Previous treatment in predicting drug-resistant tuberculosis in an area bordering East London, UK

To determine the utility of 'risk assessment' in selecting Mycobacterium tuberculosis isolates for rifampin resistance or rpoB genotyping compared to 'non-selectively' genotyping all isolates. Secondly, we examined the association between past treatment and drug resistance.

From January 2003 to December 2006, demographic, clinical, and laboratory data were prospectively collected on patients with laboratory-confirmed tuberculosis (TB). On the basis of past treatment for active TB infection or known exposure to drug-resistant TB, selected samples were sent to a mycobacterial reference laboratory for rpoB genotyping. A multivariable logistic regression model was developed to examine the association between past treatment and drug resistance, adjusted for other factors. Sensitivity, specificity, and negative and positive predictive values of past treatment as a predictor for drug resistance were determined.

There were 392 patient episodes of culture-proven TB. Thirty-three drug-resistant isolates were cultured from 30 patients: 29 (87.9%) were isoniazid-resistant, three (9.1%) were multidrug-resistant (MDR), and one (3.0%) was rifampin mono-resistant. One patient with isoniazid resistance developed recurrent disease, and two isolates, initially isoniazid-resistant, mutated and became MDR TB. Based on risk assessment, rpoB genotyping was performed on 19 samples, and two (10.5%) had mutations that predicted multiple drug resistance. Although for MDR TB, a past history of treatment predicted two out of three patients with acquired resistance, adjusted analysis did not demonstrate a significant association between previous treatment of active TB and drug resistance (odds ratio 1.5, 95% confidence interval (CI) 0.4-5.6). The positive predictive value of past treatment as a predictor for drug resistance was 12.0% (95% CI 2.6-31.2%).

Although numbers of MDR TB were too small to draw meaningful conclusions, past treatment may be useful in selecting samples for rpoB genotyping. Overall, previous treatment had a low positive predictive value for drug resistance in an area bordering East London.