Date of Award
Doctor of Philosophy (PhD)
Davis, J. Lucian
Surveillance is a cornerstone of public health, providing the data required to monitor disease trends, evaluate the impact of interventions, inform policy, and guide programmatic decision making. In order for this data to be informative and useful, however, it must be timely, accurate, and complete. In the context of tuberculosis (TB), which continues to cause millions of cases of disease and deaths each year, surveillance data is known to have gaps including undercounting cases and incomplete reporting by health facilities. To accelerate TB control and elimination, reliable data is needed to improve quality of TB care. Furthermore, challenges with unique patient identification may limit quality of care, monitoring and evaluation, and data integrity for TB in low and middle-income (LMIC) settings. This dissertation addresses these questions in the context of Uganda, a LMIC setting with a high burden of TB and HIV. Chapters 1 and 2 describe research conducted in collaboration with Uganda’s National Tuberculosis and Leprosy Programme (NTLP) to understand the quality of TB surveillance data, while Chapter 3 presents an evaluation of the delivery of a biometric technology to facilitate individual patient identification. This dissertation used both quantitative and qualitative methods not only to measure data quality, but also to characterize underlying factors that influence data collection, quality, and use. In Chapter 1, I quantitatively assessed agreement between surveillance data from the Uganda NTLP and high-fidelity data from a research study in 2017 and 2019. Agreement was measured using agreement ratios, their 95% limits of agreement, and concordance correlation coefficients, all calculated from linear mixed models. I found good overall agreement with some variation in expected facility-level agreement for smear positive diagnoses, bacteriologically confirmed treatment initiations, and TB patients who were people living with HIV. Surveillance data undercounted positive GeneXpert results, but overcounted clinically diagnosed treatment initiations and number of people taking antiretroviral therapy, relative to research data. Average agreement was similar across study years for all six measurements, but facility-level agreement varied from year to year and was not explained by facility characteristics. This chapter concluded that future research should elucidate and address reasons for variability in the quality of routine TB data in order to advance its use as a quality improvement tool. In Chapter 2, I conducted a qualitative study to answer the questions raised by Chapter 1. Specifically, I sought to understand sources of variation in the quality of routine TB data in Uganda by characterizing the experiences, processes, and perspectives of TB data collectors and users through semi-structured interviews. Together with two Ugandan researchers, I interviewed two groups of participants: programmatic stakeholders and health facility-level stakeholders including TB clinical staff and data officers. Using the Performance of Routine Information Systems Management framework, we identified four themes that explained how technical, organizational, and behavioral factors interact to influence data system processes and outcomes. Importantly, the mutually reinforcing relationship between data quality and data use relies on adequate availability of technical components, data knowledge and skill, ongoing training and engagement, and teamwork. As Uganda transitions to an electronic, case-based surveillance system for TB, addressing ongoing technical, organizational, and behavioral challenges will be key to ensuring that the new system produces data that is feasible for routine use. Finally, in Chapter 3, I conducted a mixed-methods study to understand the feasibility, acceptability, and adoption of digital fingerprinting for patient identification in a study of household TB contact investigation in Kampala, Uganda. First, I tested associations between demographic, clinical, and temporal characteristics and failure to capture a digital fingerprint, and evaluated clustering of outcomes by household and community health worker (CHW). Digital fingerprints were captured for 74% of eligible participants, with extensive clustering of failures by household arising from software and hardware failures and increasing over time. In addition, to understand determinants of intended and actual use of fingerprinting technology, I conducted in-depth interviews with CHWs and applied the Technology Acceptance Model 2. The interviews revealed that digital fingerprinting was feasible and acceptable for individual identification, but failures lowered CHWs’ perceptions of the quality of the technology, threatened their social image as competent health workers, and made the technology difficult to use. This chapter emphasizes the need for routine process evaluation of digital technologies in resource-constrained settings to assess implementation effectiveness and guide improvement of delivery. This dissertation advances the understanding of both traditional surveillance and novel approaches to collecting TB data in one high-burden setting. However, it also provides an analytic approach that can be replicated in other settings to guide quality assessments and targeted improvement of TB data systems. Finally, it highlights the importance of ongoing assessment and end-user engagement at all stages of implementation to ensure that data systems produce high-quality data that can be used to improve public health outcomes.
White, Elizabeth Bennett, "Mixed Methods Evaluation of Data Systems for Tuberculosis in Uganda" (2022). Yale Graduate School of Arts and Sciences Dissertations. 676.