Population pharmacokinetics of ritonavir-boosted saquinavir regimens in HIV-infected individuals

Abstract

Objectives

The aim of this study was to develop and validate a population pharmacokinetic model in order to describe ritonavir-boosted saquinavir concentrations dosed twice and once daily in human immunodeficiency virus (HIV)-infected patients from the UK, Uganda and Thailand and to identify factors that may influence saquinavir pharmacokinetics.

Methods

Pharmacokinetic data from 10 clinical studies were combined. Non-linear mixed effects modelling (NONMEM version V) was applied to determine the saquinavir pharmacokinetic parameters, interindividual/interoccasion variability (IIV/IOV) and residual error. Various covariates potentially related to saquinavir pharmacokinetics were explored, and the final model was validated by means of 95% prediction interval and testing the predictive performance of the model with data not included in the model-building process.

Results

Ninety-seven patients were included from the UK (n = 52), Uganda (n = 18) and Thailand (n = 27), contributing 347 saquinavir profiles (1–14 profiles per patient). A one-compartment model with zero-order absorption and lag-time best described the data with IIV/IOV on apparent oral clearance (CL/F) and volume of distribution (V/F) and with IIV on duration and absorption lag-time. The ritonavir area under the curve over the dosing interval was significantly associated with saquinavir CL/F and V/F. A typical patient from the UK had ~1.5- and 3-fold higher saquinavir CL/F compared with patients from Uganda (89.0 versus 49.8 L/h) and Thailand (89.0 versus 26.7 L/h), respectively.

Conclusions

A model to characterize ritonavir-boosted saquinavir pharmacokinetics in HIV-infected adults has been developed and validated. The model could be used for dosage adaptation following therapeutic drug monitoring and to assess patients’ suitability for once-daily boosted saquinavir therapy.

Introduction

Saquinavir is a potent protease inhibitor used as part of combination antiretroviral therapy to treat human immunodeficiency virus (HIV) and has proven efficacy in treatment-naive and -experienced patients.¹ The pharmacokinetic profile of saquinavir is characterized by poor bioavailability due to extensive first pass and hepatic metabolism, which is improved by co-administration of low-dose ritonavir to inhibit CYP3A4 activity and transport proteins. Saquinavir/ritonavir is licensed at a recommended dose of 1000/100 mg twice daily in Europe and the USA. Despite this, once-daily doses are used in clinical practice, namely 1600/100 mg (or 1500/100 mg if 500 mg film-coated Invirase® tablets are available) and 2000/100 mg. As a result of high interindividual variability, potential drug–drug interactions and use of unlicensed doses, saquinavir is a candidate for therapeutic drug monitoring (TDM).^2,3 A minimum effective concentration (MEC) to sufficiently suppress viral replication has been defined for saquinavir as a trough concentration above 0.1 mg/L,⁴ based on clinical studies in treatment-experienced HIV patients receiving twice-daily therapy.^{5 – 7}

Following approval of novel HIV drugs, the use of saquinavir in western countries has declined. However, due to lack of availability of recently approved medications in areas such as Thailand and Africa, saquinavir-boosted regimens remain a viable option. Saquinavir boosted with ritonavir is recommended as an alternative to lopinavir/ritonavir as a second-line therapy in resource-limited settings including Uganda.⁸ Furthermore, in contrast to Europe and the USA, saquinavir is approved for once-daily dosing in Thailand (1600/100 and 1500/100 mg), in addition to the standard twice-daily dose. It has also been noted that Thai patients experience significantly higher saquinavir/ritonavir concentrations compared with HIV patients from the UK administered 1600/100 mg once daily.⁹

Potential factors that can influence the pharmacokinetics of boosted saquinavir regimens are important in the clinical management of HIV infection and may aid optimal dosage selection to ensure adequate drug concentrations for viral suppression. The aim of this analysis was to develop and validate a population pharmacokinetic model in order to characterize ritonavir-boosted saquinavir concentrations in HIV-infected individuals from the UK, Uganda and Thailand.

Methods

Patients

Data obtained from HIV-infected males and non-pregnant females were pooled from 10 clinical studies evaluating saquinavir/ritonavir pharmacokinetics (saquinavir hard-gel capsules or 500 mg film-coated tablets).^{10 – 19} Boosted saquinavir was investigated either as a sole protease inhibitor or in combination with other antiretrovirals, namely a second protease inhibitor, or under various dietary conditions and at doses of 1000/100 mg twice daily, 1600/100 mg once daily or 2000/100 mg once daily. Patients were recruited and assessed at three different sites: the UK (PK Research Ltd, St Stephen’s Centre, Chelsea and Westminster Foundation Trust, London, UK), Uganda (CARE study cohort, Joint Clinical Research Centre, Kampala, Uganda) and Thailand [Thai Red Cross AIDS Research Centre and HIV Netherlands Australia Thailand Research Collaboration (HIV-NAT)]. Detailed accounts of study design, inclusion/exclusion criteria and pharmacokinetic findings of each study have been reported previously.^{10 – 19} Patients with active clinically significant diseases other than HIV, such as hepatitis infections or tuberculosis, were not allowed to participate. The intake of medications known to induce or inhibit protease inhibitor metabolism (such as non-nucleoside reverse transcriptase inhibitors or traditional medications) was not permitted, with the exception of drugs under investigation. Studies were approved by local Research Ethics Committees, and all individuals provided written informed consent.

Blood sampling and drug analysis

All patients were stable on a saquinavir/ritonavir-containing regimen at least 2 weeks prior to the start of each study and were administered saquinavir/ritonavir as part of combination antiretroviral therapy containing two nucleoside reverse transcriptase inhibitors or one nucleoside reverse transcriptase inhibitor plus one nucleotide reverse transcriptase inhibitor. In brief, on the day of pharmacokinetic sampling, drug intake was directly observed and timed and administered under fed conditions with a standard 40, 20 or 66 g fat-containing meal or under fasted conditions (UK), with a fatty African breakfast (Uganda) or a 10–15 g fat-containing meal (Thailand). Venous blood samples (7 mL) were drawn and collected into heparinized tubes at the following time points:

1. UK: pre-dose (0 h) and 0.5, 1, 2, 3, 4, 6, 8, 10 and 12 h post-dose plus a 24 h post-dose sample if a once-daily regimen was administered.

2. Uganda: pre-dose and 1, 2, 4, 6, 12 and 24 h post-dose.

3. Thailand: pre-dose and 2, 4, 6, 8, 10, 12 and 24 h post-dose or pre-dose and 0.5, 1, 1.5, 2, 2.5, 3, 4, 6, 8, 10, 12 and 24 h post-dose.

Saquinavir and ritonavir pharmacokinetics were assessed at steady state and plasma isolated (1000 g; 10 min; 4°C) within 2 h of collection and stored (−70°C) until analysed.

Quantification of saquinavir and ritonavir in plasma was performed by fully validated HPLC-tandem mass spectrometry (HPLC-MS/MS) methods or HPLC with ultraviolet detection (HPLC-UV), as illustrated previously.^{20 – 22} The majority of the plasma samples were analysed at the same laboratory in the UK with the exception of one study originating from the UK¹⁴ and those from Thailand.^10,18 However, all laboratories participate in the same external quality assurance programme (International Interlaboratory Quality Control Program for Therapeutic Drug Monitoring in HIV Infection, Nijmegen, The Netherlands) with acceptable performances. Details of each assay’s performance have been described previously.^{20 – 22}

Data analysis

Non-linear mixed effects modelling was applied using NONMEM® software (version V, level 1.1, double precision; ICON Development Solutions, Ellicott City, MD, USA)²³ by means of first-order (FO) estimation. Model fit was assessed by statistical and graphical methods. The minimal objective function value (OFV; equal to −2 log likelihood) determined by NONMEM was used as a goodness-of-fit diagnostic with a decrease in OFV of 3.84 points corresponding to a statistically significant difference between hierarchical models (P = 0.05, χ² distribution with one degree of freedom). Graphical diagnostics were performed with Microsoft® Office Excel 2003 for Windows (Microsoft Corporation, Redmond, WA, USA). The standard errors of the parameter estimates were obtained with the COVARIANCE option of NONMEM and individual Bayesian parameter and concentration estimates by the POSTHOC option.

Pharmacokinetic model building

Initially, several structural and error models were explored using a small set of data (i.e. data obtained from one study). Once an adequately fitting structural and error model was found, each study was added consecutively in order to determine whether there were any systematic differences in patients’ apparent oral clearance (CL/F) or apparent volume of distribution (V/F) between studies.

Covariate model building

The following covariates on CL/F and V/F were explored: ritonavir area under the curve over the dosing interval (AUCτ: 12 or 24 h), ritonavir clearance (dose/AUCτ), sex, body weight, study site (UK, Uganda and Thailand), concomitant protease inhibitor (fosamprenavir, 700 mg twice daily; atazanavir; 300, 200, 150 mg once daily) and concomitant tenofovir DF 300 mg once daily. Covariates were introduced separately and included in the model if they were statistically significant, i.e. associated with a decrease in OFV of 3.84 (P ≤ 0.05) and clinically relevant, which here we define as a change in the typical value of the pharmacokinetic parameter (i.e. CL/F) of 10% following the addition of a covariate. Once all significant and relevant covariates were incorporated, a backwards elimination step was then performed, removing covariates from the model one at a time. Covariates were only retained if removal from the model produced a significant increase in OFV of 6.63 (P ≤ 0.01) and were clinically relevant. The inclusion of covariates should also produce a reduction in variability.

Model validation

Using parameters defined by the model, 1000 datasets were simulated at doses of 1000/100 mg twice daily, 1600/100 mg once daily and 2000/100 mg once daily (UK study site only). From the simulated data, 95% prediction intervals (P2.5–P97.5) for each regimen were constructed. Observed data from the original dataset at the assessed doses were superimposed. At least 95% of the data points within the prediction interval, and 2.5% above and below, was indicative of an adequate model.

As an additional validation process, the predictive performance of the final model was evaluated using a set of concentration–time data from the control arm of another clinical study not included in the model-building process.²⁴ Final model parameters were used to predict individual saquinavir concentrations of the dataset and compared with the measured concentrations. Predictive performance was assessed by calculating the mean relative prediction error (%MPE) as a measure of bias and root mean-squared relative prediction error (%RMSE) as a measure of precision.²⁵

Results

Patients

Ninety-seven HIV-infected individuals (42 females) from study sites in the UK (n = 52, of which 5 were Black-African), Uganda (n = 18) and Thailand (n = 27) receiving orally administered saquinavir/ritonavir were included in the model-building process. Patient demographics and baseline characteristics are summarized in Table 1. Thai patients were lighter than UK and Ugandan patients, and patients from Uganda weighed less than UK patients (P ≤ 0.0012 for all comparisons; ANOVA with Bonferroni correction). Moreover, ritonavir AUCτ was significantly higher in Thai patients compared with those from the UK and Uganda (P < 0.0001 and P = 0.00217, respectively; ANOVA with Bonferroni correction). A median (range) of 2 (1–14) full pharmacokinetic profiles per patient were available equating to 347 profiles and 3482 concentrations in total (Figure 1).

Table 1.

Summary of patients’ demographics and baseline clinical characteristics

Parameter	Median (range)
Study site, n (%) [M/F]
UK	52 (53.6) [42/10]
Uganda	18 (18.6) [4/14]
Thailand	27 (27.8) [9/18]
Regimen, n (%) [number of profiles]
SQV/RTV 1000/100 mg bid	37 (38.1) [94a]
SQV/RTV 1600/100 mg qd	75b (77.3) [83]
SQV/RTV 2000/100 mg qd	21 (21.6) [21]
SQV/RTV/FPV 1000/100/700 mg bid	18 (18.6) [18]
SQV/RTV 1000 mg bid/100 mg qd	17 (17.5) [17]
SQV/RTV/TDF 1000/100 mg bid/300 mg qd	17 (17.5) [17]
SQV/RTV/ATV
1600/100/300 mg qd	17 (17.5) [17]
1600/100/200 mg qd	18 (18.6) [18]
1600/100/150 mg qd	18 (18.6) [18]
SQV/RTV 1000/100 mg bid fasted	22 (22.7) [22c]
SQV/RTV 1000/100 mg bid high-fat meal	22 (22.7) [22c]
Age (years)
UK	44 (22–63)
Uganda	39 (29–62)
Thailand	33 (25–48)
Weight (kg)
UK	72 (45–108)
Uganda	63 (47–86)
Thailand	49 (35–75)
BMI (kg/m2)
UK	23 (14–39)
Uganda	24 (19–34)
Thailand	20 (15–26)
RTV AUCτ (mg·h/L)
UK	7.84 (0.66–22.53)
Uganda	8.97 (3.12–20.58)
Thailand	12.76 (3.25–23.07)
Baseline CD4 cell count (cells/mm3)	387 (10–1023)
Baseline HIV-RNA (copies/mL)	3489 (<50–43 722)

n, number of patients; M, male; F, female; SQV, saquinavir; RTV, ritonavir; FPV, fosamprenavir; TDF, tenofovir; ATV, atazanavir; bid, twice daily; qd, once daily; AUCτ, area under the curve over the dosing interval (0–12 or 0–24 h).

^an = 21 evening profiles.

^bn = 30 UK patients.

^cSaquinavir 500 mg film-coated tablets.

Figure 1.

Saquinavir concentrations over time from 97 HIV-infected patients administered 1000/100 mg twice daily, 1600/100 mg once daily or 2000/100 mg once daily (3482 concentrations, 1–14 pharmacokinetic profiles per patient).

Pharmacokinetic model (FO estimation)

A one-compartment model with zero-order absorption best described the data. A one-compartment model with first-order absorption was found to be unsatisfactory (based on model-building criteria), and a two-compartment model would not run. Allowing a different CL/F value for saquinavir/ritonavir 1600/100 and 2000/100 mg once daily improved model fit (ΔOFV −52.505), as did assigning a different CL/F and V/F value, respectively, to Ugandan and Thai study sites (ΔOFV −94.131 and −27.599, respectively) and a different V/F following a morning and evening dose of saquinavir/ritonavir 1000/100 mg twice daily (ΔOFV −109.721). The addition of an absorption lag-time further improved the fit (ΔOFV −398.662). Interindividual variability (IIV) and interoccasion variability (IOV) were included on CL/F and V/F (ΔOFV −148.964, −1141.354 and −121.471 with the addition of IIV V/F, IOV CL/F and IOV V/F, respectively, compared with IIV CL/F only). IIV on duration (D1) and absorption lag further improved model fit (ΔOFV −120.362 and −22.040, respectively). IIV and IOV were described by an exponential model and residual error by a combined additive-proportional model. A bioavailability factor (F1) was introduced to explore the effect of a high-fat meal and fasted state on saquinavir pharmacokinetics. Compared with a standardized meal (F1 = 1; 10–40 g fat and an African breakfast), F was decreased by ~40% (F1 = 0.60) in the fasted state and increased by ~80% (F1 = 1.82) for a high-fat meal (66 g fat). Parameter estimates for the final model are summarized in Table 2, and the diagnostic plots are shown in Figure 2.

Table 2.

Saquinavir parameter estimates and standard errors obtained from the final population pharmacokinetic model

Parameter	Estimate	Standard error
CL/F (L/h)
1000/100 mg bid	67.6	7.8
1600/100 mg qd	93.6	8.7
2000/100 mg qd	59.0	9.4
1600/100 mg qd Uganda	59.3	25.7
1600/100 mg qd Thailand	24.9	8.7
V/F (L)
1000/100 mg bid morning dose	324	34.6
1000/100 mg bid evening dose	545	72.1
1600/100 mg qd	324	34.6
2000/100 mg qd	324	34.6
1600/100 mg qd Uganda	332	69.2
1600/100 mg qd Thailand	379	89.1
D1 (h)	3.08	0.14
Lag-time (h)	0.57	0.07
F1 (%)
fasted state	0.60	0.10
high-fat meal (66 g)	1.82	0.23
IIV CL/F (%)	41.4	27.0
IOV CL/F (%)	57.3	32.6
IIV V/F (%)	41.6	29.0
IOV V/F (%)	37.3	21.8
IIV D1 (%)	17.8	20.1
IIV lag-time (%)	46.8	52.3
Residual error
additive (mg/L)	0.26	0.05
proportional (%)	34.3	3.07
Factor associated with RTV AUCτ on SQV CL/F	−0.40	0.08
Factor associated with RTV AUCτ on SQV V/F	−0.27	0.06

bid, twice daily; qd, once daily; CL/F, apparent oral clearance; V/F, apparent volume of distribution; D1, duration of zero-order input; F1, bioavailability (relative to F1 = 1 for other dietary conditions, i.e. 10–40 g fat and African fatty breakfast); IIV, interindividual variability; IOV, interoccasion variability; RTV, ritonavir; SQV, saquinavir; AUCτ, area under the concentration–time curve over the dosing interval (12 or 24 h).

Figure 2.

Goodness-of-fit plots for the final pharmacokinetic model illustrating (a) population predictions of saquinavir versus observed concentrations, (b) individual predictions of saquinavir versus observed concentrations and (c) weighted residuals versus time post-dose. UK patients are represented by open circles, Ugandan by open squares and Thai by open triangles.

Covariate model

In the univariate analysis, ritonavir AUCτ on saquinavir CL/F and V/F gave the largest decrease in OFV (−155.509 and −50.632, respectively). A statistically significant, clinically relevant relationship was also found between body weight and saquinavir V/F. Sex, concomitant atazanavir (300 and 200 mg once daily) and tenofovir on saquinavir CL/F were also significant. The addition of covariates was described by linear models with the exception of ritonavir AUCτ, which showed a power relationship. Following multivariate analysis and backwards elimination, only ritonavir AUCτ on saquinavir CL/F and V/F remained (Table 2). An example of the equation used to describe saquinavir CL/F is shown below (similarly for saquinavir V/F), along with the equation for residual variability.

$$\mathrm{CL}∕F_{\mathit{ij}}=({\theta }_1\ast {({\mathrm{RTV}}_{\mathit{ij}}∕8.25)}^{{\theta }_2})\ast \exp ({\eta }_i+{\kappa }_{\mathit{ij}})$$

where CL/F_ij illustrates the saquinavir CL/F of the ith individual on the jth occasion, θ₁ the initial parameter estimate, θ₂ the factor associated with ritonavir AUCτ on saquinavir CL/F, RTV_ij the AUCτ of ritonavir for the ith individual on the jth occasion, 8.25 the median ritonavir (RTV) AUCτ value in mg·h/L, η_i the IIV with a mean of zero and variance ω² and k_ij the IOV with a mean of zero and variance π².

$$Y=F\ast (1+{∊}_1)+{∊}_2$$

where Y is the final prediction, F the individual prediction and ∈₁ and ∈₂ the proportional and additive model components, respectively, with a mean of zero and variance σ².

Comparison of study site

A saquinavir/ritonavir dose of 1600/100 mg once daily was common to the three evaluated study sites: UK, Uganda, Thailand. To assess the potential differences in pharmacokinetic parameters between study sites, the model was used to determine saquinavir CL/F and V/F for a typical 70 kg adult male, administered saquinavir/ritonavir 1600/100 mg once daily, without any interacting medications with a ritonavir AUC_0–24 of 8.92 mg·h/L (mean ritonavir AUCτ) and compared with the values obtained for UK patients (i.e. UK versus Uganda and UK versus Thailand; Table 3). Ratios associated with comparisons between study sites should be similar for population and individual predictions; however, based on the final model, discrepancies were observed. For example, the ratio of UK versus Thai patients for the population predictions of V/F was 0.9 in comparison with 1.4 for the individual predictions (Table 3). The agreement between population and individual predictions was improved by splitting the patients according to the study site and by obtaining parameter estimates separately for UK, Ugandan and Thai patients, in other words, using one model but with different parameter estimates for each study site (Table 3). Based on these estimates, an HIV patient from Thailand had an ~3 times lower CL/F than an individual from the UK, whereas a UK patient had ~1.5–2-fold higher saquinavir CL/F than a Ugandan (Table 3). Model parameters for each study site obtained separately are shown (Table 4).

Table 3.

Comparison of population and individual predictions of saquinavir CL/F and V/F for a typical 70 kg male receiving saquinavir/ritonavir 1600/100 mg once daily from each of the three study sites using pharmacokinetic parameters generated from the full model (top of the table) and parameters determined by splitting the model according to the study site (bottom of the table)

Study site	Population predictions		Individual predictions
	CL/F (L/h)	V/F (L)	CL/F (L/h)	V/F (L)
Full model
UK	90.8	317.2	79.2	272.9
Uganda	57.5	325.1	68.0	521.2
Thailand	24.1	371.1	23.3	195.4
ratio
UK versus Uganda	1.6	1.0	1.2	0.5
UK versus Thailand	3.8	0.9	3.4	1.4
Split model
UK	89.0	313.5	79.4	278.8
Uganda	49.8	421.0	59.0	400.2
Thailand	26.7	245.9	22.8	184.1
ratio
UK versus Uganda	1.8	0.7	1.3	0.7
UK versus Thailand	3.3	1.3	3.5	1.5

CL/F, apparent oral clearance; V/F, apparent volume of distribution.

Table 4.

Saquinavir parameter estimates and standard errors obtained from the model fitted to each study site separately

Parameter	Estimate (standard error)
	UK	Uganda	Thailand
CL/F (L/h)
1000/100 mg bid	64.8 (8.9)	—	—
1600/100 mg qd	92.7 (9.4)	52.1 (4.3)	24.9 (3.4)
2000/100 mg qd	51.3 (7.3)	—	—
V/F (L)
1000/100 mg bid morning dose	320 (27.3)	—	—
1000/100 mg bid evening dose	556 (65.4)	—	—
1600/100 mg qd	320 (27.3)	443 (67.9)	252 (31.4)
2000/100 mg qd	320 (27.3)	—	—
D1 (h)	2.99 (0.14)	3.90 (0.50)	3.05 (0.15)
Lag-time (h)	0.55 (0.06)	0.91 (0.05)	1.19 (0.13)
F1 (%)
fasted state	0.59 (0.10)	—	—
high-fat meal (66 g)	1.76 (0.21)	—	—
IIV CL/F (%)	45.5 (33.0)	20.2 (14.8)	69.1 (37.8)
IOV CL/F (%)	58.5 (35.8)	—	—
IIV V/F (%)	33.6 (21.5)	28.7 (35.2)	54.9 (34.2)
IOV V/F (%)	35.9 (20.5)	—	—
IIV D1 (%)	20.6 (17.7)	52.1 (34.4)	19.3 (20.4)
IIV lag-time (%)a	29.3 (26.6)	—	48.4 (41.2)
Residual error
additive (mg/L)	0.24 (0.04)	0.15 (0.05)	0.20 (0.11)
proportional (%)	33.8 (3.19)	51.3 (9.41)	17.3 (9.17)
Factor associated with RTV AUCt on SQV CL/F	20.53 (0.06)	20.54 (0.11)	0.87 (0.17)
Factor associated with RTV AUCt on SQV V/F	20.26 (0.06)	20.65 (0.16)	20.31 (0.13)

bid, twice daily; qd, once daily; CL/F, apparent oral clearance; V/F, apparent volume of distribution; D1, duration of zero-order input; F1, bioavailability (relative to F1 = 1 for other dietary conditions, i.e. 10–40 g fat and African fatty breakfast); IIV, interindividual variability; IOV, interoccasion variability; RTV, ritonavir; SQV, saquinavir; AUCτ, area under the concentration–time curve over the dosing interval (12 or 24 h).

^aUganda: IIV lag-time negligible, therefore removed.

Model validation

A 95% prediction interval for each saquinavir/ritonavir regimen (1000/100 mg twice daily, 1600/100 and 2000/100 mg once daily) is shown in Figure 3. They were generated from 1000 simulations of each regimen, one profile per patient with covariate values similar to individuals used in the model-building process (1000/100 mg twice daily, n = 49; 1600/100 mg once daily, n = 30; 2000/100 mg once daily, n = 21). Observed data from patients used in the model-building process were superimposed onto the prediction interval. Of 1338 concentrations for 1000/100 mg twice daily, 1.05% were above P97.5 and 1.27% were below P2.5. For 1600/100 mg once daily, 2.15% of the 418 concentrations were below P2.5, whereas no concentrations were above P97.5. In contrast, 10/230 (4.35%) were below P2.5 and 4/230 (1.74%) above P97.5 for saquinavir/ritonavir 2000/100 mg once daily. This analysis suggests that the final model provided an adequate fit to the data, with 93% to 97% of the concentration data falling within the prediction intervals for all three regimens.

Figure 3.

Ninety-five percent prediction intervals (P2.5–P97.5) on a log-scale determined from the simulated data of saquinavir/ritonavir administered to UK patients at (a) 1000/100 mg twice daily, (b) 1600/100 mg once daily and (c) 2000/100 mg once daily. Observed data are superimposed for the three regimens evaluated.

Furthermore, prediction intervals were also generated from 1000 simulations for Ugandan (n = 18) and Thai patients (n = 27) at a dose of 1600/100 mg once daily, one profile per patient. Approximately 98% of the observed saquinavir concentrations from Thai patients were within the prediction interval, whereas 92% of the concentrations from the Ugandan patients were within the 95% prediction interval (4% below P2.5 and above P97.5).

A comparison of saquinavir trough concentrations (12 or 24 h post-dose; administered twice daily or once daily, respectively) with the recommended MEC of 0.1 mg/L was also performed. Of the 347 concentration–time profiles, 65 trough concentrations were below target; 86% of which were correctly identified by the model and 14% predicted within the therapeutic range when they were, in fact, suboptimal. Of the remaining 282 trough concentrations above the MEC, the model incorrectly predicted 5% of the individual trough concentrations to be below 0.1 mg/L. The evaluation of twice- and once-daily simulated data revealed 12%, 23% and 20% of trough concentrations below the recommended MEC at a dose of 1000/100 mg twice daily, 1600/100 mg once daily and 2000/100 mg once daily, respectively, for patients from the UK. At a dose of 1600/100 mg once daily, 14% and 16% of the simulated saquinavir trough concentration data were below target for Ugandan and Thai patients, respectively.

Data from 12 additional HIV-infected individuals (1 female) receiving saquinavir/ritonavir 1000/100 mg twice daily (500 mg film-coated Invirase® tablets; 1 profile per patient, 120 concentrations) were used to further validate the model.²⁴ Median (range) age, weight, body mass index (BMI) and ritonavir AUC_0–12 were 41 years (27–61), 70 kg (64–104), 24 kg/m² (21–34) and 7.5 mg·h/L (2.0–14.7), respectively. Baseline CD4 cell count ranged between 338 and 779 cells/mm³, and 9 of the 12 patients had undetectable viraemia (<50 copies/mL); three patients had detectable HIV-RNA values of 88, 662 and 878 copies/mL. The study set out to investigate the effect of omeprazole on saquinavir/ritonavir pharmacokinetics; therefore, only data from the control phase of the study were used for validation (i.e. saquinavir/ritonavir alone). A plot of individual saquinavir predictions versus measured saquinavir concentrations is shown (Figure 4). The predictive performance of the model was acceptable, providing precise (%RMSE: 15.9%) and unbiased (%MPE, 95% CI: 1.1%, −1.8 to 4.0) predictions. Individual concentrations for one patient were poorly predicted by the model compared with other patients, removal of this individual resulted in improved predictive performance (%RMSE: 11.4%; %MPE, 95% CI: 1.4%, −0.7 to 3.6). Compared with the 11 patient profiles adequately predicted by the model, this patient had considerably lower saquinavir concentrations attributable to a higher CL/F (36–85 versus 240 L/h).

Figure 4.

Goodness-of-fit plot showing individual predictions of saquinavir versus observed concentrations of study samples used for model validation (n = 120 concentrations).

Discussion

A model to describe the pharmacokinetics of ritonavir-boosted saquinavir regimens has been developed and validated in HIV-infected patients from the UK, Uganda and Thailand. Ritonavir AUC over the dosing interval was significantly associated with saquinavir pharmacokinetics and described some of the variability in concentrations, decreasing saquinavir IIV on CL/F by 23% and IOV on CL/F and V/F by 11% and 8%, respectively.

Few publications are available evaluating the population pharmacokinetics of saquinavir, and none to date has investigated boosted regimens at the currently licensed dose (1000/100 mg twice daily) or following once-daily treatment. Furthermore, this analysis has the added advantage of determining saquinavir population pharmacokinetics in patients from three different continents, allowing exploration of ethnic differences. As a result of different doses and dosing schedules, it is difficult to compare model parameters with those outlined in the literature. Vanhove et al.²⁶ determined a CL/F of 989 L/h in HIV patients following steady-state dosing of hard gel saquinavir capsules 600 mg (unboosted) three times a day, and an analysis using MONOLIX software estimated saquinavir CL/F to be 1.26 L/h in 100 HIV patients administered a single dose of saquinavir (600 mg) and 200 mL of grapefruit juice.²⁷ However, similar to the present analysis, IIV in pharmacokinetic parameters was high (~52% to 120%).^26,27 An analysis using the Markov chain Monte Carlo approach assessed saquinavir pharmacokinetics from Phase I studies in healthy volunteers receiving saquinavir 600 mg single dose, multiple dose or 12 mg intravenously. The authors concluded that saquinavir exhibits complex and highly variable behaviour and used a two-compartment biphasic or monophasic zero-order absorption model to describe the data.²⁸ The study also estimated saquinavir clearance and volume of distribution using the bioavailability of 4% (F = 0.039).²⁸ For the present analysis, a two-compartment model was not possible, and the presence of ritonavir alters the clearance and bioavailability of saquinavir. The bioavailability of saquinavir coadministered with low-dose ritonavir is not known.

Body weight²⁶ and BMI,²⁷ respectively, were associated with saquinavir pharmacokinetics in two population pharmacokinetic studies; however, our analysis identified ritonavir AUCτ as the most important covariate. Body weight on saquinavir V/F was significant and clinically relevant in the univariate analysis; however, in the presence of ritonavir AUCτ, the addition of body weight, although significant (i.e. decreased OFV), did not really influence the predictions or improve model fit, and therefore it was not included. It is possible that in the absence of ritonavir, the effect of body weight is more important; furthermore, the analysis by Lavielle et al.²⁷ included data from a study investigating saquinavir pharmacokinetics in patients with diarrhoea and/or wasting,²⁹ so it is probably not surprising that CL/F was associated with BMI.

Some underestimation of predicted concentrations was evident at the higher concentration range. Potentially, this could be a result of the estimation method used. The model-building process was implemented with FO estimation and not first-order conditional estimation (FOCE) with interaction. The FO method can be prone to underestimation; however, we did attempt the analysis using FOCE with interaction. NONMEM struggled to converge with increasing model complexity, particularly following the addition of IIV to D1, absorption lag or both and, as a result, model fit was better using FO estimation compared with FOCE. Furthermore, saquinavir pharmacokinetics may be influenced by other covariates that were not collected or are yet to be identified. Ritonavir AUCτ was the most significant covariate; however, it was unlikely that it would describe all of the variability, particularly when dealing with a drug such as saquinavir, which is inherently poorly bioavailable and is characterized by an erratic pharmacokinetic profile. Ritonavir acts to improve saquinavir pharmacokinetics and to increase bioavailability; however, it must be noted that through this mechanism, there will be differential inhibition of saquinavir metabolism and transport, potentially complicating the system. This by no means undermines the clinical application of the model as validation was successful, and good agreement was observed between measured and individual predicted saquinavir concentrations.

The 2NN study reported ~11% and 28% higher clearance of the non-nucleoside reverse transcriptase inhibitor, nevirapine, in South American and Western HIV patients, respectively, compared with Thai and South African individuals.³⁰ A study evaluating indinavir alone and boosted with ritonavir in HIV-infected patients from Thailand observed a 20% higher indinavir maximum concentration when compared with the historical data from Caucasian patients.³¹ Furthermore, boosted saquinavir AUC_0–24 and minimum concentration (1600/100 mg once daily) were ~3-fold higher in Thai patients compared with those from the UK in an analysis performed by Autar et al.⁹ Some of the data were also included in the present population pharmacokinetic model and support the finding that saquinavir CL/F in Thai patients was three times lower than that in the UK patients. UK patients also had a higher saquinavir CL/F compared with Ugandan patients; however, this was not as pronounced as the differences observed with individuals from Thailand. Furthermore, saquinavir V/F was slightly lower in Thai patients compared with those from the UK, but the difference was not as large as that noted for saquinavir CL/F. It is well established that saquinavir absorption is improved with food,³² particularly high fat;¹⁷ however, it is highly unlikely that this contributes to higher concentrations seen in Thai patients as the fat content was lower (10–15 g) compared with the standardized meal UK patients received (20–40 g). The fat content of an African fatty breakfast is not known. As also suggested by Autar et al., the underlying mechanisms resulting in pharmacokinetic disparities between study sites are probably a combination of pharmacogenomic and environmental/lifestyle factors associated with ethnicity. Discrepancies were noted between population and individual prediction ratios when comparing CL/F and V/F between study sites. This was somewhat amended by splitting the data by study site and fitting the model to each site separately, i.e. one model but using different parameter estimates for each study site. The model still produced an adequate fit to the data as confirmed by the validation steps. IIV CL/F and V/F were highest for Thai patients and lowest for Ugandan patients. However, the duration of zero-order infusion was highly variable for Ugandans, and IIV on absorption lag was negligible and therefore removed. For all patients, few samples were taken during the absorption phase, potentially making it more difficult to fully characterize and resulting in over-prediction at earlier time points, particularly for 1600/100 and 2000/100 mg once daily. If resources permit, it may be prudent to sample more frequently during the absorption phase for drugs such as saquinavir, to allow for better characterization of earlier concentrations. The reason for the initial miss-fit of the data is not known, but could potentially be attributed to an unknown covariate skewing the population predictions. The antiretrovirals used, particularly ritonavir, require refrigerated storage, but those administered to Ugandan patients will have been subjected to erratic electricity supply or disruptions in the cold chain, potentially influencing the model predictions.

The potential for once-daily treatment still remains an important issue for antiretroviral therapy. With the exception of atazanavir, concern surrounds the lower trough concentrations obtained when moving from twice-daily to once-daily dosing of boosted protease inhibitors. A recent article assessed the pharmacokinetic forgiveness or robustness of boosted saquinavir/ritonavir regimens and showed that 1600/100 and 2000/100 mg once daily remained below the recommended saquinavir MEC (0.1 mg/L) for a longer length of time compared with 1000/100 mg twice daily.³³ The length of time below target was longest for 1600/100 mg once daily.³³ This reflects the potential window of opportunity for viral escape and resistance to develop. From the simulated data, a higher percentage of patients were below target for 1600/100 mg once daily compared with the standard twice-daily dose and 2000/100 mg once daily. It has been suggested that for some patients, once-daily dosing may not be optimal and that Thai patients may be better suited to the once-daily regimen as a result of higher concentrations. However, it should also be noted (but recognizing the high additive residual error) that ~16% of the simulated troughs from Thai patients and 14% from Ugandans were below the recommended MEC and so once-daily dosing may only be appropriate for specific individuals or where TDM services are available. Routine TDM is not currently considered a viable option in resource-limited settings; however, the development of filter paper methodologies for the collection of pharmacokinetic samples coupled with population modelling may improve the feasibility of routine TDM for some patients in developing countries.

Patient adherence or non-adherence to therapy can also influence model predictions. Adherence data were not available for the current dataset, but all clinical studies were carefully controlled, and when patients were not present at the study centre, adherence was monitored by means of pill counts and diaries. It has been shown that model estimation can be improved by incorporation of actual patient dosing history using electronic monitoring reported times.³⁴

In conclusion, a population pharmacokinetic model to predict the pharmacokinetics of ritonavir-boosted saquinavir dosed twice and once daily has been developed and validated in HIV-infected adults. Ritonavir AUCτ was significantly associated with saquinavir pharmacokinetics. Due to availability of newer agents, saquinavir is not as commonly used in clinical practice in western countries. However, use in many developing countries will continue and so the model is applicable globally. The model could be used for dosage adaptation following TDM and to assess patients’ suitability for once-daily boosted saquinavir therapy. The analysis also underscores the importance of pharmacokinetic studies in diverse patient populations.