Population Pharmacokinetics and Exposure-Response Analysis of Voriconazole in ICU and Non-ICU Hospitalized Adults

Abstract

Background

Voriconazole has marked pharmacokinetic variability and a narrow therapeutic window, making individualized dosing difficult. Published population pharmacokinetic (PopPK) models most derive from ICU or general-ward patients, and exposure-response studies are limited. This retrospective study aimed to characterize voriconazole PopPK and exposure-response analysis in real‑world hospitalized patients containing both ICU and non-ICU patients.

Methods

Adults receiving voriconazole with therapeutic drug monitoring were included. PopPK analysis used NONMEM. Demographic variables and lab biomarkers were screened. Monte Carlo simulations were used to show covariates’ effect on drug exposure under standard dosing regimen. Model-derived voriconazole exposure was linked to treatment outcome and adverse drug reactions (ADR) using logistic regression.

Results

292 patients with 473 voriconazole concentrations were included. A one-compartment model with first-order elimination adequately described the data, with estimated population clearance, volume of distribution and derived half-life of 3.3 L/h, 92 L and 19 h, respectively. Voriconazole clearance increased with body weight and albumin and decreased with age, CRP, total bilirubin, and serum creatinine. These covariates decrease clearance inter-individual variability from 44.5% to 32.6 %; all parameters fell within bootstrap 90% interval. With a standard regimen 300 mg every 12 h for first day then 200 mg every 12 h, simulations showed underexposure in younger patients without organ dysfunction and overexposure in elderly patients with elevated CRP, total bilirubin or serum creatinine. In multivariable logistic regression, excessively high average concentration (Cavg) derived by model was associated with treatment failure (adjusted odds ratio 1.45 per mg/L, 95% CI 1.07–2.03, p=0.019). Cavg was not significant, whereas older age was associated with higher ADR risk (adjusted odds ratio 1.05, 95% CI 1.01–1.10).

Conclusion

Age, CRP, albumin, total bilirubin and creatinine explain clinically relevant variability in voriconazole exposure and indicate that a fixed regimen may not benefit all patients. The association between higher Cavg and treatment failure is observational and needs further prospective studies and external validations.

Graphical Abstract

Introduction

Voriconazole (molecular formula C₁₆H₁₄F₃N₅O) is a second-generation triazole antifungal with broad-spectrum activity against Aspergillus spp., fluconazole-resistant Candida, Fusarium and Scedosporium species, among others.¹ It acts by inhibiting lanosterol 14α-demethylase, an enzyme which is key in fungal ergosterol biosynthesis, leading to destruction of fungal cell‑membrane structure and function. It can be administered by intravenous infusion or orally as tablets or suspension. Although generally effective, its clinical use can be complicated and restricted by adverse drug effects including visual disturbances, hepatotoxicity, cutaneous reactions and photosensitivity, hallucinations, and QT-interval prolongation.^2,3 In addition, voriconazole is metabolized by CYP2C19, CYP2C9, and CYP3A4 and inhibits these enzymes, particularly CYP3A4, resulting in clinically important drug–drug interactions with enzyme inhibitors, inducers, and CYP-metabolized concomitant medications. Current international guidelines endorse voriconazole as a first-line agent for the treatment of proven or probable invasive aspergillosis and as prophylaxis in selected immunocompromised hosts.⁴

Although therapeutic-drug monitoring (TDM) is widely recommended, voriconazole dosing in hospitalized patients is still challenging due to high heterogeneity of patients. Only about half of patients achieve therapeutic window, and for critically ill patients the target achievement is even worse.^5–9 Model-informed precision dosing (MIPD), most commonly implemented using population pharmacokinetic (PopPK) models, could enable individualized dosing based on patient characteristics such as body size, renal function and other covariates.^10,11 However, covariate relationships in current PopPK studies for voriconazole are inconsistent, which might be due to models being based on relatively small dataset and selected populations such as critically ill, transplant or oncology patients, and might have limited biomarkers screened in models. These limit the generalizability of models and the biomarkers which explain the voriconazole exposure heterogeneity in hospitalized patients.

In addition, there are inconsistencies among exposure-response studies.¹² Some studies have found that adequate voriconazole exposure may predict good clinical response,^13,14 while others did not find correlation between AUC and efficacy.^15,16 The exposure-response evidence based on real-world hospitalized patients are important while current relevant studies are limited.

To address these gaps, we analyzed a real-world cohort of adult patients treated with intravenous or enteral voriconazole at Changsha Central Hospital, encompassing both general-ward and ICU settings. The primary objectives of the present study are (i) to quantify the impact of clinically important biomarkers on voriconazole exposure to explain the heterogeneity among patients, and (ii) to explore the association between model-derived voriconazole exposure and clinical efficacy. These findings may help guide individualized dosing based on TDM for vulnerable hospitalized patients.

Methods

Study Design and Data Collection

This retrospective, single-center observational study was performed in the Respiratory and Critical Care Medicine Department of Changsha Central Hospital, Hunan, China. Adult patients who received voriconazole between 11/2023 and 04/2025 were screened through the electronic medical record and therapeutic drug monitoring databases. The study protocol was reviewed and approved by the Medical Ethics Sub-committee of Changsha Central Hospital (approval No. 2026-YS-110), and the study was conducted in accordance with the principles of the Declaration of Helsinki. Owing to the retrospective design and the use of de-identified clinical data, the ethics committee formally waived the requirement for written informed consent.

Patients were eligible for inclusion in the pharmacokinetic analysis if they were aged ≥18 years, had received voriconazole for treatment of invasive fungal infection, and had at least one measurable voriconazole plasma concentration. Patients were excluded if they were pregnant or lactating or had missing dosing or sampling information for pharmacokinetic analysis. Both ICU and non-ICU patients were included. For the pharmacodynamic analysis, only patients with both model-derived exposure estimates and evaluable clinical outcome or adverse drug reaction data were included.

During the treatment period, the voriconazole intravenous loading dose of 6 mg/kg q12h for the first 24 h, followed by maintenance dose of 4 mg/kg q12h. For oral administration, patients receive 200 mg q12h if body weight is ≥40kg and 100 mg q12h for those below 40 kg. For voriconazole administered by feeding tube, the same dose was used. Dose or drug modifications were performed according to physician’s experience, therapeutic drug monitoring results, organ function, detected voriconazole resistant fungi (eg, detection of Mucor or Rhizopus), and suspected toxicity. Routine therapeutic drug monitoring (TDM) was performed according to clinical practice, and most samples were collected at trough concentrations during maintenance treatment. Whole blood samples were collected in EDTA tubes, centrifuged within 30 min, and plasma was stored at −80°C until analysis. Voriconazole plasma concentrations were measured using a validated HPLC method with ultraviolet detection. The lower and upper limits of quantification were 0.16 mg/L and 20.48 mg/L, respectively.

Demographic variables were collected from the medical records included age, sex, body weight, height, and care setting (ICU or non-ICU). Laboratory variables considered as candidate covariates included platelet count, C-reactive protein (CRP), procalcitonin (PCT), interleukin-6 (IL-6), erythrocyte sedimentation rate (ESR), immunoglobulin E (IgE), albumin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin, direct bilirubin, and serum creatinine. Additional microbiological markers included blood G test, blood galactomannan, and pulmonary galactomannan. Concomitant drug–drug interaction status and immune status were also recorded. Recorded medications with potential interactions included proton-pump inhibitors (e.g, omeprazole and esomeprazole), glucocorticoids (e.g, methylprednisolone), rifampin or rifapentine, and ritonavir-containing regimens. Medications that have moderate and strong interactions were coded as ±1 and ±2, respectively. Information on documented fungal pathogens, sites of infection, underlying diseases, and clinical outcomes was retrospectively collected from the electronic medical records. Because patients frequently had multiple concurrent diagnoses, underlying diseases were grouped into clinically relevant categories. For time-varying laboratory covariates, the measurement that closest to each voriconazole concentration within the corresponding treatment episode was used. For continuous covariates with less than 40% missingness, missing values were handled using cohort median imputation and those with a high proportion of missing values were excluded from formal covariate selection. Continuous variables (demographic, clinical or laboratory) were summarized as median (range) and were compared between ICU and non-ICU patients using the Mann–Whitney U-test. Categorical variables were summarized as counts and percentages and were compared using the chi-square test or Fisher’s exact test (depend on counts number). P-value <0.05 was considered statistically significant. Statistical analyses were performed using R (4.3.1).

The primary efficacy endpoint was a composite clinical outcome based on documented discharge information. Treatment success was defined as clinical improvement, discharge, rehabilitation, or transfer for continued care after improvement. Treatment failure included death, treatment discontinuation because of lack of efficacy, discharge against medical advice in the context of ongoing treatment failure, or abandonment of treatment. This composite endpoint was used because no standardized survival endpoints available and not reliable for efficacy evaluation. The primary safety endpoint was the occurrence of voriconazole-associated adverse drug reactions, including hepatotoxicity, neurotoxicity, hallucination, arrhythmia etc, identified from medical records.

Population Pharmacokinetic Model Development

A pharmacokinetic analysis of voriconazole was conducted using NONMEM^® version 7.5 with first-order conditional estimates with interaction (FOCE-I). Pirana and R were utilized to support the NONMEM analysis and data analysis. Intravenous doses were modeled as zero-order infusions using the recorded infusion duration, whereas oral and enteral doses were modeled using first-order absorption. Because the dataset was collected by routinely TDM trough concentrations, absorption rate (ka) and oral bioavailability (F) could not be reliably estimated and were fixed to literature values with ka = 1.2 h⁻¹ and F = 0.84.^17,18,19 Inter-individual variability was explored on clearance (CL) and volume of distribution (V) (Supplemental File 1). Residual error and structural models were evaluated by considering changes in the objective function values (dOFV) and model stability. The final pharmacokinetic model was selected as the model providing the best balance between statistical fit and clinical interpretability.

Covariate Model Building

The influence of candidate covariates on pharmacokinetic parameters was evaluated in a stepwise manner. Based on clinical plausibility and prior voriconazole literature, age, body weight, inflammatory markers, liver function markers, renal function markers, and clinical setting were screened as potential predictors of parameter inter-individual variability. Continuous covariates were modeled using normalized power functions centered on the population median, whereas categorical covariates were modeled using indicator variables. Covariate selection was performed using stepwise forward inclusion and backward elimination. For forward inclusion, covariates are kept with ΔOFV > 3.84 (corresponding p < 0.05 for 1 degree of freedom), for backward elimination, covariates were removed sequentially and retained in the final model if their removal resulted ΔOFV > 6.63 (corresponding p < 0.01). Biological plausibility, parameter precision and model stability were also considered.

Model Evaluation

The adequacy of the final PK model was assessed using standard goodness-of-fit (GOF) plots by R. The relative standard error (RSE) of population parameters was set to be lower than 60% and parameter shrinkage was below 30%.^20,21 Parameter precision and robustness were further assessed using nonparametric bootstrap analysis with 1000 replicates. To further validate the model, 1000 simulations were generated by PsN and visual predictive checks (VPCs), plotted in R. Because most observed concentrations were trough samples collected within a limited time window after dosing, the VPC was interpreted primarily as an assessment of predictive performance in the trough.

Monte Carlo Simulations

To illustrate the impact of covariates on voriconazole plasma concentrations in the final model, Monte Carlo simulations were performed in NONMEM. The dosing regimen consisted of 300 mg every 12 h during the first 24 h followed by 200 mg every 12 h, and plasma concentrations were simulated over 168 h (7 days). Different patient scenarios were defined using clinically plausible covariate combinations. The resulting median concentrations versus time were plotted graphically.

Exposure-Response Relationships

To investigate exposure-response relationships, individual voriconazole exposure metrics were derived from the PK model parameters. The primary exposure variable was average concentration Cavg which was calculated by cumulative area under the concentration–time curve (cAUC) divided by the corresponding time interval (Time). Additional exposure metrics considered for safety analysis included Cmax derived using NONMEM (Supplemental File 2).

Exposure metrics were summarized at the patient level and linked to clinical efficacy and safety endpoints. Univariable logistic regression was used to evaluate the association between candidate predictors and binary efficacy and toxicity endpoints. Candidate variables for multivariable analysis included clinically relevant patient characteristics and model-derived exposure metrics, including ICU status, Cavg, albumin, CRP, age, drug–drug interaction, platelets, creatinine, immune status, total bilirubin, weight, and sex. Variables with (P < 0.1) in univariable analysis were considered for multivariable logistic regression. Final models were constructed using backward elimination.

Results

Study Population

292 patients were included in the analysis, and a total of 473 trough concentrations were sampled by routine TDM practice. Of the 473 concentrations, 459 (97.0%) were obtained 10–13 h after the preceding dose, including 435 (92.0%) at approximately 11.5 h. Of the included patients, 217 were managed in non-ICU wards and 75 were managed in the ICU. Route of administration was intravenous only in 244 patients, oral only in 1 patient, and switched between routes in 47 patients. The patient characteristics are summarized in Table 1. The median age of the overall cohort was 70 years (range 20–94), and the median body weight was 55 kg (range 30–85). The cohort was predominantly male (232/60, male/female). Compared with non-ICU patients, ICU patients showed higher CRP, lower albumin and higher creatinine levels. Additional clinical characteristics and covariates missingness are summarized in Supplementary Table S1 and S2.

Table 1.

Demographic, Clinical, and Laboratory Characteristics of the Study Population

Patient Demographics (unit)	Total Median [Range]	Non-ICU [Range]	ICU [Range]	P values
Participants (number)	292	217	75
Gender (M/F)	232/60	174/43	58/17	0.598
Age (y)	70.0 [20.0, 94.0]	70.0 [20.0, 92.0]	72.5 [39.0, 94.0]	0.296
Weight (kg)	55.0 [30.0, 85.0]	55.0 [30.0, 80.0]	55.0 [35.0, 85.0]	0.04
Height (cm)	163.0 [145.0, 182.0]	163.0 [145.0, 180.0]	163.0 [150.0, 182.0]	0.961
PLT (109/L)	222.5 [26.0, 752.0]	223.0 [53.0, 591.0]	204.0 [26.0, 752.0]	0.02
CRP (mg/L)	40.5 [0.1, 250.0]	36.5 [0.1, 249.5]	54.8 [0.1, 250.0]	0.003
PCT (ng/mL)	0.2 [0.03, 16.0]	0.2 [0.03, 16.0]	0.3 [0.03, 9.5]	<0.0001
IL6 (pg/mL)	12.6 [0.3, 678.8]	12.6 [0.3, 678.8]	12.6 [4.1, 643.0]	0.45
ESR	49.0 [3.0, 126.0]	49.0 [3.0, 126.0]	49.0 [11.0, 108.0]	0.91
IgE (IU/mL)_	119.9 [0.3, 14,323.0]	119.9 [0.3, 9900.0]	119.9 [0.3, 14,323.0]	0.78
BloodG (pg/mL)	1.0 [0.2, 1000.0]	1.0 [1.0, 1000.0]	1.0 [0.2, 1000.0]	0.64
BloodGM	0.1 [0.0, 46.8]	0.1 [0.0, 46.8]	0.1 [0.0, 3.6]	0.61
PulGM	1.0 [0.0, 59.3]	1.0 [0.0, 59.3]	1.0 [0.0, 9.2]	0.49
Albumin (g/L)	31.0 [20.1, 47.0]	32.0 [20.1, 44.3]	28.5 [22.0, 47.0]	<0.0001
AST (U/L)	23.0 [4.0, 249.0]	22.0 [9.0, 180.0]	24.0 [4.0, 249.0]	0.01
ALT (U/L)	19.0 [3.0, 205.0]	18.5 [3.0, 130.0]	21.5 [4.0, 205.0]	0.11
Total bilirubin (μmol/L)	6.9 [1.8, 47.2]	6.9 [2.2, 41.0]	6.7 [1.8, 47.2]	0.92
Direct bilirubin (μmol/L)	2.5 [0.0, 35.8]	2.4 [0.0, 29.9]	2.8 [0.1, 35.8]	0.07
Creatinine (μmol/L)	67.8 [26.0, 434.0]	67.0 [26.0, 434.0]	77.5 [37.0, 365.0]	0.006
Route (PO only/IV only/Switch)	PO:1 / IV:244 / Switch:47	PO:0 / IV:180 / Switch:37	PO:1 / IV:64 / Switch:10
Conc (number)	473	323	150
Conc range (mg/L) a	0.20–15	0.20–13.98	0.45–15

Notes: Values are median (range) unless stated otherwise; conc: concentrations. ^a minimum and maximum values are measured in all participants in the respective groups.

Abbreviations: ICU, intensive care unit; CRP, C-reactive protein; PCT, procalcitonin; IL-6, interleukin-6; ESR, erythrocyte sedimentation rate; IgE, immunoglobulin E; AST, aspartate aminotransferase; ALT, alanine aminotransferase.

Structural Model

A one-compartment model with a first order absorption, no lag time and first order elimination with a combined residual error model fitted the voriconazole serum concentration curve.

Covariate Analysis

The final pharmacokinetic model identified age, body weight, CRP, albumin, total bilirubin, and serum creatinine as significant covariates on voriconazole clearance (stepwise selection and covariates correlation in Supplemental Tables 3 and 4). Body weight and albumin were positively associated with clearance, whereas age, CRP, total bilirubin, and creatinine were negatively associated with clearance. Incorporation of these covariates reduced the inter-individual variability in clearance. Parameter estimates of the base model, final model, and bootstrap analysis are presented in Table 2.

Table 2.

Population Pharmacokinetic Parameter Estimates of the Base Model, Final Model, and Bootstrap Analysis

	Basic Model			Final Model			Bootstrap of the Final Model
Parameter	Estimate	RSE%	Shrinkage%	Estimate	RSE%	Shrinkage%	Median	90% CI (Lower)	90% CI (Upper)
Ka (1/h)-FIX	1.2			1.2			1.2
F-FIX	0.84			0.84			0.84
V (L)	97	4		92	4		92	78	107
CL (L/h)	3.52	9		3.33	10		3.3	3.1	3.5
AGE on CL				−0.54	34		−0.53	−0.86	−0.27
Weight on CL				0.92	21		0.90	0.57	1.24
CRP on CL				−0.04	34		−0.04	−0.06	−0.01
ALB on CL				0.99	23		0.99	0.62	1.4
Total bilirubin on CL				−0.21	31		−0.21	−0.31	−0.11
Creatinine on CL				−0.23	30		−0.23	−0.33	−0.10
IIV-CL (%)	44.5	8.6	22.5	32.6	15.7	29	33.5	23.1	41.1
Proportional error	0.32	16.3	16	0.34	20.9	13	0.33	0.2	0.45
Additive error	0.48	31.7	16	0.34	48.1	13	0.33	0.07	0.66

Abbreviations: RSE, relative standard error; IIV, inter-individual variability; CL, clearance; V, volume of distribution; CRP, C-reactive protein; ALB, albumin.

Model Evaluation

Goodness-of-fit plots for the final model showed acceptable agreement between observed concentrations and both population and individual predictions (Figure 1). Conditional weighted residuals and individual weighted residuals were generally centered around zero without major systematic bias over time after dose. Bootstrap analysis demonstrated that the final parameter estimates were consistent with the bootstrap median and percentile intervals, supporting model robustness. Because the observed data were mainly on trough samples, the VPC contained a limited number of bins (Figure 2). Nevertheless, the simulated prediction intervals were broadly consistent with the observed trough-domain concentrations, supporting the adequacy of the final model for exposure characterization in routine TDM practice.

Figure 1.

Voriconazole goodness-of-fit. (a) Observed concentrations versus population predictions. (b) Observed concentrations versus individual predictions. (c) Conditional weighted residuals versus time after dose. (d) Individual weighted residuals versus time after dose.

Figure 2.

Visual predictive check of voriconazole models. Open circles represent observed voriconazole concentrations. Solid and dashed lines indicate the observed median and percentile trends, and shaded areas indicate the corresponding simulation-based prediction intervals.

Simulation Results

Simulations under the standard regimen of 300 mg q12h for the first 24 h then with 200 mg q12h intravenous with 2 h infusion for a typical 55 kg patients showed significant differences in voriconazole exposure under different scenarios (Figure 3). The reference patient with median covariate values achieved exposures within the conventional therapeutic range, whereas younger patients with no inflammation and organ dysfunction tended to show lower concentrations and the trough concentration could fall below 1 mg/L. In contrast, elderly patients with high inflammatory burden, hepatic dysfunction, or renal dysfunction markers showed substantially increased exposure after 24 hours (5.5 mg/L). Across scenarios, the simulations suggested that standard dosing may be associated with underexposure in younger, less severe patients and overexposure in older, more severe patients, and the variability might be high, supporting the need for early TDM based individualizing dose adjustments. Probability of target attainment was calculated for each scenario using the simulated patients within 1.0–5.5 mg/L, which could be found in Supplementary Table 5.

Figure 3.

Simulated voriconazole concentration–time profiles under standard dosing across representative clinical scenarios. Voriconazole concentrations were simulated over 168 h for patients receiving a loading dose of 300 mg every 12 h during the first 24 h followed by 200 mg every 12 h. (a) A typical patient (age 70 yrs, weight 55 kg, CRP 39 mg/L, albumin 31 g/L, total bilirubin 7 μmol/L, creatinine 67 μmol/L); (b) Young and healthy patient (age 30 y, weight 55 kg, CRP 10 mg/L, albumin 45 g/L, total bilirubin 2 μmol/L, creatinine 30 μmol/L); (c) Elderly with organ dysfunction (age 80 y, weight 55 kg, CRP 200 mg/L, albumin 20 g/L, total bilirubin 40 μmol/L, creatinine 300 μmol/L). Shaded ribbon represents the 90% prediction interval.

Exposure-Response Analysis

Among the 292 patients for the exposure-response analysis, the median model-derived exposure metric Cavg was 3.5 mg/L in the treatment success group (220 pt) and 4.3 mg/L in the treatment failure group (72 pt), shown in Figure 4. In univariable logistic regression, factors including SETTINGS (ICU status), Cavg, albumin, CRP, and age were significantly associated with clinical outcome. In the multivariable model, settings (adjusted OR 7.84, 95% CI 4.03–15.68, P<0.001) and Cavg (adjusted OR 1.46 per 1 mg/L increase, 95% CI 1.07–2.03, P=0.019) remained independently associated with outcome, whereas CRP showed a borderline association (P = 0.08). More details can be found in Supplementary Table 6.

Figure 4.

Boxplots of model-derived voriconazole exposure (Cavg) according to clinical efficacy (A) and adverse drug reaction status (B).

Among the patients evaluable for safety, 33 patients experienced at least one recorded voriconazole-associated adverse drug reaction. Age was associated with ADR occurrence in univariable analysis (OR 1.04 per year, 95% CI 1.01–1.08; P=0.03) and in multivariable analysis (adjusted OR 1.05, 95% CI 1.01–1.10; P=0.02). In contrast, calculated Cavg and Cmax were not significantly associated with ADR in either univariable or multivariable analysis (Supplementary Table 7).

Discussion

In our retrospective real-world cohort of 292 hospitalized adults from a mixed ICU and non-ICU settings, we developed a PopPK model of voriconazole, identified six clinically relevant covariates on clearance, and used the final model to explore exposure-response relationships. Voriconazole clearance decreased with older age, higher CRP, total bilirubin, and serum creatinine, whereas it increased with body weight and serum albumin. Although these covariates have been reported previously, our study evaluated them simultaneously in cohorts including both ICU and non-ICU patients. Incorporation of the identified covariates reduced interindividual variability in clearance from 44.5% to 32.6%, supporting their relevance for explaining exposure variability in patients. In the exposure-response analysis, our results suggested higher Cavg was associated with treatment failure, and for safety analysis, neither the Cavg or Cmax were associated with voriconazole adverse reaction.

The final structural model was a one-compartment model with fixed absorption phase parameters. This is because voriconazole is mainly sparse trough concentrations obtained during routine TDM, thus there is not enough information to estimate ka, bioavailability, or more complex models. The parsimonious model focused on clearance is appropriate when the main aim is to characterize factors affecting trough concentrations and to support precision dosing in routine clinical practice.

The observed effect of age and weight was consistent with previous studies. Increasing age was positively correlated with voriconazole exposure.^22–25 For bodyweight, although most studies treat weight as a covariate on V or on allometric scaling,^26–29 several have reported that body weight is positively correlated with voriconazole clearance and associated with lower trough concentrations.^30–32 These findings suggest that age and weight could contribute to exposure variability.

The associations between clearance and CRP, albumin, total bilirubin and serum creatinine should be interpreted as reflection of patients’ physiological and disease states changes. Increasing evidence supports inflammation could increase voriconazole exposure, likely through cytokine-mediated downregulation of CYP-mediated metabolism.^33–36 Similarly, hypoalbuminemia and high bilirubin were often shown in patients with liver dysfunction and might have decreased voriconazole clearance.^8,35,37–40. Even though voriconazole is primarily metabolized hepatically, some studies had found that renal function indicators (CRCL, etc) were correlated with voriconazole exposure.^18,41,42 These covariates may reflect the relationship between disease severity and organ dysfunction and voriconazole exposure.

Current guideline-oriented TDM frameworks generally target voriconazole trough concentrations around 1.0–5.5 mg/L,^2,43 while 2–4 mg/L for vulnerable patients like transplant patients⁴⁴ and below 4 mg/L for Asian patients,⁴⁵ and higher troughs have been linked to hepatotoxicity and neurotoxicity.⁴⁶ The simulations showed that patients with different conditions may have quite different drug exposures, which indicate that these covariates might be important for early TDM‑based dose adjustment.

Our study showed that a higher model-derived average concentration (Cavg) is associated with worse clinical outcome (adjusted OR 1.46). Study has shown that maximum clinical response was achieved at Cavg levels of 3–4 mg/L, and clinical response showed U shape relationship with concentrations,⁴⁷ which is consistent with our findings. The success group has median Cavg 3.5 mg/L while the failure group has median Cavg 4.3 mg/L). In addition, the ideal target for AUC is the AUC/MIC ratio >50 (assuming the voriconazole binding is 50%). The AUC_24h for total voriconazole is above 50 when the MIC <=1 mg/L. In our study, the average AUC (Cavg*time) in both groups was adequate drug exposure if account for 24 hours (median value of 85 mg·h/L for success and 104 mg·h/L for failure group), and thus the higher exposure might not benefit treatment outcome. However, this association should not be interpreted as evidence that higher voriconazole exposure causes poorer treatment outcomes. More importantly, our data are from retrospective cohort and are confounded by disease severity. Patients with higher inflammatory burden, poorer organ function, and greater clinical severity may have both lower clearance and worse outcomes, producing an apparent negative association between exposure and success. Differences in fungal susceptibility and infection sites etc. may provide additional sources of residual confounding.

In the safety analysis, age rather than Cavg or Cmax was associated with ADR risk. Studies and reviews have reported associations between higher trough concentrations and hepatotoxicity (> 3mg/L) and neurotoxicity (>4.0 mg/L)⁴⁵ vary by population and endpoint definition.^2,46 In our study, the absence of voriconazole exposure and adverse drug reaction may be due to limited event numbers (n=33).

This study has several strengths. It reflects real-world practice, includes both ICU and non-ICU patients, incorporates route switching, and evaluates various clinical covariates. In addition, the combination of PopPK modeling, simulation, and exposure-response analysis provides a clinically useful framework for dose individualization.

At the same time, several limitations must be stated. First, model misspecification was apparent at the extremes of the observed concentration range, which may be due to voriconazole’s ability to exhibit nonlinear pharmacokinetics. The sparse trough-dominated dataset did not support reliable estimation of nonlinear Michaelis–Menten parameters or more complex compartment models. Second, the retrospective single-center design limits control over confounding and endpoint adjudication. Third, the concentration dataset was sparse and heavily trough-dominated, requiring fixed ka and bioavailability and limiting support for more complex structural models. Fourth, several clinically interesting covariates, such as microbiological susceptibility, CYP genotypes including CYP2C19, and some inflammatory markers, were either not recorded, incompletely available or not retained due to high missingness in the final model. Median imputation for missing values can reduce covariate variability and introduce bias in covariate relationship. Fifth, the efficacy and safety analyses were based on secondary patient-level endpoints rather than prospectively timed PK-PD sampling. Fifth, the information provided by this retrospective study is insufficient to classify invasive fungal disease into “proven” and “probable” categories, and minimum inhibitory concentration information is missing and might contain selection bias. Exposure‑response analyses might contain many confounders, and interpretation of analysis results should be cautious. Lastly, the patients cohort was drawn primarily from respiratory and intensive care settings; the resulting covariate model may not be directly applicable to other patient populations. Sixth

Conclusion

In conclusion, this real‑world retrospective voriconazole study showed that voriconazole exposure in hospitalized adults is affected by many factors including age, weight, CRP, albumin, total bilirubin, serum creatinine, and that a standard maintenance regimen is unlikely to be appropriate across all clinical situations. The PopPK model could aid early TDM-guided individualized dosing, especially in elder severely ill patients. Higher Cavg is associated with treatment failure, but this should not be interpreted as a causal relationship and may include confounding factors. Future studies with external model validation, richer pharmacokinetic sampling, and more accurate recording of microbiological, susceptibility, genotype, and clinical information are needed in order to translating findings into routine individualized dosing recommendations.

Funding Statement

This research received no external funding.

Data Sharing Statement

The data presented in this study is available on request from the corresponding author. The data is not publicly available due to privacy and institutional restrictions on patient-level clinical records.

Ethics Approval

The study was conducted in accordance with the Declaration of Helsinki and approved by the Medical Ethics Sub-committee of Changsha Central Hospital (protocol code 2026-YS-110, April 2026).

Informed Consent Statement

Patient consent was waived by the Medical Ethics Sub-committee of Changsha Central Hospital owing to the retrospective study design and the exclusive use of de-identified clinical data.

Author Contributions

Conceptualization: H.C. and L.L.; Patient data collection: L.Y., Y.C., H. L.; methodology: L.L.; formal analysis: L.L.; data curation, L.L., L.Y. and Y.C.; writing—original draft preparation, L.L. and H.C.; writing—review and editing, L.L., L.Y., Y.C., H.L., X.Z. and H.C.; visualization, L.L.; supervision, H.C. and X.Z.; project administration, H.C. All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.