Can a participant-referred “study buddy” increase retention of rural people who use drugs in research? A multi-site, randomized trial

Abstract

Aims:

To test the efficacy of a participant-referred “study buddy” intervention compared with standard retention strategies in retaining rural people who use drugs (PWUD) in longitudinal research.

Design:

Multi-site, two-arm, randomized controlled trial

Setting:

Rural counties in Ohio, Oregon, and Kentucky

Participants:

PROUD-R² enrolled people who were 18 years or older, resided in the study area, and used opioids or injected any drug to get high in the past 30 days between August 2020 and August 2022. Participants (n=739) were 42% female, mostly White (89%) and non-Hispanic (96%), unstably housed (57%), and reported lifetime injection drug use (93%). The most common drugs of choice were methamphetamine (42%) and heroin (38%).

Intervention and Comparator:

Participants were allocated (1:1, stratified by site), to: (1) standard retention approach involving appointment reminders and contact information updates by study staff (n=365), or (2) the intervention arm who received the standard retention approach and were asked to recruit a “study buddy” (n=374). Study buddies were invited to view a video training and instructed to encourage their intervention participant to attend follow-up appointments.

Measurements:

Intervention, control, and study buddy participants completed interviewer-administered surveys at baseline and 6- and 12-months. Outcomes included retention at 12-months (primary) and 6-months (secondary).

Findings:

Retention was 50.3% at 6 months and 46.1% at 12 months. Only 23.5% of intervention participants recruited a study buddy. In intent-to-treat analyses, the intervention did not increase retention at twelve (adjusted OR 1.08, 95% CI 0.79-1.47) or six (AOR 0.96, 95% CI 0.69-1.34) months.

Conclusions:

Recruitment of self-identified “study buddies” does not appear to significantly improve PWUD study retention at 6- and 12-month follow-up relative to standard retention approaches.

Trial registration:

ClinicalTrials.gov NCT03885024.

INTRODUCTION

People who inject drugs (PWID) in rural communities are disproportionately impacted by the substance use disorder (SUD) epidemic and less likely to benefit from advancements in treatment, yet they are rarely involved in clinical trials (1-8). Rural areas face unique challenges, including limited harm reduction services, fewer treatment options, and a scarcity of healthcare providers, which hinder effective intervention (9-11). Substance use stigma can be pronounced in these close-knit communities, with SUDs sometimes viewed as moral failings, which reduces engagement in care (12, 13). Rural PWID also face heightened social vulnerabilities, such as inadequate access to transportation (14), yet healthcare and socioeconomic support systems that are more accessible in urban areas are lacking (3, 7, 15). Rural PWID consequently remain underrepresented in clinical research, leaving a critical gap in the development and implementation of effective, evidence-based interventions to address their unique needs (16, 17).

Clinical trials are the gold standard for testing interventions aimed at reducing drug use and its consequences. Yet, clinical trials, and other longitudinal study designs are disproportionately conducted in urban medical centers, often out of reach for people in rural communities (18-20). In SUD research, poor retention poses a substantial threat to the validity of findings, especially in rural populations. Barriers to retention in rural areas can include stigma, logistical challenges such as scheduling and transportation, and general distrust of the research process, exacerbated by limited cultural competence among investigators in working with rural populations (21, 22). Successful study retention requires addressing these barriers to ensure that interventions targeting drug use in rural areas produce scientifically rigorous, meaningful, generalizable outcomes.

Networks of people who use drugs (PWUD) may hold untapped potential for helping each other engage and continue in SUD research in rural communities by fostering increased rapport and community-connectedness. While peer navigators and peer recovery support specialists can improve treatment engagement and retention in urban SUD treatment settings, these peers are typically people in recovery (23, 24). Little is known about the capacity for other study participants or their associates, regardless of active drug use, to support one another in research participation. Although the ability of participants who use drugs to recruit others who use drugs through respondent-driven sampling is well-established in substance use research (25-27), their ability to enhance study retention is less clear. Understanding how networks of people who use drugs (PWUD) may facilitate sustained study participation could offer a valuable, person-centered solution to overcoming the retention challenges that often undermine SUD research in rural settings.

Aims and hypotheses

The Peer-based Retention Of people who Use Drugs in Rural Research (PROUD-R²) study was a multi-site, randomized trial that tested the effectiveness of an intervention to improve the retention of rural PWUD in research. We hypothesized that participants who partner with a self-identified support person (hereafter, “study buddy”) to assist them in study retention would be more likely to be retained in the study at 6- and 12-months compared to participants who received standard retention approaches.

METHODS

Design

The study protocol has been registered (28) and described in detail elsewhere (29). Briefly, PROUD-R² involved participants in rural Oregon, Ohio, and Kentucky. Staff administered surveys at baseline and 6- and 12-months post-enrollment. Following the baseline survey, staff logged into the REDCap (30, 31) randomization module hosted at Oregon Health & Science University to assign participants to a study arm. Participants were allocated in a 1:1 ratio to two arms, stratified by site to ensure balance at each geographical location.

Participants

Eligibility criteria included being age 18 or older, use of opioids or injection of any drug to get high in the past 30 days and residing in the study area. The PROUD-R² study included three sites encompassing 21 counties in Appalachian Ohio (n=6) and Kentucky (n=12) and southwest coastal Oregon (n=3). All counties were non-metropolitan according to their Rural-Urban Continuum Codes (32) and/or the Oregon Office of Rural Health (33) at the time of recruitment and represented regions with high rates of opioid overdose and hepatitis C (34-37).

In Oregon and Ohio, PROUD-R² participants were recruited through respondent-driven sampling (RDS) (38) in which initial participants, or “seeds”, were recruited through community outreach. They were provided with three to four coupons to refer others and received $10 per eligible referral who enrolled. In Kentucky, community outreach was supplemented by invitations to participants who enrolled in previous studies in the study region and consented to receive contact about future research. Recruitment from prior studies accounted for 54% (n=105) of Kentucky’s analytic sample and involved outreach to participants who had been enrolled in cohort studies for an average of two years (standard deviation [SD]: 1.3).

Enrollment occurred from August 31, 2020 to August 29, 2022, with the recruitment timeline differing somewhat by site (Oregon: August 2020 – May 2022, Kentucky and Ohio: October 2020 – August 2022).

Trial Arms

Participants were randomized to receive one of two retention strategies, described below, hereafter referred to as the control and intervention conditions.

Control Condition

Control arm participants received standard retention outreach by study staff that integrated standard procedures used in longitudinal research with PWUD (39). Participants provided detailed locator form information (29) to assist with retention including name(s), personal contact information, and contact information for up to three people who could help reach the participant if staff were unable to make direct contact. Participants were contacted at 3- and 9-month post enrollment to update locator information ahead of follow-up appointments. Participants received $10 for updating or verifying locator information. Beginning one month prior to participants’ follow-up appointment date, staff began sending them appointment reminders at predefined intervals (29) and, if they were unreachable, contacted their social contacts to update contact information.

Intervention Condition

Intervention arm participants were asked to recruit a “study buddy” in addition to receiving standard retention outreach. The study buddy was a social contact who could remind them of their follow-up appointments and encourage appointment attendance. Participants who referred a study buddy who was eligible and enrolled in the study as a study buddy received $10 for the referral. The “study buddy” intervention design was inspired by the success observed in incentivized participant-participant referral methods for recruitment of rural PWUD (e.g., RDS) (40) and a desire to examine whether that phenomenon could be harnessed and organic relationships among PWUD leveraged to bolster retention when coupled with standard retention protocols.

Study buddies were required to meet the same eligibility criteria as other participants, including past 30-day drug use, until February 21, 2022, when in response to participant feedback, the drug use criterion for study buddies was removed. Most randomized participants (77.5%) were enrolled before the protocol change, though the proportion varied by site due to differences in the pace of enrollment (97.7% in Oregon, 70.6% in Kentucky, and 39.8% in Ohio).

Upon enrollment, study buddies completed a survey (described below) and were invited to view a short training video (41) developed through consultation with a community advisory board of PWUD and peer support specialists with a partner agency. The six-minute video describes the importance of substance use research and retention, and the envisioned “study buddy” role demonstrated through role play. When possible, staff showed the video to study buddies in person, but staff completed many appointments by phone due to COVID-19 social distancing measures and sent a URL to study buddies to view the video online. To capture online viewing, the video was embedded in an online Qualtrics survey that included the following post-video question: “Do you understand how to encourage the person who referred you to stay in the study? [response options: yes, no, maybe]

Staff used the same retention strategy as was used for the control arm to retain study buddies for follow-up surveys. Staff also contacted the study buddy one month prior to the follow-up appointment date of the person who had nominated them as a study buddy and two weeks afterward for missed appointments. Staff encouraged the study buddy to inform the participant that they were due for a follow-up but did not provide the participant’s contact information.

Data collection and measures

Community-based field staff conducted harmonized surveys that were programmed in REDCap, a web-based data collection system (30). Staff administered surveys in-person or by phone based on participant preference and COVID-19 restrictions for university staff. Surveys elicited data on demographic characteristics, housing insecurity, transportation access, and substance use. Staff administered follow-up surveys at 6 months and 12 months post-baseline. Study buddies completed the same data collection procedures as intervention and control participants. Ohio participants received a $40 incentive for each survey, while participants in Kentucky and Oregon received $15 per survey. The amount was higher in Ohio to improve comparability with the Kentucky and Oregon sites which offered participants an opportunity to co-enroll in other ongoing projects that offered additional incentives.

Follow-up surveys asked whether participants received encouragement from social contacts to attend their follow-up assessment; if so, from whom (name and demographic information). This information was used to decipher whether the individual’s referred study buddy encouraged their retention, in which case the study buddy received a $10 incentive.

Adverse Events

In discussion with the Data and Safety Monitoring Board (DSMB), adverse events were defined as death, non-fatal overdose, and social harms (e.g., problems with family, social contacts, or police resulting from participation). Deaths were identified during attempted retention outreach to participants and social contacts and online searches. Staff confirmed deaths by finding obituaries online and ordering death certificates. Overdoses and social harms were assessed through survey items. Data on adverse events were collected at each visit for all participants, including study buddies.

Outcomes

The primary outcome for the trial was the proportion of participants who were retained at 12 months, defined as having completed the 12-month survey between 344 and 390 days (inclusive) post enrollment. The secondary outcome, retention at 6 months, was defined similarly with a window of 164 to 210 days. Those who completed their surveys outside the window were considered not retained.

Blinding

Due to the nature of the interventions, participants and site staff administering the intervention could not Oregon Health & Science University.

Statistical methods

Sample size

The target sample size (n=700) was determined based on the number of participants needed to provide at least 80% power to detect an 11–12.5% increase in retention attributable to the intervention, assuming retention rates in the control arm ranging from 65-75%. Power was estimated via simulation using R version 4.4.1 with the ‘blme’ package (42). Generalized linear mixed models (GLMMs) with logit link and binomial distribution were fit to 2,000 simulated datasets with fixed effect terms for study group and random intercepts for the peer-referral recruitment chains. Chain lengths were simulated based on recruitment data from previous studies at each site (40). Simulations assumed a between-chain variance of 0.05 and two-sided hypothesis test at α=0.05.

Analysis

A total of 739 participants (intervention group n=374, control group n=365) and 89 study buddies were included in the primary analysis (Figure 1). Because the outcome was retention, individuals incapable of retention due to death (n=14), removal by staff for inappropriate behavior during follow-up (n=3), administrative exclusion (n=2), or withdrawal from study (n=1) were not included in the analysis. Study buddies were not included in the analyses as members of either trial arm given that they were the interventionists. Baseline participant characteristics were computed overall, by study arm and site. Descriptive statistics were also used to explore adverse events. Categorical characteristics were reported as frequency with percentage, normally distributed continuous variables as means with standard deviation (SD), and skewed continuous variables as mean and interquartile range (IQR). Standardized mean differences were reported to provide a measure of balance between arms after randomization.

Figure 1.

CONSORT diagram of flow of participants through study

*Participant followed as study buddy and was administratively excluded (n=1);

**Participant followed as control (n=1), participant did not complete baseline questionnaire/participate further and was administratively excluded, and participant randomized to another site (n=1)

Following the approach outlined in our protocol (29), statistical analysis of primary and secondary outcomes followed an intention-to-treat (ITT) analysis strategy whereby participants were classified according to their randomized study arm regardless of their adherence to study activities. The hypotheses being tested were pre-registered in the publication of our protocol (29). To evaluate the association of the intervention with retention at 12 months, we conducted mixed-effects logistic regression with the presence of a 12-month survey as the outcome. The unadjusted model included study group, and the adjusted model included study group, baseline participant demographic characteristics, substance use, problems with transportation, recent incarceration, and housing status. All covariates in the adjusted model except for problems with transportation and recent incarceration were identified a priori in our statistical analysis plan. Due to small cell size, we dichotomized education (less than a high school degree or general education diploma (GED) versus high school degree or GED or more) and combined Asian/Pacific Islanders with race reported as “Other” for the multivariable model. For gender, race, and ethnicity, we used the grand mean as the reference group.

Models for the primary outcome included a random effect for site to address clustering. We assessed whether models should account for RDS chains but found low intraclass correlation (ICC) for the chains at the sites (range: 0.0158-0.0240) using latent variable and simulation approaches to estimation in Oregon and Kentucky (43, 44). The ICC for Ohio was not estimable due to one or more variance components being estimated to be zero or near zero. Odds ratios (OR) and the proportion of participants retained at 12 months along with 95% confidence intervals (CI) and corresponding p-values were estimated from models. Models to assess our secondary outcome showed near zero variation for RDS chains in Ohio and Kentucky but not Oregon. Consequently, a random effect that accounted for RDS chains in Oregon but site for the other states was used. Statistical significance was defined at p<0.05 and tests were two-sided. The magnitude of the three site-specific intervention effects were compared to the overall effect, not to one another, and statistical tests were not reported within or between sites. Complete case analysis was performed for the multivariable models. No interim analyses were planned or conducted during the trial. ITT analyses were conducted using R 4.3.1 lme4 package (45, 46).

We also conducted per-protocol analyses with the per-protocol population defined according to intervention received (i.e., participants who recruited a study buddy were considered exposed to the intervention and those who did not were considered unexposed, regardless of the randomized arm). An “as treated” analysis was conducted, comparing those who recruited a study buddy (n=89) to those who did not (n=650). Additionally, to reduce bias associated with non-random recruitment of study buddies, we used an instrumental variables approach (47) to estimate the “complier average causal effect,” the efficacy of the intervention among those who complied with their allocated treatment assignment. Instrumental variables analyses were conducted using two-stage least squares estimation, adjusted for the same covariate set as the primary analysis, with robust standard errors to account for clustering by study site (48). Although the primary analysis was a logistic regression, because of the non-collapsibility of the odds ratio, instrumental variables analyses (and comparisons with the ITT and per-protocol analyses) are presented on the risk difference scale. A second-stage instrumental variables logistic regression model was conducted to compare with the primary analysis, although results should be interpreted with caution.

We conducted sensitivity analyses not part of the original analysis plan to assess the impact of the protocol change that occurred on February 21, 2021. Chi-squared tests were used to compare the proportion of retained participants in the intervention group pre- and post-protocol change at 6- and 12-months follow-up. We opted not to perform additional analyses given how few participants enrolled after study buddy eligibility changed (n=167), particularly in Oregon (n=9); analyses performed were considered exploratory.

RESULTS

Sample characteristics

Most participants (51%, n=380) enrolled from Oregon, followed by Kentucky (26%, n=195) and Ohio (22%, n=164). Sample characteristics by arm are described in Table 1 and by site in Table A1 in the Supplementary Appendix. Participants’ average age was 39.4 years (SD: 9.7), with most identifying as male (57%), White (89%), and non-Hispanic/Latino (96%). More than half (57%) experienced houselessness and 70% lacked transportation at least once in the past 6 months. The most common “drug of choice” was methamphetamine (42%) followed by heroin (38%). Nearly all (93%) had injected in their lifetime, and 77% had injected at least once in the past 30 days.

Table 1.

Baseline Characteristics of Participants in the Peer-based Retention Of people who Use Drugs in Rural Research (PROUD-R²) study^a

Characteristic	Overall (n = 739)	Study Group		SMD
		Control (n = 365)	Intervention (n = 374)
Site				0.00
Kentucky	195 (26.4)	96 (26.3)	99 (26.5)
Ohio	164 (22.2)	81 (22.2)	83 (22.2)
Oregon	380 (51.4)	188 (51.5)	192 (51.3)
Age, years, mean (SD)	39.4 (9.7)	38.8 (9.4)	39.9 (10.1)	−0.11
Not reported	2	0	2
Gender				0.17
Male	423 (57.2)	205 (56.2)	218 (58.3)
Female	311 (42.1)	156 (42.7)	155 (41.4)
Transgender	4 (0.5)	4 (1.1)	0 (0.0)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Race				0.19
White	661 (89.4)	323 (88.5)	338 (90.4)
African American/Black	9 (1.2)	4 (1.1)	5 (1.3)
American Indian/Alaskan Native	25 (3.4)	15 (4.1)	10 (2.7)
Asian/Pacific Islander/Native Hawaiian	3 (0.4)	2 (0.5)	1 (0.3)
Mixed race	29 (3.9)	18 (4.9)	11 (2.9)
Other	11 (1.5)	3 (0.8)	8 (2.1)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Hispanic/Latino				0.15
Yes	25 (3.4)	14 (3.8)	11 (2.9)
No	710 (96.1)	351 (96.2)	359 (96.0)
Don't know	3 (0.4)	0 (0.0)	3 (0.8)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Education				0.16
< High school	211 (28.6)	97 (26.6)	114 (30.5)
High school diploma/GED	279 (37.8)	141 (38.6)	138 (36.9)
Some college	169 (22.9)	81 (22.2)	88 (23.5)
Associate’s degree	57 (7.7)	34 (9.3)	23 (6.1)
Bachelor’s degree	22 (3.0)	12 (3.3)	10 (2.7)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Currently have driver’s license				0.12
Yes	248 (33.6)	127 (34.8)	121 (32.4)
No	486 (65.8)	235 (64.4)	251 (67.1)
Don't know	4 (0.5)	3 (0.8)	1 (0.3)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Problems with transportation in past 6 months				0.13
Never	216 (29.2)	100 (27.4)	116 (31.0)
1-2 times in past 6 months	72 (9.7)	39 (10.7)	33 (8.8)
3-4 times in past 6 months	88 (11.9)	46 (12.6)	42 (11.2)
At least once a month	72 (9.7)	36 (9.9)	36 (9.6)
At least once a week	128 (17.3)	65 (17.8)	63 (16.8)
Almost every day	160 (21.7)	78 (21.4)	82 (21.9)
Don't know	2 (0.3)	1 (0.3)	1 (0.3)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Homeless in past 6 months				0.08
Yes	422 (57.1)	207 (56.7)	215 (57.5)
No	306 (41.4)	153 (41.9)	153 (40.9)
Refuse to answer	1 (0.1)	0 (0.0)	1 (0.3)
Not reported	10 (1.4)	5 (1.4)	5 (1.3)
Incarceration in past 6 months				0.11
Yes	176 (23.8)	85 (23.3)	91 (24.3)
No	513 (69.4)	257 (70.4)	256 (68.4)
Refuse to answer	2 (0.3)	0 (0.0)	2 (0.5)
Not reported	48 (6.5)	23 (6.3)	25 (6.7)
Frequency of internet use				0.15
Never	42 (5.7)	24 (6.6)	18 (4.8)
Several times a month	56 (7.6)	29 (7.9)	27 (7.2)
Several times a week	56 (7.6)	32 (8.8)	24 (6.4)
About once a day	55 (7.4)	27 (7.4)	28 (7.5)
Several times a day	501 (67.8)	240 (65.8)	261 (69.8)
Refuse to answer	1 (0.1)	0 (0.0)	1 (0.3)
Not reported	28 (3.8)	13 (3.6)	15 (4.0)
Own a smart phone that access internet				0.07
Yes	479 (64.8)	231 (63.3)	248 (66.3)
No	219 (29.6)	114 (31.2)	105 (28.1)
Don't know	2 (0.3)	1 (0.3)	1 (0.3)
Not reported	39 (5.3)	19 (5.2)	20 (5.3)
Ever injected drugs to get high				0.13
Yes	685 (92.7)	344 (94.2)	341 (91.2)
No	53 (7.2)	21 (5.8)	32 (8.6)
Not reported	1 (0.1)	0 (0.0)	1 (0.3)
Average frequency injected any drug				0.29
More than 3 times a day	141 (19.1)	72 (19.7)	69 (18.4)
2-3 times a day	193 (26.1)	86 (23.6)	107 (28.6)
Daily	97 (13.1)	55 (15.1)	42 (11.2)
More than weekly	52 (7.0)	28 (7.7)	24 (6.4)
Weekly	49 (6.6)	21 (5.8)	28 (7.5)
More than once in the past 30 days	25 (3.4)	18 (4.9)	7 (1.9)
Once in the past 30 days	10 (1.4)	7 (1.9)	3 (0.8)
Never	83 (11.2)	33 (9.0)	50 (13.4)
Refuse to Answer	3 (0.4)	2 (0.5)	1 (0.3)
Not reported	86 (11.6)	43 (11.8)	43 (11.5)
Current drug of choiceto get high				0.17
Heroin	284 (38.4)	143 (39.2)	141 (37.7)
Street fentanyl/carfentanil powder	44 (6.0)	21 (5.8)	23 (6.1)
Opiate painkillers	32 (4.3)	13 (3.6)	19 (5.1)
Buprenorphine	10 (1.4)	5 (1.4)	5 (1.3)
Methadone	6 (0.8)	3 (0.8)	3 (0.8)
Prescription anxiety drugs	6 (0.8)	5 (1.4)	1 (0.3)
Cocaine/crack	8 (1.1)	4 (1.1)	4 (1.1)
Methamphetamine/amphetamine	308 (41.7)	149 (40.8)	159 (42.5)
Gabapentin	3 (0.4)	2 (0.5)	1 (0.3)
Otherb	24 (3.2)	14 (3.8)	10 (2.7)
Refuse to answer/not reported	14 (1.9)	6 (1.6)	8 (2.1)

Abbreviations: SMD = Standardized Mean Difference; SD=Standard Deviation, GED = General Education Development test.

^aData are n (%) unless otherwise noted.

^bIncludes marijuana (n=16), alcohol (n=4), and “other” (n=4)

Only 23.5% (n=88) of the 375 intervention participants successfully referred a study buddy into the study, including 27% (n=27) in Kentucky, 35% (n=29) in Ohio, and 17% (n=32) in Oregon. One study buddy was recruited by a participant in the control arm and considered in per-protocol but not ITT analyses. An additional study buddy was enrolled in Ohio but excluded from analyses as they were recruited by a participant who did not meet study criteria. A majority (90%) of study buddies reported past 30-day drug use at baseline. Characteristics of study buddies and index participant-study buddy pairs are described in Tables A2 and A3 in the Supplementary Appendix.

Primary outcome

Overall, 46.1% participants were retained for their 12-month appointment, including 54.4% in Kentucky, 47.6% in Ohio, and 41.3% in Oregon (Table 2). In the ITT analysis (Table 3), retention did not differ by intervention arm in unadjusted (OR: 1.03, 95% CI: 0.77 to 1.38, p=0.84) or adjusted models (adjusted OR (AOR): 1.08, 95% CI: 0.79 to 1.47, p=0.63). On the risk difference scale, assignment to the intervention increased the absolute likelihood of retention by 1.7% (95% CI: −4.9% to 8.2%, p=0.62).

Table 2.

Retention at 6- and 12-months among study buddies and by arm in intent-to-treat and per-protocol analyses

Participant group by analysis type	6-months, n (%)a	12-months, n (%)b
Intent to treat analysis
Intervention group	183/375 (48.8)	174/374 (46.5)
Control group	189/365 (51.8)	167/365 (45.8)
As treated analysis
Recruited a study buddy	49/89 (55.1)	52/89 (58.4)
Did not recruit a study buddy	323/651 (49.6)	289/650 (44.5)
Study buddies	36/89 (40.4)	36/86 (41.9)

^aDenominator at 6 months for control group represents 379 participants randomized to this arm with 14 exclusions (11 died before 6 months, 1 administratively excluded, 1 removed by study staff, and 1 withdrew); for the intervention group represents 380 participants randomized to this arm with 5 exclusions (2 died before 6 months, 1 administratively excluded, and 2 removed by study staff); and for 89 study buddies with no exclusions.

^bDenominator at 12 months for control group represents 379 participants randomized to this arm with 14 exclusions (11 died before 6 months, 1 administratively excluded, 1 removed by study staff, and 1 withdrew); for the intervention group represents 380 participants randomized to this arm with 6 exclusions (2 died before 6 months and 1 died between 6 and 12 months, 1 administratively excluded, and 2 removed by study staff); and for 89 study buddies with 3 exclusions (3 died before 12 months).

Table 3.

Unadjusted and adjusted associations between participant characteristics and participant retention at 6- and 12-months in intent-to-treat analyses¹

	6-months				12-months
	Unadjusted Model		Adjusted Model		Unadjusted Model		Adjusted Model
Characteristics	OR (95% CI)	P	aOR (95% CI)	P	OR (95% CI)	P	aOR (95% CI)	P
Study group
Intervention	0.91 (0.66, 1.24)	0.54	0.96 (0.69, 1.34)	0.81	1.03 (0.77, 1.38)	0.84	1.07 (0.79, 1.46)	0.66
Control			Reference		Reference		Reference
Age			0.98 (0.97, 1.00)	0.08			1.00 (0.98, 1.02)	0.93
Gender
Male			1.55 (0.67, 3.58)	0.31			1.18 (0.54, 2.59)	0.68
Female			1.73 (0.75, 4.01)	0.20			1.55 (0.70, 3.39)	0.28
Transgender3			0.42 (0.09, 2.09)	0.29			0.56 (0.12, 2.59)	0.46
Race
White			1.13 (0.66, 1.93)	0.65			1.29 (0.80, 2.11)	0.30
African American/Black			0.74 (0.20, 2.81)	0.66			0.92 (0.28, 3.01)	0.89
American Indian/Alaskan Native			0.98 (0.40, 2.36)	0.96			1.01 (0.46, 2.23)	0.97
Mixed race			2.25 (0.96, 5.30)	0.06			1.46 (0.69, 3.08)	0.32
Asian, Pacific Islander, Native Hawaiian, or other race			0.61 (0.20, 1.84)	0.38			0.58 (0.21, 1.65)	0.31
Ethnicity
Hispanic			3.00 (0.91, 9.87)	0.07			1.28 (0.49, 3.39)	0.61
Not/Don't Know Hispanic			Reference				Reference
Education
≥ High school degree			1.17 (0.81, 1.70)	0.40			0.94 (0.67, 1.33)	0.73
< High school degree			Reference				Reference
Housing status
Unhoused			0.58 (0.40, 0.85)	<0.01			0.68 (0.48, 0.96)	0.03
Housed			Reference				Reference
Recent incarceration
Yes			0.48 (0.32, 0.72)	<0.01			0.83 (0.57, 1.21)	0.33
No			Reference				Reference
Drug of choice to get high
Illicit opioid			0.88 (0.50, 1.54)	0.66			1.24 (0.73, 2.10)	0.42
Stimulant			0.81 (0.46, 1.45)	0.49			1.25 (0.74, 2.10)	0.41
Other named drug			Reference				Reference
Problems with transportation
At least once a week			0.96 (0.67, 1.36)	0.80			0.93 (0.67, 1.29)	0.66
Not more than once a week			Reference				Reference

Abbreviations: OR=odds ratio, CI=confidence interval, aOR=adjusted odds ratio.

¹The reference for gender and race is the grand mean

²All transgender participants are in the intervention arm.

In the “as treated” analysis, participants who chose to recruit a study buddy were 13.5% (95% CI 3.2% to 23.8%, p=0.001) more likely to be retained, although this effect cannot be attributed to the study buddy alone (Supplemental Appendix Table A4). Using instrumental variables analysis, which examines treatment efficacy among those who engaged with their treatment assignment, recruitment of a study buddy increased the absolute likelihood of retention by 7.4% (95% CI: −17.4% to 32.2%, p=0.56). This corresponds, approximately, to 34% increased odds in retention for those who recruited a study buddy versus those who did not (OR 1.34, 95% CI: 0.47-3.78, p=0.58).

Secondary outcome

Over half of participants (52.3%) who were not retained at 12-months were lost to follow-up prior to their 6-month follow-up appointment. Of the 740 participants included in the secondary outcome analysis of 6-month retention, 372 (50.3%) were retained for their 6-month appointment (Table 2). Six-month retention did not vary by arm in the unadjusted or adjusted model (Table 3; OR: 0.91; 95% CI:0.66 to 1.24, p=0.54 and AOR: 0.96, 95% CI: 0.69 to 1.34, p=0.81, respectively).

Sensitivity analyses

The proportion of intervention participants retained after the protocol change was higher than those recruited before the change at both 6- (51.0% vs. 41.2%) and 12-month (47.1% vs. 46.4%) follow-up periods (Supplemental Appendix Table A5). Neither difference was statistically significant (p=0.11 at 6-months, p=0.91 at 12-months).

Adverse Events

Eighteen deaths occurred among participants, including eleven in the control arm, three in the intervention arm, and four study buddies. Most (n=14, 77.8%) occurred between the baseline and 6-month appointments. The cause of death was obtained through death certificates or obituaries for twelve of the deaths, of which six were drug-related (i.e., overdose, drug intoxication). A total of 111 participants reported a non-fatal overdose during the study (49 in the control arm, 47 in the intervention arm, and 15 study buddies). None of the non-fatal overdoses or deaths were determined by the principal investigators or DSMB to be study related. Three participants reported potential social harms. One intervention participant reported in their 6-month interview that police had witnessed them agreeing to participate in the study and had harassed them, and two participants answered “do not know” when asked whether they had experienced problems/harms related to the study but did not specify the type of problem or harm.

DISCUSSION

The PROUD-R² study tested the effectiveness of a participant-referred “study buddy” intervention to improve study retention among rural PWUD in Oregon, Kentucky, and Ohio. The intervention did not increase participant retention at six or twelve months. Overall retention was suboptimal, consistent with COVID-19 pandemic-related challenges experienced in other clinical trials (49). The study retained 46% of participants at 12 months, overall. Several factors likely contributed to the null findings and suboptimal retention including enrollment of a highly marginalized population, limited intervention uptake, and impact of the COVID-19 pandemic.

Notwithstanding the challenges, retention in this study was similar to that observed in clinical trials for stimulant disorder treatments (50.4%) (50) and within the range of those observed in systematic reviews of National Drug Abuse Treatment Clinical Trials Network (CTN) trials (51, 52). CTN trials achieved retention rates of between 45% and 100% in mostly urban PWUD; few trials had follow-up periods longer than 6 months (51, 52). Future interventions to improve study retention might include contingency management incentives, as used in recent methamphetamine treatment trials in urban settings (53, 54).

Overall low retention rates at 12 months in both groups speak to the challenges inherent in conducting research in highly marginalized communities located in hard-to-reach rural areas. Our study intentionally recruited a highly marginalized, poorly resourced study population who had high rates of houselessness and incarceration (55-58). People who had recently been incarcerated had significantly lower retention at 6-month follow-up and those who were unstably housed had lower retention at 6- and 12-month follow-up compared to their counterparts. In urban settings, outreach at shelters and encampments may assist in overcoming these challenges, but in many rural settings, houselessness is more dispersed and service infrastructure (e.g., shelters, warming centers) is lacking. Despite the challenges, these highly marginalized rural participants are precisely the populations who might benefit the most from interventions being tested in clinical trials and more research is needed to identify effective strategies to overcome barriers to their retention.

Limited intervention uptake also contributed to null study findings: only 24% of intervention participants recruited a study buddy. The intervention originally involved staff showing the training video to study buddies during a baseline in-person appointment. However, when COVID-19 required transition to phone-based data collection, they were asked to watch the video online and uptake was suboptimal. Incomplete data on mode of survey administration prevents us from knowing the exact proportion of remotely-enrolled study buddies who watched the video, but available data indicate a possible range of 39% to 65%. Low uptake of the video combined with the limited number of study buddies recruited precludes analysis of whether the training video bolstered participant-study buddy engagement and retention.

Evidence from this trial suggests that the study’s conceptualization of who might be an effective study buddy may have been misaligned with the retention support that many participants would have preferred. During the trial implementation, participants indicated to staff that other participants who are actively using drugs may not be effective in serving in the “study buddy” role given their own challenges around criminal legal system involvement, houselessness, and ongoing substance use. In harm reduction circles, there is an ongoing debate about the role of recovery in interventions that involve people with lived/living experience with addiction, in which some experts argue that people actively using drugs may more effectively establish rapport and engage others in some harm reduction interventions (59, 60). Participants in our study indicated to staff that they preferred people who were not using drugs to assist them with study retention (e.g., family members or friends who were often in recovery from addiction but not actively using drugs). In response, we changed the study protocol to remove the recent drug use inclusion criterion for study buddies. The change was made too late in the study to examine whether study buddies who were not using drugs were more successful at encouraging retention. While the intervention was informed by investigators’ prior experience working with rural PWUD, the protocol shift might have been made sooner if PWUD were involved in study conceptualization. Future studies should engage PWUD throughout the design process to co-create interventions that optimize alignment of the role of peers or social contacts with study objectives.

Circumstances surrounding the COVID-19 pandemic greatly complicated and transformed participant engagement in clinical trials (49, 61, 62), including substance use research (63-65). Remote work requirements, social distancing measures, and changes in the life circumstances of research staff and participants (e.g., family illness, increased childcare responsibility) required shifts in our research protocols. For example, during the COVID-19 era, study teams were unable to conduct follow-up interviews in the jails due to COVID-19 precautions in place restricting visitation at the jails. This is significant given that previous research among PWUD in the rural communities inclusive of those in the PROUD-R² trial revealed that 42% were incarcerated in the past 6 months (56). While pandemic-driven changes to remote, phone-based data collection had the potential to enhance retention by making participation more convenient and accessible (63), these benefits may have been offset by limitations posed to staff’s ability to build rapport with participants, a factor critical to clinical trial retention (66). Future studies may overcome challenges in retention through the distribution of cell phones, though lost phones and theft among those who were unstably housed may continue to present barriers (65).

Among those who recruited a study buddy, “as treated” sensitivity analysis demonstrated a significant effect on study retention at 12-month follow-up; however, this finding is likely influenced by unmeasured confounders. The instrumental variables analysis, which estimates the unbiased intervention effect among those compliant to their intervention arm assignment, suggested an increase in retention attributable to the intervention of about 5-6%. Although the instrumental variables analysis has a very wide confidence interval, if the point estimate is assumed to be accurate, future retention interventions with greater adherence might be expected to have a modestly beneficial effect.

While the study had many strengths including its multi-site design, large sample size, and inclusion of highly marginalized participants often underrepresented in clinical trials, there were methodological limitations. While the inclusion of three, distinct geographic sites increases generalizability, rural populations are not monolithic and results may not be generalizable to all rural communities. The retention protocol may have been strengthened through the inclusion of more mailings and home visits following recommended protocols for longitudinal research among PWUD (39). Due to the COVID-19 pandemic and variations in institutional policies across universities, home visits for appointment reminders were restricted and the mode of administration for the surveys varied over time and across sites. Participants’ ability to participate in the study may have varied by mode of administration. The recruitment approach also varied; over half (54%) of participants in Kentucky were recruited from among those who had recently co-enrolled in another cohort study and had already been “retained” to some extent which may have contributed to the increased retention rate in Kentucky compared to other sites. In contrast, Oregon exclusively enrolled new participants from high-risk settings, which likely contributed to a lower retention rate. Finally, the absence of qualitative data on study buddy engagement presents a limitation in understanding limited uptake of the intervention.

Conclusion

The PROUD-R² study demonstrates the challenges of retaining rural PWUD in research studies. Engagement of “study buddies” did not improve retention, likely due to limited uptake and inclusion of study buddies with active drug use. Even outside the context of a pandemic, approximately 73% are retained in CTN studies of PWUD (52) and 50% are retained in trials of stimulant use disorder treatment (50). Based on findings of this trial, an intervention that involves pairing PWUD to encourage each other’s retention in a trial is unlikely to be effective in encouraging retention of rural PWUD in research. Participant feedback indicates that interventions that rely on individuals in recovery or other support people who do not use drugs may increase the effectiveness of such an approach in future research.

Data Sharing Statement

De-identified data will be made available upon written request to Co-Principal Investigator, P. Todd Korthuis.