Baseline characteristics explain differences in effectiveness of randomization to daily oral TDF/FTC PrEP between transgender women and cisgender men who have sex with men in the iPrEx trial

Introduction

Daily oral tenofovir disproxil fumarate/emtracitabine (TDF/FTC) for HIV pre-exposure prophylaxis (PrEP) nearly eliminates the risk of HIV infection when taken consistently.^1–3 PrEP could be particularly impactful for transgender women-- a key population carrying one of the highest HIV burdens globally.⁴ However, PrEP uptake in this population has been low, and this may be in part due to a lack of high quality evidence about PrEP in transgender women.⁵

iPrEx was the only placebo-controlled randomized study of daily oral PrEP that included any transgender women who have sex with men (TGW), and consequently, the trial’s results play an outsized role in our understanding of PrEP’s efficacy in TGW.⁶ Though randomization to the active arm reduced HIV incidence by 44% in the sample overall, stratified analyses found no benefit for TGW (hazard ratio 1.1, 95%CI [0.5, 2.7]).⁷ A closer look at drug levels found that tenofovir concentrations were generally lower in TGW compared to cisgender men who have sex with men (MSM), and drug was not detected at the seroconversion visit in any TGW on the active arm who became HIV positive.⁷

There are at least two possible explanations for the iPrEx results. First, there were numerous measured baseline differences between TGW and MSM. If these differences occurred across characteristics that were important modifiers of PrEP’s effectiveness—either by affecting adherence to PrEP or by modifying HIV risk—then even in the absence of any biological differences in TDF/FTC’s efficacy, the intention-to-treat (ITT) estimates of PrEP’s effectiveness might differ between the two groups.⁸ Second, there may be other unknown or unmeasured differences between TGW and MSM that might impact TDF/FTC’s effectiveness. For example, recent small pharmacological studies suggest that feminizing hormones might interfere with the ability of tenofovir to block HIV infection by lowering the available blood concentration of tenofovir diphosphate.^9,10 Understanding why randomization to PrEP was not effective in TGW in iPrEx may have useful implications for PrEP implementation.

Here, we assess to what extent differences in measured baseline characteristics between MSM and TGW could explain the observed effect heterogeneity in iPrEx. We frame this issue as a transportability¹¹ question, and estimate what the ITT effect of randomization to PrEP would have been in MSM if they had the same distribution of baseline characteristics as TGW in the study.^12,13 If this transported estimate is similar to the observed ITT estimate in TGW, then the effect heterogeneity observed in iPrEx might be due to measured population composition differences alone. If, on the other hand, the transported estimate is not similar to what was observed in iPrEx, then unique contextual or biological factors (or unmeasured differences in population composition) contributed to the effect heterogeneity in the study.

Methods

Study population and procedures

iPrEx was a placebo-controlled randomized trial of daily oral TDF/FTC PrEP conducted between 2007 and 2011 in Brazil, Peru, Ecuador, the United States, South Africa, and Thailand. iPrEx enrolled 2499 cisgender men and transgender women who have sex with men.⁶ All participants were HIV-negative at enrollment, reported risk behavior for HIV, and were assigned male sex at birth. Gender identity was recorded via a computer assisted structured interview (CASI). We additionally included any participant who reported taking feminizing hormones (irrespective of gender identity) as a TGW.⁷

The same baseline CASI also asked participants about their demographics, living situation, relationship status, recent and lifetime sexual history, and substance use. Depressive symptoms were measured via an interviewer-administered Center for Epidemiologic Studies Depression Scale (CES-D). Detailed study procedures can be found in Grant et al, 2010.⁶

Variable selection and statistical methods

We first estimated the observed intention to treat incidence rate ratio in MSM $\left(IR{R}_{msm}\right)$ and TGW $\left(IR{R}_{tgw}\right)$ using a Poisson regression including an offset for follow-up time. We excluded individuals who were HIV positive at enrollment or who contributed no follow-up time.

We estimated what the incidence rate ratio would have been in MSM had they shared the same baseline characteristics as the TGW study participants $\left({\widetilde{IRR}}_{msm}\right)$. We identified 15 candidate baseline characteristics that we hypothesized were both associated with HIV incidence and differed between MSM and TGW in iPrEx: age; total number of partners in the prior 3 months; any condomless receptive anal sex in the prior 3 months; sexual role (top, bottom, or versatile); race; ethnicity (Hispanic/Latino or non-Hispanic/Latino); country of residence; highest level of education; marital status; living situation (“With whom do you live primarily?”); past month alcohol consumption; history of transactional sex in the past 6 months; any sexually transmitted infections in the past 6 months; past month cocaine use; and past week depressive symptoms. Using a data-driven variable selection algorithm, we narrowed this list of 15 candidate covariates to include only those that both modified the ITT incidence rate ratio among MSM and differed in distribution between MSM and TGW.¹⁴

Using this reduced set of variables $\left({\mathit{W}}^{*}\right)$, we applied a generalization of the g-formula ¹⁵ to estimate ${\widetilde{IRR}}_{msm}$.^15,16 This approach is analogous to model-based direct standardization in which the MSM population is standardized to resemble the distribution of covariates observed in TGW.¹⁷ Assuming correct model specification, ${\widetilde{IRR}}_{msm}$ estimates what the ITT incidence rate ratio would have been in MSM had they shared the same distribution of baseline covariates as TGW in iPrEx. We estimated the percent of the observed effect heterogeneity between MSM and TGW that can be accounted for by differences in these baseline characteristics as$\left(\frac{\text{log}\left(IR{R}_{msm}\right)-\text{log}\left({\widetilde{IRR}}_{msm}\right)}{\text{log}\left(IR{R}_{msm}\right)-\log (IR{R}_{tgw})}*100\right)$. Analyses were performed using R v3.4.1 and STATA 15.1.^18,19

Results

Of the 2499 participants enrolled in iPrEx, 10 were HIV positive at enrollment and 44 did not return for follow-up visits. Of the remaining 2445 participants, 290 identified as trans, 29 identified as women, and 14 identified as men but reported using feminizing hormones. Together, these participants comprised the TGW group (N=333/2445 (14%)). 67 (20%) of the 333 TGW participants reported using feminizing hormones.⁷

Table 1 compares the 15 candidate baseline characteristics between MSM and TGW. The variable selection algorithm identified 6 of these 15 baseline characteristics as being necessary and sufficient for transporting the incidence rate ratio: CES-D score; number of partners in the prior 3 months; any condomless receptive anal intercourse in the prior 3 months; living situation; any history of transactional sex in the prior 6 months; and any STI diagnoses in the prior 6 months.

Table 1.

Baseline characteristics by gender

		TGW (N=333)	MSM (N=2112)	p-value
Age at baseline, mean (SD)		26 (7)	27 (9)	0.030
CESD Score, mean (SD)		17 (8)	17 (8)	0.63
Number of partners in prior 3 months, median (IQR)		15 (5, 55)	5 (3, 13)	<0.001
Any condomless receptive anal intercourse in the prior 3 months		286 (86%)^	1172 (55%)	<0.001
Country	US	6 (2%)	217 (10%)	<0.001
	Peru	184 (55%)	1192 (56%)
	Ecuador	60 (18%)	228 (11%)
	Brazil	37 (11%)	327 (15%)
	South Africa	4 (1%)	77 (4%)
	Thailand	42 (13%)	71 (3%)
Treatment assignment	Placebo	165 (50%)	1056 (50%)	0.88
	Active Arm	168 (50%)	1056 (50%)
Ethnicity	Non Hispanic/Latino	84 (25%)	597 (28%)	0.25
	Hispanic/Latino	249 (75%)	1515 (72%)
Race	Black/African American	19 (6%)	186 (9%)	<0.001
	White	38 (11%)	386 (18%)
	Mixed/Other	234 (70%)	1452 (69%)
	Asian	42 (13%)	88 (4%)
Marital Status	Single	237 (71%)	1594 (75%)	0.005
	w/Partner	95 (29%)	455 (22%)
	Married	0 (0%)	33 (2%)
	Divorced	1 (<1%)	28 (1%)
	Widowed	0 (0%)	2 (<1%)
Living Situation	With family/friends	226 (68%)	1628 (77%)	<0.001
	w/ Male partner	26 (8%)	120 (6%)
	Alone	75 (23%)	299 (14%)
	w/ Female partner	1 (<1%)	30 (1%)
	other	5 (2%)	35 (2%)
Education Level	Less than Secondary	125 (38%)	385 (18%)	<0.001
	Completed Secondary	122 (37%)	744 (35%)
	Post-Secondary	84 (25%)	960 (45%)
	No Answer/Missing	2 (1%)	23 (1%)
Sexual Role	Top	14 (4%)	609 (29%)	<0.001
	Bottom	238 (71%)	587 (28%)
	Versatile	75 (23%)	858 (41%)
	Don’t know	6 (2%)	58 (3%)
Any transactional sex in prior 6 months		214 (64%)	790 (37%)	<0.001
Any STI diagnosis in prior 6 months		126 (38%)	515 (24%)	<0.001
Alcoholic drinks per day in the past month	None/< once a month	63 (19%)	427 (20%)	0.008
	1–4 per day	67 (20%)	557 (26%)
	>=5 per day	150 (45%)	756 (36%)
	Refused/Missing/Don’t know	53 (16%)	372 (18%)
Any cocaine use in the past month		25 (8%)	105 (5%)	0.055

^{^}All variables are N (%) except where noted

In MSM, there were 77 incident HIV infections in the placebo arm and 41 infections in the active arm; in TGW, there were 10 infections in the placebo arm and 13 in the active arm. The ITT incidence rate ratio in MSM $\left(IR{R}_{msm}\right)$ was 0.53 (95%CI [0.36, 0.77]), and in TGW the $IR{R}_{tgw}$ was 1.29 (95%CI [0.24, 2.35]). After standardizing the MSM population according to the 6 selected baseline characteristics, the transported incidence rate ratio $\left({\widetilde{IRR}}_{msm}\right)$ was 1.28 (95%CI [0.12, 40.04]). This corresponds to nearly complete (99%) reduction in the observed effect heterogeneity. Overall, after accounting for baseline characteristics, the transported ITT incidence rate ratio closely resembles what was observed in TGW in iPrEx.

Discussion

Differences in baseline characteristics between MSM and TGW explained the observed effect heterogeneity in iPrEx. This finding suggests that biological differences in TDF/FTC’s efficacy in TGW or other unmeasured factors were unlikely to have been major drivers of the effect heterogeneity observed in the iPrEx trial.

Whether feminizing hormones reduce the absorption of tenofovir diphosphate enough to produce clinical differences in PrEP’s efficacy remains an important open question. Only 20% of TGW in iPrEx reported taking feminizing hormones, and there were no HIV infections among the participants taking feminizing hormones who were assigned to the placebo arm. Consequently, we cannot evaluate whether feminizing hormones reduced PrEP’s effectiveness using the iPrEx study data. Ongoing studies designed to explicitly address this question will soon provide more insight into the interaction between hormones and PrEP.

The small number of TGW included in iPrEx is a major obstacle for understanding PrEP in this key population. By using transportability, we were able to better describe the effect heterogeneity in iPrEx after accounting for numerous differences between TGW and MSM, which would have been impossible with traditional regression adjustment. Given that iPrEx is the only placebo-controlled randomized trial of PrEP that included any TGW, any insights about the effects of PrEP in this population are helpful even if substantial uncertainty remains.

Moving forward, there remains an urgent need for high-quality trans-specific research on HIV prevention strategies.²⁷ The effect heterogeneity in iPrEx exemplifies why transgender women should not be aggregated with cisgender men when conducting research, and future studies should ensure that enough transgender women are included to provide adequate power to analyze these groups separately.²⁸ Further research is also needed on PrEP for transgender men or non-binary individuals to ensure that implementation programs meet the needs of everyone who could benefit from PrEP.

Overall, our study--along with others from iPrEx and iPrEx OLE-- suggests TDF/FTC PrEP works similarly for MSM and TGW when accounting for other characteristics. PrEP should be offered to anyone at risk of HIV infection regardless of gender identity.⁷