Abstract
We present a statistical summary of results from the Model Performance Evaluation Program (MPEP) for Mycobacterium tuberculosis Drug Susceptibility Testing, 1994 to 2008, implemented by the U.S. Centers for Disease Control and Prevention (CDC). During that period, a total of 57,733 test results for culture isolates were reported by 216 participating laboratories for the first-line antituberculosis drugs used in the United States-isoniazid (INH), rifampin (RMP), ethambutol (EMB), and pyrazinamide (PZA). Using Clinical Laboratory and Standards Institute (CLSI)-recommended concentrations for one or more of three methods, agar proportion (AP), BACTEC460 (Bactec), and MGIT-960 (MGIT), yielded overall agreement of 97.0% for first-line drugs. For susceptible strains, agreement was 98.4%; for resistant strains, agreement was 91.0%, with significantly lower accuracy (chi-square test, P < 0.0001). For resistant strains, overall agreement by methods was 91.3% for AP, 93.0% for Bactec, and 82.6% for MGIT and by drugs was 92.2% for INH, 91.5% for RMP, 79.0% for EMB, and 97.5% for PZA. For some strains, performance by method varied significantly. Use of duplicate strains in the same shipment and repeat strains over time revealed consistent performance even for strains with higher levels of interlaboratory discordance. No overall differences in performance between laboratories were observed based on volume of testing or type of facility (e.g., health department, hospital, independent). By all methods, decreased performance was observed for strains with low-level INH resistance, RMP resistance, and EMB-resistant strains. These results demonstrate a high level of performance in detection of drug-resistant M. tuberculosis in U.S. laboratories.