Attrition from HIV treatment after enrollment in a differentiated service delivery model: a cohort analysis of routine care in Zambia

Abstract

Background. Many sub-Saharan Africa countries are scaling up differentiated service delivery (DSD) models for HIV treatment to increase access and remove barriers to care. We assessed factors associated with attrition after DSD model enrollment in Zambia, focusing on patient-level characteristics. Methods: We conducted a retrospective record review using electronic medical records (EMR) of adults (15+ years) initiated on antiretroviral (ART) between 01 January 2018 and 30 November 2021. Attrition was defined as lost to follow-up (LTFU) or died by November 30, 2021. We categorized DSD models into eight groups: fast-track, adherence groups, community pick-up points, home ART delivery, extended facility hours, facility multi-month dispensing (MMD, 4-6-month ART dispensing), frequent refill care (1-2 months), and conventional care (3 months). We used multivariate proportional hazards regression to assess patient-level factors associated with attrition, stratified by sex and rural/urban setting. Results: Of 517,579 eligible patients in the database, 60% (n=310,678) enrolled in DSD models, most commonly facility MMD (n=292,415, 94%) or fast track (n=13,985, 5%), with <0.5% enrolled in extended clinic hours, home ART delivery, community pick-up points, and adherence groups. Retention was high (92-96%) for all models including conventional care, with the exception of frequent refill care, where retention was substantially lower at 78%. Retention was higher in nearly all DSD models for all dispensing intervals, compared to conventional care. Rural males benefited most from DSD model enrollment, with adjusted hazard ratios for attrition that were significantly lower (0.10-0.45) than for conventional care for all DSD models except frequent refill care. Patients in frequent refill care had poorer retention than conventional care for both sexes and settings. Limitations of the study included potential selection bias into DSD models, differences in target populations for different models, and incomplete EMR data. Conclusion: Rates of retention in HIV treatment differed by DSD type, dispensing interval, and patient characteristics in Zambia. Understanding the factors that influence the retention of patients in DSD models could provide an important step towards improving DSD implementation.