Sex differences in atazanavir pharmacokinetics and associations with time to clinical events: AIDS Clinical Trials Group Study A5202

Abstract

Objectives

It is uncertain whether HIV-1 antiretroviral exposure and clinical response varies between males and females or different race/ethnic groups. We describe ritonavir-enhanced atazanavir pharmacokinetics in relation to virological failure, safety and tolerability in treatment-naive individuals to investigate potential differences.

Methods

Plasma samples were collected from participants in AIDS Clinical Trials Group Study A5202 for measurement of antiretroviral concentrations. Individual estimates of apparent oral clearance of atazanavir (L/h) were calculated from a one-compartment model and divided into tertiles as slow (<7), middle (7 to <9; reference group) and fast (≥9). Associations between atazanavir clearance and clinical outcomes were estimated with a hazard ratio (HR) from Cox proportional hazards models. Interactions between atazanavir clearance and sex, race/ethnicity and NRTIs were investigated for each of the outcomes.

Results

Among 786 participants, average atazanavir clearance was slower in females (n = 131) than males (n = 655). Atazanavir clearance was associated with time to virological failure (P = 0.053) and this relationship differed significantly by sex (P = 0.003). Females in the fast atazanavir clearance group had shorter time to virological failure (HR 3.49; 95% CI 1.24–9.84) compared with the middle (reference) atazanavir clearance group. Among males, the slow atazanavir clearance group had a higher risk of virological failure (HR 2.10; 95% CI 1.16–3.77).

Conclusions

Atazanavir clearance differed by sex. Females with fast clearance and males with slow clearance had increased risk of virological failure.

Introduction

Atazanavir is a potent once-daily PI and is recommended as a preferred active agent in combination antiretroviral regimens for HIV-infected treatment-naive individuals. Atazanavir is one of the most utilized PIs because of its efficacy, favourable tolerability, durability and dosing convenience. The recommended and most common dose of atazanavir in treatment-naive adults is 300 mg plus 100 mg of ritonavir daily.¹

Significant relationships between atazanavir plasma concentration exposure, efficacy and safety have been demonstrated but have also revealed marked interindividual variability in its concentrations.^2,3 The pharmacokinetics of atazanavir is characterized by a rapid absorption phase that is dependent on gastrointestinal pH. It is moderately bound to α-1-acid glycoprotein and albumin (89% and 86%, respectively) in plasma. Atazanavir also undergoes extensive hepatic metabolism via CYP3A4/5, with subsequent glucuronidation, and is a substrate for the drug efflux transporter P-glycoprotein.⁴

Prior studies have investigated potential sources of atazanavir variability to identify individual factors influencing its pharmacokinetics and pharmacodynamics. Among these have been the following: genetic polymorphisms of CYP3A4/5, ABCB1 and ORM1; albumin and α-1 glycoprotein acid concentrations; and antiretrovirals and other concomitant medications.^4–11 However, the relationship of atazanavir exposure with patient-specific characteristics, such as sex and race/ethnicity, has not been well delineated.^5,6,8,12 Moreover, pharmacodynamic disparities in antiretroviral exposure and clinical outcomes based on sex and race/ethnicity are also ambiguous.^13,14 This may be due to racial and ethnic minorities and women being traditionally under-represented in earlier phase studies when pharmacokinetic data are most often collected, as well as in larger HIV therapeutics clinical trials.^15,16

AIDS Clinical Trials Group Study (ACTG) A5202 (NCT00118898) randomly assigned treatment-naive participants to efavirenz- or atazanavir/ritonavir-based combination antiretroviral therapy. Plasma pharmacokinetic data were collected from the majority of study participants at multiple timepoints, providing one of the largest pharmacokinetic datasets for atazanavir in a clinical trial setting. The pre-specified pharmacology objectives were to evaluate the association of sex and self-reported race/ethnicity with antiretroviral exposure, by comparing modelled antiretroviral apparent oral clearance between men and women and between Hispanics, black non-Hispanics (blacks) and white non-Hispanics (whites); and to evaluate the association of modelled antiretroviral clearance with virological failure, tolerability and safety endpoints.

Methods

Study design and participants

Study A5202 was a Phase IIIb, randomized, partially blinded comparison study of four once-daily antiretroviral regimens in treatment-naive adults (≥16 years of age). At screening, study participants were stratified by HIV-1 RNA level (<100 000 or ≥100 000 copies/mL) and then randomized 1 : 1 : 1 : 1 to open-label efavirenz (600 mg) or atazanavir/ritonavir (300/100 mg), along with double-blinded placebo-controlled tenofovir disoproxil fumarate (300 mg) plus emtricitabine (200 mg) or abacavir (600 mg) plus lamivudine (300 mg). All research was conducted in accordance with the Declaration of Helsinki. Human subject committees of all sites approved the protocol and informed consent was obtained from all participants.

The primary efficacy, safety and tolerability results have been previously reported.^17–19 The primary efficacy outcome was time from randomization to virological failure, defined as a confirmed plasma HIV-1 RNA level ≥1000 copies/mL at or after 16 weeks and before 24 weeks or ≥200 copies/mL at or after 24 weeks. The primary tolerability outcome was assessed as time to first permanent modification of third drug (open-label atazanavir/ritonavir or efavirenz), ignoring modification of NRTI and temporary holds of antiretrovirals. The protocol originally defined the tolerability endpoint as discontinuation of any active study drug, but this endpoint was modified to just third drug after NRTI unblinding occurred for participants of the high-screening viral load stratum based on a data and safety monitoring board review.^17,18 The primary safety endpoint was time from treatment dispensation to the first development of grade 3 or 4 (i.e. severe or life-threatening) sign, symptom or laboratory abnormality that was at least one grade higher than at baseline (excluding creatinine kinase and bilirubin laboratory values).

Pharmacokinetic sampling and assays

In study ACTG A5202, blood samples for pharmacokinetic assessment were collected with the goal of obtaining drug exposure estimates for all study participants. A sparse sampling strategy was used to collect at least three blood samples for population pharmacokinetic analyses. Antiretroviral plasma concentrations assumed to have reached steady-state were obtained over one to two visits during the first 24 weeks of therapy at study weeks 4, 8, 16 or 24. For participants taking their atazanavir/ritonavir-based regimen in the morning, two clinic visits (visits A and B) were scheduled. At visit A, two blood samples were collected around an observed dose with food (one sample collected before the dose and one sample collected 3–4 h after the dose). At visit B, one sample was collected 5–12 h after a regularly scheduled dose. The order of visit A and visit B was flexible and could be combined at one visit if convenient for the participant. Participants who took their regimen in the evening were instructed to switch their dosing (both NRTIs and atazanavir/ritonavir) to the morning for the 4 days prior to their scheduled clinic visit A. Two samples were then collected around an observed dose as outlined above. A separate visit B was also scheduled in order to collect a sample 5–15 h after a regularly scheduled evening dose.

Medication adherence training was provided to participants at entry. Study participants' adherence was assessed with an ACTG self-report adherence questionnaire at weeks 8 and 24. Additionally, the dates and times of the last three antiretroviral doses were recorded in the case report forms, which were reviewed by the laboratory personnel of the University at Buffalo Pharmacology Specialty Laboratory (UB PSL) to ensure adherence prior to assaying atazanavir plasma concentrations.

Blood samples were stored on ice or isolated by centrifugation (800 g for 10 min) within 90 min of collection. Plasma samples were then aliquotted into polypropylene cryovials and stored at −70°C until shipment to the UB PSL.²⁰ Atazanavir concentrations were quantified using a validated HPLC assay consistent with the Clinical Laboratory Improvement Amendments, with the lower limit of quantification (LLQ) being 100 ng/mL. Concentrations less than the LLQ were replaced with one-half of LLQ (50 ng/mL), while samples with non-detectable concentrations were excluded from the analysis. The accuracy of concentrations was evaluated according to the ACTG and the NIH Clinical Pharmacology Quality Assurance Program.

Pharmacokinetic analysis

For the atazanavir concentration–time data, population pharmacokinetic analyses were performed with the computer program NONMEM VII (ICON Development Solutions, Ellicott City, MD, USA). The first-order conditional estimation with interaction method was used and one- and two-compartment models were tested to find a structural model that appropriately described the atazanavir concentration data without covariates. Selection of the structural model was based on successful convergence and goodness-of-fit plots (Figure S1, available as Supplementary data at JAC Online). Similar to previous ritonavir-enhanced atazanavir population pharmacokinetic studies, a one-compartment model with first-order absorption and elimination was selected (minimizations were unsuccessful with a two-compartment model).^{9–11,21–23} The model was parameterized in terms of clearance, volume of distribution and an absorption rate constant. The between-participant variability was described using an exponential random effects model; a proportional error model was used to describe residual variability. Individual Bayesian estimates of apparent oral clearance (CL/F) for atazanavir were obtained for each participant included in the pharmacokinetic model.

From the structural model, an allometric scaling model for body weight on clearance (power of 0.75) and volume (power of 1) was used to investigate the covariate relationships of sex, tenofovir disoproxil fumarate randomization, age and race/ethnicity upon atazanavir clearance. Covariates were tested using a forward inclusion–backward elimination method. Forward inclusion of a covariate required reduction in the minimum objective function value (OFV) of ≥3.84 (P < 0.05, χ² distribution, 1 degree of freedom). Covariates included during the forward step were individually eliminated and retained only if their removal increased the OFV by >10.83 (P < 0.001, χ² distribution, 1 degree of freedom). Age (continuous) was centred on the median value and effects were included in both linear and non-linear fits. Categorical covariates were tested in the model as binary or ternary values using a proportional model.

Statistical analysis

Times to virological failure as well as primary tolerability and safety outcomes were analysed as-treated (i.e. while prescribed atazanavir/ritonavir) in relation to atazanavir CL/F. Individual area under the concentration–time curves (AUCs) were not used because of possible dose–exposure non-linearity in the prescribed dosage range (AUC = dose/CL assumes dose–exposure linearity).⁴ Atazanavir CL/F association analyses were restricted to participants with estimated atazanavir CL/F values and those of white, black or Hispanic race/ethnicity (categorized according to NIH ethnic and racial definitions) due to low frequencies in other groups and as pre-specified in the study protocol. Participant-specific estimated CL/F values were evaluated by sex, race/ethnicity (three groups) and assigned NRTI using a Satterthwaite t-test and 95% CI or one-way analysis of variance.

Time-to-event survival distributions were estimated by the Kaplan–Meier method. Hazard ratios (HRs) were estimated with Cox proportional hazards models stratified by screening HIV-1 RNA stratum and adjusted for assigned NRTI and NRTI*RNA interaction term (denoted as ‘base model’), due to the interaction results previously reported during an interim analysis. During a planned data and safety monitoring board interim analysis, a shorter time to virological failure was observed with abacavir/lamivudine versus tenofovir disoproxil fumarate/emtricitabine in the high HIV-1 RNA stratum, prompting unblinding of NRTI treatment in this group of participants.¹⁷

It was pre-specified that atazanavir CL/F would be grouped into tertiles of the overall distribution. A likelihood ratio test provided evidence that the relationship between atazanavir CL/F and the outcome was not linear (e.g. numeric level 1, 2 or 3) and fit better as tertile categories (P = 0.011, virological failure; P = 0.051, tolerability; and P = 0.25, safety). For consistency, atazanavir CL/F was modelled as categories for each outcome. The middle group was defined as the reference group (slow, <7; middle, 7 to <9; and fast, ≥9 L/h). Previous pharmacokinetic modelling of HIV-infected adults receiving a 300/100 mg dose of atazanavir/ritonavir have reported mean atazanavir CL/F values ranging from 6.5 to 9.2 L/h, but most estimates have been ∼7 L/h.^7,12,21–23 Therefore, the reference group was similar to the average value reported in previous population pharmacokinetic studies.

Separate models evaluated interactions between atazanavir CL/F and each of sex, race/ethnicity and assigned NRTI on the association with the primary time to virological failure, tolerability and safety endpoints. For each endpoint, the proportional hazards assumption was tested by introducing an interaction term between atazanavir CL/F group and time (discrete time for efficacy, continuous time for safety and tolerability); this assumption was not violated (P ≥ 0.49). Multivariable analyses were stratified by screening HIV-1 RNA group and adjusted for NRTI, NRTI*RNA, sex, race/ethnicity, age, CD4+ cell count and BMI at baseline and week 8 and 24 self-reported adherence over the 1 week prior to study visit (<100% adherence at week 8 or 24 versus 100% at both visits). Sensitivity analyses evaluated associations between atazanavir CL/F and the outcomes separately within each NRTI group. Pharmacokinetic objectives were specified as secondary to study A5202 and the study was not formally powered for these objectives. Analyses were conducted with two-sided α = 0.05 with no adjustments for multiple comparisons and were carried out in SAS version 9.2 (SAS Institute, Cary, NC, USA).

Results

Study participants and pharmacokinetic collection

Of the 1857 eligible participants initially randomized in A5202, 926 (50%) were assigned to and initiated an atazanavir/ritonavir-based therapy.¹⁸ There were no plasma atazanavir concentration data for 105 (11%) participants who initiated atazanavir/ritonavir, primarily because of premature discontinuation from the study or study drug modification. Pharmacokinetic samples were available from 821 participants receiving atazanavir/ritonavir; however, due to dosing history discrepancies between the database and case report forms, population pharmacokinetic analysis was restricted to 815 (88% of 926) participants. Among these 815 individuals, 622 (76%) contributed at least three samples, 115 (14%) contributed two samples and 78 (10%) contributed one sample. Visits A and B were combined on the same day for 189 (23% of 815) participants. A total of 2195 atazanavir concentration samples were analysed, with 48 (2%) samples below the LLQ (Figure S2). Twenty-nine individuals from low frequency race/ethnicity groups were excluded, leaving 786 participants for analyses (Table 1).

Table 1.

Baseline characteristics for atazanavir/ritonavir subjects by availability of plasma pharmacokinetic atazanavir data (restricted to subjects of white, black or Hispanic race/ethnicity)

Characteristic	Included in ATV pharmacokinetic analysis (n = 786)	ATV clearance data not available or not included (n = 108)	P valuea
Assigned NRTIs, n (%)
TDF/FTC	389 (49)	55 (51)	0.780
ABC/3TC	397 (51)	53 (49)
Sex, n (%)
male	655 (83)	91 (84)	0.808
female	131 (17)	17 (16)
Age (years), median (IQR)	39 (31–45)	38 (29–46)	0.544
Race/ethnicity, n (%)b
white, non-Hispanic	336 (43)	39 (36)	0.295
black, non-Hispanic	257 (33)	43 (40)
Hispanic	193 (25)	26 (24)
Screening HIV-1 RNA, n (%)
≥100 000 copies/mL	344 (44)	40 (37)	0.185
Baseline CD4+ (cells/mm3), median (IQR)	233 (81–336)	202 (69–332)	0.368
BMI (kg/m2), median (IQR)	25 (22–28)	25 (22–28)	0.943

ATV, atazanavir; TDF/FTC, tenofovir disoproxil fumarate/emtricitabine; ABC/3TC, abacavir/lamivudine.

^aGroup differences for categorical variables were assessed with a χ² test. Differences in continuous variables were assessed using a Mann–Whitney test.

^bEighteen Asian/Pacific Islanders, three Native American/Alaskan Natives, seven subjects reporting more than one race and one subject with missing race/ethnicity information were excluded.

Atazanavir pharmacokinetic comparisons

From the one-compartment pharmacokinetic structural model, the estimate of the mean population (relative standard error expressed in percentage) first-order absorption rate constant was 0.47 h⁻¹ (11), the estimated volume of distribution was 86.7 L (5) and the estimated CL/F of atazanavir was 7.86 L/h (2). The estimated between-participant variability for CL/F expressed as a coefficient of variation was 36% (8) and the estimated residual variability was 46% (4). The estimated difference between mean atazanavir CL/F rates was 0.74 L/h (95% CI 0.22–1.27) slower in females than in males [7.63 (standard deviation 2.86) versus 8.37 (standard deviation 2.37) L/h; P = 0.006]. In the allometric scaled covariate model, sex was identified as the only covariate that was significantly associated with clearance, which was decreased in female participants by an average of 14% compared with males (Table S1). For each of the atazanavir blood samples, the ritonavir concentration was also to be measured. However, of the 2156 ritonavir concentrations, 771 (36%) were below the assay limit of detection, making it difficult to develop a model with reliable parameter estimates of ritonavir exposure. Among 734 participants of white, black or Hispanic race/ethnicity and with detectable ritonavir concentrations available, the estimated mean plasma concentration was 672 (standard deviation 461) ng/mL among males (n = 613) and 832 (standard deviation 588) ng/mL among females (n = 121; Satterthwaite t-test, P = 0.0001).

Participants randomized to abacavir/lamivudine had a mean atazanavir CL/F that was slower by 0.51 L/h (95% CI 0.17–0.86) compared with those receiving tenofovir disoproxil fumarate/emtricitabine: 7.99 (standard deviation 2.46) versus 8.51 (standard deviation 2.45) L/h (P = 0.003). Among whites, blacks and Hispanics, the estimated mean CL/F rates were 8.12 (standard deviation 2.33), 8.29 (standard deviation 2.70) and 8.42 (standard deviation 2.39) L/h, respectively, and did not differ significantly (P = 0.40). Tenofovir disoproxil fumarate randomization and race/ethnicity were not statistically significant in the allometric model.

Atazanavir oral clearance and time to virological failure

Overall, there were 106 (13.5%) protocol-defined virological failure events among 786 participants included in this analysis (Table 2). There were 44, 23 and 39 virological failures in the fast, middle and slow atazanavir CL/F groups, respectively. Atazanavir CL/F was moderately associated with time to virological failure (P = 0.053); fast versus middle atazanavir CL/F levels had an estimated HR for virological failure of 1.81 (95% CI 1.08–3.01) and slow versus middle atazanavir CL/F had an estimated HR for virological failure of 1.76 (95% CI 1.04–2.96). However, the association between atazanavir CL/F and virological failure differed by sex (P = 0.003). Females in the fast atazanavir CL/F group had the highest risk of virological failure, but the slow and middle atazanavir CL/F groups were not demonstrably different (Figure 1). In contrast, among males, the slow atazanavir CL/F group had a higher risk of virological failure and there was not a significant difference between fast and middle atazanavir CL/F groups.

Table 2.

Estimated association between atazanavir clearance level and the hazard of virological failure, tolerability and safety by sex, race/ethnicity and NRTI arm (base models^a)

Group	Comparisons by ATV clearance tertile [number of events (VF/tolerability/safety)]			Virological failure HR (95% CI)	Tolerability HR (95% CI)	Safety HR (95% CI)
Overallb				P = 0.053	P = 0.128	P = 0.22
	fast (44/72/95)	versus	middle (23/49/79)	1.81 (1.08–3.01)	1.30 (0.90–1.87)	1.06 (0.79–1.43)
	slow (39/73/100)	versus	middle	1.76 (1.04–2.96)	1.45 (1.01–2.09)	1.28 (0.95–1.72)
Sexb				P = 0.003	P = 0.070	P = 0.81
female	fast (15/12/12)	versus	middle (5/4/13)	3.49 (1.24–9.84)	3.09 (1.00–9.59)	0.84 (0.38–1.84)
	slow (8/12/28)	versus	middle	0.82 (0.26–2.54)	1.58 (0.51–4.91)	1.10 (0.57–2.14)
male	fast (29/60/83)	versus	middle (18/45/66)	1.50 (0.82–2.71)	1.15 (0.78–1.70)	1.11 (0.80–1.53)
	slow (31/61/72)	versus	middle	2.10 (1.16–3.77)	1.52 (1.03–2.24)	1.27 (0.91–1.77)
Race or ethnicityb				P = 0.085	P = 0.51	P = 0.92
white	fast (10/23/39)	versus	middle (9/27/40)	1.37 (0.55–3.39)	0.93 (0.53–1.62)	1.14 (0.73–1.78)
	slow (14/25/43)	versus	middle	1.91 (0.82–4.45)	1.05 (0.61–1.80)	1.30 (0.85–2.01)
Hispanic	fast (6/19/17)	versus	middle (8/10/15)	0.57 (0.20–1.67)	1.44 (0.67–3.10)	0.79 (0.39–1.58)
	slow (4/17/16)	versus	middle	0.53 (0.16–1.79)	1.85 (0.85–4.06)	1.04 (0.51–2.10)
black	fast (28/30/39)	versus	middle (6/12/24)	3.38 (1.38–8.25)	1.84 (0.94–3.60)	1.16 (0.70–1.93)
	slow (21/31/41)	versus	middle	2.67 (1.07–6.67)	1.89 (0.97–3.69)	1.34 (0.81–2.23)
NRTIsb				P = 0.58	P = 0.121	P = 0.31
ABC/3TC	fast (23/38/49)	versus	middle (15/29/42)	1.83 (0.95–3.54)	1.47 (0.91–2.39)	1.32 (0.88–2.00)
	slow (21/36/57)	versus	middle	1.47 (0.75–2.88)	1.19 (0.73–1.94)	1.41 (0.94–2.10)
TDF/FTC	fast (21/34/46)	versus	middle (8/20/37)	1.86 (0.82–4.22)	1.16 (0.67–2.02)	0.83 (0.54–1.28)
	slow (18/37/43)	versus	middle	2.28 (0.99–5.28)	1.84 (1.07–3.18)	1.13 (0.73–1.76)

ABC/3TC, abacavir/lamivudine; ATV, atazanavir; TDF/FTC, tenofovir disoproxil fumarate/emtricitabine; VF, virological failure.

^aEach base model was stratified by screening HIV-1 RNA stratum (< or ≥10⁵ copies/mL), and also adjusted for NRTI and NRTI by screening RNA interaction.

^bInteraction P value between group and atazanavir clearance tertile.

Figure 1.

Kaplan–Meier plots of time to virological failure by atazanavir clearance groups for males and females. Sex-by-atazanavir clearance interaction P = 0.003. ATV/r, atazanavir/ritonavir.

The association between atazanavir CL/F and time to virological failure did not differ significantly by NRTI (P = 0.58; Table 2). The atazanavir CL/F-by-race/ethnicity interaction term yielded P = 0.085 (Table 2); black participants with fast or slow atazanavir CL/F each had higher rates of virological failure compared with black non-Hispanic participants with middle atazanavir CL/F (Figure S3). The race/ethnicity interaction was not significant after adjustment for the sex-by-CL/F interaction (P = 0.19) and the CL/F-by-sex interaction remained significant in this analysis (P = 0.007, data not shown). The association between atazanavir CL/F and time to virological failure did not significantly differ by HIV-1 RNA screening stratum, baseline CD4+ cell count, age, BMI or self-reported week 8 and 24 adherence (P ≥ 0.47, data not shown). Exploratory three-way interactions between atazanavir CL/F–sex–race/ethnicity and atazanavir CL/F–sex–NRTI were not detected (P = 0.94 and P = 0.60, respectively).

In a multivariable model, the estimated association of atazanavir CL/F by sex with time to virological failure was similar to the base model results and remained significant but with slightly attenuated results (P = 0.003; Table 3). From the same multivariable model, the time to virological failure was shorter for females compared with males within the fast atazanavir CL/F group (HR 4.57; 95% CI 2.33–8.95), but was not demonstrably different within the middle atazanavir CL/F group (HR 2.10; 95% CI 0.75–5.91) or slower CL/F group (HR 0.76; 95% CI 0.34–1.73). Results were similar in an intention-to-treat sensitivity analysis of atazanavir CL/F and virological failure (data not shown).

Table 3.

Estimated association between atazanavir clearance levels on the hazard of virological failure and tolerability by sex in adjusted multivariable model^a

Sex	Comparisons by ATV clearance tertiles	Virological failure HR (95% CI)b	Tolerability HR (95% CI)c
Female	fast versus middle	2.97 (1.02–8.61)	3.14 (1.01–9.81)
	slow versus middle	0.72 (0.23–2.26)	1.63 (0.52–5.08)
Male	fast versus middle	1.36 (0.74–2.51)	1.15 (0.78–1.70)
	slow versus middle	1.98 (1.09–3.59)	1.43 (0.97–2.11)

ATV, atazanavir.

^aStratified by HIV-1 RNA screening stratum and adjusted for NRTI, NRTI-by-HIV-1 RNA stratum interaction, race/ethnicity, age, baseline CD4+ cell count, BMI and week 8 and 24 adherence.

^bSex-by-atazanavir clearance interaction P = 0.003 for virological failure.

^cSex-by-atazanavir clearance interaction P = 0.094 for tolerability.

Atazanavir oral clearance and tolerability

Among fast, middle and slow atazanavir CL/F groups, 72, 49 and 73 participants modified atazanavir/ritonavir treatment, respectively. Time to atazanavir/ritonavir modification did not differ substantially by atazanavir CL/F group [fast versus middle atazanavir clearance HR 1.30 (95% CI 0.90–1.87) and slow versus middle atazanavir clearance HR 1.45 (95% CI 1.01–2.09); P = 0.128]. The interaction between sex and the CL/F level yielded P = 0.070 (Table 2); multivariable model results were similar (Table 3). There was no significant interaction by race/ethnicity or NRTI group (Table 2).

Among females with fast atazanavir CL/F, 68% (26/38) completed atazanavir/ritonavir treatment, compared with 86% (25/29) and 81% (52/64) among females with middle and slow atazanavir CL/F, respectively. Among males, the proportions completing atazanavir/ritonavir treatment were 76% (189/249), 79% (165/210) and 69% (135/196) among those with fast, middle and slow atazanavir CL/F, respectively (Figure 2). The most commonly provided reason for atazanavir/ritonavir modification was non-compliance with study medications or visits among females (8%) and males (10%). Eighteen percent (7/38) of females categorized as having a fast atazanavir CL/F were recorded to have discontinued atazanavir for non-compliance compared with 7% (2/29) of those with middle and 3% (2/64) with slow CL/F. Among males with slow atazanavir CL/F, 14% (28/196) reported non-compliance as a reason for atazanavir modification, compared with 9% of males with either middle (18/210) or fast (22/249) atazanavir CL/F. The reasons for atazanavir/ritonavir modification are summarized in Table S2.

Figure 2.

Kaplan–Meier plots of time to atazanavir/ritonavir treatment modification by atazanavir clearance groups for males and females. Sex-by-atazanavir clearance interaction P = 0.07. ATV/r, atazanavir/ritonavir.

Atazanavir oral clearance and safety events

A total of 274 grade 3 or 4 safety events were reported: 95, 79 and 100 events in the fast, middle and slow atazanavir CL/F groups, respectively. There was not a significant association between atazanavir CL/F group and time to safety event [fast versus middle atazanavir CL/F HR 1.06 (95% CI 0.79–1.43) and slow versus middle atazanavir CL/F HR 1.28 (95% CI 0.95–1.72); P = 0.22]. There were no significant interactions between atazanavir CL/F and sex (Figure 3), race/ethnicity or NRTI (Table 2).

Figure 3.

Kaplan–Meier plots of time to safety event by atazanavir clearance groups for males and females. Sex-by-atazanavir clearance interaction P = 0.81. ATV/r, atazanavir/ritonavir; NA, not applicable.

Post hoc sensitivity analyses among LLQ samples

Unlike standard pharmacokinetic studies with intensive sampling, the sparse pharmacokinetic study design in A5202 depended on the participants' own disclosure of adherence to study medications. To further assess possible undisclosed non-adherence, the primary results were also assessed through post hoc sensitivity analyses focusing on participants with an atazanavir concentration below the LLQ by (i) excluding study participants with atazanavir concentration(s) below the LLQ prior to 20 h post-dose on visits in which there was no directly observed therapy dose (n = 9) and (ii) excluding study participants with atazanavir concentration(s) below the LLQ at any time post-dose (n = 44).

In the first analysis (i), it was hypothesized that participants with an atazanavir concentration below the LLQ at timepoints before 20 h post-dose were the most likely to represent non-adherence based on the elimination half-life of atazanavir/ritonavir (∼8 h). With their exclusion, the atazanavir CL/F-by-sex interaction remained significant for time to virological failure [among women, fast versus middle atazanavir CL/F HR 3.37 (95% CI 1.18–9.56); among men, slow versus middle atazanavir CL/F HR 1.93 (95% CI 1.06–3.52); P = 0.004] and time to tolerability [among women, fast versus middle atazanavir CL/F HR 2.95 (95% CI 0.94–9.28); among men, slow versus middle atazanavir CL/F HR 1.50 (95% CI 1.01–2.21); P = 0.085].

Furthermore, excluding all participants with an LLQ value (ii), the atazanavir CL/F-by-sex interaction remained significant with respect to virological failure. For time to virological failure among women, fast versus middle atazanavir CL/F HR was 2.07 (95% CI 0.66–6.48); among men, slow versus middle atazanavir CL/F HR was 1.97 (95% CI 1.06–3.64); P = 0.012. For time to tolerability among women, fast versus middle atazanavir CL/F HR was 2.06 (95% CI 0.60–7.04); among men, slow versus middle atazanavir CL/F HR was 1.54 (95% CI 1.04–2.30); P = 0.23. Results by race/ethnicity and NRTI treatments for time-to-event outcomes were similar to the original analyses (data not shown).

Discussion

In this report, we used atazanavir plasma concentration data collected from ACTG study A5202 to compare atazanavir pharmacokinetics and time to clinical events by sex, race/ethnicity and assigned NRTI arm. The clearance of atazanavir was reduced in females compared with males after accounting for differences in body weight. In the covariate model based on allometric principles, differences in atazanavir clearance were not apparent by randomized NRTI arm or race/ethnicity.

Sex-related variability of PI plasma concentrations has been previously reported for some PIs, including atazanavir/ritonavir. However, the pharmacodynamics and clinical implications had not been determined. According to previous studies, the mean clearance of atazanavir is reduced by ∼10%–30% in females.^6,12,23 Here, female study participants in A5202 had 8.8% lower mean individual Bayesian predicted clearances compared with males and an overall 14% lower clearance in the allometric adjusted model.

The mechanisms behind sex-based variability in atazanavir oral clearance have not been fully elucidated, but several potential factors do vary by sex.²⁴ Gastric emptying time is generally slower in females than males, presumably due to hormonal effects from oestrogen and progesterone, which may allow for greater atazanavir absorption and bioavailability.^25,26 Ritonavir exposure in females is also typically higher compared with males, which was consistent with A5202 participants who had detectable ritonavir concentrations.^27,28 This lends support to the notion that variability in ritonavir exposure is relevant to sex-related differences of PI pharmacokinetics. Unfortunately, we were unable to explicitly test this in our model due to a significant amount of missing and undetectable ritonavir concentrations. Whether the true underlying factor in atazanavir exposure variability between males and females is due to direct differences in atazanavir metabolism and disposition or is more indirectly linked to sex-related ritonavir pharmacokinetic differences requires further investigation.

There was also a race/ethnicity imbalance between females and males. Among females (males), 18% (48%) were white, 50% (29%) were black and 32% (23%) were Hispanic. According to 2011 data among adult and adolescents diagnosed with HIV infection, these reflect an over-representation of Hispanic females and white males by 15%–18% and an under-representation of black males and females by 13%.²⁹ In Figure S4, the patterns of increased virological failure among females with fast atazanavir clearance and males with slow clearance remain descriptively apparent for white and black participants (there was no evidence of an interaction between atazanavir clearance, sex and race/ethnicity; P = 0.9). These patterns appear most pronounced among blacks, particularly black females, making it difficult to exclude race/ethnicity as a key factor in the observed results.

Despite the sex-related differences of atazanavir clearance, the interpretation of virological failure and tolerability outcomes by atazanavir clearance and sex was not straightforward due to discordant results among males and females. For males with slow atazanavir clearance, there was an association with worsened virological failure and tolerability outcomes. For females, having fast atazanavir clearance was associated with faster time to virological failure and tolerability events. This pattern of tolerability in females may have been due in part to discontinuation of atazanavir/ritonavir occurring after virological failure. Among females who had an intolerability event, 6/12 with fast atazanavir clearance modified treatment after virological failure was determined, whereas most atazanavir modifications in the middle (3/4) and slow (9/12) clearance groups occurred without virological failure.

Intuitively, one could assume the overall slower atazanavir clearance among females could lead to either better virological response because of higher exposure or, if toxic levels were reached, to worse virological response from intolerability leading to non-adherence of any or all components of the antiretroviral regimen. Smith et al.³⁰ found that in A5202, females assigned to atazanavir/ritonavir had a higher risk of virological failure than men assigned to atazanavir/ritonavir, while risks of intolerability and safety events were not different by sex, and neither was self-reported adherence. Here, we show that the risk of virological failure and intolerability appears to depend on the degree of atazanavir clearance differently for males and females. Females with fast atazanavir clearance were at the highest risk of virological failure, presumably due to low drug exposure, but males with fast atazanavir clearance had reasonable efficacy.

The reasons for intolerability from males with a slower clearance of atazanavir suggest that these individuals had suboptimal adherence to study medications and procedures (Table S2). This may have been an important factor leading to their overall higher rates of virological failure. Their increased exposure to atazanavir indicates adequate periclinic visit medication adherence. However, underlying intolerability to study treatments may have been a source of undetected sporadic non-adherence over the course of the study. Therefore, a potential limitation of this study is the nature of the sparse pharmacokinetic sampling design and modelling, which required assumptions of adherence to study medications leading up to plasma concentration collections for pharmacokinetic measures. Adherence is a dynamic process, and thus this assumption may not be entirely accurate across our study population. Performing traditional intensive pharmacokinetic sampling around a directly observed dose in all study participants, however, was not an economically feasible option.

Approximately 85% of participants included in this analysis indicated 100% adherence to all reported antiretrovirals over 1 week prior to the study visit at weeks 8 and 24. The possibility of unreported non-adherence was explored through sensitivity analyses by excluding participants with atazanavir concentrations that were below the atazanavir assay limit of quantification. Results among women were attenuated for virological failure and tolerability outcomes by assuming LLQ samples were associated with non-adherence rather than rapid atazanavir clearance; however, the overall interaction of sex remained significant.

Our study did not intend to investigate every potential source of variability in atazanavir pharmacokinetics, but rather pre-selected specific patient-level factors. Therefore, differences in body composition, genetics, underlying pathophysiological processes and other concomitant medications (e.g. gastric acid reducers, CYP3A inhibitors and inducers) may also be important to consider but were not addressed here. Additionally, we evaluated the pharmacokinetics of one antiretroviral drug, atazanavir/ritonavir, while in reality the virological response is more complex due to the presence of NRTI agents in a combination regimen. Plasma concentrations of the other antiretrovirals were measured in A5202 and are planned to be evaluated in future analyses.

A major strength of this study was the large sample size compared with other previous studies and therefore a greater ability to detect differences in atazanavir pharmacokinetics between groups. With plasma concentration data collected from >130 women, ACTG study A5202 represents the largest US-based randomized study to compare atazanavir exposure in HIV-infected women. Comparison of atazanavir exposure and outcomes by sex were secondary outcomes and enrolment into A5202 was not stratified by sex. The male-to-female study participant ratio in A5202 was imbalanced (5 : 1) but reflected recently reported estimates of HIV infection diagnoses among adults and adolescents in the USA (4 : 1).²⁹

In summary, A5202 used a sparse pharmacokinetic sampling study design that allowed for the collection and analysis of the majority of participants enrolled and randomized to atazanavir/ritonavir. While there has been a lack of consistency in the understanding of sex- and race/ethnicity-related pharmacokinetic and pharmacodynamic variability of atazanavir/ritonavir, we have identified sex as an important factor in atazanavir/ritonavir pharmacokinetics and treatment outcomes.

Funding

This research was supported by grants from the National Institute of Allergy and Infectious Diseases, National Institutes of Health (AI38858, to the ACTG Central Group; AI68636, to the ACTG Network; AI68634 and AI38855, to the ACTG Statistical and Data Analysis Center; U01AI068636 and UM1AI106701-01 to the University at Buffalo Pharmacology Specialty Laboratory; AI069434, to the University of Washington; AI069477, to the University of Miami). Additional support from the General Clinical Research Center units funded by the National Center for Research Resources and P30 AI50410, to the University of North Carolina at Chapel Hill CFAR. Also, support from the NIH Experimental Therapeutics in Neurological Disorders grant T32 NS007338 at the University of Rochester.

Transparency declarations

K. M. received research support from Gilead through the University of North Carolina at Chapel Hill. E. S. D. received research grant support from Abbott, Bristol-Myers Squibb, Gilead, Merck and ViiV, and is a consultant/advisor for Abbvie, Bristol-Myers Squibb, Gilead, Janssen, Merck & Company and ViiV. P. E. S. serves as a consultant or Scientific Advisory Board member for Abbott, Bristol-Myers Squibb, Gilead, GlaxoSmithKline, Merck & Company and Janssen, and has received grant support for research from Bristol-Myers Squibb, Gilead and GlaxoSmithKline. M. F. has current research support from Merck & Company and Pfizer, and previous research support from Abbott Laboratories and GlaxoSmithKline. A. C. C. has current research support from Merck & Company and previous research support from Boehringer-Ingelheim, Gilead Sciences, Schering-Plough and Tibotec-Virco, and was a member of a DSMB for a Merck-sponsored study. A. C. C. and immediate family members previously owned stock in Abbott Laboratories, Bristol-Myers Squibb, Johnson & Johnson and Pfizer. K. Y. S. changed affiliations from Rush University Medical Center, Chicago, IL, USA to ViiV Healthcare after the completion of the study, analysis and writing of this article. K. Y. S. was a consultant/advisor for Abbott, Bristol-Myers Squibb, Gilead, GlaxoSmithKline, Janssen, Merck & Company and ViiV. C. T. is a member of a Data Monitoring Committee for Tibotec. All other authors: none to declare.

Supplementary data

Figures S1 to S4, Table S1 and Table S2 are available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/).