Mortality and retention after entry into HIV care are crucial metrics of effectiveness of services in a region. However, under routine program conditions, deaths are underreported and overall estimates of mortality and retention and site-to-site variability in these outcomes may be biased. As a result, these vital metrics of performance cannot be reliably used for program improvement practices.
We undertook a multi-stage sampling approach in which we selected a probability sample of facilities from a network of 70 facilities offering HIV care and treatment in Zambia, and then selected a simple random sample of lost patients within each facility for intensive vital status ascertainment. Deaths and in-care patients identified by tracing lost patients were used to revise overall outcome estimates through probability weights.
Among a cohort of 54,172 patients newly starting ART in 30 facilities (63% women, median age 35 years at enrollment (IQR: 29-42) and median ART initiation CD4 level of 266 cells/mm3 (IQR: 141-395)), 11,152 were lost to follow up over two years (20%). The median clinic-level loss to follow up was 26% (IQR: 22%-28%, range 20% to 40%). Among a random sample of 18% of all lost patients, 75% of outcomes were ascertained. Median clinic mortality among the lost across the 30 sites was 12% (IQR 8% to 17%, range 4% to 27%). Once outcomes among the lost were incorporated into overall estimates of mortality in the entire cohort, mortality at two years from ART rose from 2% (95% CI: 2% to 2%) to 9% (95% CI: 8% to 10%). Median site level mortality across the 30 sites was 9% (IQR: 5% to 15%, range 3% to 19%). The median ratio difference was 6% fold with IQR of 4-fold to 12-fold (Figure). Estimated retention across all sites was 61% at one year and 42% at two years, which rose to 86% and 78% upon incorporation of outcomes ascertained by tracing.
Routine program monitoring underestimated both mortality and retention outcomes, thus threatening to undermine assessments of public health effectiveness. A sampling based approach revealed that both the extent of underestimation and the actual revised mortality estimates differed markedly across facilities. Improved assessments of mortality and retention by facility can inform where to target efforts for improved outcomes.