Analytic methods

Power was calculated for the primary study outcome of undetectable HIV viral load at 12 months post randomization and assumed a two-sided alpha level of 0.05. It was expected that 240 participants would be enrolled into the trial. Based on data from the original LINC study, we expected 10% of controls to have undetectable viral load at 12 months. Given this and assuming 20% loss to follow-up (i.e., 192 evaluable subjects) the study has 80% power to detect an absolute difference of 17% (i.e., 27% vs. 10% in the intervention and control arms, respectively) using a chi-square test with continuity correction. We anticipate larger effects may be observed in our study, which would result in even higher power.

This study will use an intent-to-treat analysis that includes all randomized participants according to their assigned group. Descriptive statistics will be calculated for variables at baseline to assess whether there appear to be any differences across treatment arms.

As the primary study outcome is HIV viral load suppression at 12 months, initial analyses will be performed comparing this binary outcome between groups using a chi-square test. The primary analysis will use multivariable logistic regression analyses to control for the stratification factor, ever ART use, to improve efficiency. If there are any baseline factors that appear to differ by randomized group, additional sensitivity analyses will be conducted controlling for these factors to assess for potential confounding. The secondary outcomes of undetectable viral load at 6 months, ART initiation within 28 days, and retention in HIV care will be analyzed using the same approach described above. Change in CD4 count between baseline and 12 months will be analyzed using multiple linear regression. If the distribution of change in CD4 is skewed, transformations of the data will be considered. A median regression model will be utilized if an appropriate transformation is not identified [10, 11].

We will perform additional analyses to explore potential effect modifiers of the LINC-II intervention. The three potential effect modifiers of interest are: gender, ever ART use, and 30-day IDU. We will fit separate models including two-way interactions between randomization group and each potential effect modifier. If an interaction is significant, subsequent stratified analyses will be conducted to explore and describe the effect of the LINC-II intervention by categories of the moderator.

Exploratory analyses will be conducted to assess potential mechanisms that may drive LINC-II’s ability to improve HIV care outcomes using the Baron and Kenny approach [12]. The three potential mediators we will explore are decreases in substance use, HIV stigma, and substance use stigma. However, because the interpretation of the degree of mediation in logistic models is complicated by their inherent nonlinearity, we will conduct additional analyses using the recently developed causal inference approach to mediation (also referred to as the counterfactual framework), an approach that allows potential interactions between the intervention and mediators and derives direct and indirect effects for binary outcomes [13–15]. We will use the Stata mediation package to conduct these analyses [16, 17].

We will conduct descriptive statistics (e.g., means, medians, interquartile ranges, and confidence intervals) to assess perceptions and experiences of coordinated care using quantitative data from provider, administrator, and patient surveys over time. We will also analyze repeated measures of patients’ attitudes and experiences using mixed effects regression models controlling for randomized group to assess overall changes over time and to explore and describe potential differences between study arms. We will analyze qualitative interview data following a thematic approach [18]. Content analysis of qualitative data will reveal themes regarding care coordination and will identify best practices for LINC-II implementation in similar settings, (e.g., where addiction and HIV care systems are largely separate). Qualitative and quantitative results will be triangulated [19].

To estimate cost and cost-effectiveness, we will adapt methods developed by Rosen et al. [20, 21]. When all follow-up to the primary outcome (12-month viral suppression) has been completed, patient resource utilization will be extracted from patients’ medical records for HIV care and from study forms for narcology care in the intervention arm. Narcology care for patients in the control arm will be estimated from patient self-report at the time of 6 and 12-month outcome assessments. Unit costs will be obtained from published information, external suppliers, and the study sites’ finance and procurement records and applied to the resource usage data to provide an average cost per study participant. Costs will be measured from the provider perspective and will include the cost of all resources utilized for each study participant from the date of admission to the narcology hospital for a period of 12 months, including all drugs, laboratory tests, inpatient days, outpatient visits, case manager costs, and fixed costs such as building space, equipment, and administrative staff. For patients referred to local clinics, rather than the study hospital, for ongoing HIV care after ART initiation and/or for narcology care, fixed costs will be estimated at the facility level for a typical local clinic for each type of care.

We will estimate average cost with 95% confidence intervals to the provider per patient enrolled, per patient initiating ART, and per patient achieving viral suppression by 12 months. We will also estimate total cost per patient achieving the primary outcome, which takes into account all the costs for all the patients but divides by only the number with a successful outcome (i.e., 12-month viral suppression) and thereby links the cost of service delivery to the primary outcome. The cost-effectiveness of the intervention, compared to standard care, will also be estimated as an incremental cost per incremental outcome. The cost and cost-effectiveness results will then be evaluated in the context of existing healthcare budgets, resource availability (e.g., trained case managers), other relevant interventions that have been studied in Eastern Europe, and cost estimates for similar countries to help inform policy makers about the affordability and priority of scaling up the program.

In addition to the provider cost estimates described above, the baseline questionnaire will elicit information about patient costs of seeking care, such as transport fares, lost wages, and substitute labor costs (e.g., for childcare). The average cost to participants by arm and by outcome will be estimated and used both to help explain study results (e.g. there may be an association between patient costs and retention in care) and as a component of the overall economic evaluation.