Person:

Mitchell, Shira Arkin

Background: High quality program data is critical for managing, monitoring, and evaluating national HIV treatment programs. By 2009, the Malawi Ministry of Health had initiated more than 270,000 patients on HIV treatment at 377 sites. Quarterly supervision of these antiretroviral therapy (ART) sites ensures high quality care, but the time currently dedicated to exhaustive record review and data cleaning detracts from other critical components. The exhaustive record review is unlikely to be sustainable long term because of the resources required and increasing number of patients on ART. This study quantifies the current levels of data quality and evaluates Lot Quality Assurance Sampling (LQAS) as a tool to prioritize sites with low data quality, thus lowering costs while maintaining sufficient quality for program monitoring and patient care. Methods: In January 2010, a study team joined supervision teams at 19 sites purposely selected to reflect the variety of ART sites. During the exhaustive data review, the time allocated to data cleaning and data discrepancies were documented. The team then randomly sampled 76 records from each site, recording secondary outcomes and the time required for sampling. Results: At the 19 sites, only 1.2% of records had discrepancies in patient outcomes and 0.4% in treatment regimen. However, data cleaning took 28.5 hours in total, suggesting that data cleaning for all 377 ART sites would require over 350 supervision-hours quarterly. The LQAS tool accurately identified the sites with the low data quality, reduced the time for data cleaning by 70%, and allowed for reporting on secondary outcomes. Conclusions: Most sites maintained high quality records. In spite of this, data cleaning required significant amounts of time with little effect on program estimates of patient outcomes. LQAS conserves resources while maintaining sufficient data quality for program assessment and management to allow for quality patient care.

Rapid and accurate identification of the prevalence of schistosomiasis is key for control and eradication of this devastating disease. The current screening standard for intestinal schistosomiasis is the Katz-Kato method, which look for eggs on slides of fecal matter. Although work has been done to estimate prevalence using the number of eggs on a slide, the procedure is much faster if the laboratory only reports the presence or absence of eggs on each slide. To further help reduce screening costs while maintaining accuracy, we propose a pooled method for estimating prevalence. We compare it to the standard individualed method, investigating differences in efficiency, measured by the number of slides read, and accuracy, measured by mean square error of estimation. Complication is introduced by the unknown and varying sensitivity of the procedure with population prevalence. The DeVlas model for the worm and egg distributions in the population describes how test sensitivity increases with age of the epidemic, as prevalence and intensity of infection increase, making the problem fundamentally different from earlier work in pooling. Previous literature discusses varying sensitivity with the number of positive samples within a pool, known as the “dilution effect.” We model both the dilution effect and varying sensitivity with population prevalence. For model parameter values suited to younger age groups, the pooled method has less than half the mean square error of the individualed method. Thus, we can use half as many slides while maintaining accuracy. Such savings might encourage more frequent measurements in regions where schistosomiasis is a serious but neglected problem.

A relatively recent increase in the popularity of evidence-based activism has created a higher demand for statisticians to work on human rights and economic development projects. The statistical challenges of revealing patterns of violence in armed conflict require efficient use of the data, and careful consideration of the implications of modeling decisions on estimates. Impact evaluation of a complex economic development project requires a careful consideration of causality and transparency to donors and beneficiaries. In this dissertation, I compare marginal and conditional models for capture recapture, and develop new hierarchical models that accommodate challenges in data from the armed conflict in Colombia, and more generally, in many other capture recapture settings. Additionally, I propose a study design for a non-randomized impact evaluation of the Millennium Villages Project (MVP), to be carried out during my postdoctoral fellowship. The design includes small area estimation of baseline variables, propensity score matching, and hierarchical models for causal inference.