Vulnerable Periods: Characterizing Patterns of Sexual Risk and Substance Use During Lapses in Adherence to HIV Pre-Exposure Prophylaxis among Men who have Sex with Men

Abstract

Background:

Pre-exposure prophylaxis (PrEP) is highly efficacious, but some groups of men who have sex with men (MSM) may have difficulty adhering to daily dosing. Prevention-effective adherence suggests that PrEP’s efficacy depends on adherence at the time of HIV exposure, yet few studies have examined how exposures (i.e., high-risk sex) overlap with periods of consecutive missed PrEP doses. Substance use may also play a role in these vulnerable periods.

Methods:

We used digital pill bottles to monitor the daily adherence of 40 PrEP-experienced patients recruited from an outpatient clinic in the Northeastern US over a six-month period. Participants also completed detailed online diaries every two weeks during this time that surveyed their sexual behavior and substance use each day.

Results:

Daily adherence was high overall (M = 83.9%, SD = 18.0%), but 53% (N = 21) had a lapse of > 3 consecutive daily PrEP doses over six months. Participants’ rate of engaging in high-risk condomless anal sex (CAS) did not differ across lapse days versus continuously-adherent days. Alcohol use was not associated with engaging in CAS during a PrEP lapse. However, participants reported engaging in CAS significantly more often during a PrEP adherence lapse on days when they also used stimulant drugs.

Conclusions:

MSM may have periodic difficulty adhering to PrEP at the specific times when they are at-risk. Stimulant drug use could play an important role in increasing HIV risk specifically during adherence lapses.

1. Introduction

Although HIV incidence has slowly declined in the United States (US) in recent years, men who have sex with men (MSM) continue to account for 70% of new infections¹. Pre-exposure prophylaxis (PrEP) holds promise for achieving a significant, sustained decline in new infections², given that it can achieve over 99% efficacy in preventing HIV acquisition with optimal adherence^3-8. Although clinical guidelines currently recommend daily dosing⁹, pharmacokinetic analyses of clinical trials data have shown that fewer doses also result in high levels of protection. Efficacy declines, however, to 96% at 4 doses/week and to 76-81% at 2-3 doses/week^3,10. Studies of community-recruited MSM show that, although HIV infections are substantially reduced, many patients who start PrEP have difficulty adhering to daily dosing^4,10,11. For example, in a nationwide demonstration study of young MSM (ATN 113), only 34% of participants had drug levels consistent with taking ≥ 4 doses/week after 48 weeks¹². Likewise, over 40% of MSM in the DEMO Project had similarly low drug levels or had dropped out by 48 weeks¹¹.

Less is known about PrEP adherence in real-world settings. Moreover, typical analyses of PrEP adherence may not accurately reflect the true level of protection achieved. This is because PrEP’s effectiveness is likely to depend on patients’ adherence specifically around the times when they are exposed to HIV rather than on their adherence over broad intervals (e.g., months)¹³. This view, called prevention-effective adherence^13,14, suggests that patients’ engagement in risk behavior fluctuates over time and that understanding PrEP’s effectiveness requires exploring how adherence overlaps with potential exposures. Although some international studies suggest that adherence may improve specifically around times when patients are at highest risk^15,16, few studies have explored this in the US. Because self-reported data on PrEP adherence are generally not dependable^17-20, the majority of available US community studies have assessed adherence by analyzing dried blood spot (DBS) samples collected at regular study visits for PrEP metabolites (e.g., tenofovir diphosphate [TFV-DP])²¹. Collecting even objective adherence data at such broad time intervals, however, has drawbacks. First, although TFV-DP’s long half-life provides a stable measure of cumulative dosing over time²¹, long gaps between collection intervals may miss periods of suboptimal dosing. Second, requiring frequent study visits to collect DBS samples in-person could also provide an overly-optimistic picture of real-world adherence, because regular face-to-face contact with health professionals may itself encourage better adherence^22-24.

Alcohol and drug use are common among high-risk MSM, as they are among other groups at risk for HIV^25-26. These behaviors could pose a challenge for PrEP care, given strong evidence that alcohol and drug use are associated with poorer adherence to similar drugs used to treat HIV^27-31. However, few studies have explored the influence of alcohol and drug use specifically in the context of PrEP care. Two studies using medication refill data from large healthcare systems showed that patients with substance use diagnoses had fewer days covered by PrEP^32-33. However, neither alcohol nor other substance use were associated with PrEP drug levels over time in the iPrEx OLE or DEMO Project^10,11. Although these studies have been helpful in identifying risk factors for poor PrEP outcomes, their reliance on broad assessments of alcohol/drug use may miss time-dependent elevations in use that co-occur with disruptions in adherence. Aggregated surveys assessing typical use over broad intervals may also result in inaccuracy due to recall biases^34,35. Past research shows that self-reported alcohol/drug use is highly correlated with relevant biomarkers when assessments minimize recall intervals and ask respondents to recall exact behaviors that occurred on specific days^36-42. This approach also has the advantage of allowing researchers to assess whether the specific timing of alcohol/drug behaviors co-occurs with other behaviors (e.g., PrEP adherence, high-risk sex) on a daily basis.

In this study, we explored the co-occurrence of sexual risk behavior and substance use with lapses in PrEP adherence over six months among MSM at a community PrEP clinic.

2. Methods

Participants

Forty MSM were recruited by providers and staff at a community PrEP clinic in Providence, Rhode Island⁴³ from February to November 2017. Eligible participants were (1) 18⁺ years old, (2) assigned male sex at birth, (3) reported insertive or receptive anal sex with a man in the last 12 months, and (4) had been prescribed PrEP.

Procedures

Participants who met eligibility criteria based on their visit intake forms were approached about the study when attending routine PrEP care visits. Interested participants provided informed consent before completing baseline surveys. Participants were then provided with a digital pill bottle and were instructed to fill their medication into the bottle within a day of having their prescription refilled. Participants were also instructed to complete online dairies every two weeks over the six-month period. To facilitate this, emails with a links to the diaries were automatically sent from the study database on their due dates, with daily reminders sent for five days until completed. Surveys not completed after five days were considered incomplete. We chose online surveys because they allowed us to collect detailed daily data at intervals that minimize inaccuracies (e.g., recall bias^34,35) while also limiting interactions with research personnel that could affect adherence. As such, participants attended face-to-face appointments only at routine PrEP clinical visits (about every 90 days)⁹. Participants were paid up to $200 based on their rate of diary completion and for agreeing to use the pill bottle. All study procedures were approved by the Lifespan and Brown University Institutional Review Boards.

Measures

Digital pill bottles.

Daily PrEP adherence was continuously assessed using AdhereTech’s Generation 2 “smart” pill containers (AdhereTech New York, NY). These devices had several advantages, including: (1) a continuous internet connection via mobile networks for real-time data uploads, (2) long battery life allowing use over the entire six-month study period on a single charge, and (3) no set up or maintenance was required of participants.

Online diaries.

Participants were asked to complete an online Timeline Followback (TLFB)^44,45 survey on their sexual behavior, alcohol use, and drug use each day in two-week increments. Using a custom web application⁴², participants were asked to indicate the days on which they had oral, anal, or vaginal sex, drank alcohol, or used drugs, and each of these days were marked with a specific icon. Then, participants were asked more details about each of the behaviors that occurred on specific days. For sexual behavior, participants reported the number of partners they had that day (up to four), characteristics of each partner (e.g., gender, sexually exclusive/non-exclusive, whether they asked about HIV status, and if so, what it was), the acts they engaged in with each (e.g., oral, insertive anal, receptive anal, vaginal) and whether a condom was used for each act. For alcohol use, participants were asked about the number of standard drinks consumed that day. For marijuana, stimulant, or party drug use, participants selected the type of drug they used that day (e.g., marijuana, powder/crack cocaine, methamphetamine, ecstasy) from a list of many drugs. Finally, participants were asked whether they pocketed any PrEP doses (i.e., removed several pills from the pill bottle to store in a different container for travel, etc.) during the displayed period of time, and if so, the days on which they successfully took a PrEP pill after pocketing it. For analyses, missed doses from digital pill bottles were replaced with successfully taken doses if participants indicated taking doses on days when they reported pocketing PrEP pills. High-risk condomless anal sex (CAS) events were defined as engaging in insertive or receptive anal sex without using a condom with an unknown HIV status or sexually non-exclusive partner. Hazardous drinking was defined as 15+ drinks in an average week, or at least one occasion of 5+ drinks in a single occasion at least once in the past month⁴⁶.

Alcohol Use Disorders Identification Test (AUDIT) was used to assess alcohol-related problems in the past 12 months⁴⁷. Those who scored > 8 were classified as high-risk for alcohol-related problems⁴⁸.

Dried blood spots (DBS) were collected from participants during routine PrEP follow-up visits, typically three months after baseline. DBS were analyzed for TFV-DP, a biomarker of cumulative PrEP dosing up to three weeks prior to sampling^10,11,18,21, as well as phosphatidylethanol (PEth), a biomarker of recent alcohol use that has shown high sensitivity in detecting chronic and binge drinking episodes that occur within 1-2 weeks of sampling^49-52. TFV-DP and PEth were used to corroborate data collected via digital pill bottle and online drinking diaries and to explore pairwise associations between adherence and alcohol use.

Analysis plan

We first matched pill bottle and diary data by study day and calculated summary statistics for key variables. We then calculated Pearson correlations to explore associations between variables collected at single time points (e.g., TFV-DP, PEth). PrEP adherence lapses were defined as missing ≥ 3 consecutive doses to align with an adherence pattern that begins to reduce PrEP’s efficacy. To explore whether the rate of high-risk CAS events differed across lapses and non-lapses, we used estimated generalized estimating equations (GEEs) specifying CAS on a given day as the focal outcome, with relevant covariates and whether a given day was part of a lapse vs. a non-lapse as a key predictor (Model 1). In Model 2, we then added a categorical term reflecting daily alcohol use level (no drinking [reference group], vs. 1-4 drinks and 5+ drinks) and a two-way interaction between alcohol use level and lapse day to explore whether CAS co-occurred with alcohol use more often on an adherence lapse days versus non-lapse days. In Model 3, we added daily marijuana and stimulant drug use in a similar way. To explore whether alcohol, marijuana, or stimulant use co-occurred with lapses, we then estimated similar GEE models with PrEP lapse days versus non lapse day as a focal outcome. Since the outcomes of all models were binary, we specified binomial distributions and used a logit link function in each case.

3. Results

Attrition and Response Rates

See Table 1 for participant demographic characteristics. Six participants (15%) withdrew prior to submitting their final six-month survey, and among these participants, the average time to withdrawal was 2.83 months (SD = 0.98). One of these elected to stop PrEP during the study period, and five withdrew only from the study for unknown reasons. Those who withdrew did not differ from those retained with respect to any demographic, clinical, or behavioral characteristics. Among all participants, the overall response rate to bi-monthly TLFB surveys was 85.6% (SD = 24.5%). Among those who did not withdraw (N = 34), diary response rates were 94.4% (SD = 11.6%). All available data from participants were retained for analysis, whether or not they withdrew, providing a total of 6,013 days matched across TLFB and digital pill bottle data. Among those who did not withdraw, the mean number of matched days per participant was 156.9 (SD = 25.4). Adherence data was corrected for pocketed doses on 42 total days (0.6% of all study days).

TABLE 1.

Demographic and Behavioral Characteristics of the Study Sample (N = 40)

Characteristics	Mean (SD) or N (%)
Age (Range: 20 – 58)	38.1 (10.3)
Race
White	26 (65.0)
Black or African American	1 (2.5)
Asian	4 (10.0)
Multiracial	5 (12.5)
Other	2 (5.0)
Chose not to respond	2 (5.0)
Ethnicity (Hispanic or Latino)	6 (15.0)
Relationship status	373 (24.8)
Single	22 (55.0)
In committed relationship	1 (2.5)
In domestic partnership	6 (15.0)
Married	7 (17.5)
Divorced	3 (7.5)
Other	1 (2.5)
College degree	29 (72.5)
Low income1	8 (20.0)
Unemployed	2 (5.0)
Currently have health insurance	39 (97.5)
Number of months insured, past year	11.4 (2.2)
Depression (CES-D2)	37.9 (8.9)
Avg. time on PrEP (in years, Range: 0.25 - 3.0)	1.12 (0.62)
Self-referred for PrEP	12 (30.0)
Diagnosed with bacterial STI during study	15 (37.5)
Avg. number of high-risk CAS3 events during study period	12.7 (18.1)
“Hazardous” alcohol use	24 (60.0)
Alcohol-related problem (AUDIT4 ≥ 8)	5 (12.5)
Any stimulant/“party” drug use during study	6 (15.0)
Any marijuana use during study	17 (42.5)

Note.

¹Represents those with a household annual income <$30,000/year.

²Center for Epidemiological Studies – Depression Scale.

³Condomless anal sex.

⁴Alcohol Use Disorders Identification Test (AUDIT)

PrEP Adherence

Twenty-one participants (53%) had a lapse of ≥ 3 PrEP doses at some point over the six-month period, and of these, 11 (52.4%) had only one such lapse, seven (33.3%) had 2-3, and three (14.3%) had 4-8 lapses. The average total length of an adherence lapse that was ≥ 3 days was 6.0 days (SD = 4.9). A total of 421 study-days (7%) were considered part of such a lapse. Sixty-three percent of participants had tenofovir drug levels consistent with near-daily adherence (>1100 fmol/punch¹⁰). Only five participants had TFV-DP levels of <700 fmol/punch at the time of collection, and as assessed by digital pill bottle, these participants had a significantly lower average overall adherence rate (52.4% vs. 88.4%, p <.001) and greater number of lapses (2.4 vs. 1.0, p < .05) compared to those with ≥ 700 fmol/punch. Percent days adherent to PrEP were moderately-to-highly correlated with TFV-DP levels (fmol/punch), r=.42, p < .05. The sample’s overall adherence rate, as assessed by digital pill bottle, was 83.9% (SD = 18.0%, Range = 19.5% - 100%).

High-risk CAS, alcohol, marijuana, and stimulant drug use during PrEP lapses

Participants reported 630 total anal sex acts, of which 516 (81.9%) were without a condom, and 485 (94.0%) involved CAS with non-exclusive or unknown HIV status partners. Sixty percent of participants met criteria for hazardous drinking. However, only 12.4% had evidence of alcohol-related problems (i.e., AUDIT > 8). Participants reported a total of 1,523 drinking days, 264 (17.3%) of which involved heavy drinking (i.e., drinking 5+ drinks in a single day). Participants’ PEth values were strongly associated with self-reported drinking over the seven days prior to collecting DBS (r = .79, p < .05), suggesting that TLFB assessments of drinking were valid. Forty-two percent reported using marijuana over 756 study days. Only 15% of participants reported using stimulants or party drugs over only 51 study days, all of which involved cocaine.

The unadjusted rate at which participants reported high-risk CAS events was 6.9 per 100 person-days during periods in which participants had missed ≥ 3 PrEP doses, versus 8.1 per 100 person-days during more continuously adherent periods, IRR=0.84, p > .05, 95% CI: 0.56-1.23. In GEE models of CAS events controlling for basic covariates (Table 2), the incidence rate for engaging in high-risk CAS during a lapse was not significantly different from the rate during more continuously adherent periods. In Model 2, alcohol use level on a given day was generally positively associated with engaging in high-risk CAS that day, but the two-way interaction between drinking and lapse day was not significant (Figure 1). In Model 3, marijuana use also significantly co-occurred with CAS events during continuously adherent periods, but again the two-way interaction was not significant. However, the two-way interaction between stimulant drug use and lapse day significantly predicted CAS events, such that the rate of high-risk CAS events occurring during a lapse was 4.5 times higher on days when party drugs were concurrently used than on non-drug use days.

TABLE 2.

Generalized estimating equation (GEE) models of high-risk CAS events and days in a PrEP lapse of ≥ 3 doses

	High-risk CAS				PrEP Lapse Days
Variable	IRR	SE	p	95% CI	IRR	SE	p	95% CI
Model 1
Age	1.04	0.01	.010	1.01-1.06	0.96	0.03	.278	0.90-1.03
Single	1.75	0.60	.104	0.89-3.43	0.60	0.47	.517	0.13-3.02
College degree	1.03	0.40	.947	0.48-2.18	1.50	1.06	.563	0.36-5.99
Weekend day	1.76	0.27	<.001	1.30-2.37	1.02	0.05	.677	0.95-1.16
AUDIT1 > 8	0.91	0.42	.841	0.37-2.24	0.77	0.47	.675	0.26-1.82
DAST2 > 3	--	--	--	--	0.54	0.25	.179	0.22-1.33
Lapse day	0.93	0.24	.784	0.56-1.55	--	--	--	--
Model 2
Moderate drinking day	2.34	0.39	<.001	1.68-3.25	0.32	0.15	.012	0.13-0.78
Binge (5+) drinking day	2.67	0.69	<.001	1.61-4.42	1.06	0.49	.893	0.43-2.63
Moderate drinking x lapse day	0.51	0.34	.312	0.14-1.87	--	--	--	--
Binge drinking x lapse day	0.49	0.48	.469	0.07-3.36	--	--	--	--
Model 3
Marijuana use	2.63	0.76	.001	1.49-4.63	0.30	0.17	.034	0.10-0.91
Marijuana x lapse day	0.90	0.60	.874	0.24-3.36	--	--	--	--
Stimulant drug use	3.91	1.63	.001	1.73-8.84	4.51	2.28	.003	1.67-12.14
Stimulant drug x lapse day	4.52	2.69	0.11	1.41-14.51	--	--	--	--

Note. All GEE models specified a Poisson distribution with a log-link function and independent correlation structure. IRR = Incidence rate ratio, SE = Standard error. Models 2 and 3 contain all of the covariates in Model 1.

¹Alcohol Use Disorders Identification Test.

²Drug Abuse Screening Test.

Figure 1.

Adjusted incidence rate ratios of engaging in high-risk CAS on a given day from estimated GEE models.

Overlap association between lapses in PrEP adherence and alcohol/drugs

Overall, neither TFV-DP levels (r=0.22, p>.05) nor digital pill bottle data (r=−0.01, p>.05) were strongly associated with PEth levels. Similarly, neither TFV-DP levels (r=0.24, p >.05) nor digital pill bottle data (r=0.13, p>.05) were strongly associated with metrics of self-reported alcohol use, such as the number of drinks consumed in the 14 days prior to DBS collection. However, at the person-level, an increasingly higher percentage of those with more alcohol involvement had any PrEP lapse of ≥ 3 days and had a longer maximum length of adherence lapse (Figure 2). In GEE models of PrEP lapse days (Table 2), adherence lapse days occurred 3.1 times less often on moderate drinking days when compared to non-drinking days, and there was no difference in the rate of lapse days that occurred on binge drinking days versus non-drinking days (Figure 3). At the person-level, there was no difference in the percentage of those who had any lapse or the average maximum lapse length between those who used any marijuana during the study period and those who did not. In the GEE models, lapse days also occurred 3.3 times less often on marijuana use days.

Figure 2.

Percent of participants with any lapse in PrEP adherence of ≥ 3 days (top panel) and the average maximum length of adherence lapses in days (lower panel) by alcohol use involvement, marijuana use, and stimulant/party drug use.

Note. “Hazardous” drinker = Consuming > 14 drinks in a given week or > 5 drinks in a single day. “High AUDIT” = Score of > 8 on the Alcohol Use Disorders Identification Test.

Figure 3.

Adjusted incidence rate ratios of days in PrEP adherence lapses of ≥ 3 days from estimated GEE models.

Spearman rank-order correlations showed that TFV-DP values were not related to either the number of stimulant/party drug use days in the two weeks prior to DBS collection (r=−0.11, p>.05) or the number of such drug use days reported across the study period (r=0.23, p>.05). However, 83% of those who reported any stimulant or party drug use had a PrEP lapse of ≥ 3 days at some point during the study and had an average maximum lapse length of 6.0 days, compared to 47.1% and 4.8 days among those who reported no stimulant use. In GEE models, lapse days occurred 4.5 times more often on stimulant drug use days versus days with no such drug use.

4. Discussion

In this study of MSM recruited from a community PrEP clinic, we explored lapses in PrEP adherence (i.e., periods of consecutive missed doses that may be sufficiently long to reduce efficacy) and their overlap with relevant risk behaviors. Results suggested that, even in this older, PrEP-experienced, and highly adherent sample of MSM, adherence lapses of three or more days were common (53%), as was having more than one such lapse over a six-month period (25% overall). Consistent with past studies^10,11, objective markers of PrEP adherence collected at routine follow-up appointments were generally modestly associated with biomarkers and self-reported alcohol use, suggesting that alcohol use might have little association with tenofovir drug levels. However, using digital pill bottles that collected data continuously, our results showed that a higher percentage of those who drank heavily and screened positive for alcohol-related problems had a significant lapse in PrEP adherence and had longer lapses than moderate drinkers without alcohol problems. However, alcohol use (and heavy drinking) did not appear to occur more often specifically during these lapses in adherence. Together, these results suggest that those with heavier patterns of drinking may be at risk for lapses in adherence, possibly due to the instability it creates in their routines, but that drinking on specific days may not necessarily lead to gaps. Marijuana use was not associated with lapses in adherence, either at the person or day-level.

Our results suggest that stimulant drugs may play a more consistent role in adherence lapses. Specifically, our finding that lapse days occurred at a substantially higher rate on stimulant drug use days compared to non-drug use days suggest that stimulant drug use may be involved in clinically-meaningful lapses in PrEP adherence. These findings contrast with others studies that have shown that the use of similar drugs was not associated with tenofovir drug levels^11,53, but should be interpreted with some caution due to the low rates of drug use in this sample. Together, these results suggest that further research focused on understanding the effects of stimulants and other drugs on PrEP adherence is needed.

Our results also showed that sex events involving potential HIV exposure (i.e., engaging in CAS with a non-exclusive or HIV-status unknown partner) did not occur less often during PrEP adherence lapses when compared to periods of more regular adherence. This finding suggests that lapses in PrEP adherence were not likely due to participants taking deliberate pauses at times when they believed they would not be at risk, but rather to problems adhering to PrEP at times when they were still at risk. As such, this may signal the need for interventions that focus on helping MSM adhere to PrEP more consistently, and/or identifying triggers for periods of non-adherence. Our results also showed that, consistent with a number of past studies^54-56, the rate of engaging in high-risk CAS was significantly higher on heavy drinking days specifically during periods of regular PrEP adherence. Alcohol use did not significantly co-occur with CAS that occurred specifically during adherence lapses, suggesting that alcohol use was unlikely to have played a role in these risk windows. Also consistent with past studies^54,55, high-risk CAS events generally occurred at a much higher rate on days when participants reported using stimulant drugs. However, our results also showed that CAS was more likely to occur during a lapse when stimulant drug use also occurred than when stimulant drugs were not used. These findings suggest that use of stimulant drugs could play an important role in promoting potential HIV exposures specifically at times when patients may be less protected by PrEP, and underscore the need for interventions that focus on episodic drug use.

Limitations

Two key limitations should be noted. First, although this study design produced highly-detailed, daily data on adherence and related behaviors to understand the timing of these events, our overall sample size (N = 40) was small. As a result, data on some key predictors (e.g., stimulant/party drug use) were limited, producing sizable standard errors in our models. As such, although the consistent and sizable effects of stimulant use in these models provides strong evidence of their importance, these findings should be interpreted with caution and further research is needed to confirm their magnitude. Second, this sample consisted of a relatively older, well-educated, predominantly white, and largely PrEP-experienced group of MSM. As such, these results offer a relatively conservative picture of PrEP lapses and related risk behaviors. Future research should focus on younger, more diverse, and less PrEP-experienced MSM.

In summary, this study highlights the value of collecting detailed, daily data on PrEP adherence and associated risk behaviors over time to helping researchers understand clinically-meaningful lapses in adherence and how these lapses may affect patients’ risk for HIV. Our results showed that lapses of three or more consecutive missed doses were common, and appeared to be mostly due to problems adhering to PrEP when they were still at risk. Although alcohol use appeared to play little role in producing adherence lapses or promoting risk behaviors that occurred during lapses, stimulant drug use often co-occurred with HIV risk behavior that occurred specifically during PrEP adherence lapses, suggesting that these drugs may facilitate HIV exposure specifically when PrEP is less protective. Future research should focus on understanding the role of adherence lapses, risk behavior, and alcohol/drug use among younger, less PrEP-experienced MSM.