What You Don't Know Can Kill You: Estimating Tuberculosis Drug Resistance Rates in South Africa Using Routinely Collected Data

Every year, 8.7 million people worldwide contract tuberculosis (TB) and over 16% of them will die from it. Multi-drug resistant TB is a growing concern because it takes 4 times as long and is over 200 times more expensive to cure than drug-susceptible TB. Drug resistance testing is costly and few developing countries have routine surveillance systems to measure drug resistance rates and monitor trends. Though South Africa contributes disproportionately to worldwide increases in drug resistance it does not perform routine surveillance and its most recent surveillance survey data come from 2002 (a current surveillance study is ongoing). This study investigates the viability of using routinely collected TB test result data to estimate underlying drug resistance rates in the population of patients with suspected TB in South Africa. It employs a tuberculosis test result database collected by the National Health Laboratory Service (NHLS) which includes detailed TB test results and drug sensitivity testing data for over 37 million tuberculosis tests and 14 million patients between 2001-2011, representing approximately 85% of all TB tests performed annually in South Africa. A combination of factors related to patient history, physician decision-making and health service resource availability determine which of the patients with suspected TB are tested for drug resistance. "Naive" estimates of drug resistance that do not correct for sample selection into drug resistance testing are unlikely to produce unbiased estimates of drug resistance. This study evaluates the viability of three propensity score methods (reweighting, matching and a control function) to address the sample selection issue in the South African context, and compares estimates to "naive" estimates.