A precision randomized trial to evaluate the impact of tailored hepatitis C treatment adherence support on HCV treatment outcomes among people who inject drugs in India: design and baseline characteristics of the STOP-C trial

Abstract

Efforts to eliminate the hepatitis C virus (HCV) as a public health problem must include people who inject drugs (PWID). We describe the design and baseline characteristics of the Supporting Treatment Outcomes among PWID trial, which evaluates whether HCV treatment outcomes in PWID can be optimized by tailoring treatment support in 7 PWID-focused integrated HIV/HCV prevention/treatment centers across India. The design is a 3-arm, individual-level precision-randomized trial. Leveraging empirical data, a prediction model assigned participants as minimal or elevated risk for failure. Minimal-risk participants were randomized 3:2:1 to low-intensity (basic services), medium-intensity (patient navigation), and high-intensity (patient navigation + directly observed therapy) support, respectively. Elevated-risk participants were randomized 3:2:1 to high-, medium-, and low-intensity support, respectively. All received 12 weeks of oral direct-acting antiviral therapy. The primary outcome is sustained virologic response 12 weeks after treatment completion in an intention-to-treat analysis. Three thousand participants were randomized (2048 [68%] minimal risk, 952 [32%] elevated risk of failure). This approach will allow for the estimation of efficacy within treatment failure risk strata while preserving the ability to estimate the average treatment effect and has particular relevance with increasing emphasis on precision medicine in health care delivery.

Trial registration: Identifier: NCT04652804.

Introduction

In 2020, an estimated 58 million persons were living with chronic hepatitis C virus (HCV) and needed treatment, of whom 90% resided in low- and middle-income countries (LMICs).¹ With the expansion of oral direct-acting antivirals (DAAs)^2-7 that can achieve cure rates of 95%,^3-5 the World Health Organization (WHO), in 2016, released elimination targets for 2030⁸ calling for an 80% reduction in HCV incidence and 65% reduction in mortality. To achieve these goals, 90% of HCV-infected individuals must be diagnosed, and 80% of those diagnosed must be treated, requiring a massive scale-up in most countries. While many countries have expanded treatment access, which has been facilitated by licensing, preferential pricing, and generic DAAs that have reduced cost to ~US$150/course in settings like India,⁹ only 15 countries are on track to achieve WHO 2030 targets.¹⁰

People who inject drugs (PWID) are disproportionately affected by HCV, bearing a burden 10 to 100 times that of the general population.^11,12 For PWID, there is an opportunity to leverage existing infrastructure for HIV care and harm reduction services, including medication for opioid use disorder (MOUD) to deliver HCV care. Indeed, 3 recent meta-analyses identified that co-location of HCV testing, treatment, and other services significantly improved HCV treatment outcomes, especially among PWID.^13-15 However, these reviews also noted that despite potential synergies, positive health impact, and potential cost savings, there is limited knowledge on optimal integrated service delivery models.^13-15 Moreover, programs need to consider factors other than the provision of free medications to optimize adherence and reduce reinfection risk^16-23 while recognizing the structural (eg, infrastructure, human capital) and financial constraints to providing intensive support to all treated.^16,20 Identifying who can be cured with minimal intervention and who will need more support can improve efficiencies.

Accordingly, the Supporting Treatment Outcomes among PWID (STOP-C) trial used a novel “precision” approach to allocate intervention assignment, in this case, HCV treatment adherence support, based on need. This approach, which represents a variation on stratified randomization, allows for estimation of efficacy within strata of risk for treatment failure while preserving the ability to estimate the average treatment effect. We present details on the design and rationale of the STOP-C trial, along with selected baseline characteristics.

Methods

Trial objectives

The overall goal of the STOP-C trial was to evaluate whether HCV treatment and posttreatment outcomes in PWID could be optimized by tailoring adherence support in 7 PWID-focused integrated HIV/HCV prevention/treatment centers. The primary objective was to evaluate whether the intensity of treatment adherence support affects sustained virologic response (SVR). Secondary objectives were to evaluate (1) whether intensity of support affected HCV treatment completion and adherence, (2) incidence and correlates of HCV reinfection post-SVR, and (3) the impact of HCV cure on HIV viral suppression among PWID coinfected with HIV and HCV.

Study setting

In 2015, as part of a cluster-randomized trial, we established integrated care centers (ICCs) for PWID across 6 Indian cities,²⁴ expanding to 8 cities by 2017. Venues (nongovernmental organizations [NGOs] or government facilities) were selected for scale-up following discussions between the National AIDS Control Program, India, NGO leaders, community members, and study investigators. ICCs were scaled from existing MOUD programs to incorporate other risk reduction services, including HIV counseling and testing and field-based syringe services. Some medical services at ICCs are provided on-site, including sexually transmitted infection screening and treatment, while others are through peer-navigated referral (eg, tuberculosis treatment, antiretroviral therapy [ART]). We previously demonstrated that integration of HCV testing into these ICCs significantly improved community uptake of HCV testing, awareness of HCV-positive status, and treatment uptake, but there was no impact on SVR.²⁵ The STOP-C trial was conducted across 7 ICCs in Aizawl, Mizoram; Bilaspur, Chhattisgarh; Ludhiana, Punjab; Amritsar, Punjab; Churachandpur, Manipur; Kanpur, Uttar Pradesh; and New Delhi (Figure 1). Clients at these ICCs either sought services independently or were referred by ICC clients.

Figure 1.

Burden of hepatitis C virus (HCV) infection across STOP-C study sites. Size of circle is proportional to community HCV antibody prevalence.

Trial design

STOP-C is a 3-arm, stratified individual-level randomized clinical trial, in which treatment assignment probabilities varied by strata defined by participants’ predicted risk for failure (precision randomization). Participants in the stratum defined by lower predicted risk for failure had a higher likelihood of being allocated to a lower-intensity intervention, and participants in the stratum defined by higher predicted risk were more likely to be allocated to the higher-intensity intervention.

Prior trials evaluating the impact of HCV adherence support interventions among PWID have used standard approaches that randomize individuals to an intervention vs the standard of care, which assumes that all those assigned to an intervention have the same need and potential for benefit. In such designs, intervention assignment and individual characteristics are independent of each other, but each can affect the outcome—the impact of the intervention can be directly measured because there should be no confounding, measured or unmeasured, between intervention assignment and outcome. Under the precision randomization scenario, we intentionally create a prognostic score that lies between the causal pathway from individual characteristics to the intervention assignment (ie, confounding by design). While to our knowledge, there have not been applications of prognostic scores to trial design, prognostic scores have been used to account for confounding in observational studies similar to propensity scores.^26-30 Essentially, for each level of the prognostic score, the covariates between those who do and do not have the outcome are balanced in expectation and are therefore no longer confounding the exposure-outcome relationship.

Thus, this approach allows for estimating efficacy within strata of risk for treatment failure (minimal/elevated) while still allowing estimation of the average treatment effect and effect modification by participants’ prognostic score. This approach is similar to stratified randomization with the modification that the allocation ratio is not equal across strata.

Study population

All persons receiving care in 1 of the 7 ICCs with HCV antibodies were screened for eligibility. Participants were eligible if they (1) registered at the ICC; (2) were ≥18 years of age; (3) had a history of injection drug use; (4) had active HCV infection (detectable HCV RNA); (5) met eligibility criteria for HCV treatment according to the Indian Association for the Study of Liver disease (Figure 2)³¹; (6) provided informed consent, which required willingness to take HCV treatment and be randomized; and (7) were willing to use 1 form of contraception (participants with reproductive potential). Participants were excluded if they (1) did not provide consent; (2) were planning to migrate within 6 months; (3) had a history of prior DAA treatment; (4) had a life expectancy <1 year; (5) had decompensated cirrhosis (based on Child Turcotte Pugh score), (6) chronic hepatitis B (HBsAg positive), (7) active tuberculosis, (8) an AIDS-defining illness within 30 days prior to entry, (9) an active SARS-CoV-2 infection, (10) any allergy or hypersensitivity to hepatitis C treatment, or (11) any other acute or serious illness requiring treatment or hospitalization within 30 days of screening; or (12) were taking any contraindicated medications. Participants of reproductive potential were only included if they were willing to use contraception and were excluded if pregnant or breastfeeding. While not all of these were contraindications to hepatitis C treatment, we considered factors that would complicate the delivery of HCV treatment in a community-based setting. For example, participants were being recruited in the midst of the COVID-19 pandemic, limiting the ability to interact with participants who screened positive for SARS-CoV-2 (due to government- and research-imposed COVID-19 restrictions). Participants were eligible for rescreening (eg, when they cleared SARS-CoV-2 infection).

Figure 2.

Study design and participant flow.

Interventions

The current standard of care in India is 4-weekly dispensation of medication over a 12-week period with brief adherence counseling delivered free of charge from district government hospitals; in India, at least 1 hospital in each district provides free HCV care.

Participants in all arms without HIV infection or who were living with HIV and not on an efavirenz-containing ART regimen received a once-daily generic oral fixed-dose combination of sofosbuvir/velpatasvir (SOF/VEL) containing 400 mg SOF and 100 mg VEL for 12 weeks. If a participant was living with HIV and on an efavirenz-containing regimen, they had the option to (1) switch to a non-EFV-containing regimen (eg, dolutegravir) for at least 2 weeks prior to initiation of SOF/VEL or (2) receive generic sofosbuvir/daclatasvir (SOF/DAC) with duration guided by cirrhosis status and genotype. No participants received SOF/DAC, so all received 12 weeks of SOF/VEL treatment.

The 3 interventions targeted medication adherence during HCV treatment, all providing a level of care higher than the current standard in India, with the delivery from community-based PWID-friendly clinics vs district hospitals. Three arms were compared: arm 1, low-intensity support; arm 2, medium-intensity support; and arm 3, high-intensity support (Table 1). These interventions involve a combination of standard-of-care procedures, patient navigator (PN) support, and directly observed therapy (DOT). Interventions were chosen based on the available evidence at the time the study was designed; most data were from the pre-DAA era, but data that existed from the DAA era suggested that high adherence among PWID was possible when combined with optimal support, including MOUD, group counseling, and DOT.^6,32-35 Our choice of intervention was based on these data, as well as some evidence from the pre-DAA era, which suggested that ideal interventions included multidisciplinary approaches with intensive social support³⁶ and demonstrated the effectiveness of patient navigation support,³⁷ as well as modified DOT in multiple settings, including MOUD,^38-40 prison settings,⁴¹ and community health centers.⁴² In our setting, both PN and DOT interventions were delivered by nurses and study patient navigators, who may or may not have been peers.

Table 1.

Summary of intervention and standard packages of services by study arm.

	Arm 1: low intensity	Arm 2: medium intensity	Arm 3: high intensity
Medication delivery
4 weekly dispensation	X	X
Weekly dispensation			X
Observed doses (1-7 doses/wk)			X
Adherence support
Standard adherence counseling at treatment initiation	X	X	X
Adherence counseling at medication pick-up visits	X	X	X
Phone number provided for support as needed	X	X	X
Tailored PN support
Creation/implementation of hepatitis C care plan		X	X
Regular contact during treatment (minimum of once/2 weeks)		X	X
Medication reminders (as needed)		X
Directly observed doses (minimum of once/week)			X
Linkage to other services
Access to all ICC services (HIV testing with linkage to treatment, STI management, TB testing and linkage, counseling, condoms, MOUD, SSP)	X	X	X
Facilitated linkage to services available within the ICC and outside of the ICC	X	X	X
Participant tracking
Phone/in-person tracking for missed medication pick-ups initiated within 1 day of a missed medication pickup—at least 3 attempts	X	X	X

Abbreviations: STI, sexually transmitted infection; TB, tuberculosis; MOUD, medication for opioid use disorder; SSP, syringe services program.

Arm 1 participants received a 28-day supply of medication dispensed at baseline, week 4, and week 8, along with standard adherence counseling at every visit. Participants had access to ICC services, including facilitated linkage to referrals. Outreach workers tracked missed refill appointments. Arm 2 participants also received standard dispensation of 28-day supplies of medication but were also assigned a PN for tailored support for medication reminders, picking up refills (or home/field delivery), overcoming barriers, and facilitating service linkage. PNs collected information at entry on barriers and facilitators to adherence and collaboratively developed an HCV care plan tailored to participant needs. PNs had to contact clients at least once every 2 weeks remotely and/or in person at a location of the client’s choosing. Arm 3 participants received PN support, as well as patient-centered DOT with flexibility in frequency of pickup and site (ICC, home-based, field-based) with ≥1 observed dose per week. The frequency of pick-up and location could be changed during treatment.

Randomization

Using a “precision randomization” approach, adherence support was assigned using an unbalanced allocation method across strata, defined by the individual’s propensity for treatment failure. To determine this propensity, we developed a prediction model using data on early ART adherence from a prior cohort enrolled at the 7 ICCs from October 2017 to September 2018 who initiated ART within 90 days of enrollment (n = 431). Information on ART initiation and refills at the government ART center (ie, regimen, number of pills, date of dispensation) was abstracted from participant records. We calculated the medication possession ratio (MPR) and examined adherence within the first 90 days (~12 weeks) to mirror oral HCV treatment. The MPR was calculated as the number of tablets a client had over a particular period divided by the number of days in the period. An MPR of at least 85% was used as a surrogate for treatment success based on adherence required to achieve SVR from prior studies.⁴³

To select predictors, we used 10-fold cross-validated logistic LASSO regression models, which offer accurate prediction, provide sparse models, and manage overfitting. More than 20 predictors were explored, including sociodemographics (eg, age, sex, employment, income, homelessness), substance use (ie, alcohol and injection drug use), mental health (ie, depression and quality of life), risk behaviors (ie, sexual partners and sharing needles/syringes), and engagement in MOUD. The final model included age, sex, income, homelessness, recent injection drug use frequency, number of sexual partners, and quality of life. The model area under the receiver operating characteristic curve (AUROC) was 0.79, suggesting moderately high discrimination. Prior to randomization, each participant underwent a questionnaire, including factors identified in the prediction model. Groups were defined by a probability cutoff whereby those with a probability of treatment failure at or above 0.577 were considered at elevated risk and those below 0.577 minimal risk; in order to ensure sufficient power to detect differences within strata, we chose a cutoff that would ensure that at least one-third of participants would be identified as minimal risk. Individuals were preferentially randomized to the support level to match the predicted propensity of HCV treatment failure. Those at elevated risk were randomized 3:2:1 for arm 3, arm 2, and arm 1, respectively. Conversely, those at minimal risk were randomized 1:2:3 to arm 3, arm 2, and arm 1, respectively. Randomization allocation lists were stratified by site and risk stratum with randomly varying block sizes of 6, 12, and 18; allocations were computerized and thus concealed from site staff. Participants and staff were blinded to the risk classification but not the intervention.

Study visits

Study visits were primarily completed in person; however, flexibility was incorporated because of COVID-19, such that study visits could be conducted at participants’ homes or in a hybrid fashion. Table 2 includes details on assessments by visit. An electronic health record was in place in each ICC. Initial screening for HCV treatment eligibility among those with evidence of HCV antibodies was performed as part of the standard of care at the ICCs and included symptom screening for tuberculosis, clinical examination to rule out signs of decompensated cirrhosis, a blood draw to ascertain HCV RNA status, HBsAg status, HIV and CD4 status (if not available), additional laboratory parameters to determine treatment eligibility (complete blood count, liver function tests, renal function test, internalized normalized ratio), and a pregnancy test for women.

Table 2.

Assessments by study visit.

	Pretreatment	On treatment				Posttreatment
Characteristic	Screening	Entry week 0	Week 4	Week 8	Week 12	Week 24 (SVR 12)	Weeks 48, 72, 96, 120, 144, 168)
Clinical and laboratory assessments
Documentation of HCV RNA	X					Xa
HCV core antigen testingb							X
Documentation of HIV status	X
Brief medical history	X
Medication history	X	X	X	X	X	X
Clinical evaluation	X					X
Complete blood count	X					X
Liver function tests	X					X
Renal function tests	X					X
International normalized ratio	X			X
Pregnancy test (women)	X	X	X (if pregnancy suspected)
HBsAg testing	X					X
CD4 (if HIV antibody positive)	X
Stored plasma/serum specimen		X			Xa	Xa	X
Questionnaires
Screening questionnaire	Z
Contact/locator information	X	Z	Z	Z	Z	Z	Z
Demographics		Z				Z	Z
Substance use and risk behavior		Z	Z	Z	Z	Z	Z
HCV treatment readiness		Z
Social/family support		Z
Depressive symptoms		Z				Z	Z
Quality of life		Z				Z	Z
COVID-19 symptoms/care		Z			Z	Z	Z
Adherence assessment			Z	Z	Z
Medication side effects			Z	Z	Z
Intervention assessments
Patient navigator intake (arms 2 and 3)		Z
Hepatitis C care plan (arms 2 and 3)		Z
Patient navigation log (arms 2 and 3)		Z
DOT log (arm 3)		Z
Counseling
Pregnancy prevention counseling	X	X	Z	Z	Z
Adherence counseling		X	Z	Z
Cirrhosis/liver health counseling		X			Z	Z	Z
HCV risk reduction counseling		X	Z	Z	Z	Z	Z

X indicates that the data collection/assessment had to take place in the ICC. Y indicates that this data collection may have been collected at the ICC, at a participant’s home, or at an alternative location depending on precautions in place for COVID-19. Z indicates that the data collection may have taken place at the ICC or over the phone depending on precautions in place for COVID-19.

^aExceptions were made for participants who were unable to come to the site because of COVID-19 precautions or because they were in a rehabilitation setting. In those cases, samples could be obtained from remote locations.

^bOnly assessed on participants who achieve HCV cure.

All participants eligible for HCV treatment according to laboratory and clinical parameters were asked to provide written informed consent, after which additional study eligibility criteria were assessed using a survey. Those eligible proceeded to the entry visit, which had to occur within 90 days of the initial screening. The entry visit included additional behavioral and clinical questionnaires, a blood draw, and the precision randomization procedure. Following randomization, participants underwent additional assessments relevant to their intervention arm assignment. All participants were asked to return for on-treatment visits every 4 weeks (2 total). The end-of-treatment visit was conducted 12 weeks after treatment initiation and the SVR visit 12 weeks after treatment completion (window 10-60 weeks). Participants who were living with HIV or had achieved SVR were followed post-SVR at 6-month intervals.

Data collection

Surveys included questions that have been previously used in studies among PWID in India. Substance use was captured using a questionnaire adapted from a variety of sources (WHO ASSIST,⁴⁴ PhenX toolkit, AUDIT, existing surveys) for use in India and included information on type, route, and frequency of drug and alcohol use. Depressive symptoms were assessed using the patient health questionnaire - 9 (PHQ-9)⁴⁵ and quality of life using a modified EuroQol-5 Dimension. Social support was captured using the Medical Outcomes Study Social Support Survey.⁴⁶ Adherence was captured using the visual analog scale, in which participants estimated the percentage of doses taken over the prior 30 days. Readiness for HCV treatment was captured using an instrument adapted from the HIV treatment beliefs component of an HIV treatment readiness scale.⁴⁷

Safety and monitoring

Participant safety was monitored using standard assessments at follow-up and interim visits. Adverse events were graded according to the Division of AIDS Table for Grading the Severity of Adult and Pediatric Adverse Events.⁴⁸ Serious adverse events were reported to the Johns Hopkins Medicine and YR Gaitonde Centre for AIDS Research and Education Institutional Review Board within 7 days. The study was overseen by an independent Study Monitoring Committee, which reviewed adverse event data semi-annually through SVR.

Outcome measures

Table 3 describes all primary and secondary outcomes. The primary outcome is SVR, defined as HCV RNA less than the lower limit of quantification (LLOQ), measured 12 weeks (range, 10-60) after completion of treatment (cure). Secondary outcomes included (1) treatment completion; (2) treatment adherence, defined using self-reported and objective data; (3) HCV reinfection; and (4) HIV viral suppression. Additional exploratory outcomes included (1) medication for opioid use disorder, (2) quality of life, and (3) all-cause mortality.

Table 3.

Definitions of primary and secondary outcomes.

Characteristic	Definition	Data source
Primary
Sustained virologic response (SVR)	HCV RNA < lower limit of quantification (LLOQ, 30 IU/mL) measured 12 weeks (range 10- 60 weeks) after treatment completion	Laboratory specimen collected at SVR visit
Secondary
Treatment completion	Completing the prescribed course of treatment (12 weeks)	Study records on premature treatment discontinuation
Treatment adherence
Self-reported	Percentage of doses taken (visual analog scale)	Visual analog scale
Self-reported	>90% of doses reported as taken (visual analog scale)	Visual analog scale
Refills completed/pill counts	Medication possession ratio	Medication dispensation records, including refill completion dates and pill counts (at 4-week intervals for arms 1 and 2 and 1-week intervals for arm 3)
Refills completed/pill counts	Medication possession ratio >90%	Medication dispensation records, including refill completion dates and pill counts (at 4-week intervals for arms 1 and 2 and 1-week intervals for arm 3)
HCV reinfection	Testing positive for HCV core antigen after achieving SVR	Laboratory specimens collected at post-SVR follow-up visits
HIV viral suppression	HIV RNA less than the LLOQ	Government ART books

Statistical analysis

The primary outcome analysis compares the relative risk (RR) of achieving SVR across level of support (low, medium, high intensity) within the 2 strata of risk (minimal and elevated) independently using separate log-binomial models (or their extensions, ie, Poisson regression with robust variance estimation), including adjustment for site. The reference group is arm 1 for the minimal-risk stratum and arm 3 for the elevated stratum. Primary comparisons are intention to treat. Additional analyses for the primary outcome include a per-protocol (PP) analysis, which excludes those who died prior to SVR assessment, are alive but lost to follow-up, or prematurely discontinued treatment. Additional sensitivity analyses consider (1) alternative analytical methods to account for site, (2) accounting for covariates prognostic for outcome, and (3) missing data/nonadherence related to the primary outcome. Noninferiority analyses are planned if superiority is not demonstrated in the minimal-risk stratum. Subgroup analyses explore whether the effect of the intervention in either stratum varies by the prognostic score itself and other covariates of interest, including age, employment, HIV status and treatment, drug use, injection drug use, alcohol use, homelessness, depressive symptoms, MOUD use, distance to study site, readiness for HCV treatment, and study site. Additional exploratory analysis incorporates phylogenetic analysis of baseline and SVR samples to reclassify persons with evidence of reinfection (vs treatment failure) as having achieved SVR. Additional details on the analysis of the primary outcome, analysis of secondary and exploratory outcomes, and considerations related to missing data can be found in the Supplementary Material, Statistical Analysis Plan.

Of particular relevance to the precision randomized design are analyses to estimate heterogeneity in treatment effect by the prognostic score and the average treatment effect, which estimate the effect of the support strategies in the overall population. Heterogeneity by level of prognostic score provides a more nuanced assessment of heterogeneity as compared to conducting subgroup analysis with one variable at a time.⁴⁹ Under the unbalanced randomization scenario, the prognostic score lies between the causal pathway from individual characteristics to the treatment assignment, thus introducing an informed confounding (ie, individual’s propensity for treatment failure). The precision randomization can then be used to reweight the sample back to the initial sample as if the randomization allocation were equivalent to all levels of the prognostic score. That is, by using inverse probability of treatment weights, the relationship between the prognostic score and treatment assignment is removed to estimate the average treatment effect and provide answers about whether the high- and moderate-intensity interventions resulted in superior outcomes in the overall population. Further analyses will use information from the prognostic score and the observed data to identify the optimal threshold for assigning treatment support.

Power calculations

The anticipated sample size for this trial was 3000, with approximately 420 recruited at each site over 18 to 24 months. Sample size calculations were based on the primary outcome (SVR). Initial calculations considered that strata would be defined by the median of the prognostic score (1500 per stratum), accounting for variability depending on the cutoff. As such, power was calculated considering 1000 and 1500 persons within the stratum. We calculated the minimal detectable RR assuming a 2-sided α = 0.05 and 80% power. In the elevated-risk stratum, assuming an SVR of 85% in those receiving the low/medium-intensity intervention, an RR of 1.06 comparing SVR in those receiving the high- vs medium-intensity intervention and 1.08 comparing those receiving the high- vs low-intensity intervention could be detected with a stratum size of 1500. Power was slightly reduced with a stratum sample size of 1000. In the minimal-risk stratum, if 95% SVR was observed in the low-intensity group, an RR of 1.03 in the medium- vs low-intensity intervention and an RR of 1.04 in the high- vs low-intensity intervention could be detected with a stratum size of 1500.

Results

Enrollment

Trial recruitment started on January 16, 2021, when the first site, Ludhiana, was activated. The last site, Aizawl, was activated on January 28, 2022, and the last person was randomized on December 13, 2022. Over 24 months of recruitment, a total of 4905 initiated screening, of whom 4431 completed all screening procedures; 424 of the 468 with incomplete screening were potentially eligible for HCV treatment. Of the 4431, 3000 were eligible for the STOP-C study and randomized. The primary reasons for not being eligible (n = 1432) were ineligibility for HCV treatment because HCV RNA was <30 IU/mL (n = 1087) or failure to meet Indian National Association for the Study of the Liver (INASL) guidelines (n = 397) (Figure 2). In total, 2048 (68%) were identified as being at minimal risk of failure and 952 (32%) at an elevated risk of failure. Of the 3000 randomized, 6 were subsequently excluded (4 found ineligible after randomization, 2 nonviable specimens), leaving an analytical sample of 2994.

While the trial start was initially delayed because of COVID-19 and national lockdowns, recruitment was completed within the original timelines. Additional partial lockdowns occurred during the COVID-19 Delta wave (April-August 2021) among 3 sites active during this time (Bilaspur, Kanpur, and Ludhiana) and the Omicron wave, which primarily affected the site in Aizawl. While recruitment slowed substantially during the Delta wave and minimally during the Omicron wave, sites on average recruited 25 participants per month for a total of 193 participants per month when all sites were active.

Baseline characteristics

Table 4 provides characteristics of participants by study site. There was substantial heterogeneity in demographic and behavioral characteristics of participants by study site. Compared to participants in other sites, those in the northeastern sites of Aizawl and Churachandpur had higher educational attainment but were more likely to be unemployed. Compared to no one reporting living on the street in Aizawl, 29.0% reported living on the street or in a slum in Delhi. Compared to only 2.1% reporting daily injection in Aizawl, 47.0% reported daily injection in New Delhi. Similarly, the proportion of participants living with HIV varied from 3.8% in Churachandpur to 54.1% in New Delhi. Accordingly, the distribution of the prognostic score also varied substantially by site, with lower scores in northeastern cities and the highest scores in northern and central India (Figure 3).

Table 4.

Baseline characteristics of PWID randomized to the STOP-C trial by site and overall.

Characteristic	Aizawl	Churachandpur	Bilaspur	Kanpur	Delhi	Amritsar	Ludhiana	Overall
N	338	344	557	315	338	487	615	2994
Prognostic score (probability) (IQR)	0.18	0.24	0.55	0.54	0.77	0.42	0.44	0.43
	(0.13, 0.26)	(0.13, 0.36)	(0.41, 0.66)	(0.37, 0.71)	(0.55, 0.94)	(0.26, 0.57)	(0.29, 0.60)	(0.25, 0.62)
Sociodemographics
Median age (IQR)	30 (25, 36)	29 (24, 38)	27 (22, 33)	34 (27, 42)	28 (24, 33)	30 (26, 34)	29 (26, 33)	29 (25, 35)
Sex
Male	323 (95.6)	323 (93.9)	557 (100)	315 (100)	338 (100)	483 (99.2)	613 (99.7)	2952 (98.6)
Female	15 (4.4)	21 (6.1)	0 (0)	0 (0)	0 (0)	4 (0.8)	2 (0.3)	42 (1.4)
Median income, rupees (IQR)	0 (0, 0)	7000 (4000, 10 000)	8000 (6000, 9000)	5000 (3000, 7500)	5000 (3000, 7500)	9000 (7000, 12 000)	12 000 (9000, 15 000)	7000 (0, 10 000)
Living on the street or in a slum	0 (0)	0 (0)	2 (0.4)	27 (8.6)	98 (29.0)	1 (0.2)	0 (0)	128 (4.3)
Marital status
Never married	163 (48.2)	168 (48.8)	306 (54.9)	159 (50.5)	214 (63.3)	229 (47.0)	304 (49.4)	1543 (51.5)
Currently married	80 (23.7)	132 (38.4)	238 (42.7)	124 (39.4)	91 (26.9)	242 (49.7)	280 (45.5)	1187 (39.7)
Widowed/divorced/separated	95 (28.1)	44 (12.8)	13 (2.3)	32 (10.2)	33 (9.8)	16 (3.3)	31 (5.0)	264 (8.8)
Educational attainment
No schooling	0 (0)	3 (0.9)	29 (5.2)	79 (25.1)	53 (15.7)	25 (5.1)	44 (7.2)	233 (7.8)
Primary school	3 (0.9)	49 (14.2)	155 (27.8)	64 (20.3)	95 (28.1)	91 (18.7)	82 (13.3)	539 (18.0)
Secondary school	141 (41.7)	166 (48.3)	309 (55.5)	125 (39.7)	164 (48.5)	225 (46.2)	338 (55.0)	1468 (49.0)
High school and above	194 (57.4)	126 (36.6)	64 (11.5)	47 (14.9)	26 (7.7)	146 (30.0)	151 (24.6)	754 (25.2)
Employment
Unemployed	284 (84.0)	306 (89.0)	51 (9.2)	53 (16.8)	47 (13.9)	37 (7.6)	45 (7.3)	823 (27.5)
Daily wages	2 (0.6)	30 (8.7)	314 (56.4)	199 (63.2)	239 (70.7)	192 (39.4)	171 (27.8)	1147 (38.3)
Weekly/monthly wages	52 (15.4)	6 (1.7)	192 (34.5)	63 (20.0)	52 (15.4)	258 (53.0)	398 (64.7)	1021 (34.1)
Don’t know/refused	0 (0)	2 (0.6)	0 (0)	0 (0)	0 (0)	0 (0)	1 (0.2)	3 (0.1)
Lifetime history of incarceration	5 (1.5)	7 (2.0)	116 (20.8)	46 (14.6)	163 (48.2)	105 (21.6)	182 (29.6)	624 (20.8)
Last time incarcerated
Within last 6 months	0 (0)	0 (0)	17 (14.7)	4 (8.7)	6 (3.7)	7 (6.7)	12 (6.6)	46 (7.4)
More than 6 months to 1 year ago	0 (0)	0 (0)	2 (1.7)	4 (8.7)	16 (9.8)	4 (3.8)	14 (7.7)	40 (6.4)
More than 1 year ago	5 (100)	7 (100)	97 (83.6)	38 (82.6)	141 (86.5)	94 (89.5)	156 (85.7)	538 (86.2)
Substance use
Injection frequency last 3 months
None	320 (94.7)	218 (63.4)	47 (8.4)	95 (30.2)	20 (5.9)	218 (44.8)	363 (59.0)	1281 (42.8)
Some days	11 (3.3)	105 (30.5)	142 (25.5)	151 (47.9)	159 (47.0)	251 (51.5)	250 (40.7)	1069 (35.7)
Every day	7 (2.1)	21 (6.1)	368 (66.1)	69 (21.9)	159 (47.0)	18 (3.7)	2 (0.3)	644 (21.5)
Drugs injected in prior 3 months
Heroin	7 (2.1)	116 (33.7)	0 (0)	85 (27.0)	116 (34.3)	269 (55.2)	250 (40.7)	824 (27.5)
Stimulants	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
Buprenorphine	0 (0)	0 (0)	508 (91.2)	206 (65.4)	307 (90.8)	1 (0.2)	0 (0)	1022 (34.1)
Painkillers	0 (0)	0 (0)	1 (0.2)	0 (0)	1 (0.3)	0 (0)	0 (0)	2 (0.1)
Allergy medicines	0 (0)	0 (0)	503 (90.3)	216 (68.6)	315 (93.2)	0 (0)	1 (0.2)	1035 (34.6)
Sedatives	0 (0)	0 (0)	0 (0)	0 (0)	1 (0.3)	0 (0)	0 (0)	1 (0.0)

Figure 3.

Prognostic score for treatment failure by study site. The score was developed using a prediction model using data on early ART adherence, defined as a medication possession ratio ≥85% within the first 90 days (~12 weeks) to mirror oral HCV treatment. The final model included age, sex, income, homelessness, recent injection drug use frequency, number of sexual partners, and quality of life.

Table 4.

Continued

Characteristic	Aizawl	Churachandpur	Bilaspur	Kanpur	Delhi	Amritsar	Ludhiana	Overall
Shared needles in prior 3 months	2 (0.6)	16 (4.7)	55 (9.9)	37 (11.8)	119 (35.2)	12 (2.5)	41 (6.7)	282 (9.4)
Injection partners in prior 30 days
0	335 (99.1)	234 (68.0)	157 (28.2)	275 (87.3)	233 (68.9)	475 (97.5)	544 (88.5)	2253 (75.3)
1	3 (0.9)	110 (32.0)	296 (53.1)	35 (11.1)	47 (13.9)	7 (1.4)	64 (10.4)	562 (18.8)
More than 1	0 (0)	0 (0)	104 (18.7)	5 (1.6)	58 (17.2)	5 (1.0)	7 (1.1)	179 (6.0)
Syringe services program in prior 6 months	3 (0.9)	34 (9.9)	383 (68.8)	6 (1.9)	53 (16.7)	22 (4.5)	160 (26.0)	661 (22.1)
MOUD in prior 6 months	71 (21.0)	23 (6.7)	200 (35.9)	83 (26.4)	221 (65.4)	254 (52.2)	449 (73.0)	1301 (43.5)
Harmful or hazardous alcohol use (AUDIT-3)	6 (1.8)	69 (20.1)	59 (10.6)	41 (13.0)	78 (23.1)	101 (20.7)	69 (11.2)	423 (14.1)
Noninjection drug use in prior 3 months	0 (0)	8 (2.3)	464 (83.3)	251 (79.7)	331 (97.9)	386 (79.3)	85 (13.8)	1525 (50.9)
Psychosocial
Moderate/severe depressive symptomsa	1 (0.3)	0 (0)	2 (0.4)	44 (14.0)	34 (10.1)	0 (0)	9 (1.5)	90 (3.0)
Self-reported health state (out of 100)b	90 (87, 91)	95 (95, 95)	80 (70, 90)	80 (60, 95)	60 (50, 70)	80 (70, 80)	80 (60, 95)	68 (61, 74)
Median social support scorec	5 (5, 5)	4 (3.9, 4.4)	3.3 (2.9, 3.5)	3.9 (3, 4.3)	4.1 (1.1, 5)	5 (5, 5)	4.6 (4.3, 4.8)	4.4 (3.5, 5)
Sexual behaviors
Sexual partners
0	220 (65.1)	160 (46.5)	242 (43.5)	202 (64.1)	255 (75.4)	204 (41.9)	279 (45.4)	1562 (52.2)
1	111 (32.8)	151 (43.9)	297 (53.3)	104 (33.0)	76 (22.5)	277 (56.9)	316 (51.4)	1332 (44.5)
2	5 (1.5)	17 (4.9)	9 (1.6)	7 (2.2)	5 (1.5)	5 (1.0)	14 (2.3)	62 (2.1)
≥3	2 (0.6)	16 (4.7)	9 (1.6)	2 (0.6)	2 (0.6)	1 (0.2)	6 (1.0)	38 (1.3)
HIV
Living with HIV	72 (21.3)	13 (3.8)	95 (17.1)	97 (30.8)	183 (54.1)	57 (11.7)	124 (20.2)	641 (21.4)
On antiretroviral therapy (among PLHIV)	72 (100)	13 (100)	90 (94.7)	96 (99.0)	91 (49.7)	53 (93.0)	120 (96.8)	535 (83.5)
HCV
HCV treatment readinessd	60 (58, 60)	48 (46, 51)	46 (45, 47)	47 (45, 50)	45 (43, 47)	47 (47, 47)	53 (51, 54)	47 (46, 53)

Abbreviations: AUDIT, Alcohol Use Disorders Identification Test; PLHIV, people living with HIV.

Sample n = 2994; 6 participants were excluded (4 ineligibility after randomization, 2 missing lab results).

^aModerate to severe depressive symptoms defined as the sum score of individual questions from the Patient Health Questionnaire (PHQ-9) being ≥15; Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med. 2001;16(9):606-613.

^bAdapted from the EuroQol-5 Dimension.

^cMean across all questions (range from 1 to 5; 1 = support none of the time, 5 = support all of the time; range in sample: 1 to 5).

^dSum score of individual questions (range from 12 to 60; range in sample: 32 to 60).

Discussion

We describe a novel, precision-randomized design in which arm allocation probabilities varied by estimated risk for failure. Using this design, we were able to recruit 3000 people with a history of injection drug use from 7 diverse cities across India in the middle of a global pandemic. Flexibility was required to ensure the safety of study staff and participants in light of the COVID-19 pandemic. We observed significant heterogeneity in characteristics of participants and the prognostic score by study site, which may have implications for outcomes.

The STOP-C study represents one of the largest trials of HCV treatment among predominantly actively injecting PWID globally and one of the only trials from an LMIC. The success of HCV elimination efforts will depend largely on their ability to reach PWID, particularly those actively using and not engaged in health services, including MOUD. Yet, to date, most of the data on the treatment of PWID derive from small samples from the United States,^43,50,51 Australia,^6,32,52,53 and Western Europe,^54-56 with limited data from LMICs.^57,58 The majority in those other studies were on MOUD and older (median age >40 years). The STOP-C study represents a unique sample in that it includes predominantly younger PWID (75% are under 30 years of age), most of whom are actively injecting and are in and out of care for HIV and opioid use disorder. Even with a limited sample size, multiple studies have demonstrated poorer outcomes among those younger than 40 years, even in non-substance-using populations.^43,59

The prognostic score generated using preliminary data from our own clinical settings on early ART adherence had only moderately high predictive accuracy. Moreover, while we chose a cut-point to ensure that at least one-third of the population was included in the minimal-risk stratum, we actually observed the reverse—that two-thirds of the population were classified as minimal risk. This likely reflects the overall higher risk profile of PWID living with HIV⁶⁰ who were included in score development. Importantly, subsequent analyses can further examine the performance of the score, determine whether an alternative cutoff or approach might have been more appropriate, and can externally validate the score within other populations, which will improve generalizability. Importantly, by applying inverse probability of treatment weights, we can still answer questions about the overall population—for example, is high-intensity support a superior intervention to low-intensity support in all PWID, regardless of propensity for failure?

In summary, the design of the STOP-C trial will allow important questions to be answered about the optimal strategy to ensure HCV cure among PWID. Moreover, this precision-randomized design represents a unique approach with likely increased applicability. As resources become increasingly limited in the setting of persistent and/or new public health challenges, designs such as this offer an important opportunity to maximize efficiencies while still allowing for the estimation of efficacy. This design offers a particularly relevant option as medicine and public health embrace the use of precision medicine and artificial intelligence in health care.

Supplementary material

Supplementary material is available at the American Journal of Epidemiology online.

Funding

This work was supported by the National Institutes of Health (K24DA035684, P30AI094189, R01AI145555, R01DA041034, DP2DA040244).

Conflict of interest

S.H.M. received materials support from Abbott. D.L.T. reports medical editing for UpToDate and scientific consulting for Merck, Excision Bio, and Evrys; none are related to this study. S.S.S. reports grants and products from Gilead Sciences to institutions not related to this study and products from Abbott Laboratories to institutions related to this study, reports honoraria from Gilead Sciences and Abbott Laboratories, and serves as the Managing Trustee of the YR Gaitonde Medical Educational and Research Foundation and on the Board of Directors of the Serious Fun Children’s Network. Other authors: none declared.

Data availability

The data that support the findings of this study are available from the corresponding author (S.H.M.) upon reasonable request.