We investigated the use of DNA amplification by the polymerase chain reaction (PCR) for detection of Mycobacterium tuberculosis from clinical specimens. Two-thirds of each sample was processed for smear and culture by standard methods, and one-third was submitted for DNA extraction, amplification of a 317-bp segment within the insertion element IS6110, and detection by agarose gel electrophoresis, hybridization, or both. DNA was prepared from over 5,000 samples, with 623 samples being culture positive for acid-fast bacilli. Of 218 specimens that were identified as M. tuberculosis, 181 (85%) were positive by PCR. In the M. tuberculosis culture-positive group, PCR was positive for 136 of 145 (94%) and 45 of 73 (62%) of the fluorochrome smear-positive and -negative specimens, respectively. Of 948 specimens that were either culture positive for mycobacteria other than M. tuberculosis or culture negative, 937 specimens were negative by PCR and 11 (1%) specimens initially appeared to be false positive for M. tuberculosis. The reasons for discrepant results varied; some errors were traced to the presence of an inhibitor in the specimen (7.3% in unselected specimens), nucleic acid contamination, low numbers of organisms in the specimen, antituberculosis therapy, and possible low-level nonspecific hybridization. In comparison with culture, the sensitivity, specificity, and positive predictive value were 83.5, 99.0, and 94.2%, respectively, for PCR. When PCR was corrected for DNA contamination, the presence of inhibitor, and culture-negative disease, the values became 86.1, 99.7, and 98.4%, respectively. If the results for multiple specimens submitted from the same patient are considered, no patient who had three or more sputum specimens tested would have been misdiagnosed.
