Efficacy estimates of oral pre-exposure prophylaxis for HIV prevention in cisgender women with partial adherence

Abstract

Pre-exposure prophylaxis (PrEP) with daily oral tenofovir disoproxil fumarate and emtricitabine (TDF/FTC) is effective for HIV prevention in both men and women with sufficient adherence, however the adherence-efficacy relationship for cisgender women has not been well established. We calculated the adherence-efficacy curve for cisgender women using HIV incidence and plasma tenofovir concentrations (TFV) from three trials (FEMPrEP, VOICE, and Partners PrEP). We imputed tenofovir diphosphate (TFV-DP) concentrations, a measure of long term adherence, from TFV quantification using data from HIV Prevention Trials Network 082, which measured both TFV-DP and TFV. Two, four, or seven pills per week reduced HIV incidence by 59.3% (95% CI: 29.9% - 95.8%), 83.8% (95% CI: 51.7% - 99.8%), and 95.9% (95% CI: 72.6% - 100%), respectively. Our adherence-efficacy curve can be validated and updated by HIV prevention studies measuring TFV-DP directly. It suggests high adherence confers high protection in cisgender women. However, the lower efficacy with partial adherence highlights the need for new PrEP products and interventions to increase adherence.

Editor summary:

Pre-exposure prophylaxis with daily antiretrovirals to prevent HIV-1 acquisition has not proved consistently effective in cisgender women. Modeling adherence to daily PrEP clarifies how robust protection can be achieved.

Introduction

Pre-exposure prophylaxis (PrEP) with daily oral tenofovir disoproxil fumarate and emtricitabine (TDF/FTC) is safe and highly effective for HIV prevention.^1–4 However, some studies of TDF/FTC in cisgender women have shown little or no efficacy, attributed to low product adherence.^5,6 Achieving high protection requires regular pill taking during periods of exposure. Adherence is typically quantified in PrEP users with measurements of intraerythrocytic tenofovir diphosphate (TFV-DP) or plasma tenofovir (TFV) concentrations.^7,8 TFV-DP has a half-life of 17 days when measured in red blood cells via dried blood spots (DBS) and 4–5 days when measured in peripheral blood mononuclear cells (PBMC). TFV-DP can be used to estimate average pill taking over the previous 1–2 months. By contrast, TFV has a relatively short half-life of around 3 days, and therefore its quantification can indicate whether a dose was taken within the last week but is not informative of adherence prior to the last dose.

In men and transgender women who have sex with men (MSM/TGW), the relationship between PrEP adherence and efficacy has been defined by comparing TFV-DP concentrations in individuals newly diagnosed with HIV and those without HIV to those with known adherence to TDF/FTC PrEP from directly observed dosing.^7,9 These methods suggest that partial adherence leads to reductions in HIV incidence of 76% (95% confidence interval (CI): 56% - 96%), 96% (95% CI: 90% - >99%), and 99% (95% CI: 96% - >99%) with two, four, and seven pills per week when TFV-DP was measured in PBMCs, respectively, and 84% (95% CI: 12% - 99%) with two to three pills and >99% (95% CI: 75% - >99%) with four to six pills per week when measured in DBS.^7,9 Repeating this analysis in cisgender women has been complicated by the lack of large placebo-controlled efficacy trials measuring TFV-DP. Recent analysis using data from HIV Prevention Trials Network (HPTN) 084, which used TDF/FTC as an active control to assess efficacy of cabotegravir for long-acting PrEP, showed a reduction in HIV incidence of 80% (95% CI: 32% –97%), 88% (95%: CI 43% - 99%), and 99% (95%: CI 0% - 99%) in those taking two to three, four to six, and seven weekly pills, respectively, relative to those with no quantifiable TFV-DP, however there was no placebo control arm in that study so these estimates may be confounded by different rates of HIV exposure between adherence and non-adherent participants.^10,11

Randomized clinical trials (RCTs) of PrEP enrolling cisgender women have had mixed results, due to variable adherence in trial cohorts. The Partners PrEP and Botswana TDF2 studies found 66% (95% CI: 28% - 84%) and 49.4% (95% CI: −21.1% - 80.1%) PrEP efficacy among women, respectively.^3,4 In Partners PrEP, quantifiable plasma TFV was associated with a 94% (95% CI: 41% - 99%) reduction in HIV incidence in women.¹² By contrast in FEMPrEP, HIV prevention with TDF/FTC showed only 6% efficacy (95% CI: −52% - 41%) although this improved to 18% (95% CI: −36% to 51%) when excluding follow-up time where PrEP was unavailable to the participant either due to a missed visit or PrEP discontinuation.⁵ The VOICE trial reported a similarly low efficacy for TDF/FTC of −4.4% (95% CI: −49% - 27%).⁶ Adherence, measured by plasma TFV, was low in both the FEMPrEP and VOICE studies; only 21% of those newly diagnosed with HIV in FEMPrEP had quantifiable TFV at the first visit following acquisition and only 37% of those newly diagnosed with HIV in the oral arm of VOICE had quantifiable TFV at any visit. Participants without HIV had quantifiable plasma TFV in 35% and 52% of samples in FEMPrEP and VOICE, respectively, although the difference in concentration between cases and controls was not significant in either study.

Estimates of the association between PrEP adherence and PrEP efficacy are vital for informed PrEP use and projecting its population impact. Given the lack of data on the relationship between product adherence and efficacy in cisgender women (henceforth women), researchers either projected impact assuming comparable overall adherence and effectiveness to that observed in Partners PrEP and TDF2,^13–17 assumed “all-or-nothing” adherence and an efficacy based on the protection afforded in those with quantifiable plasma TFV,^18–21 or a combination of both²². The former approach does not account for the variability seen in PrEP adherence and persistence among different populations, while the latter may not properly account for the partial protection afforded by imperfect adherence.

We combined HIV incidence and plasma TFV concentrations from three efficacy studies enrolling women: FEMPrEP, VOICE, and Partners PrEP with pharmacological data (TFV and TFV-DP) from another study in cisgender women (HPTN 082) to define the relationship between TFV-DP daily oral adherence and efficacy in reducing HIV incidence in cis-gender women in sub-Saharan Africa.²³

Data and adherence-efficacy curve calibration

We calibrated an adherence-efficacy curve mapping TFV-DP concentration in DBS to the reduction in HIV incidence relative to placebo in cisgender women (Fig. 1). We used data from three randomized placebo-controlled efficacy studies of TDF/FTC in adult cisgender women in sub-Saharan Africa (FEMPrEP, VOICE, and Partners PrEP). The studies did not measure TFV-DP concentrations but did assess adherence via plasma TFV quantification. We imputed TFV-DP values in the three efficacy studies using data from HPTN 082 (a TDF/FTC implementation study in African cisgender women which measured both plasma TFV and TFV-DP). We fit the adherence-efficacy curve to the number of new HIV diagnoses with and without quantifiable plasma TFV in the active arms of the three studies, accounting for the difference in placebo incidence in each study (see Supplementary Fig. S1). As a sensitivity analysis, we considered three functional forms of the adherence-efficacy curve each making different assumptions. We also considered models that predicted active arm HIV incidence directly from plasma TFV without imputation, but these models performed slightly worse by deviance information criterion (Supplementary Table S1). For more details, please see the Methods.

Figure 1: TDF/FTC efficacy estimation flowchart:

Our methodology relies on data from three randomized clinical trials (RCTs): FEMPREP, VOICE and Partners PrEP, as well as TFV-DP measurements from HPTN 082. For each RCT, we first impute TFV-DP measurements using the HPTN 082 data, then compute TDF/FTC efficacy using the imputed values. Next, we combine the computed efficacy with the placebo arm HIV incidence from each RCT to predict the HIV incidence stratified by tenofovir detectability. Finally, this HIV incidence is compared to the observed incidence to estimate a likelihood of the observation. Using this likelihood function, we calibrate the model via Markov Chain Monte Carlo

PrEP efficacy by TFV-DP concentration

We next estimated the reduction in HIV incidence imparted by TDF/FTC as a function of TFV-DP DBS concentration measured in fmol per punch, the standardized paper disc used in DBS assays. In our base model we assumed that 1) HIV incidence among those not adherent to TDF/FTC is the same as in the placebo arm and that 2) HIV incidence approaches zero with increasing TDF/FTC use (Fig. 2). Under these assumptions we estimated that DBS TFV-DP concentrations of 350, 700, and 1250 fmol/punch were associated with 53.8% (95% CI: 26.1% - 94.1%), 78.7% (95% CI: 45.3% - 99.7%), and 93.7% (95% CI: 66.0% - 100.0%) reductions in HIV incidence.

Figure 2: Efficacy of TDF/FTC by TFV-DP measurement.

Reduction in incidence in HIV as a function of tenofovir diphosphate (TFV-DP) levels in dried blood spots measurement according to model ℎ₁. Grey ribbon = IQR, Dashed line = 95% Credible Interval. Colored rectangles represent the IQR of TFV-DP measurements associated with two pills per week (red), four pills per week (blue) and seven pills per week (green) derived from directly observed dosing.⁸

These values were robust to the model assumptions (Extended Data Table 1). Using our first alternative model, which allowed HIV incidence among those with no adherence to TDF/FTC to differ from the placebo group (Extended Data Fig. 1), we estimated that DBS concentrations of 350, 700, and 1250 fmol/punch were associated with 48.4% (95% CI: 19.4% - 91.5%), 75.7% (95% CI: 44.8% - 99.5%), and 92.7% (95% CI: 66.6% - 100.0%) reductions in HIV incidence. Using our second alternative model, which allowed maximum efficacy to be less than 100% (Extended Data Fig. 2), we estimated that DBS TFV-DP concentrations of 350, 700, and 1250 fmol/punch were associated with 52.4% (95% CI: 28.5% - 83.8%), 76.8% (95% CI: 48.6% - 97.1%), and 92.0% (95% CI: 67.7% - 99.8%) reductions in HIV incidence.

PrEP efficacy by pill taking frequency

Finally, we estimated the reduction in HIV infection associated with specific weekly pill-taking frequencies. We sampled across the range of TFV-DP DBS measures by dose as established by directly observed dosing.⁸ Assuming a log-normal distribution, for women of African descent the TFV-DP DBS ranges were 407 (IQR 344–481), 840 (IQR 710–993), and 1507 (IQR 1275 – 1782) fmol/punch for two, four, and seven pills per week, respectively (Fig. 2 colored rectangles). Reduction in HIV incidence with each dose, using all available data from the efficacy trials, is estimated by averaging the adherence-efficacy curve across the associated TFV-DP range estimated from directly observed dosing.

Using the base model, as defined in the previous section, we estimate two, four, and seven pills per week were associated with 59.3% (95% CI: 29.9% - 95.8%), 83.8% (95% CI: 51.7% - 99.8%), and 95.9% (95% CI: 72.6% - 100%) reductions in HIV incidence. These estimates were again robust to the model structure (Extended Data Table 1). Using our first alternative model (ℎ₂), we estimate two, four, and seven pills per week were associated with 54.3% (95% CI: 25.3 – 93.8), 81.6% (95% CI: 52 – 99.7), and 95.3% (95% CI: 72.8 – 100) reductions in HIV incidence (Extended Data Fig. 1). Using our second alternative (ℎ₃), we estimate two, four, and seven pills per week were associated with 57.8% (95% CI: 32.6% - 87.3%), 82% (95% CI: 54.9% - 98.2%), and 94.5% (95% CI: 72.7% - 99.9%) reductions in HIV incidence (Extended Data Fig. 2).

Discussion

Oral PrEP with TDF/FTC reduces HIV incidence in both MSM/TGW and women who have high adherence. The relationship between pills taken per week and PrEP efficacy has been quantified for MSM/TGW using TFV-DP in DBS as an objective surrogate measure for PrEP adherence. Recently, data has emerged on the relationship between adherence and efficacy among women. We calibrated an adherence-efficacy curve in women by combining 1) incidence and plasma TFV quantification data from three placebo-controlled efficacy trials in women with 2) pharmacological data from HPTN 082 on the relationship between plasma TFV and TFV-DP in DBS. The framework allows for easy updates when new evidence becomes available.

We estimated that women with TFV-DP levels associated with two, four, or seven pills per week had 59%, 84% and 96% lower HIV incidence, respectively, compared to those not taking PrEP. This reduction is only slightly lower than that estimated in MSM/TGW at the same level of adherence,^7,9 suggesting lower observed efficacy in PrEP trials in women is primarily due to adherence. This finding is consistent with sub-group analyses in Partners PrEP and HPTN 084 that found high effectiveness in women with quantifiable plasma TFV and TFV-DP concentrations consistent with two or more pills per week, respectively,^11,12 and with a recent meta-analysis that found high levels of protection in women consistently taking four to six pills per week in PrEP demonstration projects.²⁴

On an individual level, the number of pills taken per week does not uniquely determine the reduction in HIV incidence conferred. The efficacy of TDF/FTC relies upon the accumulation the active metabolites tenofovir-diphosphate (TFV-DP) and emtricitabine-triphosphate (FTC-TP) into HIV target cells systemically and in mucosal tissues, for exposures via sexual contact, or in the blood, for exposures via injection drug use.^25–27 Pharmacokinetic modeling of TDF/FTC predicts that active metabolites reach protective levels in colorectal tissues and peripheral blood with only two weekly doses, whereas three may be required to reach protective levels in the female genital tract. Therefore, both weekly adherence and the route of exposure likely play a significant role in the efficacy of TDF/FTC. Consistent with our findings, these models predict greater forgiveness for missed doses in MSM/TGW, where exposure is likely to be in colorectal tissues, than in cisgender women, where exposure is likely to be in the female genital tract.^27,28 Currently, on-demand “2-1-1” PrEP consisting of two pills prior to and two pills following sex is not recommended for cisgender women due to a longer interval before detection and lower concentrations in the female genital tract although there have been no “2-1-1” efficacy trials in cisgender women.^27,29 Although our results indicate that four weekly pills can provide upwards of 80% reduction in HIV incidence, we did not investigate the impact of pill timing relative to sexual acts. Women taking on-demand PrEP will likely have better protection than predicted by our curve as effective use only requires high adherence during periods of exposure, so individuals with infrequent exposure may be highly protected despite low TFV-DP measures cumulative adherence measures. Pharmacokinetic modeling will be crucial for projecting efficacy of TDF/FTC in non-daily PrEP regimens and in understudied populations such as heterosexual men and people who inject drugs.²⁸

Our adherence-efficacy curve will aid the evaluation of new PrEP products. Given the high effectiveness of daily oral PrEP, clinical trials of daily oral tenofovir alafenamide fumarate (TAF), long-acting cabotegravir (CAB-LA), and long-acting lenacapavir (LEN), are evaluated in superiority and/or non-inferiority trials against an active control regimen of TDF/FTC. The preventive efficacy of these new products can only be derived from counterfactual estimates of placebo incidence.³⁰ Such counterfactual placebos often rely on baseline estimates of recent infection in the enrolled population,³¹ markers of HIV exposure such as STIs,³² or propensity scores.³³ For MSM/TGW, counterfactual estimates can also be constructed from the active placebo arm using the adherence-efficacy relationship;³⁴ our work will allow that method to be extended to women’s studies. In brief 1) incidence in the TDF/FTC arm should be the product of the counterfactual placebo arm and the hazard ratio (1 – efficacy) due to product use 2) this hazard ratio can be estimated using the TFV-DP measures from a representative sample of individuals in the TDF/FTC arm, 3) counterfactual placebo incidence can then be estimated by dividing the incidence in the TDF/FTC arm by the calculated hazard ratio.

Our adherence-efficacy curve can also be incorporated into projections of PrEP impact. Mathematical models of PrEP-based initiatives aimed at reducing HIV incidence rely upon assumptions about efficacy in the target population^13–22. In both controlled trials and demonstration projects, cisgender women who take PrEP often have only partial adherence^11,23,24. Our curve will allow for future modeling studies of PrEP to 1) account for HIV prevention among cisgender women with partial adherence and 2) estimate the impact of interventions intended to increase PrEP adherence in this population.

This work has two major limitations. First, the efficacy trials in women did not collect TVF-DP DBS measurements from participants, so we relied on data from HPTN 082 to impute TFV-DP measurements from plasma TFV quantification. This assumes the same distribution of TFV-DP conditional on plasma tenofovir. Pill-taking patterns may differ across populations due to geographic, temporal, or demographic factors. In this case the population from HPTN 082 is comparable to the populations in the efficacy studies with regards to region and time but is younger. Second, there is evidence that trial participants may adjust TDF/FTC usage depending on HIV exposure;^35,36 however, we assumed uniform exposure at all levels of TDF/FTC usage. Incorporating self-reported behavioral data from study participants could identify and correct this potential bias.

Increased PrEP uptake is an important part of ending the HIV epidemic and will likely depend on the availability of multiple PrEP options.³⁷ Providing the best PrEP option for those in high incidence populations requires knowledge of HIV preventive efficacy at different levels of adherence. Building on previous work that showed the high efficacy of oral PrEP among women with quantifiable plasma TFV,¹² we developed an adherence-efficacy curve based on TFV-DP, to better quantify HIV prevention benefits for women at all levels of TDF/FTC adherence. Our results indicate that the relationship between PrEP adherence and efficacy in MSM/TGW and cisgender women may be more similar than has been appreciated.

Online Methods

Preventive efficacy TFV-DP Models

We assumed that the protection afforded by TDF/FTC against HIV acquisition could be predicted based on the TFV-DP DBS measure, $X_D$, via a TFV-DP model $f\left(X_D\right)$. Following previous work, we assumed that $f$ is an exponentially decaying function and considered three different models relating HIV incidence reduction to DBS concentrations (Extended Data Fig. 3).⁷ Our base model, $f_1$, assumed that individuals with no quantifiable TFV-DP have the same HIV incidence as individuals from the trial population concurrently randomized to placebo and that HIV incidence approaches zero with increasing TFV-DP drug concentrations. The second model, $f_2$, allowed individuals with low concentrations of TFV-DP to have greater or lower HIV incidence than individuals on placebo, reflecting findings from prior studies. For example, MSM/TGW in the iPrEx open label extension with the lowest PrEP adherence had higher HIV incidence than those not taking PrEP,⁷ but women have been found to decrease pill taking during periods of low potential exposure.^35,36 The third model, $f_3$, allowed for the possibility that even high PrEP adherence may not lead to complete protection, so efficacy approaches $k_3$ with full adherence.

$$\begin{array}{rr}f\left(X_D,\mathbf{k}\right)& \equiv \frac{\text{Hazard}\text{rate}\text{in}\text{individuals}\text{on}\text{PrEP}\text{with}\text{TFV-DP}\text{of}{X}_D}{\text{Hazard}\text{rate}\text{off-PrEP}}\hfill \\ f_1\left(X_D\mid k_1\right)& =\mathrm{e}\mathrm{x}\mathrm{p}\left(-X_D\mathrm{l}\mathrm{n}2/k_1\right)\hfill \\ f_2\left(X_D\mid k_1,k_2\right)& =\mathrm{e}\mathrm{x}\mathrm{p}\left(-X_D\mathrm{l}\mathrm{n}2/k_1\right)k_2\hfill \\ f_3\left(X_D\mid k_1,k_3\right)& =\mathrm{e}\mathrm{x}\mathrm{p}\left(-X_D\mathrm{l}\mathrm{n}2/k_1\right)\left(1-k_3\right)+k_3\hfill \end{array}$$

The values $k_1$, $k_2$, and $k_3$ are free parameters which need to be estimated. For each model, the parameter $k_1$ represents the IC50, i.e., the level of TFV-DP required to achieve half the maximum theoretical reduction in HIV incidence. That is

$$f(0)-f\left(k_1\right)=\left(f(0)-\underset{X_D\to \mathrm{\infty }}{\mathrm{l}\mathrm{i}\mathrm{m}}\left(f\left(X_D\right)\right)\right)/2$$

Imputation of TFV-DP values from plasma detectability

The three randomized, placebo-controlled PrEP efficacy trials in women — FEMPrEP (Clinicaltrials.gov identifier: NCT00625404), Partners PrEP (NCT00557245), and VOICE (NCT00705679) — measured product adherence via detection of plasma TFV, which reflects pill-taking within the two weeks. Each trial dichotomized plasma TFV concentrations according to a different quantification threshold (Extended Data Table 2) ranging from >0.3 ng/ml at any visit to >10 ng/ml at a given visit. In contrast, HPTN 082 (NCT02732730) longitudinally measured both plasma TFV as well as intraerythrocytic TFV-DP in DBS but was not placebo-controlled.²³ To determine HIV incidence with increasing TFV-DP concentrations we used data from the three placebo-controlled efficacy trials and applied our model by imputing efficacy trial participants’ TFV-DP measures based on data from HPTN 082. For each trial we used its TFV quantification threshold to categorize HPTN 082 participants into positive or negative quantification groups (Extended Data Fig. 4). We then sampled from these groups to impute TFV-DP levels for the three placebo-controlled trial participants. We denote the distribution of TFV-DP measures associated with plasma quantification $P\in \{+,-\}$ and trial $T$ as ${\Delta }_{PT}$ (see Extended Data Fig. 4).

Combining the above imputation with the TFV-DP models ($f_1$, $f_2$, and $\left.f_3\right)$ generated three quantification models of PrEP efficacy as a function of plasma quantification $(P\in \{+,-\left\}\right)$ and the trial $\left(T\right):h_1,h_2$ and $h_3$. None of the three randomized placebo-controlled trials that measure TDF/FTC efficacy in women collected TFV-DP DBS measurements. Therefore, the above estimates rely on imputed DBS measurements given plasma TFV quantification from HPTN082. We therefore evaluated whether a more direct approach — estimating PrEP efficacy from plasma TFV quantification alone — might predict PrEP efficacy in trial cohorts better than imputed DBS measures. Specifically, we considered two additional, simplified, quantification models that directly used plasma TFV quantification without imputing a TFV-DP concentration $\left(h_4\right.$ and $\left.h_5\right)$.

$$\begin{array}{rr}h(P\mid T,\mathbf{k})& \equiv \frac{\text{Hazard}\text{rate}\text{in}\text{individual}\text{on}\text{PrEP}\text{with}\text{quantifiability}\text{of}\mathrm{P}}{\text{Hazard}\text{rate}\text{off-PrEP}}\hfill \\ h_i(P\mid T,\mathbf{k})& ~f_i\left(X_D\mid \mathbf{k}\right)i\in \{\mathrm{1,2},3\}{X}_D~{\Delta }_{PT}\hfill \\ h_4(P\mid T,\mathbf{k})& =\left\{\begin{array}{ll}{k}_4& P=+\\ k_5& P=-\end{array}\right.\hfill \\ h_5(P\mid T,\mathbf{k})& =\left\{\begin{array}{ll}{k}_4& P=+\\ 1& P=-\end{array}\right.\hfill \end{array}$$

Calibration of efficacy parameters

Each quantification model was calibrated using data from three randomized placebo-controlled efficacy trials (Extended Data Table 2) using Markov Chain Monte Carlo (MCMC) to estimate the unknown model parameters $\mathbf{k}$. In this procedure, the likelihood of a given set of proposed values for $\tilde{\mathbf{k}}$ and a model $h$ must be computed at each step (see supplementary methods). In total, we ran 100 MCMC chains of 1100 steps each for a total of 110000 parameter sets for each model. Extended Data Table 3 shows the median and 95% Bayesian credible interval (CI) estimated for each model parameter. As these estimates rely on prior assumptions about the joint distribution of TFV-DP and TFV concentrations, our uncertainty intervals for both parameters and model outputs cannot be viewed as confidence intervals but instead Bayesian credible intervals.

We compared the five quantification models — three using imputed DBS and two using plasma TFV quantifiability only — to reproduce the number of HIV diagnosis, stratified by plasma TFV quantification (Supplementary Fig. S1). By deviance information criterion (DIC) which accounts for both the goodness of model fit and model complexity, the three models using imputed DBS outperformed those using plasma TFV only (Supplementary Table S1).

Likelihood estimation

The likelihood was estimated via the following procedure:

1. From the total number of new HIV diagnoses and person-quarters of follow-up time in the placebo arm of each efficacy trial, $S_{0T}$ and $Q_{0T}$, we randomly sampled the probability that an individual will have an HIV diagnosis during a given quarterly visit, $r_T$, from a beta distribution.

   $$r_T~\beta \left(1+S_{0,T},1+Q_{0,T}-S_{0,T}\right)$$
2. From the number of samples with and without quantifiable TFV among those with no HIV diagnosis, $Y_{+,T}$ and $Y_{-,T}$, and the total amount of followup in the active arms of each trial, $Q_{1,T}$, we randomly sampled the total amount of quarterly visits with and without quantifiable TFV.

   $$Q_{+,T}~Q_{1,T}\beta \left(1+Y_{+,T},1+Y_{-,T}\right)$$

   $$Q_{-,T}=Q_{1,T}-Q_{+,T}$$
3. For each trial, we calculated the HIV hazard ratio of individuals with and without quantifiable TFV compared to the placebo group, $h^{(\pm \mid T,\tilde{\mathbf{k}})}$ of HIV. For models $h_1$, $h_2$ and $h_3$, we imputed values of TFV-DP concentration as described above. We drew $Q_{+,T}$ and $Q_{-,T}$ values for follow-up time with and without quantifiable TFV, respectively. The probability of HIV acquisition with and without quantifiable TFV in each trial is then averaged across all follow-up time.

   $$p_{\pm ,T}=\left(\sum _{j=1}^{Q_{\pm ,T}}1-{(1-r_T)}^{h^{\left(j\right)}(\pm |\tilde{\mathbf{k}},T)}\right)/Q_{\pm ,T}$$
   For models $h_4$ and $h_5$, there is no imputation step, so the hazard ratio is completely determined by TFV quantification.

   $$p_{\pm ,T}=1-{\left(1-r_T\right)}^{h(\pm |\tilde{\mathbf{k}},T)}$$
4. The likelihood is computed by comparing the rates, $p_{P,T}$, to the observed number of seroconversions in the active arm of each trial $S_{P,T}$ and amount of follow-up time.

   $$L(\tilde{\mathbf{k}},\text{Data})=\mathrm{e}\mathrm{x}\mathrm{p}\left(\prod _{P,T}\mathrm{P}\mathrm{B}\mathrm{i}\mathrm{n}\mathrm{o}\mathrm{m}\left(S_{P,T};Q_{P,T},p_{P,T}\right)\right)$$

5. Steps 1–4 were repeated 10 times and the final likelihood is the average across each repetition.

Statistics and reproducibility

In this work, we reanalyzed HIV incidence and plasma TFV quantifiability data from three randomized, blinded placebo-controlled clinical trials — FEMPrEP, Partners PrEP, and VOICE — and plasma TFV and TFV-DP concentrations from the implementation study HPTN 082, whose statistical methods have been described elsewhere.^3,5,6. No data were excluded from these analyses. No statistical method was used to predetermine sample size. Sample sizes were instead fixed according to the endpoints of the original studies. Uncertainty intervals were reported as 95% credible intervals, which are calculated as the 2.5% and 97.5% quantiles for each estimand across all MCMC-derived samples. Data calibration was performed in R (version 4.1.2) using the adaptMCMC package (version 1.4).

Extended Data

Extended Data Fig. 1. Efficacy of TDF/FTC by TFV-DP measurement in model $h_2$.

Reduction in incidence in HIV as a function of tenofovir diphosphate (TFV-DP) levels in dried blood spots measurement according to model $h_2$. Grey ribbon = IQR, Dashed line = 95% Credible Interval. Colored rectangles represent the IQR of TFV-DP measurements associated with two pills per week (red), four pills per week (blue) and seven pills per week (green) derived from directly observed dosing.⁸

Extended Data Fig. 2. Efficacy of TDF/FTC by TFV-DP measurement in model $h_3$.

Reduction in incidence in HIV as a function of tenofovir diphosphate (TFV-DP) levels in dried blood spots measurement according to model $h_3$. Grey ribbon = IQR, Dashed line = 95% Credible Interval. Colored rectangles represent the IQR of TFV-DP measurements associated with two pills per week (red), four pills per week (blue) and seven pills per week (green) derived from directly observed dosing.⁸

Extended Data Fig. 3. Adherence-efficacy model curves.

Assumed functional forms for PrEP efficacy as a function of intraerythrocytic tenofovir diphosphate, testing assumptions that individuals with no quantifiable TFV-DP have the same HIV incidence as individuals concurrently randomized to placebo and that HIV incidence approaches zero with increasing TFV-DP. $f_1$ includes both assumptions. $f_2$ challenges the first assumption by allowing individuals with low PrEP adherence to have lower (or higher) efficacy $k_2$ than those on placebo. $f_3$ challenges the second assumption as efficacy approaches $k_3$.

Extended Data Fig. 4. Imputation of TFV-DP concentration from plasma.

TFV quantifiability Measurements of intraerythrocytic TFV-DP among HPTN 082 participants with quantifiable or unquantifiable plasma TFV using the quantification threshold of each trial (see Extended Data Table 2). Plots are based on a total of N = 1083 samples, which are divided into positive and negative depending on the quantification threshold. For FEMPrEP and partners PrEP, the total is only 1081 because for two samples the quantification threshold could not be evaluated due to missing data. Width = frequency, white dot = median, black rectangle = interquartile range, black line = upper and lower adjacent values.

Extended Data Table 1: Estimates of Efficacy.

Estimated reduction in HIV incidence TDF/FTC at select concentrations of TFV-DP concentration and adherence. Table shows median values and 95% credible intervals

	Efficacy estimates in women			Previous results in MSM/TGW
TFV-DP/adherence	h 1	h 2	h 3	Anderson et al9	Grant et al7
350 fmol/punch	53.8% (26.1 – 94.1)	48.4% (19.4 – 91.5%)	52.4% (28.5 – 83.8)		68.0% (17.3 – 87.7)
700 fmol/punch	78.7% (45.3 – 99.7)	75.7% (44.8 – 99.5%)	76.8% (48.6 – 97.1)		93.5% (75.1 – 98.2)
1250 fmol/punch	93.7% (66.0 – 100.0)	92.7% (66.6 – 100.0)	92.0% (67.7 – 99.8)		99.4% (96.0 – 99.9)
Two pills per week	59.3% (29.9 – 95.8)	54.3% (25.3 – 93.8)	57.8% (32.6 – 87.3)	76% (56 – 96)
Four pills per week	83.8% (51.7 – 99.8)	81.6% (52.0 – 99.7)	82% (54.9 – 98.2)	96% (90 – >99)
Seven pills per week	95.9% (72.6 – 100)	95.3% (72.8 – 100)	94.5% (72.7 – 99.9)	99% (96 – >99)	100% (57 – 100)

Extended Data Table 2: Data from three randomized placebo-controlled efficacy trials on daily oral PrEP in women.

In each trial the quantification of plasma TFV was assessed in each newly diagnoses individual as well as a subset of the individuals without HIV in treatment arm. HIV incidence is per 100 person-years. Each trial used a slightly different of plasma quantification threshold.

	Quantifiable plasma TFV			Incidence (per 100 Person-Year)
Trial	HIV diagnosis	No HIV diagnosis	Quantification Threshold	Placebo	Treatment
FEMPrEP	7/33 (21%)	35/95 (37%)	>10 ng/ml at visit	5.0	4.7
Partners PrEP	1/8 (13%)	135/175 (77%)	>0.3 ng/ml at visit	2.8	1.0
VOICE	24/61 (39%)	77/148 (52%)	>0.3 ng/ml at any visit	4.2	4.7

Extended Data Table 3: Model Parameters.

Parameter values with medians and credible intervals as estimated using MCMC. IC50 = TFV-DP DBS measure required to achieve half the maximum theoretical incidence reduction. RR = relative incidence compared to the placebo arm.

Model	Parameter	Description	Median (95%CI)
h 1	k 1	TFV-DP DBS IC50	299 (126 – 801)
h 2	k 1	TFV-DP DBS IC50	314 (86 – 803)
	k 2	RR of those with unquantifiable TFV-DP DBS	1.17 (0.85 – 1.65)
h 3	k 1	TFV-DP DBS IC50	322 (120 – 702)
	k 3	Minimum possible RR	0 (0 – 0.16)
h 4	k 4	RR of those with quantifiable plasma TFV	0.49 (0.32 – 0.74)
	k 5	RR of those with unquantifiable plasma TFV	1.2 (0.87 – 1.65)
h 5	k 4	RR of those with quantifiable plasma TFV	0.47 (0.31 – 0.7)

Data Availability Statement

The data used in this study are available at github.com/FredHutch/PrEPCiswomen.