INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The recent resurgence of tuberculosis complicated by the AIDS epidemic has refocused attention on the need for more rapid and accurate diagnostic tests, particularly those using molecular techniques. Such tests would facilitate early isolation of potentially infectious patients and prompt institution of antituberculosis chemotherapy and contact investigation. A valid negative rapid test would free up expensive and scarce respiratory isolation facilities in hospitals burdened with large populations of high-risk patients.

Currently available radiometric culture methods for tuberculosis diagnosis (1, 2) with nucleic acid probes for identification still require from 10 d to 3 wk, and conventional techniques require 3 to 6 wk. Sputum smears are rapid but insensitive, and they are not specific for tuberculosis. Therefore, decisions regarding respiratory isolation and institution of therapy are still based largely on clinical grounds. Polymerase chain reaction (PCR) and other nucleic acid amplification methods may dramatically increase the sensitivity and specificity of laboratory tests aimed at detecting Mycobacterium tuberculosis. New automated commercial systems may be rapid enough to enable clinicians to make important decisions within hours of admission (3).

Previous studies of PCR have focused mainly on cultured specimens obtained selectively from microbiology laboratories, and they have reported specimen-specific yields (4). Some studies have been carried out with large proportions of acid-fast bacilli (AFB) smear-positive specimens than commonly found in clinical populations (5, 6, 8, 9). In addition, many studies have used culture positivity as the reference and left out clinical tuberculosis. (3, 6). To our knowledge, no study has prospectively enrolled patients, collected at least two sputum specimens within 24 h, and studied the feasibility of PCR as a rapid tool to diagnose or rule out tuberculosis at the time of first patient contact.

Many amplification targets for M. tuberculosis have been reported. One of the more common PCR targets is the repetitive sequence IS6110, specific for M. tuberculosis complex. An alternative approach utilizes the FDA-approved Roche Amplicor M. tuberculosis kit, which targets the 16S ribosomal RNA gene for amplification with subsequent detection of the products using an M. tuberculosis specific probe.

We performed this prospective study to compare utility of both techniques with that of AFB smears, culture, and clinical diagnosis of pulmonary tuberculosis.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Population

From December 1994 through June 1995 all patients referred from the emergency room, wards, or clinics of Cook County Hospital who required admission to isolation rooms to rule out pulmonary tuberculosis were referred to the pulmonary service for evaluation. History, physical examination, and chest radiography results were reviewed. Chest radiographs were scored by two pulmonologist investigators (R.A.C. and S.M.). The chest radiographs were divided into four zones: above the clavicles, and upper, middle, and lower zones. Chest radiographs demonstrating nodular, alveolar, or interstitial infiltrates predominantly affecting the zones above the clavicles or upper zones, or a miliary pattern, were classified as "typical," whereas those with other patterns were classified as "atypical." Normal radiographs or inactive processes (i.e., calcified granuloma, old rib fracture) were classified as "negative." Patients with typical chest radiographs were automatically isolated and offered enrollment. Symptomatic patients (cough for more than a week, 10-lb. weight loss, night sweats, or fever) with atypical chest radiographs were also isolated and offered enrollment. Symptomatic patients with known infection caused by HIV were isolated and offered enrollment even if their chest radiographs were negative. Patients who were in respiratory failure, required ICU admission, were prisoners, or in whom smear- or culture-positive tuberculosis had been diagnosed within the previous year were excluded. All patients gave informed consent to participate in the study, which was approved by the institutional review boards of Cook County Hospital and the University of Illinois.

Face-to-face interviews were conducted by trained research nurses using a standardized questionnaire eliciting information on symptoms of tuberculosis, risk factors, social history, history of previous tuberculosis, tuberculosis treatment, and tuberculin skin testing. Tuberculin skin testing was performed intradermally with 5 TU according to the Mantoux technique. All patients were offered HIV testing.

Tuberculosis was defined by a positive culture for M. tuberculosis on any specimen, including specimens obtained outside the study protocol. Patient medical and pharmacy records were reviewed by the two pulmonologist investigators (R.A.C. and S.M.) to determine if they met criteria for clinical tuberculosis. They were so classified if they had resolution of symptoms and infiltrates on chest radiographs after 3 mo of antituberculosis chemotherapy and no alternative etiology was identified.

Sputum Collection

Trained research assistants collected sputum specimens every 4 h for a total of three specimens within the first 12 h of arrival to the hospital. These research assistants were available on call 24 h daily during the study period, exclusive of weekends and holidays. Eligible patients presenting during weekends and holidays were not enrolled. One specimen was then collected every morning for the next 3 d, for a total of six specimens per patient. Normal saline induction was performed for patients unable to expectorate spontaneously. Specimens were decontaminated by the N-acetyl-L-cysteine-NaOH method and concentrated by centrifugation (13). Smears for microscopic examination were prepared from the concentrated specimens and stained using auramine-rhodamine and examined using fluorescence microscopy. Smears demonstrating acid-fast bacilli were reviewed after staining by the Kinyoun staining method (14, 15). A portion of sputum sediment was then inoculated onto two slants of Lowenstein-Jensen and Middlebrook 7H10 (Difco Laboratories, Detroit, MI) media; smear-positive specimens were also inoculated in a BACTEC vial containing 7H12 medium. Cultures were incubated for a total of 8 wk. Nucleic acid probes were used to identify M. tuberculosis. Nontuberculous mycobacteria were identified by conventional methods (13). The remaining portion of the pellet was frozen at zero degrees centigrade and then transported to the molecular biology laboratory at Evanston Hospital for PCR analysis, which was performed 1 to 4 wk after patient enrollment. PCR results were not made available to clinicians.

PCR Assays

Reagent and sample preparation, PCR amplification, and product detection were performed in separate rooms using dedicated equipment, positive-displacement pipettors, and unidirectional work flow. In addition, dUTP was incorporated into the amplicons so that they could be inactivated by UNG nuclease to prevent carryover contamination. Positive and negative controls for both the sample preparation and PCR processes were utilized in each experiment, and they included both known positive and negative sputa as well as a dilution of tuberculosis DNA as a positive control and water blanks as negative controls. All samples also underwent amplification of a fragment of the human p53 gene as a control for amplification inhibitors. The in-house assay was based on the work of Eisenach and colleagues (9, 16) and has been previously reported. Two hundred fifty microliters of decontaminated sputum was used for the in-house PCR assay, 100 µl for the Amplicor assay and the remaining specimen frozen at 70° C for repeat testing if needed, and quality control. The in-house assay consisted of mycobacterial lysis, performed according to the method of Eisenach (16), amplification, and amplicon detection using hybridization in solution to a radioactively labeled oligonucleotide probe specific for the IS6110 product, as previously described (17). PCR amplification was also performed using the Roche Amplicor PCR amplification/ detection kits, following the directions provided by the manufacturer.

Follow Up

The city of Chicago Department of Health maintains a data base of all patients suspected to have, or diagnosed with, tuberculosis. This data base was reviewed 1 and 2 yr after study completion. Information was retrieved for smear, culture, and final diagnosis for all patients evaluated for this study.

Statistical Methods

Univariate statistics were generated for all demographic, clinical, and laboratory variables. Frequencies and cross-tabulations were performed on categorical data grouped according to smear, culture, and PCR results using SPSS statistical software package (18). The sensitivity and specificity of the two methods for PCR were compared using the weighted least-squares method for the analysis of categorical data (19).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Six hundred five patient referrals were made to the pulmonary service during the study period. Four hundred nine were isolated to rule out tuberculosis. Of these, 26 patients had a history of a positive smear or culture for tuberculosis within the last year and were ineligible for the study. Sixty-one patients were ineligible because they were referred from the Cook County Jail. Of these, six patients proved to have tuberculosis. Of the remaining 322 patients, 162 were not felt to be at very high risk based on a pulmonologist's review of their history and chest radiograph. Their work-up proceeded according to usual hospital routine. Only one of these patients proved to have tuberculosis, culture-positive only, on a specimen obtained 4 mo after his evaluation.

This left a possible 160 patients who were eligible for the study. Sixty-eight of those eligible were not enrolled because they were admitted on holidays and weekends, four patients refused, and one patient was transferred to ICU. Nine of the eligible, but nonenrolled, patients proved to have newly diagnosed tuberculosis or a rate of 16%. Characteristics of these groups are shown in Table 1. Although the rate of tuberculosis was less in the nonenrolled patients, this difference was not statistically significant.

Two of the 87 patients did not have adequate specimen volume for PCR processing. Specimens from the remaining 85 patients were analyzed by PCR. Five-hundred ten specimens (6.0 per patient) were processed for routine mycobacteriology, and 316 (average, 3.7 per patient; range, 2 to 6) were processed for PCR.

Our population was mainly black men, and it is reflective of our tuberculosis population as a whole (see Table 1). About one-third of the patients were infected with HIV, one-third were negative, and the remainder had unknown status and refused HIV testing. Twenty-seven patients had culture-positive tuberculosis (32%). Two of these patients were classified as having tuberculosis based on positive cultures from nonstudy specimens obtained during their admission. Twelve of these had at least one smear positive for tuberculosis and were thus considered to have smear-positive tuberculosis. Cultures from four patients, all smear-negative, yielded mycobacteria other than tuberculosis (MOTT) from sputum specimens, including M. avium, M. fortuitum, and M. gordonae.

PCR and Conventional Mycobacteriology

The results of PCR as compared with conventional mycobacteriology are presented in Table 2. The PCR results from the specimens obtained within the first 24 h identified nearly all patients positive with PCR results on any specimen. Therefore the results of the first two PCR specimens collected within the first 24 h of admission became the focus of this analysis.

For the 12 patients with smear-positive tuberculosis, PCR was positive by either the in-house or the commercial technique on all specimens. No patient in our study had smear-positive MOTT infection.

Overall, the in-house technique was more sensitive (85 versus 74%) but less specific (88 versus 93%) than the Roche Amplicor for a final diagnosis of tuberculosis, although these differences were not statistically significant. As expected, on smear-negative patients the sensitivity fell (73 and 53%) for the in-house and Roche assays, respectively. The specificity for this group of patients was unchanged (see Table 3).

The specificity of the in-house technique could be improved by 5% if both of the two sputum specimens were required to be PCR-positive. However, this requirement resulted in a substantial fall in the overall sensitivity for both techniques. The requirement of two positive PCR tests resulted in an even more dramatic drop in sensitivity for smear-negative patients (see Table 3).

Incremental Yield of Two Specimens for Tuberculosis

The first specimen was positive in 70% of all patients with tuberculosis by the in-house technique and 55% using the Roche kit (see Table 4). The diagnostic yield increased with the additional specimen to 85 and 74%, respectively, The yield for smear-negative patients was 47% by the in-house and 20% by the Roche assay; this increased to 73 and 53% with the addition of a second specimen. We then excluded patients with paucibacillary disease, defined retrospectively as those with 20 colonies or less on final culture, from the analysis. The in-house method detected 100%, and Roche detected 95% of patients with more than paucibacillary disease.

Analysis of Discrepant Samples

A total of seven patients had apparent false positive PCR results with the in-house assay; three of these seven were also positive by the Roche assay. Repeat PCR analysis of these samples yielded repeatedly positive results. Three of these patients actually had a history of pulmonary tuberculosis, two occurring 2 yr or more prior to their current admission. The third had a history of tuberculosis diagnosed at another hospital 4 mo prior to the study; had this history been clear he would not have been included. The positive PCR results of these patients (all by the in-house assay, two by the Roche assay) indicates the potential for residual nucleic acid in nonviable organisms to generate a positive PCR result. The false positive results of the remaining four patients are unexplained, but they are felt to most likely be cross-contamination of negative samples with a strongly positive sample during the sample preparation process. Another possibility is contamination of the samples with nonviable organisms prior to sample preparation for PCR. Amplicon carryover has been a very rare event in our laboratory, and periodic testing of the UNG enzyme has not revealed failure.

Two of the five patients who had cultures that yielded MOTT (M. gordonae and M. fortuitum) gave false positive results on PCR. One of these was positive by both PCR assays, the other by the in-house method only. In studies by this laboratory and others, both PCR targets are quite specific for M. tuberculosis complex organisms only, so there is no clear explanation of these positive results. These isolates were unavailable for follow-up PCR testing.

A final two patients were positive by PCR with negative smears, cultures, PPDs, histories, and chest radiographs. Again, no clinical explanation for these results was found.

Five patients, all of whom had smear-negative tuberculosis, were false negative by in-house and Roche techniques. One additional patient was false negative by Roche only. The five patients false negative using both techniques had only one or two out of six cultures positive for fewer than 20 colonies of M. tuberculosis. One patient, false negative by Roche, had only one out of six cultures 1+ positive for M. tuberculosis. None of these patients had cavitary disease on chest radiograph.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Our study attempted to provide some answers to the questions of diagnostic yield of PCR prospectively in a blinded study comparing PCR results with cultures and clinical diagnosis of tuberculosis. We performed PCR on as many as six specimens per patient and found that the first two specimens obtained within the first 24 h of hospital arrival identified nearly all patients with positive PCR results on any specimen. Testing beyond two specimens therefore did not appear to enhance the results meaningfully, and it is probably not clinically or economically practical.

The sensitivity of two PCR specimens obtained within the first 24 h of arrival to the hospital was 100% for patients with smear-positive tuberculosis. In addition, the sensitivity of PCR, although less than 100% for patients with smear-negative tuberculosis, increased to 100% when patients with paucibacillary disease are not considered false negative. Because patients with paucibacillary disease (defined as  20 colonies on all cultures) were less likely to have been infectious, the negative predictive value of two PCR specimens was 100% for the identification of patients most likely to be infectious. This information may be useful to clinicians attempting to make difficult decisions about the need for isolation.

The clinical interpretation of positive PCR results is less clear in patients with negative sputum smears. Although PCR identified 53 to 73% of smear-negative patients with tuberculosis, the clinical utility of these findings is reduced by the presence of substantial number of false positives, resulting in positive predictive values of only 61 to 67%. Our results were derived from a highly selected patient population with a high prevalence of tuberculosis. If PCR were applied to a population with lower prevalence, the positive predictive values would likely decline further; however, the negative predictive value would increase.

Because of financial constraints precluding sample collection on weekends and holidays, we were only able to enroll approximately half of the eligible patients. There were no statistically significant differences in patient characteristics between the enrolled and nonenrolled groups (see Table 1). The lower rate of tuberculosis in the nonenrolled group may be explained by the less vigorous work-up these patients received, possibly reducing the number of cases of tuberculosis.

Financial constraints also necessitated the storage of samples for batch analysis, rather than immediate analysis as they came into the laboratory. As commercial kits and automated amplification/detection equipment becomes available, it will become more practical and cost-effective for laboratories to perform nucleic acid amplification for tuberculosis detection on a more rapid basis. Institution-based cost effectiveness studies would need to be performed to evaluate the expense of running PCR assays 7 days a week compared with possible savings of isolation days and hospital days. Additionally, automation will make such testing available in a wider number of laboratories, and standardization will reduce the interlaboratory variation currently found (20).

We evaluated two PCR methods, an in-house technique, and a commercially available kit. There was a tendency for the commercial kit to be less sensitive but more specific than our in-house assay; however, these differences did not prove to be statistically significant. Similar differences have been reported elsewhere (21). The differing sensitivities between in-house and Roche tuberculosis assays may be attributable to a number of factors, including sample preparation methods, the choice of PCR primers and the PCR assays themselves, the differing detection methods, and the use of a large sample volume (22) in the in-house assay.

A strength of our study is that it assesses the potential role of PCR as it might be used in a high volume clinical setting on specimens collected within 24 h of patient arrival to the hospital. There have been few studies evaluating the efficacy and yield of PCR testing in such a context (23). Many studies relied on laboratory specimens only and did not prospectively evaluate the diagnostic utility of direct PCR on sputum on patients newly admitted to the hospital with suspected pulmonary tuberculosis (4). Other studies (6, 11, 24) have compared PCR using in-house and commercial systems with culture results on individual sputum samples and not the diagnostic yield among patients. They have looked at clinical data only to resolve cases that were false positive. Unlike our study, the specimens have come from mixed populations of newly diagnosed and treated patients. Protocols for specimen collection were not controlled, and most clinical information was obtained retrospectively.

Prospective studies of similar design to ours compared PCR results with a clinical diagnosis of tuberculosis with careful follow-up to determine the patient's final tuberculosis status. Beige and colleagues (28) prospectively studied 103 non-HIV-infected patients admitted to rule out tuberculosis. They compared PCR using a DNA-based in-house technique with a patient-based diagnosis of tuberculosis classified according to the American Thoracic Society (29) criteria. They found PCR 98% sensitive and 70% specific. They had quite a few false positives in patients who were PPD positive and no evidence of active tuberculosis. Bradley and colleagues (30) studied respiratory tract specimens from 421 patients with a broad range of risk factors for pulmonary tuberculosis. They found the RNA-based Gen-Probe M. tuberculosis Direct Test to be 93.6% sensitive, 70% sensitive in smear-negative sputum samples. In contrast to the work of Beige and colleagues, they found a specificity of 96.8% overall. Compared with our population, only 5% of their study population had tuberculosis and nearly half of these patients were receiving treatment. Chin and colleagues (31) evaluated the utility of PCR in respiratory tract specimens from 227 patients and compared the results with a rigorous clinical diagnosis of tuberculosis. Like most investigators they found PCR 100% sensitive in smear-positive patients but only 50% sensitive in smear-negative patients. They had a very low false positive rate of less than 1%. The variation in amplification assays and current lack of standardization among in-house molecular tests may have contributed to the variation in sensitivity and specificity results reported.

The potential for false positives results in amplification assays is a great concern. We had a rate of 7 to 12% among our patients. These occurred despite the use of proper laboratory technique and chemical means to inactivate amplicons, which should reduce the danger of amplicon carryover and reamplification as a cause of false positives. It seems most likely that cross-contamination of samples during sample preparation may have occurred, as samples yielding false positive results continued to be positive on repeat testing. The multistep nature of sample preparation used in the in-house assay may have contributed to the higher number of false positives compared with the Roche kit and is likely to be the basis of the enhanced sensitivity of the in-house assay. Another possible source of error is contamination of samples with nonviable organisms. PCR assays directed at DNA targets cannot distinguish viable from nonviable organisms and therefore cannot differentiate patients with clinically active disease from those who have been treated for tuberculosis or have resolved their tubercular disease without treatment. The observance of PCR-positive sputa from patients long after successful treatment of tuberculosis is well known (20). Furthermore, contamination of samples with nonviable organisms within the laboratory could present another cause for false-positive results that needs to be recognized; amplifiable tuberculosis DNA has been recovered from sterilized bronchoscopes (17).

In summary, our findings indicate that PCR testing of two sputum samples obtained within 24 h of arrival to the hospital may prove diagnostic in all patients who have smear-positive tuberculosis and 53 to 73% of smear-negative pulmonary infections. PCR testing on two specimens, rather than just one, allowed the detection of 75 to 85% of patients, missing only those patients with paucibacillary disease. Further study with large numbers of patients would be required before this approach could be adopted widely in clinical practice. The relatively low positive predictive value in smear-negative patients makes interpretation of a positive test less certain. These findings are consistent with a recent comprehensive review (32), and the current recommendations of the Centers for Disease Control and Prevention (33).