Measuring Outcomes of Antiretroviral Treatment in Africa

Baryahikwa travels to remote areas of the countryside in an effort to measure the success of one of the world's greatest public health challenges of this generation: treating millions of HIV-infected persons with lifesaving, but complicated and potentially toxic regimens of antiretroviral therapy in resource-limited countries. More than 3 million persons have started therapy for HIV treatment in the last five years in the antiretroviral scale-up, and 10 million others will start in the next several years. Now that therapy is getting into the hands of patients who need it the most, the questions turn to what impact these treatments are having. Understanding outcomes in the scale-up, however, turns out to be more difficult than it appears. Throughout treatment programs in Africa, many patients are lost to follow-up, meaning that they have not returned to the clinic where they received their initial therapy. Whether these patients have transferred their care elsewhere, stopped taking their medications or have died is not known. From a public health point of view, the success of the scale-up programs will be judged by how many patients survive and how many die. But when losses are upward of 20 percent in one year, and known deaths are less than 5 percent, the true number and percentage of deaths are unknown. Without this information, public health evaluation of the scale-up programs faces a critical barrier. At UCSF, Elvin Geng, MD, a fellow in the Division of Infectious Diseases — along with David Glidden, PhD, professor, and Jeffrey Martin, MD, MPH, associate professor, Department of Epidemiology and Biostatistics — is working on a simple technique to address this problem. In essence, they are using the tried and true approach of sampling. For example, consider that there are 5,000 patients in a treatment program, and 1,000 of them are lost. It would be impossible in resource-limited settings to look for all 1,000 patients. But if 100 in a random sample are sought for and their current status determined, then their outcomes are an accurate representation of all 1,000 lost patients. “If one adds the deaths in the lost patients to the deaths that are already known to the clinic, one can obtain a true understanding of the number of patients who have died in a program,” explains Geng. Understanding Survival in Uganda Serving the local, regional and global communities to eliminate health disparities is part of the vision outlined in the UCSF Strategic Plan, which was unveiled in June 2007. Specifically, UCSF aims to position UCSF as a leader in global health and to foster research collaborations with other academic institutions. Working with investigators from the Mbarara University of Science and Technology in Mbarara, Uganda, the UCSF team applied their sampling approach to 3,628 HIV-infected adults who had begun antiretroviral therapy between Jan. 1, 2004, and Sept. 30, 2007. Of these 3,628 patients, 829 patients became lost to follow-up. A sample of 128 of the lost was sought by a tracker, and the current status, alive or dead, of 111 (87 percent) was ascertained. Applying a probability weight to allow the sampled patients to represent all lost patients changed the estimate of survival markedly. Without the sample, the one-, two— and three-year death rates were 1.7 percent, 2.1 percent and 2.3 percent, respectively. But with the sample correction, the estimates became 7.5 percent, 10.3 percent and 12.2 percent. These changes are likely to significantly alter our understanding of how scale-up programs are doing. “Without understanding survival, it becomes impossible to move forward,” says Martin. “Since survival is the more important measure of how successful the treatment programs are, a valid estimate is necessary to inform the next generation of public health strategies.” The motivation for arriving at a valid answer came from the UCSF Clinical and Translational Science Institute (CTSI), which seeks to use multidisciplinary science to translate established approaches into real-life practice. Indeed, the project blended together clinical, epidemiological and biostatistical sciences. Added Glidden, “The technique of using sampling to address this problem had been described in the statistical literature in the past, but had never been applied in a real-life public health setting of this importance.” “The success of this approach depends on how many of the looked-for patients the tracker can find,” Geng said. “In fact, a health educator at the clinic looks for these lost patients on a motorcycle. Few patients have phones and roads are limited. Many people might wonder whether any patients can be found at all. It turns out that the skill and knowledge of the patient tracker, as well as the tight social networks in the countryside in Uganda, make finding patients possible.” The next step, according to Geng, is to use the sampling technique to determine factors related to favorable outcomes while taking antiretroviral therapy. Certain kinds of losses to follow-up can alter apparent associations between predictors and survival. These might lead to an incorrect understanding of the predictors of survival and outcome and misinform public health leaders. For example, without adjusting for deaths in the sampled patients, women appeared to have a high rate of death in this clinic. But once the number of sample-based deaths was used, the association went away. These kinds of “errors” are particularly dangerous, since they could lead to misguided strategies and policies. The findings were published in the August 6, 2008, HIV/AIDS theme issue of the Journal of the American Medical Association. Other UCSF investigators participating in the project were David Bangsberg, MD, MPH, and Steven Deeks, MD, of the Department of Medicine.