UCLA

Tuberculosis is a global public health issue with more than 2 billion people infected worldwide. It is also a serious public health concern within the United States with 9,557 cases of active disease diagnosed in 2015 alone. In the U.S., specific sub-groups, such as foreign-born persons, persons with diabetes or persons living with HIV or other immunocompromising conditions are known to be at higher risk of TB disease. Among foreign-born residents in the U.S., persons born in high-morbidity countries are known to be at even higher risk of developing the disease. Yet, TB disease incidence rates by country of birth are not reported at the local, state or national level despite these large, known differences in risk by country of birth. This is part due to the complications of using country-of-specific population estimates and technical challenges of using standard distributions to model a communicable disease. This thesis aims to call attention to this notable gap and, in part, to fill it.

Data on 5,447 diagnosed TB cases from the Los Angeles County Department of Public Health TB Control Program were combined with stratified population estimates available from the Public Use Microdata Survey to calculate incidence rate of TB disease for the years 2005 through 2011, stratifying by country of birth and other demographic factors. Bayesian models were used to account for the uncertainty in the number of diagnoses and the population estimates. Extending these models into spatial analysis required the use of a hierarchical Bayesian model. Prediction models were constructed using bootstrap backward elimination and stochastic variable selection.

We estimated that the unadjusted incidence rate among persons born in the Philippines was 44.3 per 100,000 person-years and among persons born in Vietnam 38.7 per 100,000 person-years in comparison to 2.3 per 100,000 for U.S.-born persons. In spatial analysis, TB disease incidence was found to be spatially heterogeneous within Los Angeles County and remained so within high-risk countries of birth and when accounting for age, sex and years in residence. In prediction modeling, we found the addition of PUMA-level ecological variables did not improve the prediction of TB disease incidence beyond models using age, sex, country of birth and years in residence. With these three analytical approaches–non-spatial, spatial and prediction–we confirmed that TB disease incidence rates varied markedly by country of birth and showed that issues arising from the technical challenges of dependent outcomes, sparse data and uncertainty in population estimates can be ameliorated.