Changes in Glomerular Kidney Function among HIV-1 Uninfected Men and Women Receiving Emtricitabine/Tenofovir Disoproxil Fumarate Pre-exposure Prophylaxis: A Randomized Placebo-controlled Trial

Abstract

Importance

Tenofovir disoproxil fumarate (TDF) use has been associated with declines in the estimated glomerular filtration rate (eGFR) when used as part of antiretroviral treatment by HIV-1 infected persons, but limited data are available for risk when used as pre-exposure prophylaxis (PrEP) for HIV-1 prevention.

Objective

To determine whether TDF-based PrEP causes eGFR decline in HIV-1 uninfected adults.

Design, Setting, and Participants

A per-protocol safety analysis of changes in eGFR in the Partners PrEP Study, a randomized, placebo-controlled trial of daily oral TDF and emtricitabine (FTC)-TDF PrEP among African heterosexual HIV-1 uninfected members of serodiscordant couples conducted from 2008 to 2012.

Main Outcomes and Measures

Pre-defined outcomes of this analysis were mean eGFR change and a ≥25% eGFR decline from baseline. eGFR was calculated using Chronic Kidney Disease Epidemiology Collaboration.

Results

Of 4640 subjects randomized and followed on either once daily TDF (n=1548), FTC-TDF (n=1545), or placebo (n=1547), 63% were male. At enrollment, median age was 35 years (range 18–64) and mean eGFR was 130 mL/min/1.73m². During a median follow-up of 18 months (interquartile range 12–27), mean within-group eGFR change from baseline was +0.14 mL/min/1.73m² for TDF, −0.22 mL/min/1.73m² for FTC-TDF, and +1.37 mL/min/1.73m² for placebo, translating into average declines in eGFR attributable to PrEP versus placebo of −1.23 mL/min/1.73m² (95% CI −2.06, −0.40; p=0.004) for TDF and −1.59 mL/min/1.73m² (95% CI −2.44, −0.74; p<0.001) for FTC-TDF. The difference in mean eGFR between PrEP and placebo appeared by one month after randomization, was stable through twelve months, and then appeared to wane thereafter. The proportion of persons who developed a confirmed ≥25% eGFR decline from baseline by 12 and 24 months was 1.3% and 1.8% for TDF and 1.2% and 2.5% for FTC-TDF, and these frequencies were not statistically different compared to placebo (0.9% and 1.3% by 12 and 24 months).

Conclusion and Relevance

In this large randomized, placebo-controlled trial among heterosexual persons, with median follow-up of 18 months and maximum follow-up of 36 months, daily oral TDF-based PrEP resulted in a small but non-progressive decline in eGFR that was not accompanied by a substantial increase in the risk of clinically relevant (≥25%) eGFR decline.

Trial Registration

Clinicaltrials.gov Identifier: NCT00557245

Introduction

Antiretroviral pre-exposure prophylaxis (PrEP) with tenofovir disoproxil fumarate (TDF) alone or in combination with emtricitabine (FTC-TDF) has demonstrated protection against HIV-1 acquisition in diverse geographical and at-risk populations^1–4, with effectiveness of 44–75% in randomized, placebo-controlled comparisons and ~90% in subset analyses of adherent participants.

Among HIV-1 infected individuals receiving antiretroviral therapy, studies have consistently demonstrated a significantly higher frequency of kidney dysfunction, including decline in estimated glomerular filtration rate (eGFR), in patients receiving TDF-containing regimens compared to those receiving regimens not containing TDF.^5–9 Extrapolating results from these studies to the PrEP context, however, is potentially confounded by HIV-1 infection and concomitant use of other antiretroviral medications. In PrEP clinical trials,^1–4,10 PrEP exposure was not associated with overt kidney toxicity. However, whether TDF exposure among HIV-1 uninfected adults causes more subtle but still clinically relevant declines in eGFR requires exploration. Use of PrEP with FTC-TDF is now recommended by the US Centers for Disease Control and Prevention and the World Health Organization,^11,12 lending greater importance to profiling the safety signals of FTC-TDF in HIV-1 uninfected persons.

We investigated the effect of daily oral TDF-based PrEP on eGFR in HIV-1 uninfected adults in a placebo-controlled trial of PrEP in which PrEP adherence was high.

Methods

Study design and participants

Data were from the Partners PrEP Study,^1,13 a phase III, randomized, placebo-controlled trial of daily oral TDF and FTC-TDF PrEP among heterosexual HIV-1 uninfected members of HIV-1 serodiscordant couples (Clinicaltrials.gov number NCT00557245). Between July 2008 and November 2010, 4747 HIV-1 serodiscordant heterosexual couples were enrolled at nine research sites in Kenya and Uganda. Eligible HIV-1 uninfected participants were ≥18 years of age, did not have active hepatitis B infection, were sexually active, were not pregnant or breastfeeding, had normal renal function (defined by serum creatinine ≤1.3 mg/dL for men / ≤1.1 mg/dL for women and Cockcroft-Gault calculated creatinine clearance of ≥60 mL/min), not receiving ongoing therapy with agents with known significant nephrotoxic potential, and did not have diabetes requiring hypoglycemic medication or active and clinically significant cardiac disease. HIV-1 uninfected partners were randomly assigned in a 1:1:1 ratio to one of the three study groups: TDF, FTC-TDF, or an inert placebo. TDF and FTC were dosed at 300 mg daily and 200 mg daily, respectively; these doses are also the standard for treatment of HIV-1.¹⁴

HIV-1 uninfected partners were followed monthly up to 36 months with HIV-1 testing, study medication refill for 30 days, collection of the prior month’s unused medication, and adherence counseling. Adherence to study medication was assessed by pill counts of returned bottles at each monthly visit. Laboratory safety, including serum creatinine, was evaluated at baseline, month 1 and quarterly thereafter. Grading of adverse events was based on the 2009 DAIDS grading systems adapted to local laboratory reference ranges.¹⁵ Study medication was permanently discontinued in subjects who experienced HIV-1 acquisition and was withheld in women who became pregnant for the duration of pregnancy and breastfeeding. Additionally, study medication was temporarily withheld if a participant had a confirmed creatinine abnormality (i.e., confirmed with repeat testing, ideally completed within 7 days) defined as serum creatinine increase 1.1 times upper limit of normal and or >1.5-fold change from baseline. Study drug could be restarted if serum creatinine returned to normal or within 1.3-fold of the baseline value. Study drug was permanently discontinued with a confirmed ≥grade 2 creatinine abnormality (defined as ≥1.4 times the upper limit of normal or a Cockcroft-Gault calculated creatinine clearance <50 mL/min).

The study protocol was approved by the University of Washington Human Subjects Review Committee and ethics review committees at each of the study sites. All participants provided written informed consent. Study progress was reviewed by an independent Data and Safety Monitoring Board (DSMB); in July 2011 the DSMB recommended that the placebo arm be discontinued, due to definitive demonstration of PrEP efficacy against HIV-1 acquisition. Additionally, the DSMB recommended continued blinded follow-up of the active arms to garner additional data on safety and efficacy of FTC-TDF vs TDF.¹⁶

Assessment of GFR

The eGFR was calculated from serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) Equation.¹⁷ The CKD-EPI equation has recently been validated in African populations and provides more accurate estimates for eGFR values in the normal range than both the Modification of Diet in Renal Disease Study and Cockoft-Gault equations when compared to a direct measure of GFR by iohexol clearance.^18,19 Serum creatinine was measured at baseline, month 1, and quarterly thereafter. For this analysis, predefined study outcomes were mean eGFR change from baseline and a decline in eGFR of ≥25% compared to baseline, as confirmed by a second measurement obtained prior to study drug discontinuation. The cutoff of ≥25% decline in eGFR was adapted from established criteria for the diagnosis of acute kidney injury²⁰; eGFR decline of this magnitude has been associated with increased morbidity and mortality.^21–23 eGFR values >200 were imputed to 200 mL/min/1.73m² consistent with the range of GFR values in the CKD-EPI study.¹⁷ All site laboratories participated in regular proficiency testing.

Statistical analysis

The primary analysis was a per-protocol safety analysis, censoring participants’ visits occurring after >4 consecutive weeks off study drug for any reason (including protocol required safety hold, missed visits, and HIV-1 seroconversion). Our aim was to estimate the effect of continuous PrEP use on eGFR; recognizing that full adherence is naturally impractical in clinical settings, the per-protocol analysis is a robust approach to address drug safety.²⁴ The primary analyses were conducted using data collected from November 2008 through July 2011, when the placebo arm of the trial was discontinued.

To assess the study endpoints (absolute mean eGFR difference from baseline and time to first confirmed ≥25% eGFR decline from baseline), we used standard regression methods as the primary approach, and marginal structural models weighted with inverse probability of censoring weights in sensitivity analyses.^25,26 Marginal structural models have been proposed as one method to address potential selection bias or confounding that can result from post-randomization nonadherence and censoring.^25,26 We also evaluated treatment effects among subgroups of sex and age. For all analyses, each active PrEP arm (TDF and FTC-TDF) was compared separately to placebo; additional testing compared TDF versus FTC-TDF. The net mean eGFR difference associated with PrEP was computed as the difference in mean eGFR change from baseline between the TDF or FTC/TDF groups and the placebo group. Outcome and person-time were evaluated at 1 month and then quarterly.

For the absolute mean change in eGFR, linear regression and marginal structural linear models were fit using generalized estimating equations, with time since enrollment fitted using 3 knot restricted cubic spline (details provided in etable 1a). Models were adjusted for site, sex, and age with an independent correlation structure and robust standard errors to correct for the within-person correlations.²⁷ Treatment effects by visit-month since randomization were generated in a separate model fitted with treatment by categorical time interaction.

For the analysis of time to the first confirmed decline in eGFR of ≥25%, we used right-censored Kaplan-Meier methods to estimate the cumulative probability, standard Cox proportional hazards models to estimate relative hazard rates,²⁸ and marginal structural Cox proportional hazards models with time-dependent inverse probability-of-censoring weights ^25,26 with robust standards errors derived by the Lin and Wei variance estimator²⁹ to control for within-person correlation (see etable 1b for details of weight estimation). Cox proportional hazard models were adjusted for baseline eGFR as 3 knot restricted cubic spline and stratified according to site, age groups, and sex. Ties were handled using the Efron approximation method.^30,31

Three additional sensitivity analyses were conducted: 1) a time to event analysis of the repeated events of a ≥25% eGFR decline using the Andersen-Gill counting process approach in Cox regression models under the per-protocol approach³²; 2) an intention-to-treat analysis including all randomized persons with at least one post-randomization creatinine measurement regardless of time off study medication using data collected through July 2011; and 3) an intention-to-treat approach including the additional follow-up of the two active PrEP arms after the suspension of the placebo arm in July 2011.

Finally, in exploratory analysis, we evaluated baseline factors associated with a ≥25% decline in eGFR from baseline. Additionally, we evaluated the frequency of a >1.5 fold increase in serum creatinine above baseline and study medication discontinuation related to creatinine abnormalities. Analyses were conducted using SAS software (version 9.3, SAS Institute).

Results

Of the 4747 HIV-1 uninfected individuals enrolled in the Partners PrEP Study, 4640 (98%) were included in the primary per-protocol safety analysis: 1548 in the TDF group, 1545 in the FTC-TDF group, and 1547 in the placebo group (Figure 1). Of 107 excluded, 51 did not have any post-randomization serum creatinine measurement and 56 were off study medication >4 consecutive weeks by their first creatinine measurement, generally due to treatment refusal, missed visits, or pregnancy. Of the 4640 participants included in the primary analysis, 63% were male and mean age at enrollment was 35 years (range 18–64). Baseline characteristics were comparable across the three treatment groups (Table 1). Overall, 6548.8 person-years were accrued during median follow-up of 18 months (interquartile range 12 to 27) for this per-protocol safety analysis, representing approximately 88% of the total person-years collected in the study [i.e., 12% of person-years were excluded from this per-protocol analysis due to post-randomization censoring, mostly due to missed visits (5%), pregnancy (2.5%), and treatment refusal (1.4%)]. The distribution of triggers for censored person-time were no more frequent in the active PrEP arms than in the placebo group including that resulting from creatinine abnormality-related study medication hold (0.7% overall: 0.6% for TDF, 0.8% for FTC-TDF, and 0.6% for placebo; p>0.05 for both TDF and FTC-TDF vs placebo). Because of the truncated follow-up of the placebo group, few participants (n=718) contributed ≥30 months of follow-up in the primary per-protocol analysis. An additional 2638 person-years were accrued in the active PrEP arms after July 2011 and contribute to the sensitivity analyses using the intent-to-treat approach. Overall, including the additional follow-up of the active PrEP arms beyond July 2011 in the sensitivity analysis, participants were followed for a median of 30 months (interquartile range 24 to 36); with the TDF and FTC-TDF arms observed for a median of 36 months (interquartile range 27 to 36).

Figure 1. Study Participants.

Overall, 4640 (98%) HIV-1 uninfected persons contributed to the primary per-protocol safety analysis. 107 participants were not included in the primary analysis: 51 because they did not have any post-randomization creatinine measurement to permit estimation of glomerular filtration rate and 56 because were >4 consecutive weeks off study medication by the time of their first post-randomization creatinine measurement mostly due to missed visits and drug refusal. The intention-to-treat analysis included all randomized participants with at least one post-randomization serum creatinine measurement regardless of time off study medication (n=4696). FTC denotes emtricitabine and TDF denotes tenofovir disoproxil fumarate.

Table 1.

Enrollment characteristics according to treatment group

Characteristic	FTC-TDF (n=1545)	TDF (n=1548)	Placebo (n=1547)
Age (years)
Mean (range)	35 (18–64)	34 (18–64)	35 (18–64)
≤24	11%	12%	11%
25–34	44%	45%	43%
35–44	32%	30%	33%
≥45	13%	13%	13%
Male	64%	62%	61%
Creatinine (mg/dL)	0.78 ± 0.15	0.78 ± 0.15	0.78 ± 0.15
eGFR (mL/minute/1.73m2)
Mean	129 ± 17	130 ± 17	129 ± 17
eGFR ≥90	98%	97%	98%
Weight (kg)
Mean	61 ± 10	61 ± 10	61± 11
>50kg	87%	86%	87%
Systolic blood pressure ≥140mmHg	5%	5%	6%
Diastolic blood pressure ≥90mmHg	3%	3%	5%

Unless stated, statistics are mean ±standard deviations for continuous covariates and percentages for categorical covariates. FTC denotes emtricitabine TDF denotes tenofovir disoproxil fumarate

Effect of TDF and FTC-TDF PrEP on absolute mean eGFR change from baseline

Overall, mean eGFR at baseline was 130 mL/minute/1.73 m² for the TDF group, 129 mL/minute/1.73m² for the FTC-TDF group, and 129 mL/minute/1.73 m² for placebo group. During randomized treatment, PrEP was associated with a small but statistically significant decline in eGFR. During a median 18 months of PrEP treatment, the mean within-group eGFR change from baseline was +0.14 mL/min/1.73 m² for the TDF group, −0.22 mL/min/1.73 m² for the FTC-TDF group, and +1.37 mL/min/1.73m² for placebo, representing absolute mean eGFR change associated with PrEP of −1.23 mL/min/1.73 m² (95%CI −2.06, −0.40; p=0.004) for TDF and −1.59 mL/min/1.73 m² (95%CI −2.44, −0.74; p<0.001) for FTC-TDF (Table 2). Compared to baseline eGFR, the estimated differences in mean eGFR change from baseline between PrEP and placebo translated into a 0.9% and 1.2% decline in eGFR that was associated with TDF and FTC-TDF, respectively. The difference between PrEP and placebo in eGFR changes from baseline appeared by 4 weeks after randomization (−1.70 mL/min/1.73 m², p=0.001 for TDF vs. placebo and −2.42 mL/min/1.73 m², p<0.001 for FTC-TDF vs. placebo), was stable to 12 months, and then appeared to gradually wane thereafter (at 24 months: −0.81 mL/min/1.73 m², p=0.31 for TDF vs. placebo and −0.42 mL/min/1.73 m², p=0.63 for FTC-TDF vs. placebo). The pattern of change over time in crude mean eGFR difference from baseline had upper limits of the 95% confidence intervals under 3 mL/min/1.73 m² through 36 months post-randomization with the additional follow-up of the two active PrEP arms (Figure 2). Overall, PrEP effects were consistent among subgroups of age and gender and in all sensitivity analyses including marginal structural models.

Table 2.

Estimated mean eGFR difference from baseline, according to treatment group

	Number of individuals evaluated; within group estimated mean eGFR change from baseline, (mL/min/1.73 m2)d						Attributable mean eGFR difference mL/min/1.73 m2 (95%CI)d
Characteristic	FTC-TDF		TDF		Placebo		TDF vs. placebo	FTC-TDF vs. placebo
A. Primary analysis- overall treatment effect
Overall per-protocol	1545;	−0.22	1548;	+0.14	1547;	+1.37	−1.23 (−2.06, −0.40); p=0.004	−1.59 (−2.44, −0.74); p<0.001
B. Sensitivity analyses for overall treatment effect
MSMa per-protocol	1545	−0.21	1548	+0.15	1547	+1.38	−1.23 (−2.06, −0.40); p=0.004	−1.59 (−2.44, −0.74); p<0.001
Intention-to-treatb	1558;	−0.17	1568;	+0.18	1570;	+1.41	−1.23 (−2.06, −0.41); p=0.003	−1.59 (−2.44, −0.74); P<0.001
ITT extended follow-upc	1558;	−0.08	1568;	+0.32	1570;	+1.28	−0.96 (−1.78, −0.14); p=0.02	−1.36 (−2.20, − 0.52); p=0.001
C. Treatment effect among subgroups (per-protocol approach)
Sex
Male	988;	+0.25	962;	+0.66	936;	+1.75	−1.09 (−2.09, −0.08); p=0.03	−1.50 (−2.53, −0.49); p=0.004
Female	557;	−0.69	586;	−0.43	611;	+1.04	−1.47 (−2.92, −0.02); p=0.05	−1.73 (−3.23, −0.23); p=0.02
Age
18–34yrs	846;	−0.39	879;	+0.29	834;	+1.28	−0.99 (−2.19, 0.21); p=0.10	−1.67 (−2.88, −0.46); p=0.007
35–44yrs	491;	−0.21	471;	+0.33	508;	+1.78	−1.45 (−2.87, −0.02); p=0.05	−1.99 (−3.45, −0.54); p=0.007
≥45yrs	208;	+0.27	198;	−0.82	205;	+0.76	−1.58 (−3.49, 0.34); p=0.11	−0.49 (−2.56, 1.58); p=0.64
D: Treatment effect by month since randomization (per-protocol approach)
1 month	1545;	−0.19	1548;	+0.53	1547;	+2.23	−1.70 (−2.68, −0.72); p=0.001	−2.42 (−3.42, −1.43); p<0.001
3 months	1495;	−0.58	1481;	+0.21	1476;	+1.43	−1.22 (−2.25, −0.19); p=0.02	−2.01 (−3.08, −0.95); p<0.001
6 months	1428;	+0.09	1402;	+0.44	1410;	+1.51	−1.07 (−2.14, <−0.01); p=0.05	−1.42 (−2.51, −0.33); p=0.01
12 months	1203	−0.41	1159	+1.03	1173	+1.60	−0.57 (−0.72, 0.58); p=0.33	−2.01 (−3.20, −0.83); p<0.001
15 months	1078;	−0.08	1031;	+0.42	1046;	+0.90	−0.48 (−1.71, 0.74); p=0.44	−0.98 (−2.25, 0.28); p=0.13
18 months	905;	+0.90	888;	+0.66	899;	+1.16	−0.50 (−1.89, 0.88); p=0.47	−0.26 (−1.65, 1.13); p=0.71
24 months	614;	−1.01	589;	−1.40	621;	−0.59	−0.81 (−2.39, 0.77); p=0.31	−0.42 (−2.11, 1.27); p=0.63

The primary approach was a per-protocol analysis censoring any visits occurring after >4 consecutive weeks off study medication.

^aMSM: Marginal structural models. The marginal structural models used time-dependent stabilized weights (mean 1.00, range 0.99–1.03), which were estimated using pooled logistic regression models with censoring indicator as the outcome and prior visit histories of the time-varying covariates, creatinine clearance and serum creatinine, as 3 knot restricted cubic splines at 5th, 50th, and 95th percentiles plus treatment arm, sex, site, age, baseline eGFR, and time since randomization as part of the numerator model.

^bIntention-to-treat analysis includes all randomized participants with at least one post-randomization serum creatinine measurement regardless of time off study medication, for data collected through July 2011.

^cITT extended follow-up analysis is an intention-to-treat approach that includes the additional follow-up time of persons randomized to TDF and FTC-TDF groups beyond July 2011 when the placebo arm was discontinued. During the extended follow-up, both investigators and participants remained blinded to the type of PrEP drug the participant was receiving. Placebo groups contributed records only up to July 2011.

^dAttributable mean eGFR difference represents the difference in mean eGFR change from baseline between the respective active PrEP treatment group and the placebo group. All subgroups and the treatment by month estimates are for the per-protocol approach. Estimates were generated from linear regression fit using generalized estimating equations. Models were adjusted for sex, indicator for site to account for heterogeneity in serum creatinine assaying, continuous age and time since randomization as 3 knot restricted cubic splines at 5th, 50th, and 95th percentiles. Baseline body mass index, elevated blood pressure or non-steriodal inflammatory drug use did not modify PrEP effects (p>0.05 for all) and their addition as covariates did have substantial effect on the estimates. Treatment effects by visitmonth since randomization were generated from treatment by categorical time interaction. P-values are two-sided testing the null hypothesis of no treatment effects. FTC denotes emtricitabine TDF denotes tenofovir disoproxil fumarate

Figure 2. Variation over time in crude means eGFR change from baseline, according to treatment group.

Plot 2A represents all data collected through July 2011, when the trial’s placebo arm was discontinued; because of truncation of follow-up time in July 2011, few participants had achieved >30 months of follow-up. Plot 2B represents crude mean eGFR changes from baseline that includes additional follow-up of the TDF and FTC-TDF arms beyond July 2011. The placebo group contributed person-time only up to July 2011. Whiskers represent 95% confidence intervals. FTC denotes emtricitabine and TDF denotes tenofovir disoproxil fumarate.

Overall, confirmed CKD-eGFR decline to <60 mL/min/1.73 m² was recorded in two participants, both in the TDF group. First, a 58 year-old, 61 kg male with baseline CKD-eGFR of 99 mL/min/1.73m² had CKD-eGFR of 10 mL/min/1.73 m² (serum creatinine 7.2 mg/dL) at 36 months with concurrent 2+ dipstick proteinuria, grade 4 liver transaminases, and clinical features suggestive of acute hepatitis. There was no concurrent nephrotoxic medication. Study drug was permanently discontinued and eGFR returned to >60mL/min/1.73 m² within 4 weeks. Second, a 34 year-old, 58 kg male with baseline eGFR of 154 mL/min/1.73 m² had CKD-eGFR of 57 mL/min/1.73 m² (serum creatinine 1.53 mg/dL) at 30 months with history of recent relocation to a hot and dry region. Urine dipstick and liver enzymes were normal and there was no concomitant medication. Study drug was discontinued permanently and eGFR returned to >60mL/min/1.73 m² within 2 weeks. Both events were conservatively managed.

Effect of TDF and FTC-TDF on a ≥25% eGFR decline from baseline

Overall, confirmed ≥25% eGFR decline was rare (Table 3). A total of 72 events occurred in the study, 68 during the per-protocol observation period and 4 during the censored period. Of these 68 events, 23 were in the TDF group (incidence rate=1.08 per 100 person-years), 27 were in the FTC-TDF group (incidence rate=1.24 per 100 person-years), and 18 were in the placebo group (incidence rate=0.83 per 100 person-years), representing attributable incidence rate difference of 0.41 per 100 person-years (95% CI −0.19, 1.01) for FTC-TDF and 0.25 per 100 person-years (95% CI −0.33, 0.83) for TDF alone, neither of which was statistically different than placebo. The proportion of persons who developed a ≥25% eGFR decline from baseline was 1.3% for TDF, 1.2% FTC-TDF, and 0.9% for the placebo by 12 months; 1.8% for TDF, 2.5% FTC-TDF, and 1.3% for the placebo by 24 months; and 1.8% for TDF, 2.5% for FTC-TDF, and 2.2% for placebo by 36 months (Figure 3). Compared to placebo, the adjusted relative hazards for a confirmed ≥25% eGFR decline from baseline associated with active PrEP was 1.33 (95% CI 0.71, 2.48; p=0.37) for TDF alone and 1.45 (95% CI 0.79, 2.64; p=0.23) for FTC-TDF (Table 3). In exploratory analysis, older age, female gender, and higher baseline eGFR appeared to be independently associated with increased likelihood for ≥25% eGFR decline from baseline (p <0.05 for all). Overall, PrEP effects were consistent among subgroups of age and gender and in all sensitivity analyses including marginal structural models.

Table 3.

Absolute incidence rates and hazard ratios for a ≥25% eGFR decline from baseline overall and among subgroups, according to treatment study group.

	No. of events/total person-years			Absolute incidence rate difference per100 person-yearsa (95%CI)		Adjusted hazard ratioe (95%CI); p-value
Approach	FTC-TDF	TDF	Placebo	FTC-TDF vs. Placebo	TDF vs. Placebo	TDF vs. Placebo	FTC-TDF vs. Placebo
A. Primary per-protocol analysis -overall treatment effect
Per-protocol	27/2184.1	23/2133.3	18/2174.3	0.41 (−0.19, 1.01)	0.25 (−0.33, 0.83)	1.33 (0.70, 2.48); p=0.37	1.45 (0.79, 2.64); p=0.23
B. Sensitivity analyses
MSM per-protocol	27/2184.1	23/2133.3	18/2174.3	0.41 (−0.19, 1.01)	0.25 (−0.33, 0.83)	1.33 (0.71, 2.51); p=0.38	1.45 (0.80, 2.63); p=0.23
Repeated eventsb	37/2206.9	31/2154.3	26/2188.1	0.49 (−0.22, 1.20)	0.25 (−0.43, 0.93)	1.22 (0.55, 2.67); p=0.63	1.38 (0.68, 2.79); p=0.37
Intention-to-treatc	28/2445.2	25/2431.6	19/2460.7	0.37 (−0.18, 0.92)	0.26 (−0.28, 0.79)	1.37 (0.75, 2.50); p=0.31	1.44 (0.80, 2.59); p=0.22
ITT extended follow-upd	37/3731.2	32/3733.0	19/2460.7	0.22 (−0.25, 0.69)	0.09 (−0.37, 0.54)	1.38 (0.78, 2.46); p=0.27	1.54 (0.88, 2.70); p=0.13
C. Treatment effect among subgroups (per-protocol approach)
Sex
Male	12/1392.1	8/1349.2	8/1362.4	0.27 (−0.36, 0.91)	0.01 (−0.57, 0.58)	1.04 (0.39, 2.78); p=0.94	1.41 (0.5, 3.45); p=0.46
Female	15/792.0	15/784.0	10/811.9	0.66 (−0.56, 1.89)	0.68 (−0.55, 1.91)	1.51 (0.68, 3.38); p=0.31	1.56 (0.70, 3.48); p=0.28
Age groups
18–34yrs	9/1065.1	15/1089.5	7/1056.9	0.18 (−0.56, 0.92)	0.35 (−0.43, 1.12)	1.54 (0.60, 3.98); p=0.37	1.37(0.5, 3.67); p=0.54
35–44yrs	13/747.1	9/714.7	9/755.1	0.55 (−0.68, 1.77)	0.07 (−1.07, 1.20)	1.07 (0.42, 2.69); p=0.89	1.56 (0.67, 3.67); p=0.30
≥45yrs	5/371.9	3/329.1	2/362.3	0.79 (−0.61, 2.20)	0.35 (−0.92, 1.64)	1.46 (0.24, 8.76); p=0.68	2.11 (0.4, 10.94); p=0.37

The primary approach was a per-protocol analysis censoring visits occurring after >4 consecutive weeks off study medication.

MSM: Marginal structural models weighted by inverse probability censoring weights (mean weight 1, range 0.99, 1.02); details of weight estimation provided in etable 1b.

^aAbsolute incidence rate difference represents the difference of incidence rate in the placebo arm from the incidence rate of the respective active PrEP arm. The rates and rate differences reported for the marginal structural models results are unweighted as in the primary approach.

^bAnalysis of repeated events of ≥25% eGFR decline in a per-protocol approach using Andersen Gill counting process approach in Cox- regression model. Given that changes in eGFR is a continuum of cumulative biological process for which the true start or end of at-risk periods for each repeated event is nearly impossible to establish, the reported rates in the repeated events approach should be interpreted rates of episodes of ≥25% eGFR decline but not as true incidence rates.

^cIntention-to-treat analysis included all randomized participants with at least one post-randomization serum creatinine measurement regardless of the time off study medication, for data collected through July 2011.

^dITT extended follow-up is an intention-to-treat approach that includes additional follow-up time of persons randomized to TDF and FTC-TDF groups beyond July 2011 when the placebo arm was discontinued by the DSMB. The proportion of patients who developed a ≥25% eGFR decline from baseline was consistent with that recorded in the primary per-protocol analysis (i.e 1.3% for TDF, 1.3%, and 0.9% for the placebo, by 12 months; 1.9% for TDF, 2.3%, and 1.3% for the placebo by 24 months; and 2.2% for TDF, 2.8%, and 2.2% by 36 months). During the additional follow-up both the investigators and participants remained blinded to the type of PrEP drug the participant was receiving. Placebo groups contributed records only up to July 2011.

^eHazard ratios estimated using Cox proportional hazards models stratified according sex, age groups, and site to account for laboratory heterogeneity in serum creatinine estimation. For subgroups, the group evaluated was dropped from the stratification. All subgroups estimates are for the per-protocol approach. Models were adjusted for baseline eGFR as 3 knot restricted cubic splines at 5^th, 50^th, and 95^th percentiles. All p-values are two-sided testing the null hypothesis of no treatment effects. FTC denotes emtricitabine and TDF denotes tenofovir disoproxil fumarate.

Figure 3. Cumulative probability of a ≥25% estimated glomerular filtration rate decline from baseline, according to study treatment.

Plot 3A represents estimates for the primary per-protocol safety analysis including data accrued up to July 2011 when the placebo arm was discontinued by the DSMB. The proportion of persons who developed a ≥25% eGFR decline from baseline was 1.3% for TDF, 1.2% for FTC-TDF, and 0.9% for placebo by 12 months; 1.8% for TDF, 2.5% for FTC-TDF, and 1.3% for placebo by 24 months; and 1.8% for TDF, 2.5% for FTC-TDF, and 2.2% for placebo by 36 months. Plot 3B represents estimates for the sensitivity analysis that included additional follow-up of the TDF-and FTC-TDF arms beyond July 2011 with placebo arm data truncated at July 10, 2011. The proportion of persons who developed a ≥25% eGFR decline from baseline was 1.3% for TDF, 1.3% for FTC-TDF, and 0.9% for placebo by 12 months; 1.9% for TDF, 2.3% for FTC-TDF, and 1.3% for placebo by 24 months; and 2.3% for TDF, 2.8% for FTC-TDF, and 2.2% for placebo by 36 months. Failure function was calculated over full data and evaluated at indicated times; it is not calculated from aggregates of number of persons shown on the x-axis plots. FTC denotes emtricitabine and TDF denotes tenofovir disoproxil fumarate.

Frequency of a >1.5-fold serum creatinine increase above baseline

Overall, a total of 451 unconfirmed events of serum creatinine increase >1.5-fold above baseline were recorded (n=237 participants). Of these, 159 (35%) were confirmed on repeat measurement from 47 (1% of 4696 total subjects regardless of time off study medication) participants: 63 events were in the TDF group, 60 in the FTC-TDF group, and 36 in the placebo. Study medication was permanently discontinued in 5 of these subjects per protocol specificatin (2 each in the TDF and FTC-TDF groups and one in placebo; all had borderline creatinine clearance at baseline range 60 to 72 mL/min).

Discussion

In this safety analysis from a large randomized placebo-controlled trial, daily oral TDF-based PrEP resulted in a small but statistically significant decrease in estimated glomerular filtration rate – specifically, a change relative to baseline <1.5%, which was non-progressive for 36 months and was not accompanied by a significant increase in the likelihood of a clinically-relevant change in eGFR (i.e., ≥25%). The observed results were consistent in different subgroups and in multiple statistical approaches to evaluate the treatment causal effects. To our knowledge, this is the largest randomized trial to quantify the magnitude of subclinical eGFR decline in the presence of high adherence to PrEP in both men and women and across a broad range of ages.

Glomerular filtration rate is easily estimated from serum creatinine using prediction equations that take into account age, sex, and race or body weight, and provides a more reliable and accurate index for detection and monitoring of glomerular kidney dysfunction compared to serum creatinine alone. Age-related decline in GFR has been considered part of the normal aging process declining by approximately 1 mL/min/1.73 m² per year beginning after 40 years of age.^33,34 However, the clinical significance of drug-related subclinical eGFR decrease in healthy HIV-1 uninfected adults is unknown. In the current study, we observed small subclinical declines in mean eGFR with upper bounds of the 95% confidence intervals in the range of 1–3 mL/min/1.73 m²; PrEP effects were reversible after drug discontinuation. Because PrEP use is a time-dependent intervention for months or years of greatest HIV-1 risk and not life-long, the clinical significance of the observed changes in eGFR may be quite small. Early TDF-induced nephrotoxicity appears to be reversible in both HIV-1 infected and uninfected persons after TDF discontinuation.^35,36 In our study, an increase in the within-group eGFR over-time for the placebo and TDF groups is likely a regression to the mean rather than a true biological effect³⁷ and the between-group differences represent unbiased estimates of PrEP effects; analysis of covariance yielded similar between-group estimates. Mean eGFR decline appeared to be non-progressive to a period of 36 months, as assessed with the additional follow-up of the two active PrEP arms beyond July 2011. The majority of creatinine elevations observed were self-limited and were not confirmed on subsequent measurement, and the occurrence of clinically relevant decline in eGFR (i.e. ≥25% eGFR decline from baseline) was low. In the two subjects who developed eGFR <60 min/mL/1.73 m², eGFR rebounded to >60 min/mL/1.73 m² within four weeks after drug discontinuation. There was no evidence of substantial increase in clinically relevant eGFR decline related to PrEP compared to placebo although, given the 95% confidence intervals, an increase in absolute rate of a ≥25% eGFR decline as high as 1% per year that could be attributed to PrEP cannot be ruled out.

Drug exposure is an important determinant of both PrEP efficacy and assessment for safety. Adherence in the Partners PrEP Study was the highest of any published PrEP clinical trial:^1,38 tenofovir was detectable in plasma in 82% of a randomly selected cohort of subjects and 17% of those samples with no detected drug were a result of protocol-defined drug holds.³⁹ Our findings, which enriched for drug exposure in the primary analysis by limiting to per-protocol periods, are thus encouraging in demonstrating that clinically relevant eGFR decline was rare in the context of high exposure to PrEP.

Our study provides both new and complimentary evidence to the recent analysis from the iPrEx study,⁴⁰ a PrEP trial among men who have sex with men, in which FTC-TDF PrEP was associated with a small but statistically significant decrease in calculated creatinine clearance. However, an important limitation of that analysis was that PrEP adherence, based on detection of tenofovir in plasma, was estimated to be only ~50% in iPrEx.

The results should be interpreted in light of the following limitations. First, creatinine-based GFR estimating equations are less accurate in persons with low creatinine generation, including those with low muscle mass, muscle wasting, or reduced meat intake, which may be more common in African individuals. The CKD-EPI equation has demonstrated high accuracy in African populations, and intra-individual changes in eGFR are less susceptible to this limitation of creatinine-based estimates. Second, long-term treatment effects beyond the study period cannot be ascertained. However, it is reassuring that in a large observational study with long-term follow-up (median: 7.9 years) of HIV-1 infected individuals on TDF-containing combination antiretroviral therapy, most of the observed eGFR loss occurred during the first year of TDF exposure and stabilized after 2 years⁷. In our study, mean eGFR decline appeared to stabilize after the first year of observation and then waned over time. Third, post-randomization censoring has the potential to introduce selection bias and/or confounding. However, the consistency of the primary analysis estimates with marginal structural models estimates lends confidence to our findings. Fourth, against a low background level of a ≥25% eGFR decline (i.e. 0.83% per year recorded in the placebo group) we had only the ability to detect large increases in the risk of ≥25% eGFR decline. However, the low absolute rates of ≥25% eGFR decline recorded in the active arms with additional follow-up (median of 36 months in the active PrEP arms) is encouraging. Fifth, the study required persons with normal renal function at entry, and PrEP effects among subpopulations with co-morbidities or concurrent nephrotoxic medications could not be fully evaluated. Lastly, the current study did not evaluate changes in proximal tubular function, another potential consequence of TDF exposure. A recent sub-study in iPrEx did not show evidence of nephrotubulopathy,⁴⁰ and we observed no significant difference in graded abnormalities in serum phosphorus between the PrEP and placebo.¹ Whether TDF-based PrEP causes early proximal tubular injury in HIV-1 uninfected individuals warrants additional evaluation.

In conclusion, in this large randomized placebo-controlled trial among uninfected African men and women, with median follow-up of 18 months and maximum follow-up of 36 months, daily oral TDF-based PrEP was associated with a small but non-progressive decline in eGFR that was not accompanied by a substantial increase in the risk of clinically relevant eGFR decline. Our data support the safety of TDF-based PrEP in heterosexual populations as part of a comprehensive HIV-1 prevention package.

Partners PrEP Study Team

University of Washington Coordinating Center and Central Laboratories, Seattle, USA

Connie Celum (principal investigator, protocol co-chair), Jared M. Baeten (medical director, protocol co-chair), Deborah Donnell (protocol statistician), Robert W. Coombs, Lisa Frenkel, Craig W. Hendrix, Jairam R. Lingappa, M. Juliana McElrath.

Study sites and site principal investigators

Eldoret, Kenya (Moi University, Indiana University): Kenneth H. Fife, Edwin Were; Kabwohe, Uganda (Kabwohe Clinical Research Center): Elioda Tumwesigye; Jinja, Uganda (Makerere University, University of Washington): Patrick Ndase, Elly Katabira; Kampala, Uganda (Makerere University): Elly Katabira, Allan Ronald; Kisumu, Kenya (Kenya Medical Research Institute, University of California San Francisco): Elizabeth Bukusi, Craig R. Cohen; Mbale, Uganda (The AIDS Support Organization, CDC-Uganda): Jonathan Wangisi, James D. Campbell, Jordan W. Tappero; Nairobi, Kenya (University of Nairobi, University of Washington): James Kiarie, Carey Farquhar, Grace John-Stewart; Thika, Kenya (University of Nairobi, University of Washington): Nelly R. Mugo; Tororo, Uganda (CDC-Uganda, The AIDS Support Organization): James D. Campbell, Jordan W. Tappero, Jonathan Wangisi. Data management was provided by DF/Net Research, Inc. (Seattle, USA) and site laboratory oversight was provided by Contract Laboratory Services (CLS) of the Wits Health Consortium (University of the Witwatersrand, Johannesburg, South Africa).