Early outcomes after implementation of treat all in Rwanda: an interrupted time series study

Introduction: Nearly all countries in sub-Saharan Africa have adopted policies to provide antiretroviral therapy (ART) to all persons living with HIV (Treat All), though HIV care outcomes of these programmes are not well-described. We estimated changes in ART initiation and retention in care following Treat All implementation in Rwanda in July 2016. Methods: We conducted an interrupted time series analysis of adults enrolling in HIV care at ten Rwandan health centres from July 2014 to September 2017. Using segmented linear regression, we assessed changes in levels and trends of 30-day ART initiation and six-month retention in care before and after Treat All implementation. We compared modelled outcomes with counterfactual estimates calculated by extrapolating baseline trends. Modified Poisson regression models identified predictors of outcomes among patients enrolling after Treat All implementation. Results: Among 2885 patients, 1803 (62.5%) enrolled in care before and 1082 (37.5%) after Treat All implementation. Immediately after Treat All implementation, there was a 31.3 percentage point increase in the predicted probability of 30-day ART initiation (95% CI 15.5, 47.2), with a subsequent increase of 1.1 percentage points per month (95% CI 0.1, 2.1). At the end of the study period, 30-day ART initiation was 47.8 percentage points (95% CI 8.1, 87.8) above what would have been expected under the pre-Treat All trend. For six-month retention, neither the immediate change nor monthly trend after Treat All were statistically significant. While 30-day ART initiation and six-month retention were less likely among patients 15 to 24 versus >24 years, the predicted probability of both outcomes increased significantly for younger patients in each month after Treat All implementation. Conclusions: Implementation of Treat All in Rwanda was associated with a substantial increase in timely ART initiation without negatively impacting care retention. These early findings support Treat All as a strategy to help achieve global HIV targets. To study the impact of Treat All implementation in July 2016, we conducted an interrupted time series analysis of clinical data from July 2014 through November 2017. We utilized routinely collected data from an open observational cohort of patients receiving HIV care at ten Rwandan health centres affiliated with the Central Africa International epidemiologic Databases to Evaluate AIDS (CA‐IeDEA; www.iedea-ca.org). CA‐IeDEA is a multi‐country project that collects secondary data from patients receiving HIV care and treatment in the Central African region and is one of seven regions that comprise the global IeDEA network (www.iedea.org). The ten health centres in Rwanda have been previously described 17. All research was conducted according to the principles of the Helsinki Declaration and was approved by the Rwanda National Ethics Committee and the Albert Einstein College of Medicine Institutional Review Board, both of which waived written or verbal patient consent because the data were de‐identified prior to extraction into research database. This study is reported in accordance with the STROBE statement for reporting of observational studies (Table S1). We included all persons ≥15 years of age newly enrolling in care at health centres affiliated with CA‐IeDEA from 1 July 2014 through 13 September 2017 (90 days prior to the close of dataset). Because we focused on patients newly initiating HIV care, persons known to have transferred from another facility (N=323), as well as participants receiving HIV care >30 days prior to enrolment, and thus likely to be transfers in (N=415), were excluded (Figure S1). In July 2014, Rwanda had fully implemented guidelines recommending provision of ART to all adults (≥15 years) with CD4 count <500 cells/mm3, as well as all pregnant or breastfeeding women and all patients co‐infected with tuberculosis or viral hepatitis 18. In July 2016, national HIV treatment guidelines were expanded to include ART for all persons with HIV regardless of disease stage or CD4 count 19; all health centres included in this analysis reported implementation of these guidelines in July 2016. The 2016 guidelines also recommended ART initiation within seven days of diagnosis, attending medical consultations every three months, and monthly ART pick‐up from health centre pharmacies. Each participating health centre routinely collects demographic, clinical and laboratory data as part of clinical care using standardized paper forms; these data are regularly entered into electronic databases. Patient data were de‐identified prior to extraction into the research database. Health centre characteristics were obtained as part of a site assessment periodically conducted at all sites participating in the global IeDEA network 20. Two primary outcomes were considered in this analysis: ART initiation within 30 days of enrolment and six‐month retention in care. We used the date of enrolment into HIV care as specified in health centre data; we defined ART initiation date as the date of the first ART prescription ordered after enrolment. All patients were included in analyses of ART initiation. We defined six‐month retention as having at least one health centre visit within five to nine months after enrolment. All patients whose enrolment was more than nine months before the close of the dataset and were not known to have died or transferred out prior to the six‐month visit window were included in analyses of this outcome. As secondary outcomes, we also examined the proportion of patients ever initiating ART and the number of days between enrolment and ART initiation. Because viral load measurement was performed on <10% of patients who entered care in the pre‐Treat All period and <50% of those entering care during the Treat All period, we did not analyse viral suppression as an outcome. Baseline demographic and clinical variables included sex (female or male), age, body mass index (<18.5 vs. ≥18.5 kg/m2), referral source into HIV care (voluntary counselling/testing programme (VCT), maternal/prenatal care, other), WHO stage (I‐II vs. III‐IV) and CD4 count (categorized as <200, 200 to 349, 350 to 500 and ≥500 cells/μL), measured up to 90 days after enrolment. Health centre characteristics included location (urban vs. peri‐urban), size (≥2000 vs. <2000 patients with HIV in care), whether adolescents and adults were seen in separate clinics or patients of all ages were seen together, availability of physicians, mid‐level providers and adherence counsellors (all or some of the time vs. none of the time), whether sites provided incentives (such as mobile phone airtime vouchers or transportation costs) for early enrolment in care, the number of pre‐ART counselling sessions typically occurring at the health centre (four or more vs. less than four), and availability of adherence support including medication review and referrals to mental health counselling or peer support. For all variables, missing values were categorized as such. We defined two periods for the study: the pre‐Treat All period (July 2014 through June 2016), and the Treat All period (July 2016 through November 2017). Baseline characteristics of patients enrolling in the pre‐Treat All and Treat All periods were compared using bivariate logistic regression models that accounted for clustering within health centres. We used the Kaplan–Meier method to estimate median time from enrolment to ART initiation. For the interrupted time series analysis, we used segmented linear regression models to estimate the predicted probability of initiating ART within 30 days of enrolment and six‐month retention in care among patients entering care in each month. To do this, we fit a separate model for each outcome as follows: In these models, Yt is the independent outcome (predicted probability), β 0 estimates the baseline level at the beginning of the study period, β 1 estimates the linear trend before Treat All implementation, β 2 estimates the immediate level change (i.e. jump) after Treat All implementation and β 3 estimates the change in linear trend after Treat All implementation relative to the pre‐Treat All trend. Models accounted for clustering within health centres and did not include data on patients enrolling in care from 1 June to 31 July 2016 to account for a two‐month transition period of Treat All guideline implementation. For both primary outcomes, we plotted the proportion of patients enrolling in each month who achieved each outcome as well as fitted values from the segmented regression models described above. We also calculated counterfactual values by extending the pre‐Treat All regression models (i.e. not including the Treat All change and Treat All trend terms). We then calculated differences between the observed and expected (counterfactual) outcomes at the last month of follow‐up. We calculated 95% confidence intervals (CIs) using the bootstrap method 21. Sub‐analyses were performed using similar models to examine levels and trends by sex, age, referral source and baseline CD4 count. Finally, we examined predictors of initiating ART within 30 days and of six‐month retention in care among patients enrolling after Treat All guidelines. For these analyses we utilized modified Poisson regression models with robust variances to calculate crude and adjusted risk ratios (RRs), with generalized estimating equations to account for clustering within health centres. Multivariate models were adjusted for all individual demographic and clinical characteristics but were not adjusted for health centre characteristics given the relatively small number of centres (N=10). Data were analysed using SAS 9.4 (SAS Institute Inc., Cary, NC); statistical significance for all tests was two‐sided at p < 0.05. In sensitivity analyses, to account for differences in ART eligibility criteria, we examined the proportion of patients initiating ART and time from enrolment to ART initiation excluding patients enrolling in care before Treat All who were not eligible for ART (i.e. those with CD4 500 cells/mm3). We then repeated the segmented regression analysis of ART initiation within 30 days limited to patients eligible for ART. Similarly, to account for the potential influence of ART initiation on retention in care, we repeated the segmented regression analysis of six‐month retention in care limited to patients who initiated ART. Finally, to determine whether use of missing indicators biased results, we modelled predictors of ART initiation within 30 days and six‐month retention in care using a complete case analysis that excluded missing values. The funders of the study had no role in the study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.