Using Routinely-Collected Data To Estimate Prevalence Rates of Multi-drug Resistant Tuberculosis

Using Routinely-Collected Data To Estimate Prevalence Rates of Multi-drug Resistant Tuberculosis


Accurate information on disease prevalence is needed to target limited health resources in order to maximize overall population health. Disease surveillance studies can provide accurate estimates of disease prevalence, but they are infrequently conducted because they entail significant financial and time costs. Between surveillance studies rigorous statistical methods applied to routinely collected data can produce an economical unbiased estimate of disease prevalence. This is an important issue for multi-drug resistant tuberculosis (MDR-TB), which is costly to treat, requires a long treatment duration and is likely fatal if left untreated.

We use high-quality 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 simulated maximum likelihood estimation (SMLE) method and a method of simulated moments (MSM) to generate estimates. Our identification strategy relies on exogenous institutional variation in drug resistance testing rates induced by time period and national health policy changes.

We estimate that at least one-third of MDR-TB cases went undiagnosed between 2004-2011 and that the official World Health Organization (WHO) estimate of 2.5% is therefore too low. Consistent with prior estimates, women have slightly higher rates of MDR-TB than men. We also 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 increased investment in early detection of MDR-TB, such as the ongoing implementation of Xpert MTB/RIF PCR technology, and more effective treatment, such as new antibiotics. This study demonstrates the value of routinely collected data for disease surveillance in developing countries where resources are extremely limited and the potential welfare cost to poor resource allocation is high. Using routinely collected data to monitor disease prevalence is an effective strategy to drive evidence-based health policy in low resource settings.


ReSEP members involved: Rulof P. Burger

Collaborators: Zoë M. McLaren