Integrated care van delivery of evidence-based services for people who inject drugs: a cluster-randomized trial

Abstract

Background and Aims:

People who inject drugs (PWID) are at risk for adverse outcomes across multiple dimensions. While evidence-based interventions are available, services are often fragmented and difficult to access. We evaluated the effectiveness of an integrated care van (ICV) that offered services for PWID.

Design:

Cluster-randomized trial

Setting:

Baltimore, Maryland

Participants:

Prior to randomization, we used a research van to recruit PWID cohorts from 12 Baltimore neighborhoods (sites), currently served by the city’s mobile needle exchange program.

Intervention and Comparator:

We randomized sites to receive weekly visits from the ICV (n=6) or to usual services (n=6) for 14 months. The ICV offered case management, buprenorphine/naloxone, screening for HIV, hepatitis C virus, and sexually transmitted infections, HIV pre-exposure prophylaxis, and wound care.

Measurements:

The primary outcome was a composite harm mitigation score that captured access to evidence-based services, risk behaviors, and adverse health events (range 0 to 15, with higher numbers indicating worse status). We evaluated effectiveness by comparing changes in the composite score at 7 months versus baseline in the two study arms.

Findings:

We enrolled 720 cohort participants across the study sites (60 per site) between June 2018 and August 2019: 38% women, 73% black, 85% urine drug test positive for fentanyl. Over a median of 10 months, the ICV provided services to 734 unique clients (who may or may not have been cohort participants) across the 6 intervention sites, including HIV/hepatitis C virus testing in 577 (79%) and buprenorphine/naloxone initiation in 540 (74%). However, only 52 (7.2%) of cohort participants received services on the ICV. The average composite score decreased at 7 months relative to baseline, with no significant difference in the change between ICV and usual services (difference in differences: −0.31; 95% confidence interval: −0.70, 0.08; P=0.13).

Conclusions:

We found no significant difference in composite score changes in PWID recruited from neighborhoods assigned to the ICV or to usual services. The ICV successfully served large numbers of clients, but unexpectedly low use of the ICV by cohort participants limited our ability to detect meaningful differences.

INTRODUCTION

The toll from injection drug use extends over several dimensions, including the psychosocial disruption of addiction, life-threatening acute and chronic infections, and, most pressingly, fatal overdose. In 2021, 106,699 people died from drug overdose in the United States, the highest number ever recorded and a 14% increase over the preceding year (1). While evidence-based interventions are available for people who inject drugs (PWID) – including needle exchange programs (NEP), medications for opioid used disorder (MOUD), hepatitis C virus (HCV) cure with direct acting agents, and HIV pre-exposure prophylaxis (PrEP) – increasing the availability of and access to these services has been a stubborn challenge. In 2019, only an estimated 15% persons in the US with past year opioid use disorder received MOUD (2), with important and persistent racial disparities in access (3).

Data suggest that integration and co-location of substance use disorder and medical treatment services, typically in clinic settings, may improve outcomes relevant to PWID (4–8). Recently, the National Academies of Science, Engineering, and Medicine published a consensus report proposing ways to better integrate services for opioid use disorder and infectious diseases (4). Mobile clinics offer a further opportunity to meet people where they are and provide services. However, few studies have formally evaluated the effectiveness of mobile PWID service delivery.

The launch and evaluation of a PWID-focused integrated care van (ICV) was a public health-academic partnership between the Baltimore City Health Department and Johns Hopkins University. The ICV visited neighborhood sites weekly and offered a broad range of PWID-focused services in a welcoming atmosphere (9, 10). The objective of the study was to determine if the availability of ICV services improved neighborhood-level PWID outcomes. We hypothesized that service delivery by the ICV would improve access to services, reduce risky behavior, and improve health outcomes among PWID in the neighborhood, compared with neighborhoods assigned to usual services.

MATERIALS AND METHODS

Study design

To compare the effectiveness of the ICV intervention with usual services over 14 months, we conducted a 2-arm, matched-pair, cluster-randomized trial (Figure 1) in 12 drug- and overdose-affected neighborhoods in Baltimore, Maryland. We chose this design, in which clusters (neighborhood sites) were randomized rather than individual participants, to avoid denying ICV services to individuals according to randomized treatment assignment. This trial is registered with ClinicalTrials.gov (study identifier NCT03567174).

Figure 1.

CONSORT flow diagram for cluster randomized trial comparing integrated care van to usual services for people who inject drugs, Baltimore, Maryland, June 22, 2018, through March 20, 2020. NEP, needle exchange program.

Site selection, pairing, and randomization

We considered approximately 16 neighborhood sites in Baltimore that were known for drug use activity and were served by the city’s mobile NEP. With the goal of minimizing contamination, we considered distance between sites and health department data on the degree of client overlap between sites. For sites that were particularly close to one another or had high client overlap, we excluded one of the sites. After 12 study sites were selected (Supplemental Figure 1), we matched sites into pairs, considering numbers of clients served at individual sites and client demographics. To protect confidentiality, we used codes for individual neighborhood sites (N1-N12) throughout the manuscript. Once site pairs were finalized, a statistician, who was not involved in the trial, used a statistical software package to randomly assign one site in each pair to the intervention and the other to usual services. We enrolled study cohorts from 4 sites (2 pairs) at a time in 3 phases (Supplemental Figure 2). As we completed cohort enrollment in each pair, the statistician announced the site assignments for the pair to the researchers so that intervention roll-out could begin. The nature of the intervention precluded masking.

Study participants

Prior to randomization, we recruited cohorts of 60 participants per site. We used a dedicated research van to recruit participants from the 12 neighborhoods, using study flyers and word-of-mouth. Inclusion criteria were i) 18 years of age or older, ii) history of injection drug use, and iii) injection drug use on 4 or more days in the prior 30 days, or needle/syringe sharing in the prior 6 months, or confirmed HIV-positive status (we included this last criterion to maximize inclusion of HIV-positive persons with any history of injection drug use, even if not using currently). Exclusion criteria included i) incompetence to provide informed consent or participate in the study, and ii) unwilling or unable to provide a blood sample. Participants provided written informed consent and the study was approved by the Johns Hopkins Medicine Institutional Review Board and by a Public Health Review by the Baltimore City Health Department.

Procedures

Neighborhood sites assigned to the intervention arm received weekly ICV visits and continued to receive standard NEP services. The ICV was a 36-foot recreational vehicle that was outfitted with two exam rooms, a phlebotomy space, a bathroom, and a waiting area for clients. ICV clients were referred by the mobile NEP and by word of mouth. Each session was staffed by a multidisciplinary team that included two clinicians (physicians or nurse practitioners), a case manager, and a certified phlebotomist. The ICV provided services free-of-charge under the auspices of the Baltimore City Health Department. Clients did not need identification or insurance to receive services. Services were targeted to people who use drugs, but anyone could receive services on the ICV. Case managers conducted a social needs assessment of all new clients and assisted with insurance enrollment and linkage to primary care or mental health services. The ICV offered the following services: i) point-of-care serologic testing for HIV and HCV, ii) buprenorphine/naloxone (BUP/NX) initiation and maintenance, iii) sexually transmitted infection screening/treatment, iv) antiretroviral therapy and direct acting agents for clients with HIV and HCV infection, respectively, v) PrEP, vi) naloxone overdose kits, and vii) wound care (9). Clinicians on the ICV were waivered to prescribe BUP/NX and were trained in the management of sexually transmitted infections, PrEP, HIV, and HCV treatment.

Neighborhood sites assigned to usual services continued to receive standard mobile NEP van visits at least weekly. In addition to clean needles, the NEP provided naloxone overdose kits. Access to MOUD and low-income primary health care is generally good in Baltimore City. The ICV did not provide services in these neighborhoods.

Participant follow-up

Cohort participants completed a baseline visit and follow-up visits at 7 and 14 months in the research van. We used multiple approaches to foster retention, including reminder letters, phone calls, check-in visits between study visits to update contact information, and a missed visit tracking protocol. Study visits included biometric identification, a behavioral and service access survey administered by trained interviewers, a blood draw, and a urine sample. At the baseline visit participants had point-of-care HIV testing (INSTI HIV-1/HIV-2 Antibody Test, bioLytical Laboratories, Richmond, BC, Canada) with pre-and post-test counseling. Other laboratory testing included HCV serology (Roche Cobas e801 analyzer with anti-HCV Generation II reagent, Roche Diagnostics, Rotkreuz, Switzerland), HCV RNA (Abbott RealTime HCV Viral Load kit, Abbott Laboratories, Abbott Park, IL, USA and Roche cobas 6800 with Roche cobas reagent, Roche Diagnostics, Rotkreuz, Switzerland), HIV RNA (RealTime HIV-1, Abbott Laboratories, Abbott Park, IL, USA), CD4 cell counts by flow cytometry, and urine testing for a panel or more than 40 drugs and metabolites by targeted liquid chromatography-tandem mass spectrometry.

Outcomes

The ICV provided several distinct PWID-focused evidence-based services. We developed an a priori composite harm mitigation score to capture and quantify PWID indicators across multiple domains (Table 1). The scoring rubric was based on consideration of World Health Organization guidance for evidence-based PWID services (11), a predictive risk model for HIV seroconversion among PWID (12), gaps in the HCV care continuum (13), and overdose risk (14). The score ranged from 0 to 13 or 15 (depending on HIV and HCV status at baseline), with higher scores indicating a poorer status. The composite harm mitigation score included factors in three areas: 1) use of beneficial services (NEP, MOUD, recent HIV or HCV testing if not known to be positive, engagement in the HIV or HCV care continua if infected, etc.), 2) risky behaviors (injection drug use, sharing injection equipment), and 3) adverse events. There were two categories of adverse events in the scoring rubric. The first category included events that could be assessed cross-sectionally at any visit (e.g., recent non-fatal overdose, emergency department use) and a second category of events that could only be assessed at follow-up visits with a change in status from baseline (HIV seroconversion, HCV seroconversion, and death).

Table 1.

Components and definitions of the composite harm mitigation score¹

Score component	Applicable group (denominator)	Scores and category definitions
Service access
HIV care continuum	HIV-positive	0 – Suppressed viral load (HIV RNA <200 c/mL) 1 – Not suppressed and either i) took antiretroviral therapy in prior 30 days or ii) had visit with HIV provider in prior 6 months 2 – Not suppressed and did not take antiretroviral therapy in prior 30 days and did not have visit with HIV provider in prior 6 months
HIV testing	HIV-negative	0 – Had HIV test in past 6 months 1 – Did not have HIV test in past 6 months
PrEP use	HIV-negative	0 – Used PrEP in prior 6 months 1 – Did not use PrEP in prior 6 months
HCV care continuum	HCV-positive i. Detectable HCV RNA or ii. HCV seropositive, undetectable HCV RNA, and history of HCV treatment	0 – Prior HCV treatment and undetectable HCV RNA (<15 IU/L) 1 – Detectable HCV RNA and either treated for HCV or evaluated by HCV provider in prior 6 months 2 – Detectable HCV RNA and not treated for HCV or evaluated by HCV provider in prior 6 months.
HCV testing	HCV-negative i. HCV seronegative or ii. HCV seropositive, undetectable HCV RNA, and no history of HCV treatment	0 – Had HCV test in past 6 months 1 – Did not have HCV test in past 6 months
MOUD use	All	0 – Used MOUD in past 6 months 1 – Did not use MOUD in past 6 months
NEP use	Injected in prior 6 months	0 – Used NEP in past 6 months 1 – Did not use NEP in past 6 months
Available naloxone kit	All	0 – Has naloxone kit on person or where drugs used 1 – Does not have accessible naloxone kit
Risk behaviors
Injection drug use	All	0 – No injection drug use in prior 6 months 1 – Injection drug use in prior 6 months
Recent drug use	All	0 – Urine drug test negative for drugs2 1 – Urine drug test positive for one or more drugs2
Sharing injecting equipment	All	0 – No sharing syringe/works or not injecting in prior 6 months 1 – Sharing works (cotton/cooker) only in prior 6 months 2 – Sharing needle/syringe in prior 6 months
Adverse outcomes
Non-fatal overdose	All	0 – No overdose in prior 6 months 1 – One or more overdose in prior 6 months
Emergency department use	All	0 – No emergency department visits in past 6 months 1 - One or more emergency department visits in past 6 months
Clinical status change (assessed at follow-up only)
HIV seroconversion	HIV-negative	0 – No HIV seroconversion 2 - HIV seroconversion occurring between baseline and follow-up
HCV seroconversion	HCV-negative	0 – No HCV seroconversion 1 - HCV seroconversion occurring between baseline and follow-up
Death	All	0 – Alive Maximum points – Death confirmed by National Death Index

PWID, people who inject drugs; HCV, hepatitis C virus; PrEP, pre-exposure prophylaxis; MOUD, medication for opioid use disorder; NEP, needle exchange program.

¹Score ranges from 0 to 13 or 15 (depending on HIV and HCV status) with higher scores indicating worse status.

²Includes fentanyl (or metabolite), heroin (morphine or 6-monoacetylmorphine), cocaine (or metabolite), amphetamine, or methamphetamine.

Most scoring components had a binary categorization (i.e., 0 or 1) and no component had more than three categories (i.e., 0, 1, or 2). Most components were based on self-report, but objective data were included whenever feasible. For example, successfully treated HIV and cured HCV infections were determined by HIV RNA and HCV RNA measurements, respectively, and deaths were confirmed in the National Death Index. Recent drug use was determined by urine drug testing. The primary outcome was the difference in the composite harm mitigation score (7-month follow-up minus baseline), with differences below and above zero indicating favorable and unfavorable changes, respectively. Secondary outcomes included individual components of the composite score.

We determined the number of research participants that received services from the ICV during follow-up by matching names and dates of birth with clients that visited the ICV (at any site). As the ICV was a service-based intervention, we did not systematically query and distinguish recent injection vs. non-injection drug use in ICV clients. However, to estimate the percentage of ICV clients that were of PWID with recent injection, we defined recent injection drug use as either a documented visit to the mobile NEP 3 months before or after their encounter (by record match) or clinician mention of injection drug use in the intake history and physical.

Modifications due to the Coronavirus disease 2019 (COVID-19)

The COVID-19 pandemic, which struck in March 2020, led to a prolonged suspension of both the ICV intervention and follow-up visits on the research van due to the inability to attain social distancing on the vans. The ICV continued to serve clients though telemedicine and with in-person visits to the health department clinic as needed, particularly to maintain continuity for clients who were receiving BUP/NX (15). To the extent possible, we attempted to follow research participants during COVID-19, initially by telephone, and later by resuming in-person follow-up visits on the research van (Supplemental Figure 2). Originally, the primary outcome for the study was to be assessed at both the 7-month and the 14-month follow-up visits, using methods for repeated observations. At the time research activities were suspended, we had completed 7-month follow-up visits at all sites but had completed 14-month visits at only 4 of 12 sites. Consequently, we revised our statistical analysis protocol and restricted the primary outcome assessment to data collected at the 7-month visit to permit a comparison unaffected by COVID-19.

Power and statistical analysis

To estimate the power to correctly reject a null hypothesis, we used simulation methods assuming a normally distributed composite harm mitigation score. We adapted the methods of Arnold et al. (16) for using mixed effects logistic regression with random intercepts for cluster (SD = .22) and participant (SD = 1.05) with one baseline and one follow-up assessment, 80% retention, and a similar treatment effect at each follow-up. With an initial sample size of 720 participants, we would have 80% power if the true person-level composite scores were 0.33 standard deviations higher in one arm compared with the other, controlling for composite scores at baseline, random variation across study sites, and random variation across individuals.

The effect of the intervention on the composite score was estimated using mixed effects linear regression with data from the baseline assessment and 7-month follow-up assessment. The model included random intercepts for study participant and study site, and fixed effects for group assignment, assessment, and group-by-assessment interaction term. The interaction term provided the estimate of the effect of assignment to the intervention group on the composite score. Although we randomized by matched-pairs for logistical reasons, we “broke the pairing” in the analysis by omitting a coefficient for matched-pair. This approach can be more statistically powerful when the matching accounts for little variability among clusters (17). In a sensitivity analysis, we also estimated the effect of the intervention on the composite score using an analysis of covariance (ANCOVA) approach, as described in the supplementary materials.

A variety of statistical approaches were used to estimate the effects of the intervention on the 16 secondary outcomes based on the nature and distribution of the outcomes. Associations between group assignment and follow-up only outcomes (HIV seroincidence, HCV seroincidence, and mortality) were evaluated using Fisher’s exact test and reporting exact 95% CIs. Dichotomous and trichotomous secondary outcomes measured at baseline and follow-up were analyzed using mixed effects logistic regression and ordinal logistic regression, respectively, with the same random effects and fixed effects as in the primary outcome analysis. PrEP use in the prior six months could not be analyzed using mixed effects logistic regression due to the very small number of PrEP users, and the odds of PrEP use by group assignment at follow-up was evaluated using Fisher’s exact test and reporting Cornfield 95% CIs (18). We conducted sensitivity analyses in which the baseline composite score and other baseline factors, which were associated with the outcome, were included as covariates in mixed models. We also repeated analyses using multiply imputed datasets to explore whether differential attrition affected outcome analyses. We used Stata software (version 17; StataCorp, College Station, TX, USA) for statistical analyses.

RESULTS

Participants and follow-up

Between June 22, 2018, and August 14, 2019, we recruited 720 participants, 60 at each of the 12 neighborhood sites. Participants in the two arms were similar by gender and race, although participants in the ICV arm were older than those in the usual services arm (Table 2). Almost all participants reported injecting in the prior 30 days; over half reported injecting daily, and approximately one-third reported receptive needle/syringe sharing in the prior 6 months. HCV seropositivity was higher in the ICV arm than in the usual services arm (66.7% vs. 53.1%). HIV prevalence was similar in the two arms (−11%). The mean composite mitigation score was 7.35 overall and was similar in the two arms. Supplemental Table 1 shows baseline characteristics stratified by site.

Table 2.

Baseline characteristics of participants enrolled in a cluster-randomized trial evaluating an integrated care van for people who inject drugs, Baltimore, Maryland, June 2018 to August 2019

Characteristic1	Overall (n=720)	Usual services (n=360)	Integrated care van (n=360)
Age, years	50 (39, 55)	48 (38, 54)	51.5 (41, 57)
Gender2
Male	444 (61.7)	218 (60.6)	226 (62.8)
Female	276 (38.3)	142 (39.4)	134 (37.2)
Race
Black	523 (72.6)	253 (70.3)	270 (75.0)
White	163 (22.6)	88 (24.4)	75 (20.8)
Other	34 (4.7)	19 (5.3)	15 (4.2)
Hispanic	17 (2.4)	10 (2.8)	7 (1.9)
Marital status
Never married	457 (63.5)	226 (62.8)	231 (64.2)
Married	90 (12.5)	53 (14.7)	37 (10.3)
Divorced/separated	134 (18.6)	63 (17.5)	71 (19.7)
Widowed	39 (5.4)	18 (5.0)	21 (5.8)
Education
Less than high school	267 (37.1)	123 (34.2)	144 (40.0)
High school graduate or equivalency	303 (42.1)	160 (44.4)	143 (39.7)
At least some college	150 (20.8)	77 (21.4)	73 (20.3)
Currently employed	91 (12.6)	45 (12.5)	46 (12.8)
Currently have health insurance	654 (90.8)	319 (88.6)	335 (93.1)
Current housing
Own house or apartment	347 (48.3)	155 (43.2)	192 (53.3)
Staying with family or friends	294 (40.9)	164 (45.7)	130 (36.1)
Residential drug treatment	10 (1.4)	4 (1.1)	6 (1.7)
Homeless	68 (9.5)	36 (10.0)	32 (8.9)
High risk alcohol use3	363 (53.0)	177 (50.3)	186 (55.9)
Age when first injected drugs, years	21 (17, 27)	22 (17, 29)	21 (17, 26)
Injected drugs in prior 30 days	707 (98.2)	354 (98.3)	353 (98.1)
Number of days injected in prior 30 days	30 (21, 30)	30 (23, 30)	30 (20, 30)
Used a needle or syringe after someone else in prior 6 months	256 (35.6)	122 (33.9)	134 (37.2)
Receiving MOUD	261 (36.3)	97 (26.9)	164 (45.6)
Methadone	213 (29.6)	74 (20.6)	139 (38.6)
Buprenorphine	48 (6.7)	23 (6.4)	25 (6.9)
Used HIV PrEP in prior 6 months	5 (0.7%)	3 (0.8%)	2 (0.6%)
Urine drug test positivity
Fentanyl4	613 (85.1)	330 (91.7)	283 (78.6)
Heroin4	258 (35.8)	147 (40.8)	111 (30.8)
Cocaine4	573 (79.6)	288 (80.0)	285 (79.2)
Amphetamine or methamphetamine	12 (1.7)	5 (1.4)	7 (1.9)
Methadone4	322 (44.7)	121 (33.6)	201 (55.8)
Buprenorphine4	29 (4.0)	14 (3.9)	15 (4.2)
HCV seropositive	431 (59.9)	191 (53.1)	240 (66.7)
HCV RNA detectable5	290 (65.9)	117 (59.4)	173 (71.2)
HIV-positive	81 (11.2)	40 (11.1)	41 (11.4)
HIV RNA >200 copies/mL6	37 (46.0)	19 (48.0)	18 (44)
Composite harm mitigation score, mean (SD)7	7.35 (1.82)	7.32 (1.89)	7.37 (1.74)

HCV, hepatitis C virus; MOUD, medication for opioid use disorder; PrEP, pre-exposure prophylaxis; PWID, people who inject drugs; SD, standard deviation

¹Categorical and continuous variables shown as n (%) and median (25^th percentile, 75^th percentile), respectively, unless otherwise specified.

²Includes male-to-female and female-to-male transgender persons.

³High-risk alcohol use defined by score ≥ 3 for women or ≥ 4 for men on Alcohol Use Disorders Identification Test-Concise (AUDIT-C).

⁴Detection of fentanyl or norfentanyl considered fentanyl positive; detection of morphine or 6-acetylmorphine considered heroin positive; detection of cocaine or benzoylecgonine considered cocaine positive; detection of methadone or 2-ethlene-1,5-dimethyl-3,3-diphenylpyrrolidine (EDDP) considered positive for methadone; detection of norbuprenorphine or buprenorphine-glucuronide considered positive for buprenorphine.

⁵Percentages reflect number with detectable HCV RNA among HCV seropositive individuals.

⁶Percentages reflect number with HIV RNA >200 copies/mL among HIV seropositive individuals.

⁷Score based on service access, risk behaviors, and adverse events relevant to PWID (Table 1) with a range of 0 to 13 or 15 (depending on HIV and HCV status), where higher scores indicate poorer service access/risk/health outcome indicators

Over three-quarters of participants had urine drug tests positive for fentanyl. Urine drug test positivity for both fentanyl and heroin were lower in the ICV arm than in the usual services arm. Among cohort participants, 6.7% and 29.6% reported currently attending a MOUD program based on BUP/NX and methadone, respectively. Similarly, only 4.0% had baseline urine toxicology positive for BUP (or metabolites), while 44.7% were positive for methadone (or metabolite). Participants in the ICV arm were more likely than those in usual services to report current engagement in MOUD (with methadone specifically) and to have urine drug testing positive for methadone.

Participants were followed through March 2020, when routine follow-up was disrupted by COVID-19. Seven-month follow-up, which was not affected by COVID-19, was 75% and 69% in the ICV and usual services arms, respectively (Figure 1). In contrast, only approximately 20% of participants completed the 14-month follow-up visit due to large numbers of censored visits.

Integrated care van service delivery

The ICV began providing services to individual neighborhood sites (5-hour session once a week per site) in a gradual roll-out between September 2018 and July 2019. ICVs provided services for a median of 10.4 months (site range: 3.9, 18.2) as of suspension of activities in March 2020 (Table 3). ICV clients accrued quickly at 5 of the 6 intervention sites. However, the N3 site failed to attract a substantial number of clients despite multiple ICV visits to the site. This ICV intervention was discontinued at N3 after 6.9 months, although the site remained in the analysis. Across the 6 intervention sites, the ICV provided services to 734 unique individuals (site range: 33, 248). The demographics of ICV clients – median age 48 years, 33% women, 22% white – were like those of research participants. However, only 126 (17.2%) ICV clients met criteria for recent injection drug use. This differed from the research cohort, where virtually all participants were actively injecting drugs, consistent with inclusion criteria.

Table 3.

Characteristics of clients and services provided by the integrated care van in 6 neighborhoods in Baltimore, Maryland from September 2018 to March 2020.

Characteristics and services	Overall	N1	N31	N6	N7	N10	N12
ICV factors
Date services started	SEP 20182	SEP 2018	SEP 2018	NOV 2018	FEB 2019	JUL 2019	NOV 2019
Date services ended	MAR 20202	MAR 2020	APR 2019	MAR 2020	MAR 2020	MAR 2020	MAR 2020
Months of service, n	18.32	18.2	6.9	15.4	12.4	8.3	3.9
Total client visits, n	4,0782	1,535	68	1,139	683	450	203
Unique clients, n	7342	248	33	184	126	94	49
Client characteristics
Age (years), median (P25, P75)	48.4 (34.9, 55.5)	45.0 (33.6, 54.9)	40.8 (31.8, 55.0)	48.4 (36.4, 55.0)	50.4 (40.0, 56.8)	51.0 (36.8, 56.4)	50.9 (34.7, 56.4)
Female, n (%)	242 (33.0)	82 (33.1)	19 (57.6)	65 (35.3)	34 (27.0)	27 (28.7)	15 (30.6)
Race, n (%)
Black	573 (78.1)	151 (60.9)	18 (54.6)	171 (92.9)	103 (81.7)	89 (94.7)	41 (83.7)
White	145 (19.7)	89 (35.9)	14 (42.4)	9 (4.9)	20 (15.9)	5 (5.3)	8 (16.3)
Other	16 (2.2)	8 (3.2)	1 (3.0)	4 (2.2)	3 (2.4)	0 (0.0)	0 (0.0)
Hispanic, n (%)	6 (0.8)	4 (1.6)	0 (0.0)	0 (0.0)	2 (1.6)	0 (0.0)	0 (0.0)
Visits attended, n (%)
1–2	333 (45.4)	102 (41.1)	29 (87.8)	73 (39.7)	65 (51.6)	39 (41.5)	25 (51.0)
3–8	256 (34.9)	90 (36.3)	2 (6.1)	70 (38.0)	34 (27.0)	43 (45.7)	17 (34.7)
>8	145 (19.7)	56 (22.6)	2 (6.1)	41 (22.3)	27 (21.4)	12 (12.8)	7 (14.3)
Active injection drug use3	126 (17.2)	80 (32.3)	7 (21.2)	12 (6.5)	7 (5.6)	18 (19.2)	2 (4.1)
Unique clients receiving services
POC HIV test, n (%)	577 (78.6)	177 (71.4)	22 (66.7)	146 (79.4)	110 (87.3)	79 (84.0)	43 (87.8)
HIV positive test, n (%)	19 (2.6)	3 (1.2)	1 (3.0)	2 (1.1)	4 (3.2)	3 (3.2)	6 (12.2)
New HIV diagnosis, n (%)	3 (0.4)	0 (0.0)	1 (3.0)	1 (0.5)	1 (0.8)	0 (0.0)	0 (0.0)
Treated for HIV on ICV, n (%)	4 (0.5)	4 (1.6)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
HCV antibody test, n (%)	525 (71.5)	153 (61.7)	18 (54.6)	145 (78.8)	102 (81.0)	73 (77.7)	34 (69.4)
Positive HCV antibody, n (%)	119 (16.2)	50 (20.2)	2 (6.1)	18 (9.8)	28 (22.2)	13 (13.8)	8 (16.3)
HCV RNA measured, n (%)	109 (14.9)	43 (17.3)	3 (9.1)	14 (7.6)	27 (21.4)	10 (10.6)	12 (24.5)
HCV RNA detectable, n (%)	70 (9.5)	25 (10.1)	2 (6.1)	8 4.4)	19 (15.1)	8 (8.5)	8 (16.3)
Treated for HCV on ICV, n (%)	21 (2.9)	11 (4.4)	1 (3.0)	1 (0.5)	5 (4.0)	1 (1.1)	2 (4.1)
Screened for STI4, n (%)	602 (82.0)	190 (76.6)	20 (60.6)	162 (88.0)	108 (85.7)	80 (85.1)	42 (85.7)
Tested positive for STI, n (%)	60 (8.2)	20 (8.1)	3 (9.1)	18 (9.8)	9 (7.1)	7 (7.5)	3 (6.1)
Treated for STI, n (%)	42 (5.7)	15 (6.1)	1 (3.0)	12 (6.5)	7 (5.6)	5 (5.3)	2 (4.1)
Wound care provided, n (%)	35 (4.8)	19 (7.7)	1 (3.0)	8 (4.4)	4 (3.2)	1 (1.1)	2 (4.1)
PrEP prescribed, n (%)	18 (2.5)	6 (2.4)	0 (0.0)	5 (2.7)	6 (4.8)	1 (1.1)	0 (0.0)
Naloxone overdose kit distributed, n (%)	342 (46.6)	73 (29.4)	10 (30.3)	94 (51.1)	70 (55.6)	58 (61.7)	37 (75.5)
Initiated BUP/NX, n (%)	540 (73.6)	177 (71.4)	16 (48.5)	152 (82.6)	92 (73.0)	70 (74.5)	33 (67.4)

N1, N3, N6, N7, N10, N12: neighborhood sites assigned to integrated care van; ICV, integrated care van; P₂₅ and P₇₅, 25^th percentile and 75^th percentile, respectively; POC, point-of-care; HCV, hepatitis C virus; STI, sexually transmitted infection; PrEP, pre-exposure prophylaxis; BUP/NX, buprenorphine/naloxone

¹Site N3 was slow to accrue clients and service to this site was discontinued approximately 6 months after initiating.

²Median across sites

³Active injection drug use defined as notation of recent injection drug use in the baseline ICV clinical assessment or a visit to the health department’s syringe services program within 3 months before or after an ICV visit.

⁴Screening for sexually transmitted infection included serologic syphilis testing and urine nucleic acid amplification for N. gonorrhea and C. trachomatis

Over 70% of clients visiting the ICV were tested for HIV, HCV, and sexually transmitted infections (9). Nearly half of clients were given naloxone overdose kits. Remarkably, 74% of clients initiated BUP/NX and the ICV quickly became principally identified for this service among clients (10). Other services that were provided to smaller numbers of clients included HCV treatment with direct acting agents (n=21), PrEP prescription (n=18), and wound care (n=35).

Only 52 of 720 (7.2%) study participants received services from the ICV: 35 (9.7%) recruited from intervention sites and 17 (4.7%) recruited from usual service sites. The number of study participants that visited the ICV ranged from 0 to 11 (18.3%) across sites (Supplemental Table 2).

Primary outcome

Compared with baseline, the composite harm mitigation score declined in both arms (indicating a favorable change in status) by the 7-month visit: −1.93 and −1.63, in ICV and usual services arms, respectively (Figure 2). The difference between temporal score changes in the two arms (difference in difference) was not statistically significant (−0.31 in ICV versus usual services; 95% CI: −0.70, 0.08; P value 0.13) (Supplemental Tables 3 and 4). Sensitivity analyses, in which the baseline composite score and other baseline factors were included as covariates in mixed models, yielded effect estimates similar to that of the primary analysis (Supplemental Table 5). Effect estimates were also little changed in analyses that use multiple imputation to assess for outcome sensitivity to differential attrition (Supplemental Table 6).

Figure 2.

Model-based change in the composite harm mitigation score between baseline and 7-month follow-up in neighborhood sites randomized to usual services or the integrated care van in Baltimore, Maryland. Composite scores range from 0 to 15, with lower scores indicating more favorable status. Vertical bars represent 95% confidence intervals. Compared with baseline, the composite harm mitigation score declined in both arms by the 7-month visit (indicating a favorable change in status): −1.93 and −1.63, in ICV and usual services arms, respectively. The difference between temporal score changes in the two arms (difference in difference) was not statistically significant (−0.31 in ICV versus usual services; 95% CI: −0.70, 0.08; P value 0.13)

Secondary outcomes

Secondary outcomes include the components from the composite harm mitigation score (Figure 3 and Supplemental Table 7). During follow-up there were 2 HIV seroconversions, 7 HCV seroconversions, and 11 deaths. For many outcomes, confidence intervals around point estimates were wide due to small numbers of events (e.g., HIV seroconversions) or analyses restricted to subgroups (e.g., HIV care cascade only evaluable in subset living with HIV). Of 16 secondary outcomes, 3 differences between study arms were statistically significant at the P <0.05 level. Compared with usual services, participants in the ICV arm were less likely to use PrEP, less likely to use MOUD, and were less likely to share needles/syringes at the follow-up visit. The P values for these three outcomes ranged from 0.044 to 0.048.

Figure 3.

Odds ratios and 95% confidence intervals (error bars) for secondary outcomes in a cluster-randomized trial. The relative odds for the intervention versus the usual services arms are shown for each outcome. See text for methods. OR, odds ratio; CI, confidence intervals; HCV, hepatitis C virus; PrEP, pre-exposure prophylaxis; MOUD, medications for opioid use disorder;

DISCUSSION

Herein, we present the findings of a cluster-randomized trial to evaluate the impact of mobile, low-barrier access to multiple evidence-based PWID services in an urban setting. Because the goal of increasing access to services and reducing risk is multifaceted, we used a composite harm mitigation score for the primary outcome to capture changes between baseline and follow-up across multiple dimensions. The composite score was constructed of individual indicators of recent service use, risk behaviors, and adverse events. We found that the composite score decreased between baseline and the 7-month follow-up visit in both arms (indicating improved status). However, there was no evidence that the change in score differed in those recruited in neighborhoods assigned to ICV or usual services.

Consistent with the primary outcome, 13 of 16 secondary outcomes did not differ significantly by study arm. There were marginally significant differences by study arm for three secondary outcomes: use of PrEP in the prior 6 months (favoring the usual services arm), use of MOUD in the prior 6 months (favoring the usual services arm), and sharing injection equipment in the prior 6 months (favoring the ICV arm). In the context of the overall null finding for the composite score, the absence of a compelling explanation for these associations, and marginal statistical significance, we believe these associations are likely to be false detection (type 1) errors.

For over 20 years, researchers have demonstrated the use of mobile clinics to increase access to medical services (hepatitis vaccination, tuberculosis screening, HIV and HCV testing, wound care) among PWID (19–23). One study, which used an observational pre-post design, found that use of medical van services was associated with a 21% decrease in emergency department use (19). In our trial, we found no difference in emergency department use by study arm. We are not aware of any completed randomized studies of mobile clinic effectiveness. Of note, HIV Prevention Trial Network (HPTN 094) is an active multi-site randomized trial comparing integrated services from mobile health vans versus referral to community services with peer navigation for PWID (ClinicalTrials.gov identifier NCT04804072). This trial differs from ours in that randomization is being done at the individual- rather than cluster-level.

Four aspects of the study should be considered to interpret this trial’s negative results. First, COVID-19 prevented completion of the final (14-month) follow-up visit. This reduced the ability to detect longer-term intervention effects on study outcomes. Second, and of key importance, in the initial 7 months of the trial only 7.2% of study participants visited the ICV for services, an overlap that was unexpectedly low given that the ICV delivered services at sites where the research van had recruited participants. In addition, there was some contamination – 4.7% of participants in the usual services sites received services from the ICV (compared with 9.7% of participants from the intervention sites). The low exposure to ICV services in the study cohort limited the ability of an intervention, even a highly efficacious one, to produce a measurable effect in the study cohort. This effectively reduced the power of the study to detect differences between arms. Third, HIV point-of-care testing at the baseline cohort visit may have reduced demand for HIV testing at the ICV in the short term

Fourth, although similar demographically, there was evidence that cohort participants differed from ICV clients behaviorally. Whereas 98% of cohort participants reported injection drug use in the prior 30 days (consistent with inclusion criteria), less than 20% of ICV clients were people with recent injection drug use, with most being people with non-injection drug use. Moreover, whereas BUP/NX was the clear driver of service demand on the ICV, BUP/NX use (inside or outside a treatment program) was uncommon among cohort participants relative to methadone. Only 4.0% of cohort participants had a urine drug test positive for BUP at baseline, while nearly half of cohort participants had a positive urine drug test for methadone, acknowledging that the latter drug has a longer serum half-life than the former. This is relevant because PWID who were taking methadone would have no reason to seek BUP/NX, which is incompatible with methadone due to precipitated withdrawal.

It is important to note that low ICV exposure in the study cohort was not due to lack of interest in ICV services in the community (except at a single site). Over approximately 18 months of serially expanding service across 6 neighborhoods, the ICV provided PWID services to hundreds of clients, summarized in Table 3 and in an earlier manuscript (9). Moreover, in a separate qualitative evaluation of the ICV, we found high acceptability and satisfaction with the service model (10). Additional research is needed to distinguish the services that are most and least suitable for mobile clinics. Our experience was that buprenorphine was the principal demand-driver (hook) on the ICV, but this opened the door to infectious disease screening, medical evaluation, harm reduction services, interaction with a case manager, and even HCV treatment. Considered in total, our experience with the ICV was favorable, despite the absence of a neighborhood-level effect on PWID outcomes in the setting of low intervention coverage in the study cohort. Baltimore City Health Department has elected to continue the ICV and to expand services to additional neighborhoods. In retrospect, this suggests that our focus on PWID in the study cohort may have been too narrow to capture benefits in people with non-injection drug use.

Our trial has strengths, including a public health-academic joint venture, a rigorous design, and follow-up rates above 70% for street-recruited PWID. In addition to challenges discussed above, this trial had other limitations, including conduct in a single city with findings that may not generalize to other locations in the United States, some intervention contamination (participants in the usual services arm visiting the ICV), and modestly higher 7-month follow-up rates in the intervention than the usual services arms.

In conclusion, in this cluster-randomized trial, we found no evidence that weekly neighborhood visits from a mobile health van improved access to services and outcomes among PWID in the neighborhood. However, a reduction in follow-up time due to COVID-19, low exposure to the ICV intervention in the research cohort, and evidence that clients visiting the ICV differed in relevant ways from the research cohort limited our ability to assess the potential benefits of the ICV. However, data from this analysis and earlier publications indicate that the ICV provided access to buprenorphine, infectious disease testing, and other evidence-based services to substantial numbers of people who use drugs.

Supplementary Material

Suppl. Materials

Supplemental Figure 1 Baltimore map with site locations

Supplemental Figure 2 – Enrollment/recruitment follow-up & COVID timeline

Supplemental Table 1. Baseline participant characteristics, stratified by matched pairs and study arm

Supplemental Table 2. Composite PWID harm mitigation scores at baseline and follow-up and access of ICV services, stratified by neighborhood site.

Supplemental Table 3. Composite harm mitigation scores at baseline and follow-up, stratified by study arm.

Supplemental Table 4. Results of mixed effects linear regression of composite harm mitigation score with group, time, and group-by-time fixed effects, and random intercepts for study clusters and participants (n = 720)

Supplemental Table 5. Sensitivity analyses in which the baseline composite score and other baseline factors were included as covariates in mixed models

Supplemental Table 6. Sensitivity analyses in which multiple imputation was used to assess for outcome sensitivity to differential attrition.

Supplemental Table 7. Crude data and odds ratios for secondary outcomes among study participants at ICV vs. usual services study sites