Background
An estimated 35.3 million people globally are living with HIV (UNAIDS, 2013). A number of prevention methods are available, from condoms to male circumcision, prevention of mother-to-child transmission to clean needles, but to date these have not been sufficient to stop the epidemic. In 2012 alone, an estimated 2.3 million people became newly infected (UNAIDS, 2013). Additional safe and effective approaches to HIV prevention are urgently needed.
People who inject drugs (PWID) have a disproportionate burden of HIV. Existing methods of HIV prevention for PWID include approaches used across populations to reduce sexual transmission as well as approaches specific to PWID to reduce HIV transmission through sharing unclean needles and other injection equipment. However, political and structural barriers prevent access to needle and syringe programs and opioid substitution therapy in many settings, and additional prevention modalities would be helpful for these populations.
PrEP is the use of an antiretroviral drug to block the acquisition of HIV infection by uninfected people. Proof of concept has long been established in the laboratory by animal studies and in real world application by the prevention of mother-to-child transmission and post-exposure prophylaxis. The safety of the drugs being considered for PrEP, tenofovir and emtricitabine, has been established through their use for treatment and in safety trials in uninfected people (Peterson et al., 2007). Five trials of effectiveness (Phase IIb and Phase III) of oral PrEP have been conducted in the last decade. These have examined the effectiveness of PrEP among PWID, serodiscordant couples, heterosexual women and high risk men who have sex with men (MSM).
Of the five effectiveness trials, only one trial examined efficacy among PWID: the Bangkok Tenofovir Study (Choopanya et al., 2013). This Phase III clinical trial tested whether daily tenofovir disoproxil fumarate (TDF) could safely and effectively prevent HIV infection among PWID in Bangkok, Thailand. Over 2400 PWID were enrolled in the study and randomly assigned to daily TDF or placebo. Participants were also provided regular HIV testing and risk reduction counseling. The primary outcome of the trial was HIV incidence, which was 0.35 per 100 person-years (py) in the TDF group (17 infections) and 0.68 per 100 py in the placebo group (33 infections), indicating a 48.9% reduction in HIV incidence related to PrEP (95% CI: 9.6, 72.2; p=0.01). Serious adverse events were not statistically significantly different between the two groups (p=0.35). Trial findings led the U.S. Centers for Disease Control and Prevention (CDC) to recommend PrEP be considered “as one of several prevention options for persons at very high risk for HIV acquisition through the injection of illicit drugs” (CDC, 2013).
This systematic review examined evidence to answer the following PICO question: Should oral PrEP (containing tenofovir (TDF)) be used for HIV prevention among people who inject drugs (PWID)? In addition, we reviewed the values and preferences about PrEP among people who use drugs and considered studies of cost and feasibility for the GRADE process.
Methods
PICO question
PICO 1: Should oral PrEP (containing tenofovir (TDF)) be used for HIV prevention among people who inject drugs (PWID)?
P: People who inject drugs
I: Oral PrEP (containing tenofovir (TDF))
C: Placebo
O: (1) HIV infection, (2) any adverse event, (3) any stage 3 or 4 adverse event, (4) condom use, (5) number of sexual partners, (6) injection frequency, (7) needle/syringe sharing
Inclusion criteria
To be included in the review, an article had to meet the following criteria:
Randomized controlled trial evaluating the use of oral PrEP (containing tenofovir (TDF)) to prevent HIV infection among PWID.
Measured one or more of the following key outcomes: (1) HIV infection, (2) any adverse event, (3) any stage 3 or 4 adverse event, (4) condom use, (5) number of sexual partners, (6) injection frequency, (7) needle/syringe sharing
Published in a peer-reviewed journal, or presented as an abstract at a scientific conference, between January 1, 1990 and January 1, 2014.
Only studies among people who inject drugs were included; studies among people who use, but do not inject, drugs were excluded, as HIV risk and transmission modalities differ between these groups. However, both terms were used in the search.
No restrictions were placed based on location of the intervention. No language restrictions were used on the search. Articles in languages other than English were translated where necessary.
Following the GRADE approach, if direct evidence from PWID populations was limited for one or more of the key outcomes, indirect evidence from other populations (men who have sex with men, or heterosexual men or women) would be used instead, but downgraded for indirectness. If evidence from other populations was limited, evidence from non-randomized but controlled studies would be used instead, but also downgraded for directness.
Search strategy
The following electronic databases were searched using the date ranges January 1, 1990 to January 1, 2014: PubMed, CINAHL (Cumulative Index to Nursing and Allied Health Literature), and EMBASE. Secondary reference searching was conducted on all studies included in the review. Further, selected experts in the field were contacted to identify additional articles not identified through other search methods.
Abstracts from the following conferences were searched from January 1, 1990 to January 1, 2014: International AIDS Conference (IAC) and IAS Conference on HIV Pathogenesis, Treatment, and Prevention (IAS). We had planned to search the Conference on Retroviruses and Opportunistic Infections (CROI) as well, but abstracts from this conference were no longer available online to the public at the time the search was conducted.
Search terms
The following terms were entered into all computer databases:
(“people who use drugs” or PWUD or “people who inject drugs” or PWID or “drug users” or IDU or IDUs) AND (“pre-exposure prophylaxis” or PrEP or tenofovir or TDF) AND (HIV OR AIDS)
These search terms were used both for the main systematic review (PICO question) and for the values and preferences review.
The search for abstracts was more difficult given the search engines available on conference websites. For each conference, a search was first conducted for all abstracts including the word “PrEP”. These search results were then further searched for keywords regarding PWID.
Screening abstracts
Titles, abstracts, citation information, and descriptor terms of citations identified through the search strategy were screened by two reviewers. Full text articles were obtained for all selected abstracts and both reviewers independently assessed all full-text articles for eligibility to determine final study selection. Differences were resolved through consensus.
Articles not meeting the inclusion criteria for the review, but presenting potentially interesting background information relevant to PrEP among PWID, including review articles, qualitative studies, cost or cost-effectiveness analyses, or descriptions of interventions without an evaluation component, were included in an annotated bibliography of additional articles.
Data extraction and management
Data were extracted independently by two reviewers using standardized data extraction forms. Differences in data extraction were resolved through consensus and referral to a senior team member from WHO when necessary. Study authors were contacted when additional information or data were needed.
The following information was gathered from each included study:
Study identification: Author(s); type of citation; year of publication
Study description: Study objectives; location; population characteristics; description of the intervention; study design; sample size; follow-up periods and loss to follow-up
Outcomes: Analytic approach; outcome measures; comparison groups; effect sizes; confidence intervals; significance levels; conclusions; limitations
Risk of bias was assessed using the Cochrane Collaboration's tool for assessing risk of bias (Cochrane Handbook, chapter 8.5 – Higgins & Green, 2011). This tool assesses random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias) incomplete outcome data (attrition bias), and selective reporting (reporting bias). Methodological components of the studies were assessed and classified as high, low, or uncertain risk of bias.
Data analysis
Data were analyzed according to coding categories and outcomes. If multiple studies reported the same outcome, meta-analysis would have been conducted using random-effects models to combine effect sizes with the program Comprehensive Meta-Analysis (CMA). Data were summarized in GRADE tables, summary of finding tables, and risk/benefit tables.
Results
Our initial database search yielded 183 citations and 243 conference abstracts; no additional studies were identified through other means (). Once duplicates were removed, 392 records were reviewed and 131 article citations and 236 abstracts were excluded for being unrelated to the study topic. After review of the remaining 17 articles and 7 abstracts by two independent screeners, 16 articles were excluded for not meeting the study design criteria and were coded as background or values and preferences, while 6 abstracts were excluded for providing additional information on the included trial, but without reporting key outcomes. The remaining study (with data for PICO outcomes reported in one article and one conference abstract) was deemed eligible for inclusion in our review.
Disposition of citations during the search and screening process.
The one study that met all inclusion criteria was the Bangkok Tenofovir Study (Choopanya et al., 2013; Vanichseni et al., 2013). This study was a randomized controlled trial to assess whether daily oral use of tenofovir disoproxil fumarate (tenofovir) can reduce HIV transmission in injecting drug users. The trial was conducted in Bangkok, Thailand, where 2413 total participants were recruited from 17 drug treatment clinics. Participants' ages ranged from 20 to 59 years (mean=32.4), 80% were male, and 63% reported injecting drugs in the past 12 weeks.
Using the Cochrane Risk of Bias tool, the study was judged to have low risk of bias across all of the following categories: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), and selective reporting (reporting bias). For selective reporting (reporting bias), the study was initially judged to have uncertain risk of bias. The study protocol was available, and all of the study's pre-specified primary outcomes of interest were reported in the pre-specified way. However, for two secondary outcomes, condom use and number of sexual partners, outcomes that were predefined in the protocol were not available, or not available in the pre-specified way, in published reports. After contacting the study authors for additional information, data on these outcomes were shared with the review team that have not yet been published given the recent conclusion of the trial. Therefore, the judgment on selective reporting was changed to low risk. For incomplete outcome data (attrition bias), we noted that loss to follow-up was high relative to the number of events. Loss to follow-up was 14.9% in the PrEP group and 14.6% in the placebo group; additional participants from both groups withdrew from the study, died, or otherwise stopped follow-up. Although there were no differences in follow-up time, withdrawal, or loss to follow-up between treatment groups, GRADE guidance notes that “large loss to follow-up in relation to the number of events always... raises the issue of a serious threat of bias” (Guyatt et al., 2011). Further, GRADE generally urges caution classifying a single RCT in a single location as an overall high quality of evidence (Guyatt et al., 2011). For these reasons, we made a judgment of high risk for this measure and downgraded the quality of evidence for potential risk of bias across all study outcomes (as most outcomes had relatively high loss to follow-up relative to the number of events, and the single trial was the only evidence across all outcomes). Finally, we calculated relative risks and 95% confidence intervals for outcomes where effect size estimates were not presented. Based on this analysis, in GRADE, we downgraded three outcomes – condom use, number of sexual partners and needle/syringe sharing – for imprecision because the 95% CI includes appreciable benefit or harm according to the GRADE general guideline of a relative risk of under 0.75 or over 1.25.
The study measured all seven key outcomes for this review: (1) HIV infection, (2) any adverse event, (3) any stage 3 or 4 adverse event, (4) condom use, (5) number of sexual partners, (6) injection frequency, and (7) needle/syringe sharing. Results for each outcome are presented below.
HIV infection
Incident HIV infection was significantly reduced among participants in the tenofovir study arm as compared to the control arm using both an intention-to-treat analysis and a modified intention-to-treat. In the intention-to-treat analysis, there were 17 incident cases of HIV infection out of 1204 participants in the tenofovir study arm and 35 incident HIV infections out of 1209 participants in the control group, resulting in a 51.8% reduction in HIV incidence (95% confidence interval (CI): 15.3-73.7, p=0.01). In the modified intention-to-treat analysis (excluding 2 control participants who were HIV-positive at enrolment), there were 17 incident cases of HIV in the tenofovir group out of 4843 person-years (py) for an incidence of 0.35 per 100 py and 33 incident cases of HIV in the control group out of 4823 py for an incidence of 0.68 per 100 py. Thus, in the modified intention-to-treat analysis, there was a 48.9% reduction in HIV incidence (95% CI: 9.6–72.2, p=0.01).
In age-stratified analyses, PrEP was effective in those age 40 and older, while there was no significant difference between PrEP and placebo in the age groups 20-29 or 30-39 (although results trended in a positive direction). For participants age 40 and older, the number of incident infections overall was small, with 1 incident infection out of 1066 py in the tenofovir group for an incidence of 0.09 per 100 py (95% CI: 0.002-0.52) and 9 incident infections out of 1052 py in the control group for an incidence of 88.9 per 100 py (95% CI: 41.1-99.4); this difference was statistically significant (p=0.01). For participants age 20-29, there were 11 incident infections out of 1976 py in the tenofovir group for an incidence of 0.56 per 100 py (95% CI: 0.28-1.00), versus 17 incident infections out of 1993 py in the control group for an incidence of 0.85 per 100 py (95% CI: 0.50-1.37); this difference was not statistically significant (p=0.30). For participants age 30-39, there were 5 incident infections out of 1801 py in the tenofovir group for an incidence of 0.28 per 100 py (95% CI: 0.09-0.65), versus 7 incident infections out of 1778 py in the control group for an incidence of 0.39 per 100 py (95% CI: 0.16-0.81); this difference was not statistically significant (p=0.55).
Any adverse event
There was no statistically significant difference in reported adverse events between the two study arms. In the tenofovir arm, 91% of participants (1098/1204) had an adverse event (10965 events total). In the placebo arm, 90% of participants (1083/1209) had an adverse event (11550 events total). This difference between arms was not statistically significant (p=0.46).
Any stage 3 or 4 adverse event
Both study arms also reported similar rates of stage 3 and 4 adverse events. In the tenofovir arm, 13% of participants (156/1204) had a stage 3 or 4 adverse event (414 events total). In the placebo arm, 13% of participants (160/1209) had a stage 3 or 4 adverse event (389 events total). This difference between arms was not statistically significant (p=0.89).
Condom use
Condom use data were not reported in any published articles or abstracts at the time of the search, but the Bangkok Tenofovir Study authors were contacted and provided additional unpublished data for condom use outcomes. Participants who self-reported sex with a live-in or casual partner were then asked questions about condom use. A skip pattern error in the initial years of the study made data on condom use with casual partners unreliable. Therefore, we present data on condom use with live-in partners. At baseline, 6.5% (34/526) of tenofovir study arm participants with live-in partners reported always using condoms with those partners (vs. less than always condom use), compared with 8.5% (44/518) of placebo arm participants. At 12-month follow-up, these changed to 11.1% (41/369) in the tenofovir group and 11.3% (44/388) in the placebo group. At 12-month follow-up, this translates to a relative risk (RR) of 0.979 (95% CI: 0.656 to 1.463).
Number of sexual partners
Self-report of sex with more than one partner in the previous 3 months was 22% at enrollment across both study arms and dropped to 11% at the 12-month follow-up and 6% at the 72-month follow-up. Additional unpublished data were also shared by the study investigators. At the 12 month follow-up, reported number of sexual partners in the previous 3 months was not statistically significant between the tenofovir and placebo arms (p=0.181). At the 12 month follow-up, 44.9% (413/919) of tenofovir arm participants and 41% (394/960) of placebo arm participants reported no sexual partners in the past 3 months, 45% (414/919) and 47% (451/960) respectively reported 1 partner, 6.4% (59/919) and 7.5% (72/960) respectively reported 2 partners, 1.5% (14/919) and 2.1% (20/960) respectively reported 3 partners, and 2.1% (19/919) and 2.4% 23/960) respectively reported 4 or more partners. In regression analyses there were no interactions between time and treatment group for this outcome.
Injection frequency
Self-report of injecting drugs in the previous 3 months was 63% at enrollment across both study arms and dropped to 23% at the 12-month follow-up and 18% at the 72-month follow-up. Additional unpublished data were also shared by the study investigators. At the 12 month follow-up, reported injecting drugs in the previous 3 months was 22.1% (203/919) for the tenofovir arm and 23.3% (224/960) for the placebo arm; this difference was not statistically significant (p=0.520). In regression analyses there were no interactions between time and treatment group for this outcome.
Needle/syringe sharing
Self-report of needle sharing in the previous 3 months was 18% at enrollment across both study arms and dropped to 2% at the 12-month follow-up and 1% at the 72-month follow-up. Additional unpublished data were also shared by the study investigators. At the 12 month follow-up, reported needle sharing in the previous 3 months was 2.3% (21/919) for the tenofovir arm and 2.4% (23/960) for the placebo arm; this difference was not statistically significant (p=0.874). In regression analyses there were no interactions between time and treatment group for this outcome.