AUC of 0.773 (95% CI: 0.689-0.857), with a sensitivity of 0.848 and specificity of 0.571, indicating a robust discriminative ability. The calibration curve for the test set displayed close alignment with the ideal line, reflecting minimal deviation between predicted and observed outcomes, thereby confirming good calibration performance (Figure 4E). DCA further demonstrated that the model provided substantial net clinical benefit at low-risk thresholds; however, its utility diminished progressively with increasing threshold probabilities, suggesting limited clinical applicability in high-risk scenarios (Figure 4F). Nomogram for predicting the risk of voriconazole-induced liver injury. All binary clinical variables are coded as 0 or 1, where 0 represents "No" (or absence) and 1 represents "Yes" (or presence). Specifically for the variables "ezetimibe," and a value of 0 indicates and 1 indicates total cholesterol The a score for each the total to the value on the which is then converted to the risk of liver injury in Pharmacology frontiersin.org et Voriconazole, is indicated for IFD et al., et al., 2020). However, the increasing clinical use of voriconazole has been associated with a in reported including DILI, neurotoxicity, and visual liver injury is of the most common and severe as a for the of drug development in clinical and on use et al., et al., 2016). hepatotoxicity for of drug between and and highlighting its significant on drug safety and et al., This study analyzed the risk factors for ADRs in patients voriconazole treatment or The overall incidence of ADRs was with DILI in of the observed DILI in cohort the incidence of hepatic reported in et al., et al., This may be attributed to genetic in a key involved in voriconazole a higher prevalence of resulting in a voriconazole of concentrations, and plasma levels et al., et al., The concomitant use of (e.g., voriconazole plasma concentrations, with (e.g., or (e.g., may drug thereby increasing the risk of hepatotoxicity et al., ADRs hepatic dysfunction, gastrointestinal disturbances, neurotoxicity, and visual which typically within of therapy Early hepatic injury is by elevated followed by in levels The in may as an for while elevation reflects hepatocellular is typically attributed to mitochondrial dysfunction, of bile acid homeostasis and oxidative (Watkins, 2019;Wu et al., 2020). This study identified TC and the concomitant use of ezetimibe, caspofungin, and glucocorticoids as independent risk factors for DILI. research indicates that voriconazole therapy serum and TC to and baseline with with voriconazole plasma et al., 2021). a of may hepatic the metabolic of voriconazole and increasing to DILI. cholesterol and function, in and et al., 2021). TC levels may also hepatotoxicity through and oxidative stress, both of which are to liver injury. cholesterol primarily via mitochondrial and cell In it function, and glutathione, et al., 2020). cholesterol and in and liver injury et al., study ezetimibe as a significant risk factor for voriconazole-induced liver injury. an cholesterol ezetimibe may bile or and drug (e.g., or voriconazole plasma and hepatotoxicity and ezetimibe is primarily via its on use of a may to increased plasma of ezetimibe, its hepatic and the metabolic on thereby or liver injury has been reported to severe cholestatic hepatitis, which after drug discontinuation. the of ezetimibe, ezetimibe is a of the responsible for biliary of bilirubin from with bilirubin for thereby bilirubin and to et al., studies have that a of of caspofungin is associated with a lower risk of hepatic it a for invasive et al., 2022). However, the high incidence of hepatotoxicity observed in cohort may toxicity between caspofungin and further studies to study included patients from the hematology where voriconazole was primarily for the prophylaxis of IFD in high-risk patients cell utilized as therapy for may with an increased risk of DILI. Previous studies that significantly voriconazole plasma concentrations, with the most in voriconazole levels and increasing the incidence of et al., et al., 2021). may voriconazole resulting in diminished plasma may to voriconazole to therapeutic however, (e.g., to drug and hepatotoxicity in patients. voriconazole via (e.g., may such as voriconazole which can or therapy may liver injury through such as which via and the of fatty et al., or through mitochondrial et al., their with may oxidative stress, while may such as thereby hepatocellular et al., study a significant between use and the risk of voriconazole-induced hepatotoxicity = = studies have demonstrated that a hepatoprotective effects in toxin-induced liver injury models, independent of its et al., hepatic injury and in a model of fatty liver by 3 et al., et al., 2022). Additionally, in and hepatic thereby oxidative stress, and in liver injury models et al., 2016). Furthermore, have been to hepatic in patients with by and enhancing et al., have demonstrated in oxidative in models of liver and their DILI remains studies are to the effects of on oxidative stress, responses, and in voriconazole-induced hepatotoxicity using animal models. Additionally, analysis of clinical cohort data is to the observed between use and voriconazole-related liver with that are by the represents an independent risk factor for DILI. therapy regimens, it is to the potential hepatotoxicity of each as well as the and to ensure both therapeutic safety and study developed a nomogram prediction model for voriconazole-induced DILI by identifying key predictors through binary logistic regression analysis. The model demonstrated predictive with AUC of 0.728 and 0.773 for the training and test revealed minimal deviation between predicted and observed with mean absolute of and DCA further the clinical utility of the model, substantial net across a range of threshold This study was the to identify elevated TC levels as a of voriconazole-induced characteristics, the nomogram for to stratify DILI risk and therapeutic such as or treatment However, study has trough plasma of which have been linked to hepatotoxicity in were excluded as a due to insufficient therapeutic drug data at trough plasma in research the model's predictive Additionally, a key for cholestatic liver injury, was included in the analysis of the records ALP Furthermore, study included a limited of cases (n = representing a ezetimibe was associated with a significantly increased risk = = in the model, such events are to selection studies with are to the between ezetimibe and voriconazole-associated DILI and to further its relationship. the of genetic in and critical voriconazole be assessed due to the of These may to in hepatotoxicity risk by drug and the concomitant use of ezetimibe, glucocorticoids, or caspofungin is significantly associated with an increased risk of voriconazole-induced DILI. on these identified risk factors, developed a nomogram prediction model for voriconazole-related DILI. This model the identification of high-risk treatment strategies to mitigate hepatotoxicity and the incidence of voriconazole-related adverse
Feng et al. (Fri,) studied this question.