Correction of estimates of retention in care among a cohort of HIV-positive patients in Uganda in the period before starting ART: a sampling-based approach

OBJECTIVE: The aim of this study was to use a sampling-based approach to obtain estimates of retention in HIV care before initiation of antiretroviral treatment (ART), corrected for outcomes in patients who were lost according to clinic registers. DESIGN: Retrospective cohort study of HIV-positive individuals not yet eligible for ART (CD4 >500). SETTING: Three urban and three rural HIV care clinics in Uganda; information was extracted from the clinic registers for all patients who had registered for pre-ART care between January and August 2015. PARTICIPANTS: A random sample of patients who were lost according to the clinic registers (>3 months late to scheduled visit) was traced to ascertain their outcomes. OUTCOME MEASURES: The proportion of patients lost from care was estimated using a competing risks approach, first based on the information in the clinic records alone and then using inverse probability weights to incorporate the results from tracing. Cox regression was used to determine factors associated with loss from care. RESULTS: Of 1153 patients registered for pre-ART care (68% women, median age 29 years, median CD4 count 645 cells/µL), 307 (27%) were lost according to clinic records. Among these, 195 (63%) were selected for tracing; outcomes were ascertained in 118 (61%). Seven patients (6%) had died, 40 (34%) were in care elsewhere and 71 (60%) were out of care. Loss from care at 9 months was 30.2% (95% CI 27.3% to 33.5%). After incorporating outcomes from tracing, loss from care decreased to 18.5% (95% CI 13.8% to 23.6%). CONCLUSION: Estimates of loss from HIV care may be too high if based on routine clinic data alone. A sampling-based approach is a feasible way of obtaining more accurate estimates of retention, accounting for transfers to other clinics.