Triaging Clients at Risk of Disengagement from HIV Care: Application of a Predictive Model to Clinical Trial Data in South Africa

Abstract

Purpose: To reach South Africa's targets for HIV treatment and viral suppression, retention on antiretroviral therapy (ART) must increase. Here, we aim to successfully identify ART clients at risk of loss from care prior to disengagement.

Patients and methods: We applied a previously developed machine learning and predictive modelling algorithm (PREDICT) to ART client data from SLATE I and II trials. The primary outcome was interruption in treatment (IIT), defined as missing the next scheduled clinic visit by >28 days. We tested two risk triaging approaches: 1) threshold approach classifying individuals into low, moderate, or high risk of IIT; and 2) archetype approach identifying subgroups with characteristics associated with risk of ITT. We report associations between risk category groups and subsequent IIT at the next scheduled visit using crude risk differences and relative risks with 95% confidence intervals.

Results: SLATE datasets included 7199 client visits for 1193 clients over ≤14 months of follow-up. The threshold approach consistently and accurately assigned levels of IIT risk for multiple stages of the care cascade. The archetype approach identified several subgroups at increased risk of IIT, including those late to previous appointments, returning after a period of disengagement, living alone or without a treatment supporter. Behavioural elements of the archetypes tended to drive the risk of treatment interruption more consistently than demographics; eg adolescent boys/young men who attended visits on time experienced the lowest rates of treatment interruption (10% PREDICT datasets; 7% SLATE datasets), while adolescent boys/young men returning after previously disengaging had the highest rates of subsequent treatment interruption (31% PREDICT datasets; 40% SLATE datasets).

Conclusion: Routinely collected medical record data can be combined with basic demographic and socioeconomic data to assess individual risk of future treatment disengagement. This approach offers an opportunity to prevent disengagement from HIV care, rather than responding only after it has occurred.

Trial registration: SLATE I trial: Clinicaltrials.gov NCT02891135, registered September 1, 2016. First participant enrolled March 6, 2017, in South Africa and July 13, 2017, in Kenya. SLATE II trial: Clinicaltrials.gov NCT03315013, registered 19 October 2017. First participant enrolled 14 March 2018.

Keywords: HIV service delivery; machine learning; predictive modelling; retention; risk triaging.

Figure 1

Proportion of visits classified as IIT stratified by risk threshold approach and time on ART (PREDICT model data).

Figure 2

Relative risk of interruption in treatment (IIT) at next scheduled visit stratified by risk archetype and demographic strata.

Figure 3

Schematic of intervention mapping guided by behavioural archetype.