The paradox of HIV prevention: Did biomedical prevention trials show how effective behavioral prevention can be?

In the past decade, behavioral HIV prevention has labored under the presumption that it is not very effective. This presumption may be justified, but at best only partly so. Our team, with support from the National Institute of Mental Health (R01MH090173), has spent a decade summarizing evidence of the effectiveness of behavioral HIV prevention interventions [1–19]. Clearly the various interventions for HIV prevention do not have the effectiveness expected of a vaccine – the still-elusive single, one-time intervention to prevent HIV – but the presumption that they have no or limited effectiveness is not justified. Undoubtedly these interventions have contributed to the decline of HIV over the past two decades [20]. On the other hand, the promise of biomedical interventions for HIV – whether treatment or prevention – imagined in many randomized controlled trials is rarely fully realized in real-life application. Common thinking about HIV prevention through behavior change may be characterized by unfounded pessimism, while biomedical interventions may be rewarded by unreal optimism.

Our systematic reviews have convinced us of two things: that HIV prevention through behavior change has an effect in containing the further spread of HIV infection, and that abandoning behavioral HIV prevention efforts entirely and replacing them with biomedical interventions alone would be a risky bet.

Our belief in HIV prevention through behavior change has been strengthened by a close examination of the trials of oral pre-exposure prophylaxis, or PrEP. These trials of the effectiveness of tenofovir disproxil fumarate (TDF), with and without emtracitibine (FTC), as a daily oral preventive medication proved effectiveness, with sufficient adherence nearly completely blocking HIV transmission [21].

On careful review, we observed that, in some studies, both the treatment and control groups reduced risk behaviors, changes which cannot be attributed to PrEP alone. We present this evidence and why we think the behavioral interventions in these trials were effective.

We focused our attention on the initial six Phase 3 randomized controlled trials of oral PrEP [22–27] (one assessed topical and oral PrEP). We reviewed the articles that reported primary trial results and all supplementary materials also made available, including study protocols obtained online, as well as subsequent published analyses.

The history and context of the oral PrEP trials

The trials of oral PrEP were born in controversy. Skeptics raised numerous objections. Some of these focused on the biological aspects: that the risk of providing potent drugs with possible side effects to HIV-uninfected people was unacceptable [28], or that using FTC/TDF for prevention could result in drug resistance, compromising its use for treatment [29]. Others focused on the possible behavioral effects: that people might increase risk taking, believing they were protected from HIV [30]. Still others focused on what they perceived as the exploitation of people at risk of HIV in low-income countries, doubting these participants would ever be recipients of the expensive drug if it were found to be effective [31]. While the first two fears have largely proved to be baseless, the last objection does have some merit: many people at risk in the countries where these trials were conducted are still waiting for their chance to use PrEP.

Against this background, initial feasibility and acceptability studies of oral PrEP were undertaken in Cambodia [32], Cameroon, Ghana, and Nigeria [33], primarily among female sex workers. These studies faced intense criticism from community groups in three of these sites [34]. All but the Ghana study were stopped [35]. The results of one initial study showed some promise [33], but Phase 3 effectiveness trials were stalled following this controversy [36]. Once Phase 3 trials did start, it was with intense scrutiny from both an academic and a community perspective.

PrEP trials were eventually initiated among men who have sex with men (largely in Latin America but with participants from Thailand and the United States) [22], serodiscordant couples in Kenya and Uganda [23], high-risk heterosexuals in Botswana [24], people who inject drugs in Thailand [25], and high-risk women in sub-Saharan Africa [26,27]. These trials began with equipoise on the question of PrEP effectiveness, making double-blind placebo-controlled trials necessary. All participants were to receive the standard HIV prevention package available in their countries. In reality, they received much more.

The trials were “event-driven”: investigators calculated the number of incident HIV infections needed to detect a significant difference between the treatment and control arms. They then calculated the total number of participants assumed to be needed to reach that number of events in the shortest time possible, based on available data on HIV incidence. The calculated number of participants were then recruited and followed until the targeted number of events was reached, when the studies were unblinded and examined for differences. With the number of events and the number of participants both fixed, the only variable subject to change once a trial started was the time needed for recruitment and for follow-up and completion of the trial.

The number of events needed was based on the difference between the expected HIV incidence and the observed HIV incidence in the trial. The number of new infections needed to detect a difference in these trials seemed modest, ranging from as few as 40 to roughly 300 in the largest. The estimated number of participants recruited for these studies to reach this number of events ranged from 1200 to 5000. Some of these trials experienced prolonged follow-up periods, as the targeted number of events occurred more slowly than originally envisioned. In the most extreme case, a trial began enrollment in June 2005 and continued enrolling until July 2010, following some participants as long as 72 months, though only 40 events were targeted [25]. The HIV incidence in the placebo arm of this trial, 0.68 per 100 person years, was far less than the 3.4 per 100 person years used in trial planning, a seemingly reasonable estimate based on a previous HIV vaccine trial in the same research clinics [37]. The number of events needed in another trial was based on estimated HIV incidence of 2.75 per 100 person years, ultimately finding 1.99 per 100 person years in the placebo arm [23]. An assumed annual incidence of 5 per 100 person years was used in another trial, but the measured HIV incidence in the placebo arm at study completion was only 3.1 per 100 person years [24].

Behavior changes in the PrEP trials

Some differences between projected and observed HIV incidence in these trials could be simple miscalculation: investigators may have used overestimates of HIV incidence to calculate the number of events needed. It is also possible that the behavioral HIV prevention interventions provided in these trials to both arms may have decreased risk-taking behaviors and made HIV infections less likely. Data found in results of these trials, challenging to disentangle, appear to support this latter explanation.

In assessing PrEP effectiveness, the investigators needed equivalency of the treatment and control groups. Any major difference would jeopardize the investigators’ ability to reject their null hypotheses. Investigators paid close attention to the “between-group” comparisons at baseline and over time to demonstrate equivalency. In this regard, the studies were generally exemplary. The investigators were able to recruit high risk people, randomize them to two (or more) conditions without introducing any potentially distorting differences, and follow the majority to the conclusion. The two trials that did not obtain conclusive results encountered difficulties in promoting adherence to the study medication [38,39], not in these other aspects of good trial conduct.

One result of this close focus on demonstrating equivalency of the arms was a focus on “between-group” comparisons in the analyses, to the near exclusion of any examination of changes over time experienced by both groups equally. Nonetheless, some elements of these changes can be detected.

In the iPrex trial, for example, observed rates of syphilis among study participants dropped by half in both arms, and other STIs decreased even more [22,40]. The reported number of partners with whom participants engaged in receptive anal sex dropped by half as well, again in both arms. Likewise, the reported percentage using condoms for receptive sex increased from approximately 50% at baseline to around 80% at follow-up. While self-reported behavioral data can always be questioned, the correlation between the reported decreased risk behavior and observed decreases in sexually transmitted infections makes these data difficult to doubt.

In the Partners PrEP trial, more than a quarter of all participants at enrollment reported sex without condoms with their HIV-infected partner in the prior month; this decreased to 13% at 12 months and 9% at 24 months [23]. Even a trial that did not demonstrate effectiveness of PrEP showed modest but significant reductions in some key behaviors: number of sex partners, vaginal sex acts, and sex acts without a condom all decreased over time [26].

Some subsequent analyses of behavior changes in these trials were published, primarily focused on the possible effect of PrEP on subsequent risk-taking, or disinhibition. Results compared three different conditions under which behavior changes may have occurred: by participants’ perception of to which arm they were randomized in the iPrEx trial [40], before and after unmasking of the Partners PrEP trial [41], or over time in one extended study [42]. In the third case, investigators reported marked decreases in sexual risk-taking behaviors over the course of the study. For example, the percentage of participants who reported sex with more than one partner in the previous three months dropped by two thirds from baseline to the 72-month follow-up (21.7% to 6.5%). Casual partner sex declined from 38% at baseline to 14.5% at 72 months [42]. The limited behavioral data reported suggest a pattern of decreasing risk-taking behaviors over the course of the double-blind studies regardless of trial arm, perhaps in response to the comprehensive, behaviorally based package of HIV prevention interventions provided to all participants. Behavioral changes reported in studies are sometimes believed to be a product of the Hawthorne effect [43], in which awareness of being studied alters the natural behaviors of research subjects, or of social desirability bias [44]. In the latter, the desire for social approval is thought to affect accurate reporting of sensitive behaviors. Both of these sources of bias exist in these PrEP trials, though the thoroughness of the protocols and the care taken in data collection suggest these biases were not a major threat.

What made these studies effective, in addition to the PrEP effect?

In trying to understand what made the HIV prevention packages apparently more effective than they are generally believed to be, we examined the content of the interventions offered. Trial participants actually received considerably more than the standard prevention provided in their countries. A key difference was the frequency of the interventions: participants were seen monthly for risk-reduction counseling and support, free condoms, adherence support, and follow-up.

Contrast this to the current WHO recommendations for HIV testing and counseling [45], the key element of HIV prevention used in most countries. This intervention consists of testing those at risk of HIV, identifying those who are infected, with referral to care and treatment, including antiretroviral therapy, but little for the uninfected: “Individuals who test HIV-negative should receive brief health information about their test results. Research to date has not demonstrated that a lengthy counselling session is needed or is beneficial” (p.24). The recommended frequency of contact – an opportunity to provide the type of counseling and support that was part of the standard prevention package in most of these PrEP trials – is also limited: “WHO recommends that people in high-risk categories retest at least annually” (p.24).

In standard practice, then, the opportunity to intervene more meaningfully with those not infected but at risk of HIV infection is often missed. A comparison of some of the key components of behavioral prevention provided in the PrEP trials and the current WHO guidelines is shown in Table 1.

	HIV testing and counseling	Risk reduction counseling	Condom provision	STI assessment and treatment	PrEP adherence counseling
Partners PrEP [23,41]	Monthly	Monthly	Monthly	Every 3 months	Monthly
Bangkok Tenofovir [25,42]	Monthly	Monthly	Monthly or every 3 months	N/A	Monthly or more frequently
iPrEX [22,40]	Monthly	Monthly	Monthly	Every 24 weeks	Monthly
VOICE [27]	Monthly	monthly	Monthly	At enrollment, annually, or when indicated	Monthly
Botswana TDF-FTC [24]	Monthly	Every 3 months (or on request)	Each visit	Monthly	Monthly
FemPrEP [26]	Monthly	Monthly	Monthly	On entry and final visit	Monthly
WHO Consolidated guidelines on HIV testing services 2015 [45]	Annually	Brief information on test results	N/A	N/A	N/A
WHO PrEP Implementation Tool 2017 [46]	Every 3 months	First visit	First visit	Every 3 to 6 months	Every visit (or every 3 months)

Trial participants received more frequent and more intense behavioral interventions than the standard HIV prevention package offered in the countries where these trials took place.

Though they appear to have been effective in the PrEP trials, one of the criticisms of behavioral interventions is effectiveness requires intensity and reinforcement. It is also becoming clear, however, that biomedical interventions too require reinforcement to be effective, either to achieve the adherence that makes PrEP effective and or the retention in treatment that leads to viral suppression. Effective HIV prevention through any of the means currently available, behavioral or biomedical, requires a greater intensity of efforts than is often allocated to the task.

Perhaps because they were concerned about loss to follow up and potential criticism like that levelled at the initial feasibility studies, trial investigators provided higher quality of HIV prevention than current guidance recommends. We contend that these trials, beyond proving PrEP to be effective, have also demonstrated that the HIV prevention model used in most countries can be more effective if provided with greater frequency and intensity.

It has become a truism in HIV that the dichotomy between biomedical and behavioral prevention is false, and that all biomedical prevention interventions require a behavioral response. We could be criticized for making a distinction between the two that many would like to believe is passé. In response, we would point out that the integral nature of the biomedical and behavioral is not well conveyed in the presentation of these and many other studies: as can be seen, extracting the behavioral components of these trials’ outcomes is difficult to do. Including a more robust presentation of all data, behavioral as well as biomedical, not limited to “between-group” comparisons, would make evaluation of trial results fairer and more balanced. As a result, a more robust picture of the potential impact of behavioral interventions to prevent HIV may emerge.

We admit that none of the evidence we have gleaned from these studies is conclusive. We present it as highly suggestive that behavioral HIV prevention deserves more credit for effectiveness. This realization is not new: during the (often long) course of these PrEP trials, a common joke was, “We know how to prevent HIV… Just enroll people in PrEP trials!” In the intervening years since these trials were published, it appears that the kernel of truth in that joke has been forgotten and that purely behavioral HIV prevention is afforded less and less prominence in HIV priorities worldwide. It is time for that to change.