The science of implementation begins with the ability to measure important occurrences in routine, day-to-day clinical care. Although the provision of care and treatment for HIV-infected patients in Africa over the last five years has scaled up rapidly, loss to follow-up (i.e., unknown outcomes) commonly exceeds 40% at 1 year. By precluding knowledge of key events, such as mortality and retention in care, losses systematically impair our ability to measure occurrences in clinical care for the conduct implementation science.

In response, our group has developed a novel sampling-based approach to manage the impact of losses to follow up on the conduct of implementation science. This approach (a) recognizes that the high absolute number of patients lost to follow-up precludes complete outcome ascertainment; (b) identifies outcomes in a representative but numerically small sample of lost patients in the community; (c) represents these outcomes through probability weights in epidemiologic analyses and (d) provides corrected estimates of both the magnitude and the determinants of outcomes such as mortality and retention in care. Over the last 18 months, we have employed this approach in a large HIV clinic in rural Uganda and shown that estimates of mortality are vastly underestimated, retention in care is also underestimated and, subsequently, that analyses to identify determinants of mortality are grossly biased.

Before, however, this approach can be more widely recommended as a strategy to strengthen the evaluation of HIV programs in Africa, we must first establish its generalizability. To this end, this CFAR supports an effort to implementing this approach in a network of HIV care and treatment sites in the Family AIDS Care and Education Services (FACES) program in Kenya (PI: Dr. Craig Cohen), which has worked closely with the AIDS Services, Prevention, Intervention, Research and Education (ASPIRE) program (PI: Dr. Diane Havlir) to provide HIV care and services to over 78,561 patients in Western Kenya over the last 5 years in order to obtain sample corrected estimates of mortality and retention in care. We will also extend our understanding of the tracking by collecting data on the process itself, such as data on the tracker's, characteristics, locator information, the time used to seek lost patients, the number of informants interviewed in the path, and the distances covered, to evaluate the determinants of successful tracking.

By providing for accurate estimates of the outcomes of the ART roll out as they occur in everyday practice in Africa, our work forms the epidemiologic cornerstone for subsequent "T2" translational research involving implementation and dissemination sciences. If this sampling-based approach can be shown to be widely applicable in resource-limited settings, it may become the standard for scientists seeking to understand the effect of specific programmatic practices.