Statistical Analysis Plan for the Primary and Selected Secondary Endpoints in the ENHANCED-SPS Study

1.1 Study design

ENHANCED-SPS is an Implementation Science study, designed to rigorously evaluate the effects of a multi-component intervention, including viral load counseling and peer support, on care outcomes among pregnant and breastfeeding women with HIV in South Western Uganda. Since the intervention involves clinic-level changes, 14 government-run HIV clinics were randomized to the intervention or the standard-of-care. Randomization was conducted by an independent statistician using a computer-based random number generator and within strata defined by distance, patient burden (HIV prevalence and the number of new HIV diagnoses), and other socioeconomic characteristics (e.g., main occupation of the catchment area of the study clinics).

To facilitate examination of secular trends, retrospective data were collected from all trial clinics (regardless of randomization arm) in the 18-months leading up to intervention initiation. For simplicity, we refer to the pre-intervention phase as “Phase 1” and the intervention phase as “Phase 2”. This document focuses on analyses of Phase 2 data only; analyses of Phase 1 data will be described elsewhere.

The primary objective of this study is to determine whether the intervention improved viral suppression after 12 months of follow-up. Key secondary endpoints include disclosure of HIV status, ART adherence, and retention in care. We also plan to evaluate predictors of these endpoints and describe changes in ART regimen during follow-up.

1.2 Description of population and context

Throughout, the target population is pregnant and postpartum women with HIV and accessing care from public health facilities in South Western Uganda. Women with HIV accessing care at the study clinics were eligible for trial enrollment if they were (1) pregnant, previously diagnosed with HIV, and attending the antenatal clinic (ANC) for the first time; (2) pregnant and newly diagnosed with HIV at their first ANC visit; or (3) breastfeeding and newly diagnosed with HIV at the postnatal visit. All eligible and consenting women were consecutively enrolled until the desired clinic-specific sample size was attained (Appendix A.2).

We will provide a participant flow diagram (i.e., a CONSORT diagram) and describe the study population with summary measures of baseline participant characteristics, including age, marital status, enrollment group (pregnant and previously diagnosed; pregnant and newly diagnosed, or breastfeeding and newly diagnosed), ART regimen, and HIV viral suppression status (HIV RNA< 1000 c/mL). Throughout, “baseline” refers to the time of enrollment into Phase 2, when intervention delivery began. We will provide these summary measures overall and by arm. We may also provide these summary measures by clinic.

2.1 Missing data

Despite regular viral load monitoring, offered every 6 months or more frequently for pregnant and breastfeeding women, viral load data may be subject to missingness at baseline and at endline. For some women, this missingness may reflect being out of care. For others, this missingness may reflect shortages of viral load assays. Yet, for others, this missingness may reflect gaps in the data system. Given our reliance on routinely collected viral loads and on the EMR system, we anticipate missingness at baseline and endline will be greater in the control arm than the intervention arm.

Our ability to adjust for differential measurement is attenuated by the magnitude of missing outcomes. As a result, we pre-specify the following approach:

1. If 50% or greater of control participants have an endline viral load, we will compare endline viral suppression by arm, as detailed below.

2. If fewer than 50% of control participants have an endline viral load, we will compare the change in viral suppression from baseline to endline in the intervention arm only, as detailed below.

The approach taken for viral suppression, the primary endpoint, will determine the approach taken for the secondary endpoints, specified below.

2.2 By arm comparison

If there are sufficient data to support a by arm comparison of endline viral loads (Section 2.1), we will conduct the following analysis. Using targeted minimum loss-based estimation (TMLE),^1–5 we will compare the proportion of participants with viral suppression after 12 months of follow-up between arms. TMLE accounts for clustering and allows for flexible adjustment for the ways in which participants with measured outcomes could differ from participants missing outcomes. Our adjustment set will consist of study arm, age, enrollment group, and baseline viral suppression status. To minimize the risk of model misspecification bias, we will use Super Learner, an ensemble machine learning method, to combine predictions from generalized linear models, stepwise regression, and the mean.⁶ In the primary analysis, we will use a Single-stage TMLE that pools across clinics to estimate and compare the proportion of participants with viral suppression between arms. In a sensitivity analysis, we will use Two-Stage TMLE to first estimate the proportion with viral suppression in each clinic, accounting for missing outcomes, and then compare the estimated clinic-specific proportions by arm. In an additional sensitivity analysis, we will conduct an unadjusted, complete-case analysis. Specifically, we will restrict to participants with measured outcomes and compare the proportion with viral suppression between arms. All analyses will exclude participants who are missing both baseline and endline viral loads.

All analyses will also account for dependence of participants within clinics. Specifically, for statistical inference, we will aggregate the influence curve to the clinic level,^4,5 and use the Student’s t-distribution with 14-2=12 degrees of freedom as finite sample approximation to the standard normal distribution.⁷ We test the null hypothesis that the ENHANCED-SPS intervention did not improve viral suppression, as compared to the standard-of-care, with a one-sided test at the 5% significance level. We will report point estimates and two-sided 95% confidence intervals for the arm-specific endpoints. The primary analysis will be on the difference scale and for the study population (i.e., the sample average treatment effect [SATE]).^2,8 We will also examine relative effects in secondary analyses.

2.3 Within arm comparison

If there are not sufficient data to support a by arm comparison of endline viral loads (Section 2.1), we will conduct the following analysis within the intervention arm only. In the primary analysis, we will use a Single-Stage TMLE to compare the proportion with viral suppression at endline versus baseline, accounting for clustering and for the ways in which participants with measured viral loads could differ from participants with missing viral loads. As described in Section 2.2, we will use Super Learner for flexible estimation; here, our adjustment set will include age and enrollment group. In a sensitivity analysis, we will use Two-Stage TMLE to first estimate the proportion with viral suppression at baseline and endline in each clinic (accounting for missing viral loads) and then compare the estimated clinic-specific proportions over time. In an additional sensitivity analysis, we will conduct an unadjusted, complete-case analysis. As before, all analyses will exclude participants who are missing both baseline and endline viral loads. We will use the same approach for statistical inference and test the null hypothesis that the ENHANCED-SPS intervention did not improve viral suppression over time with a one-sided test at the 5% significance level.

2.4 Subgroups

To understand effect heterogeneity, we will repeat the above analyses within the following pre-specified baseline subgroups: age group (15-24 years, 25-34 years, >35 years), enrollment group, and marital status. If conducting a by arm analysis, we will also include baseline viral suppression status as a subgroup. Additionally, given the rollout of dolutegravir (DTG) during the study, we will repeat these analyses stratified by women who did and did not switch to DTG during follow-up. Participants with missing data on the variable determining their subgroup classification will be excluded from that analysis.

2.5 Predictors

In addition to subgroup analyses, we will report the number and proportion of participants who were not suppressed at baseline and achieved viral suppression at endline – overall and within subgroups. Likewise, we will report the number and proportion of participants who sustained viral suppression (i.e., had viral suppression at both baseline and endline). Additionally, we may conduct the following analyses to understand predictors of viral non-suppression at endline. Within each arm separately, we will estimate unadjusted associations and adjusted associations of viral non-suppression using as predictors: age, enrollment group, marital status, baseline ART regimen, and baseline HIV viral suppression. All analyses will be implemented with the Single-Stage TMLE and yield associations on the relative scale (statistical analogs of the causal risk ratio).