Population Pharmacokinetics and Pharmacodynamics of Itraconazole for Disseminated Infection Caused by Talaromyces marneffei

ABSTRACT

First-line treatment of talaromycosis with amphotericin B deoxycholate (DAmB) is labor-intensive and toxic. Itraconazole is an appealing alternative antifungal agent. Pharmacokinetic data were obtained from 76 patients who were randomized to itraconazole in the Itraconazole versus Amphotericin B for Talaromycosis (IVAP) trial. Plasma levels of itraconazole and its active metabolite, hydroxyitraconazole, were analyzed alongside longitudinal fungal CFU counts in a population model. Itraconazole and hydroxyitraconazole pharmacokinetic variability was considerable, with areas under the concentration-time curve over 24 h (AUC₂₄) of 3.34 ± 4.31 mg·h/liter and 3.57 ± 4.46 mg·h/liter (mean ± standard deviation), respectively. Levels of both analytes were low; itraconazole minimum concentration (C_min) was 0.11 ± 0.16 mg/liter, and hydroxyitraconazole C_min was 0.13 ± 0.17 mg/liter. The mean maximal rates of drug-induced killing were 0.206 and 0.208 log₁₀ CFU/ml/h, respectively. There were no associations between itraconazole C_min/MIC and time to sterilization of the bloodstream (hazard ratio [HR], 1.01; 95% confidence interval [CI], 0.99 to 1.03; P = 0.43), time to death (HR, 0.99; 95% CI, 0.96 to 1.02; P = 0.77), or early fungicidal activity (EFA) (coefficient, −0.004; 95% CI, −0.010 to 0.002; P = 0.18). Similarly, there was no relationship between AUC/MIC and time to sterilization of the bloodstream (HR, 1.00; 95% CI, 0.99 to 1.00; P = 0.50), time to death (HR, 1.00; 95% CI, 0.99 to 1.00; P = 0.91), or EFA (coefficient, −0.0001; 95% CI, −0.0003 to 0.0001; P = 0.19). This study raises the possibility that the failure of itraconazole to satisfy noninferiority criteria against DAmB for talaromycosis in the IVAP trial was a pharmacokinetic and pharmacodynamic failure.

TEXT

Talaromyces marneffei is a thermally dimorphic fungus with endemicity limited to Southeast Asia (northern Thailand, Vietnam, and Myanmar), South Asia (northeastern India) and East Asia (southern China, Hong Kong, and Taiwan) (1). In these regions, talaromycosis is the third most common opportunistic infection after tuberculosis and cryptococcal meningoencephalitis and a leading cause of morbidity and mortality among people living with HIV/AIDS (2, 3). Mortality rates are as high as 30% at 6 months, despite modern antifungal chemotherapy and supportive care (3–5). Talaromycosis is also increasingly reported in patients with underlying immunosuppressive conditions other than HIV (6). Disseminated infection is the most common form and manifests as fever, bone marrow involvement (anemia, leukopenia, and thrombocytopenia), skin lesions, weight loss, lymphadenopathy, hepatosplenomegaly, respiratory failure, and circulatory collapse (3, 7).

Itraconazole is an orally bioavailable broad-spectrum antifungal agent with a relatively favorable safety profile in comparison to other systemic antifungal agents (8). It is used for the prevention and treatment of a wide range of fungal diseases, including aspergillosis, candidiasis, and those caused by dimorphic fungi, such as histoplasmosis, blastomycosis, and talaromycosis (9–13). Itraconazole is lipophilic, poorly soluble at physiological pH, and highly protein bound (14). It partitions into lipid-rich tissues, and drug exposure increases at the effect site in the setting of tissue infection and inflammation (9, 15). Higher exposures are associated with greater clinical response but also increased likelihood of toxicity (11, 16–22). Itraconazole was recently shown in the Itraconazole versus Amphotericin B for Talaromycosis (IVAP) trial to be inferior to amphotericin B deoxycholate (DAmB) for the induction phase of treatment for talaromycosis, with a risk of death at week 24 of 21.0% compared to 11.3% in the amphotericin B group (P < 0.001) (8).

This study investigated the population pharmacokinetics (PK) and pharmacodynamics (PD) of itraconazole for patients with talaromycosis. The PK-PD study was performed as a substudy of the IVAP trial (8). The pharmacodynamics of itraconazole were estimated using serial quantification of fungal CFU in the bloodstream of patients who were fungemic.

RESULTS

Study participants.

PK data were obtained for a randomly selected subgroup of 76 patients in the itraconazole treatment arm; PD data were available for 65 of these. All 76 patients had culture-positive disseminated talaromycosis, with T. marneffei isolated from blood, skin lesions, lymph nodes and/or serous fluid.

Forty-three percent of patients were female. The median age was 33 years (interquartile range [IQR], 29 to 36), weight was 45 kg (IQR, 40 to 50), body mass index was 17.1 kg/m² (IQR, 15.6 to 19.0), and estimated glomerular filtration rate (eGFR) using the abbreviated Modification of Diet in Renal Disease Study (MDRS) calculation was 113.9 ml/min/1.73 m² (IQR, 86.7 to 139.1). All patients had advanced HIV disease, with a median CD4 cell count of 9 cells/μl (IQR, 4 to 20).

Pharmacokinetic data.

The data set included 1,316 itraconazole observations and 1,314 hydroxyitraconazole (OH-itraconazole) observations, with an arithmetic mean of 17.3 observations of each analyte per patient. A total of 95 itraconazole observations and 106 OH-itraconazole observations were below the lower limit of quantification (LLQ). The median ratio of OH-itraconazole concentration to itraconazole concentration per time point was 0.905. This ratio did not change significantly over time (ratio = 1 + 0.0003 × time; r² = 0.03; P = 0.32). Figure 1 shows the raw concentration data for both analytes.

FIG 1.

Drug concentrations in 76 patients. Black diamonds represent itraconazole concentrations. White triangles represent hydroxyitraconazole concentrations. Arrows represent approximate times of itraconazole administration. The median ratio of the concentration of hydroxyitraconazole to itraconazole per time point is 0.905.

Population pharmacokinetic analysis.

The PK model was built in two stages. First, the parent drug (itraconazole) was modeled in isolation with a saturable term for drug clearance. The second stage of model building involved adding metabolite (OH-itraconazole) data into the model. The saturable clearance mechanism of the parent drug fed into the central compartment of the metabolite. An acceptable fit of the base model to the data was achieved with first-order clearance of the metabolite from the central compartment.

The potential impact of covariates on the PK was assessed. Multivariate linear regression of covariates was performed using the PMstep command in Pmetrics; this did not reveal any significant associations between the Bayesian posterior PK estimates (i.e., clearance and volume) versus age, sex, weight, body mass index (BMI), renal function (creatinine level and eGFR), or CD4 cell count. Hence, further model building was not performed. The final PK model comprised 5 compartments, representing the gastrointestinal tract, the parent drug in the central compartment (circulation), the metabolite in the central compartment, the parent drug in the peripheral compartment, and the metabolite in the peripheral compartment (Fig. 2).

FIG 2.

Structure of the pharmacokinetic-pharmacodynamic model for itraconazole in talaromycosis. After ingestion, itraconazole is absorbed from the gastrointestinal tract into the bloodstream according to the absorption rate constant, K_a. Saturable hepatic metabolism of itraconazole results in the presence of hydroxyitraconazole in the bloodstream. Both itraconazole and hydroxyitraconazole undergo bidirectional transfer between the central and peripheral compartments. Hydroxyitraconazole is partially removed from the central compartment through first-order clearance. The pharmacodynamic effect on the burden of talaromycosis in the bloodstream is produced by the additive effect of itraconazole and hydroxyitraconazole in the bloodstream. Black dashed arrows indicate clearance mechanisms. Gray dashed arrows indicate the pharmacokinetic compartments that produce pharmacodynamic effects. The single asterisk indicates saturable clearance of parent drug by hepatic metabolism, determined as V_max/{V_max /[K_m × V_cp + X(2)]}. The double asterisk shows first-order clearance of metabolite (CLm/V_cm). GI, gastrointestinal; PD, pharmacodynamic; X(1), amount of itraconazole in the gut; X(2), amount of itraconazole in the bloodstream; X(3), amount of itraconazole in the peripheral compartment; X(4), amount of hydroxyitraconazole in the bloodstream; X(5), amount of hydroxyitraconazole in the peripheral compartment; K23, K32, K45, and K54, first-order transfer constants between central and peripheral compartments; N, number of CFU in the bloodstream.

The observed-versus-predicted values for the plasma concentrations of itraconazole and OH-itraconazole are shown in Fig. 3A, and plots of weighted residuals against predicted concentrations and time are displayed in Fig. 3B. Parameter values for the final model are summarized in Table 1. Mean predicted parameter values described the observed values better than medians and were used in subsequent modeling and analyses. The mean area under the concentration-time curve from 0 to 24 h (AUC_0–24) of itraconazole was 3.34 mg·h/liter (standard deviation [SD], 4.31 mg·h/liter; coefficient of variation [CV], 129%), and median AUC_0–24 was 1.91 mg·h/liter. For hydroxyitraconazole, the mean AUC_0-24 was 3.57 mg·h/liter (SD, 4.46 mg·h/liter; CV, 125%), and median AUC_0–24 was 2.27 mg·h/liter. The mean minimum concentration (C_min) of itraconazole was 0.11 mg/liter (SD, 0.16 mg/liter; CV, 147%), and median C_min was 0.06 mg/liter. For OH-itraconazole, the mean C_min was 0.13 mg/liter (SD, 0.17 mg/liter; CV, 132%), and median C_min was 0.08 mg/liter.

FIG 3.

(A) Scatterplots of observed versus predicted values for the chosen population pharmacokinetic model after the Bayesian step. (Left) Itraconazole concentrations. r² = 0.69; intercept = −0.03 (95% confidence interval, −0.09 to 0.03); regression slope = 1.01 (95% CI, −0.95 to 1.06). (Right) OH-itraconazole concentrations. r² = 0.73; intercept = −0.05 (95% CI, −0.12 to 0.00); regression slope = 0.99 (95% CI, 0.94 to 1.04). (B) Each panel displays the weighted residual error values against predicted concentrations in the scatterplot on the left and against time in the center. On the right is a histogram of residuals with the normal curve superimposed. (Top) Itraconazole concentrations. Mean weighted residual error, 0.01 (P = 0.78; standard deviation = 0.77). Shapiro-Wilk test for normality: P = 0. (Bottom) OH-itraconazole concentrations. Mean weighted residual error, 0.11 (P = 0.04; standard deviation = 1.35). Shapiro-Wilk test for normality: P = 0.

TABLE 1.

Parameter estimates for the final pharmacokinetic model

Parameter (units)a	Mean	Median (95% credibility intervalb)	SD
Ka (h−1)	1.781	0.238 (0.103–0.397)	4.275
Vcp (liters)	783.762	668.104 (500.251–1,060.468)	287.724
K23 (h−1)	16.200	20.831 (1.782–28.103)	12.804
K32 (h−1)	6.449	1.152 (0.411–3.131)	9.554
Vmax (mg/h)	55.836	37.767 (25.013–65.382)	36.512
Km (mg/liter)	0.426	0.223 (0.162–0.435)	0.473
K45 (h−1)	1.562	0.005 (0.005–0.005)	5.501
K54 (h−1)	25.713	29.986 (29.977–29.990)	9.191
SCLm (liters/h)	133.351	135.899 (100.738–199.961)	70.765
Vcm (liters)	866.057c	738.255c (552.777–1,171.817)	287.724

^aK_a, absorption rate constant from the gut to the central compartment; V_cp, volume of the central compartment for itraconazole; K23, first-order transfer constant of itraconazole from the central to the peripheral compartment; K32, first-order transfer constant of itraconazole from the peripheral to the central compartment; V_max, maximal rate of enzymatic metabolism of itraconazole; K_m, concentration of itraconazole in the central compartment at which clearance is half maximal; K45, first-order transfer constant of hydroxyitraconazole from the central to the peripheral compartment; K54, first-order transfer constant of hydroxyitraconazole from the peripheral to the central compartment; SCLm; first-order clearance of hydroxyitraconazole from the central compartment; V_cm, volume of the central compartment for hydroxyitraconazole.

^bUsed in Bayesian statistics to represent the interval within which an unobserved value falls with a 95% probability.

^cFixed as 1.105 × V_cp.

In vitro susceptibility tests.

MICs against itraconazole were determined using Clinical and Laboratory Standards Institute (CLSI) methods (M38) (23) for isolates from 69 patients. Of these, 70% had an MIC of 0.008 mg/liter, 27% had an MIC of 0.016 mg/liter, and 3% had an MIC of 0.03 mg/liter.

Population pharmacodynamic modeling.

PD data (CFU per milliliter of blood) were available from 65 patients who received itraconazole. Of those 65 patients, 52 were fungemic at baseline. In total, 452 quantitative cultures were obtained, with a mean of 7 observations per patient. There was a large degree of variation in the time to sterilization of blood cultures, with a mean of 330 h and a range of 13 to 3,306 h. Early fungicidal activity (EFA) was calculated by performing a linear regression of log₁₀ CFU/ml versus day of blood culture per patient. EFA was defined as the slope of the regression line. The median EFA was −0.3 (range, −1.6 to 0.1) log₁₀ CFU/ml/day.

Of the 452 quantitative culture results available, 162 were below the LLQ. Handling values below the LLQ as LLQ/2 provided the best model fit to the data, with acceptable levels of bias and imprecision (Fig. 4). The parameter estimates for the population PD model are summarized in Table 2. Mean parameter values predicted the observed values better than median values. After completion of the 14-day induction phase of treatment, 25 of the 52 patients who were fungemic at baseline remained fungemic (48%) (Fig. 5A). The time course of the reduction in fungal burden over the first 14 days of itraconazole treatment in all 65 patients for whom PD data were available is displayed in Fig. 5B.

FIG 4.

Scatterplots of observed versus predicted values for the chosen population pharmacodynamic model after the Bayesian step. For the linear regression, r² = 0.68; intercept = −0.07 (95% confidence interval, −0.09 to 0.22); slope = 0.99 (95% CI, 0.92 to 1.07).

TABLE 2.

Parameter estimates for the final pharmacodynamic model

Parametera	Mean	Median	SD
Kkillmaxp (log10CFU/ml/h)	0.206	0.010	0.436
Hp	2.194	2.999	1.176
EC50p (mg/liter)	13.449	14.976	3.222
Kkillmaxm (log10CFU/ml/h)	0.208	0.055	0.422
Hm	1.325	0.671	0.957
EC50m (mg/liter)	8.640	6.697	3.515
IC (CFU/ml)	1,442.141	5.909	4,412.916

^aKkill_max, maximum rate of drug-induced killing of T. marneffei; H, Hill/slope function; EC₅₀, plasma concentration of drug that induces a half-maximal killing rate; IC, estimated fungal density just prior to initiation of itraconazole. Parameters with the suffix “p” describe the parent drug, itraconazole; those with the suffix “m” refer to the metabolite, hydroxyitraconazole.

FIG 5.

(A) Kaplan-Meier plot of the time to sterilization (limited to the 14-day induction phase of treatment). (B) Time course of reduction in fungal burden for the 65 patients who provided PD data. Triangles are observed data points from individual patients; lines are model estimates of each patient’s PD profile.

Pharmacodynamics.

We explored both AUC/MIC and C_min/MIC as measures of drug exposure. Associations between drug exposure and various PD endpoints, including time to sterilization, time to death, and EFA in the first 14 days, were evaluated. Higher baseline fungal burden was significantly associated with a longer time to sterilization (hazard ratio [HR], 0.25; 95% confidence interval (CI) 0.17 to 0.37; P < 0.001). There was a trend toward an association between higher baseline fungal burden and time to death (HR, 1.47; 95% CI, 0.97 to 2.19; P < 0.1). All subsequent analyses were adjusted for baseline fungal burden.

Cox proportional hazard models revealed that there were no associations between C_min/MIC and time to sterilization (HR, 1.01; 95% CI, 0.99 to 1.03; P = 0.38) or time to death (HR, 0.99; 95% CI, 0.96 to 1.02; P = 0.71). Similarly, there were no associations between AUC/MIC and time to sterilization of the bloodstream (HR, 1.00; 95% CI, 0.99 to 1.00; P = 0.46) or time to death (HR, 1.00; 95% CI, 0.99 to 1.00; P = 0.99) (Fig. 6). Linear regression following adjustment for baseline fungal burden revealed that C_min/MIC had no significant impact on EFA [EFA (log₁₀ CFU/ml/day) = −0.64 to 0.004 × (C_min/MIC); P = 0.18]. There was also no significant relationship between AUC/MIC and EFA [EFA (log₁₀ CFU/ml/day) = −0.64 to 0.0001 × (AUC/MIC); P = 0.19]. These associations are shown in Fig. 7. To test whether these negative findings were a function of the fact that a small number of patients achieved rapid sterilization of the bloodstream, data from individual patients with the greatest EFA values and fastest time to sterilization were examined closely for higher AUC and C_min values and for higher AUC/MIC and C_min/MIC values. No such correlation was found.

FIG 6.

Cox model predictions of hazard ratios depending on PD index. All models are adjusted for the median baseline fungal burden of 2.2 log₁₀ CFU/ml. (A) The hazard ratio for time to sterility with increasing C_min/MIC is 1.01 (95% CI, 0.99 to 1.03); P = 0.38. (B) The hazard ratio for time to sterility with increasing AUC/MIC is 1.00 (95% CI, 0.99 to 1.00); P = 0.46. (C) The hazard ratio for time to death with increasing C_min/MIC is 0.99 (95% CI, 0.96 to 1.02); P = 0.71. (D) The hazard ratio for time to death with increasing AUC/MIC, adjusted for the fungal burden, is 1.00 (95% CI, 0.99 to 1.00); P = 0.99.

FIG 7.

Relationship between pharmacodynamic indices and early fungicidal activity, adjusted for a median baseline fungal burden of 2.2 log₁₀ CFU/ml. (A) Predicted log₁₀ EFA = −0.64 to 0.004 × (C_min/MIC); P = 0.18. A one-unit increase in C_min/MIC decreases the log₁₀ EFA by −0.004 CFU/ml/day (95% CI, −0.010 to 0.002). (B) Predicted log₁₀ EFA = −0.64 to 0.0001 × (AUC/MIC); P = 0.19. A one-unit increase in AUC/MIC decreases the log₁₀ EFA by −0.0001 CFU/ml/day (95% CI, −0.0003 to 0.0001).

DISCUSSION

Itraconazole is attractive as a potential agent for the treatment for talaromycosis because of its potent in vitro activity against T. marneffei (24, 25), oral bioavailability, improved tolerability profile, and improved access compared with DAmB. Multiple large case series have demonstrated outcomes from talaromycosis treated with itraconazole that are comparable to those obtained with treatment with DAmB (3, 26, 27). However, itraconazole induction therapy was shown in the IVAP trial to be associated with excessive mortality from talaromycosis at 6 months and significantly reduced EFA compared with DAmB (8). Our study raises the possibility that the poor clinical outcomes from itraconazole are related to concentration-dependent therapeutic failure.

Our PK parameter estimates for itraconazole are in keeping with those described in previous population PK models (28). Itraconazole is extensively metabolized in the liver with negligible renal clearance. Renal function did not account for any portion of PK variability in the present analysis. There was very modest variability in weight among study participants; this was insufficient to fully explore the potential impact of weight on PK. The rates of killing induced by itraconazole and OH-itraconazole were similar. This in vivo finding is consistent with comparable in vitro potencies of itraconazole and OH-itraconazole, which are seen against a large range of fungal pathogens (29).

Drug exposure targets for itraconazole have been established for oropharyngeal candidiasis, invasive aspergillosis, cryptococcosis, and histoplasmosis based on observations that patients tend to have better clinical outcomes with trough concentrations of at least 0.5 to 1 mg/liter (11, 13, 20, 30, 31). The appropriate PD target for patients with talaromycosis is not known. Our study did not provide further insight into this issue because an association between drug exposure and microbiological and/or clinical response was not evident, despite exploration of multiple different PK and PD measures and indices. There were too few patients with drug exposures high enough to elicit maximal antifungal activity and thereby separate the population into groups with a high and low probability of therapeutic success. We were also unable to explore models of multiexponential decline in fungal burden, despite this rich data set, because so few patients mounted an appreciable PD response. At the end of the 14-day induction period, approximately 50% of patients were still fungemic. There was extensive variability in the PK, such that one would expect an association between the PK and the PD to be apparent if it exists. The MICs of itraconazole were uniformly low and cannot be implicated in the poor PD response observed. Only 3 of 76 patients (4%) achieved a C_min of 0.5 mg/liter. Therefore, almost all patients were at the lower end of the exposure-response curve as defined for other fungal pathogens. Our data are subject to a potential limitation: due to biosafety regulations in handling T. marneffei, drug was extracted from samples in Vietnam prior to shipment to the United Kingdom. Extraction of calibration curves and quality control assays were then performed at the University of Liverpool. The potential for data discrepancies induced by these methods was limited by freezing the calibration curve and quality control samples following extraction, so that they were subjected to the same conditions as patient samples. It is nevertheless reassuring that our estimates for the population PK are consistent with those described by others (28). A potential limitation in our PK-PD model is that we fixed the volume of OH-itraconazole as a scalar of itraconazole volume. This approach fails to account for the molecular masses of each compound, which are nonidentical.

The concept of concentration-dependent therapeutic failure is well understood for the triazoles. This results from a number of issues common to this class of antifungals. First, oral bioavailability is frequently suboptimal. In the case of itraconazole, dissolution and absorption depend on an acidic environment (pK_a value, 3.7). In healthy volunteers, itraconazole absorption from capsule formulations has been improved by 80% by coadministration of cola (pH 2.5) (32), and maximum concentration (C_max) increased by approximately 70% after a meal (33). Patients in the IVAP trial were given itraconazole after a meal or a cola drink. However, since the basis for these recommendations was data collected from studies of healthy volunteers, it is possible that these are insufficient or ineffective approaches to gastric acidification in the presence of HIV-associated achlorhydria or gastrointestinal disease (34). The second contributor to concentration-dependent therapeutic failure in triazoles is significant PK variability, principally related to variation in oxidative metabolism (35). Itraconazole is extensively metabolized by cytochrome P450 (CYP450) 3A4 isoenzymes, the phenotype of which varies significantly between individuals. In our model, estimates of AUC_0–24 were highly variable, with CV values of 129% and 125% for the parent drug and the metabolite, respectively. Similarly, CV values for estimates of C_min were 147% and 132% for the parent and the metabolite, respectively. Third, drug-drug interactions are common among the triazoles, and we were unable to account for these in this analysis. First-line antiretroviral treatment in Vietnam at the time of the trial consisted of tenofovir, lamivudine, and efavirenz. Efavirenz in an inducer of numerous hepatic enzymes, including CYP3A4. Coadministration of itraconazole (200 mg twice daily) and efavirenz (600 mg once daily) decreases itraconazole C_max, AUC, and C_min by 37%, 39%, and 44%, respectively, and decreases hydroxyitraconazole C_max, AUC, and C_min by 37%, 35%, and 43%, respectively (36). Finally, the formulation of itraconazole is known to have significant impact on serum drug concentrations, the oral bioavailability of capsule formulations being approximately 30% that of oral solutions (19, 33). Patients in the described cohort were administered a capsule formulation of itraconazole from Stada (now Stellapharm), Vietnam. To the best of our knowledge, there are no published data on the bioequivalence of this formulation versus other formulations of itraconazole.

The large PK variability of itraconazole and the capacity for drug interactions mean that therapeutic drug monitoring (TDM) is widely advocated in clinical practice to achieve therapeutic levels. This study represents an opportunity to define target levels for TDM, yet it is unable to do so due to the universally low levels of drug exposure achieved and consequent lack of PD effect produced in the study population. Moreover, treatment guidelines for talaromycosis were recently updated as a result of the IVAP trial, to state that all patients with talaromycosis should receive amphotericin B induction therapy regardless of disease severity (1). The evidence for this has been designated the highest grade possible, since data demonstrating the inferiority of itraconazole were obtained from a large randomized controlled trial. This could deprioritize the future question of the role of itraconazole for talaromycosis.

It remains possible that a different formulation, dosage, and/or mode of administration of itraconazole would have provided higher systemic drug exposure and led to better mycological and clinical outcomes. This PK-PD substudy illustrates the importance of a deep understanding of dose-exposure-response relationships for any drug-pathogen combination to adequately interpret the conclusions of late-phase clinical trials.

MATERIALS AND METHODS

Clinical study.

The PK and PD data were collected during a substudy of a multicenter prospective randomized clinical trial (Itraconazole versus Amphotericin B for Talaromycosis [IVAP] trial; ISRCTN59144167), which compared clinical response and mortality following treatment with itraconazole (300 mg every 12 h [q12h] for 3 days followed by 200 mg q12h for 11 days) to DAmB (0.7 mg/kg/day) for induction therapy for HIV-associated talaromycosis (8). Patients were recruited between October 2012 and December 2015 from the 5 hospitals across Vietnam. Patients in the itraconazole arm were asked to take a small meal or drink or cola prior to drug administration, which was directly observed during the 14-day induction period of treatment. Patients over 18 years of age with culture-confirmed talaromycosis and HIV infection were eligible for the trial. Exclusion criteria included infection of the central nervous system, pregnancy, a liver transaminase level of >400 U/liter, an absolute neutrophil count of <500 cells/mm³, a creatinine clearance of <30 ml/min, or existing prescription of any antifungal therapy for more than 48 h. IVAP trial participants at the Hospital for Tropical Diseases in Ho Chi Minh City were invited to participate in the PK-PD substudy. Ethical approval was granted by the Hospital for Tropical Diseases, the Oxford University Tropical Research Ethics Committee, and the Vietnam Ministry of Health.

Pharmacokinetic and pharmacodynamic sampling.

A total of 76 patients randomized to receive itraconazole agreed to participate in the PK-PD substudy. For the PK, 15 patients underwent intensive sampling, with samples taken at 0, 0.5, and 2 h postdose on day 1; 1, 3, 4, and 12 h on day 2; and 0, 0.5, 1, 2, 3, 4, 6, and 12 h on day 8. The remaining 61 patients underwent sparse sampling at 0 h on day 1, followed by 1 sample on each of days 1 to 4, days 8 to 10, and day 12 and at each of their follow-up visits during weeks 4, 8, 12, and 24 of the study. For each PK sample, 2 ml of blood was collected in heparinized collection tubes and placed immediately on ice. Within 30 min of collection, samples were centrifuged at 2,000 rpm for 15 min, and the plasma was stored at −80°C until analysis. Itraconazole and OH-itraconazole were extracted on site in Ho Chi Minh City (extraction procedure described below). Samples containing acetonitrile as internal standard were plated onto Sabouraud dextrose agar in three independent experiments to confirm sterility. Samples containing extracted drug were stored at −80°C until shipment to the University of Liverpool for analysis.

For the PD analysis, blood was collected for quantitative culture on a daily basis for the first 4 days of treatment and then on alternate days for the remainder of the first 14 days of treatment, until there was no microbial growth. Quantitative culture was performed by serially diluting 100 μl of blood 10-fold and plating onto Sabouraud dextrose agar. Plates were incubated at 37°C for quantification of fungal burden.

Bioanalysis of PK samples.

Itraconazole and OH-itraconazole concentrations in plasma were measured using liquid chromatography-tandem mass spectrometry (LC-MS/MS) (1260 Agilent ultraperformance liquid chromatograph [UPLC] coupled to an Agilent 6420 Triple Quad mass spectrometer; Agilent Technologies UK Ltd., Cheshire, UK). Itraconazole was extracted in Vietnam by protein precipitation. In total, 300 μl of acetonitrile containing 6,7-dimethyl-2,3-di-(2-pyridyl)-quinoxaline (10 ng/ml) was added to 100 μl of matrix. Samples were vortexed thoroughly and then centrifuged at 13,600 rpm for 3 min. Three hundred microliters of supernatant was removed and placed in a 500-μl Eppendorf tube for storage at −80°C prior to shipping to the University of Liverpool. Samples were thawed and vortexed before 150 μl supernatant was transferred to a 96-well autosampler plate. Thirty microliters was injected on an Agilent Zorbax C₁₈ rapid-resolution high-definition (RRHD) column (2.1 by 50 mm; 1.8 μm) (Agilent Technologies UK Ltd., Cheshire, UK).

Chromatographic separation was achieved using a gradient consisting of 60% A–40% B (0.1% aqueous trifluoroacetic acid [TFA] as mobile phase A and 0.1% TFA in acetonitrile as mobile phase B). The mass spectrometer was operated in positive ion mode, and a multiple-reaction monitoring (MRM) method was used for optimum sensitivity and selectivity. The limit of quantitation of both itraconazole and OH-itraconazole was 0.005 μg/ml. The intraday coefficient of variation (CV) for itraconazole was <13.5%, and the interday CV was <10.5%, over the concentration range 0.005 to 8.0 μg/ml. For OH-itraconazole, the intraday CV was <9.0% and the interday CV was <8.7% over the same concentration range.

MIC testing.

The MICs of itraconazole against Talaromyces marneffei were determined in duplicate using the standardized CLSI method for yeasts (23).

Population PK modeling.

The PK-PD model was fitted to the data in two steps. First, the PK was solved. The mean Bayesian estimates for each individual’s PK were fixed and taken forward for the PD modeling. The PD model was then solved by supplying each patient’s PK posterior estimates as covariates alongside the dosing history and individual PD data. Concentration-time data for itraconazole in plasma were modeled using the nonparametric adaptive grid parameter estimation function in Pmetrics (version 1.5.0) (37).

The base PK model was itself constructed in a 2-step process, since itraconazole has an active metabolite, OH-itraconazole. First, a PK model was developed to describe the PK of the parent drug. Three clearance models were tested: linear clearance only, Michaelis-Menten clearance (concentration-dependent, saturable clearance), and a combination of these mechanisms. The final base model for the PK of the parent drug (indicated by the suffix “p”) took the form:

$$\frac{\mathit{dX}\left(1\right)}{\mathit{dt}}=-K_a\times X\left(1\right)$$ (1)

$$\frac{\mathit{dX}\left(2\right)}{\mathit{dt}}=K_a\times X\left(1\right)-\left(K23+\frac{V_{\text{max}}}{K_m\times V_c\text{p}+X\left(2\right)}\right)\times X\left(2\right)+K32\times X\left(3\right)$$ (2)

$$\frac{\mathit{dX}\left(3\right)}{\mathit{dt}}=K23\times X\left(2\right)-K32\times X(3)$$ (3)

where equations 1, 2, and 3 describe the rate of change in amount of itraconazole in milligrams in the gut, central, and peripheral compartments, respectively. K_a is the absorption rate constant from the gut to the central compartment. X(1), X(2), and X(3) are the amounts of itraconazole in the gut, central and peripheral compartments, respectively, in milligrams. K23 and K32 represent first-order transfer constants connecting the central and peripheral compartments. V_max is the maximal rate of enzymatic metabolism of itraconazole (in milligrams per hour) and K_m (mg/liter) is the concentration of itraconazole in the central compartment at which enzyme activity is half maximal. V is the volume of the central compartment, in liters.

The same variations of clearance mechanism were investigated to incorporate the OH-itraconazole (metabolite; indicated with the suffix “m”) data in the PK model. In this case, solely linear clearance, without a saturable clearance component, provided the best fit to the data. The following differential equations were added to the PK model:

$$\frac{\mathit{dX}\left(4\right)}{\mathit{dt}}=\left(\frac{V_{\text{max}}}{K_m\times V_c\text{m}+X\left(2\right)}\right)\times X\left(2\right)-K45\times X\left(4\right)+K54\times X\left(5\right)–\left(\frac{\text{SCLm}}{V_c\text{m}}\right)\times X(4)$$ (4)

$$\frac{\mathit{dX}\left(5\right)}{\mathit{dt}}=K45\times X\left(4\right)-K54\times X(5)$$ (5)

where equations 4 and 5 describe the rate of change in amount of OH-itraconazole in milligrams in the central and peripheral compartments, respectively. Accordingly, X(4) and X(5) are the amounts of OH-itraconazole in those compartments, in milligrams, with K45 and K54 being the first-order intercompartmental rate constants. SCLm is the first-order clearance of OH-itraconazole from the central compartment (in liters per hour), and V_cm is the volume of the central compartment of OH-itraconazole, in liters. V_cm was fixed as a ratio of V_cp, taken from the median ratio of parent to metabolite concentrations at each time point in the data.

Multivariate bidirectional linear regression of each subject’s covariates against the posterior parameter values was performed to determine whether any clinical variables impacted PK parameters. The fit of the model to the data was assessed using a visual inspection and linear regression of the observed-predicted scatterplots both before and after the Bayesian step. Measures of precision and bias were assessed and weighted residuals were plotted against predicted concentrations and time. Models were compared by assessing 2× difference in log-likelihood values evaluated against a chi-square distribution with the appropriate number of degrees of freedom (difference in number of parameters between candidate models). Information loss was estimated using the Akaike information criterion. Predictive performance was evaluated in terms of bias and precision through calculation of the mean weighted error and the mean weighted squared error, respectively.

There is some uncertainty surrounding the most appropriate PD target for itraconazole, and C_min is generally adopted as a pragmatic target, the PK profile of itraconazole being relatively flat (38). We quantified drug exposure in terms of both C_min and AUC. Since the data were collected in a real-world clinical environment, precise drug administration and blood sampling times varied between individuals. Estimates of drug exposure in uniform time intervals across individuals were therefore not possible. The C_min for each patient was calculated as the mean of the lowest model-estimated PK output per day, over the time frame for which there were data (and therefore model estimates) for that patient. The AUC was calculated as the total average AUC for the treatment course divided by the number of 24-h intervals for which data were available per patient. This was done in Pmetrics from each patient’s posterior mean parameter estimates using the trapezoidal rule in the function MakeAUC (37).

Pharmacodynamic modeling.

The population PK model described above was used to obtain the mean Bayesian estimates for each patient’s PK parameters. These were fixed for each patient and input to the maximum likelihood estimator in ADAPT 5 (39) in order to define the PD parameters and estimate the PD weighting functions. The weighting functions were estimated using the following variance model: variance = [intercept + slope × fb]², where fb is the fungal burden measured from the quantitative cultures. Each patient’s PD data were fitted to the PD model one individual at a time, employing the following structural model:

$$\frac{\mathit{dN}}{\mathit{dt}}=-\left\{\left[K{\text{kill}}_{\text{max}}\text{p}\times \left(\frac{\frac{X{\left(2\right)}^{H\text{p}}}{V\text{p}}}{\frac{X{\left(2\right)}^{\mathit{Hp}}}{V\text{p}}+{\text{EC}}_{50}{\text{p}}^{H\text{p}}}\right)\right]+\left[K{\text{kill}}_{\text{max}}\text{m}\times \left(\frac{\frac{X{\left(4\right)}^{H\text{m}}}{V\text{m}}}{\frac{X{\left(4\right)}^{H\text{m}}}{V\text{m}}+{\text{EC}}_{50}{\text{m}}^{H\text{m}}}\right)\right]\right\}\times N$$

In this model, N is the number of CFU in the bloodstream, t is time and dN/dt is the rate of change of fungal burden in the bloodstream. Kkill_max, EC₅₀, and H are the maximal rate of fungal killing, the concentration of drug that induces the half-maximal rate of killing, and the Hill (slope) function, respectively. The model enabled itraconazole and OH-itraconazole to affect the PD simultaneously and independently: parameters with the suffix “p” refer to the parent drug, itraconazole; those with the suffix “m” refer to the metabolite, OH-itraconazole. As above, X(2) and X(4) are the amounts of itraconazole and OH-itraconazole in the central compartment, respectively, in milligrams. The initial condition (IC) represents an estimate of the pretreatment fungal density in the bloodstream. These PD parameters were estimated for each patient alongside the weighting functions (intercept and slope) from the variance model. These weighting functions were then transcribed into the PD data file for Pmetrics, and the model was run in Pmetrics to arrive at a solution for the population PD. Bayesian posterior estimates of the population PD parameters were then obtained from the final PD model. Population PD model fit was determined according to the criteria used for the population PK model. Internal PK-PD model validation by means of Monte Carlo simulation and visual predictive check demonstrated that 77.8% of observed CFU values fell within the 5th and 95th simulated percentiles (P < 0.05).

In building the PD model, several methods for handling data below the LLQ were investigated (40). This was necessary because the quantitative cultures were performed by serially diluting 100 μl of blood and the lowest fungal count recorded was 0.699 log₁₀ CFU/ml; that is, 5 CFU/ml. It is possible that there were samples with CFU counts below 5 CFU/ml but that these colonies were not picked up in the 100 μl of blood plated and were therefore recorded as zero. Thus, there is a degree of uncertainty inherent in measurements toward the lower values of the measurement, as is true for many laboratory assays. The PD model was run with these “zero” CFU counts supplied to Pmetrics as 1 (i.e., unchanged; 0 log₁₀ CFU/ml) and as LLQ/2 (0.350 log₁₀ CFU/ml) and by discarding these data points altogether, to determine which of these 3 methods provided the best model fit. EFA was calculated by performing a linear regression on log₁₀ CFU/ml versus day of cerebrospinal fluid (CSF) culture, taking the slope of the regression line as the EFA for each patient.

Statistical modeling.

For patients who had both PK and PD data available, Cox proportional hazard models were fitted to examine the effect of AUC/MIC and C_min/MIC for itraconazole on the time to sterilization of fungal cultures and the time to death. The Cox models took the form h(t) = h₀(t)exp[(β₁ × PDI) + (β₂ × BFB)] where t is time to event, h(t), is the hazard function and h₀(t) is the baseline hazard. β₁ and β₂ are the coefficients for regression. The hazard ratio is estimated by exp(β_i). PDI and BFB are the pharmacodynamic index (either AUC/MIC or C_min/MIC) and the baseline fungal burden, respectively. Baseline fungal burden was stratified according to its mean due to violation of the proportional hazard assumption, although this did not alter the nonstratified effect sizes of either pharmacodynamic index. The relationship between each pharmacodynamic index and EFA was assessed using a linear regression model, which took the form: EFA = β₀ + (β₁ × PDI) + (β₂ × BFB) + ε, where β₀ is the intercept, β₁ and β₂ are the regression coefficients for PDI and BFB, respectively, and ε is the model error term.