Background
Disengagement from antiretroviral treatment undermines HIV programme success. Current approaches targeting interventions at specific demographic groups miss most people out of care. Classifying patterns of engagement could use precision public health principles to identify unique care needs, and inform service development targeted at patients with similar experiences.
Methods
Using routine health data from the Western Cape Provincial Health Data Centre, South Africa, we analysed individuals ≥15 years old who initiated treatment after 01-Sep-2016, sought care in low-resource, high-HIV burden settings, and had ≥5 years follow-up before database closure (30-Sep-2022). Group-based trajectory-modelling identified subgroups with distinct longitudinal engagement patterns from initiation (disengagement defined using pharmacy refill data as a treatment interruption >90 days). We assessed subgroup associations with viral suppression ≤1000 copies/mL with logistic regression.
Results
This cohort included 26422 individuals: 71% female with median of age 31 years (IQR 26-38) at initiation. The best-fitting model identified five subgroups (figure): 1) Optimal engagement (38%); 2) Early disengagement (25%); 3) Delayed disengagement after two years (15%); 4) Disengagement with early return by two years (12%); and 5) Disengagement with delayed return by five years (10%). No variables were strongly predictive of subgroup membership, and subgroups were similar in age, sex, comorbidities, baseline CD4 count, regimen, and duration on treatment. All suboptimal patterns were strongly associated with lower odds of suppression at five years than the optimally engaged subgroup (adjusted odds ratios [95% confidence interval]: 0.05 [0.04-0.06], 0.15 [0.13-0.17], 0.64 [0.56-0.73], and 0.33 [0.29-0.39] for subgroups 2, 3, 4, and 5 respectively), with stronger associations compared to demographic and clinical variables.
Conclusions
Disengagement was common, and most interrupted treatment (>90 days) within five years from initiation. However, engagement was not binary: people disengaged in various ways. Patterns of disengagement behaviour were more predictive of viral suppression than demographic or clinical factors, and could prove useful to target resources. It was difficult to predict subgroup membership with routine data, so further exploration of lived experiences behind each pattern could help identify what each subgroup needs to support sustained engagement, and facilitate more person-centred service delivery.