Abstract:
Tuberculosis (TB) kills more than 1.5 million people annually and now ranks alongside HIV as the leading infectious disease cause of death worldwide. There have been calls for more drug-resistant tuberculosis (DR-TB) prevalence studies yet only 8.3% of the population in high burden countries live in regions with at least two data points to estimate trends. We develop an innovative, low-cost method for estimating disease prevalence that relies solely on routinely-collected data from diagnostic laboratories, provides continuous surveillance and accounts for under-detection. We use TB test result data from South Africa’s National Health Laboratory Service, which includes test records from over 11 million patients, to estimate the prevalence of MDR-TB between 2004-2011. We develop a theoretical model of the clinician’s decision to perform drug susceptibility testing and use a simulated maximum likelihood estimation (SMLE) method and a method of simulated moments (MSM) to estimate model parameters. We estimate that at least one-third of MDR-TB cases went undiagnosed between 2004-2011. The MDR-TB rate in South Africa could be as high as 3.29 - 3.37% and the official World Health Organization estimate of 2.5% based on notification rates is therefore too low. We find that clinician behavior in the province of Mpumalanga is consistent with a sub-optimal approach of MDR testing only after observing the failure of first-line drug therapy. These findings highlight the need for investment in early detection of MDR-TB and more effective treatment. Using routinely collected data to monitor population prevalence rates is an effective low-cost strategy to guide health policy in low resource settings