INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The search for a rapid, accurate, yet inexpensive test for the diagnosis of active tuberculosis (TB), begun almost a century ago (1), has become the equivalent of the search for the holy grail. At present the most widely used rapid test is the direct microscopic examination of a smear of sputum for acid-fast bacilli (AFB smear). However the preparation and reading of the smear are time consuming and detect only 40-80% (2) of pulmonary TB cases, and only the more advanced cases (4). Diagnosis of patients at an earlier stage, while still smear negative, would be advantageous because they are less contagious (5, 6) and have lower morbidity and mortality (7). Mycobacterial cultures are highly sensitive, but take at least 2 wk, or longer if solid media are used, and culture facilities are not available in many countries. Chest X-ray is commonly used for the diagnosis of active TB, but has poor sensitivity, specificity, and reproducibility (7). Tuberculin skin testing has very poor specificity (8), and will be false negative in 20-30% of patients newly diagnosed with active TB (9, 10). Recently developed nucleic acid amplification techniques have specificity of more than 95%, and are more than 95% sensitive in smear-positive specimens (11- 13). However these tests are expensive, require sophisticated technology, and have sensitivity of only 50-71% in patients with smear negative active TB (12)the clinical setting in which a rapid diagnostic test other than the AFB smear is most needed. Serologic tests using an enzyme-linked immunosorbent assay (ELISA) technique to detect antibodies to Mycobacterium tuberculosis are relatively simple and inexpensive but to date have had poor sensitivity and specificity (1), particularly in smear negative cases (15, 16), although the use of a combination of several antigens appears promising (17).

We have conducted a prospective study to evaluate the sensitivity, specificity, and clinical utility of nucleic acid amplification tests and a new commercially available multiple antigen serologic test for the diagnosis of minimal pulmonary TB.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient Population, Clinical Information

Between July 1995 and June 1998 we prospectively studied patients at the Montreal Chest Institute suspected of active pulmonary TB. Patients were included if they were referred for sputum induction either because their spontaneous sputum was AFB smear negative or was not obtainable. Patients were excluded if they had received any antituberculous medications in the previous 14 d or had a recent exacerbation of asthma.

After providing informed consent, participants underwent sputum induction on one or more occasions. This was performed for up to 15 min with hypertonic (3%) saline administered by an ultrasonic nebuliser (De Vilbiss Ultra-Neb 90), which produced 5-6 ml/min (maximum 90 ml). Participants also had blood drawn for serology, and were tuberculin skin tested with the Mantoux technique using 5-TU of PPD-T (Tubersol; Connaught Laboratories, Toronto, ON, Canada). Additional investigations, such as bronchoscopy, were performed at the discretion of the treating physician. Hospital records were reviewed to abstract demographic data and clinical information.

All chest X-rays were reviewed by one of the investigators (D.M.) who estimated the overall likelihood of active TB and percentage of total lung parenchyma affected, and classified the abnormalities using a categorization scheme developed for Immigration Canada. Where pairs of chest X-rays taken 2-3 mo apart were available, they were interpreted as better, worse, or unchanged. All such readings were made without knowledge of the dates of the radiographs, final diagnosis, or other clinical information.

Microbiologic Tests

All respiratory specimens (one or more per patient) were digested and decontaminated with a solution of 2% sodium hydroxide (NaOH) and N-acetyl-L-cysteine (NALC), then centrifuged for 15 min at 3000 rpm. The supernatant was removed and the remaining sediment was mixed in a 1:10 dilution with sterile water. The processed sample was stained with Auramine O fluorochrome and examined with fluorescent microscopy. Ziehl-Nielson acid-fast staining was used to confirm the presence of AFB. Mycobacterial cultures were performed by inoculating 0.1 ml of the processed sample onto two tubes of Loewenstein Jensen media and into the liquid media of the BACTEC 460 system according to the manufacturer's instructions. Positive culture results were sent to the provincial reference laboratory (Laboratoire de Sante Publique de Quebec) for confirmation and drug susceptibility testing.

Aliquots of 100 µl for polymerase chain reaction (PCR) were prepared from all specimens submitted by study participants, if a sufficient concentrated decontaminated specimen remained after smear and culture were performed. Patients were considered smear, culture, or PCR positive if any smear, culture, or PCR, respectively, were positive.

Nucleic Acid Amplification and Detection Techniques

The Roche Amplicor Mycobacterium test (Amplicor MTB) was performed according to the manufacturer's instructions (Roche Diagnostics, Canada). Briefly, 100 µl of concentrated decontaminated specimen (prepared as above) was added to 500 µl of wash buffer, then centrifuged at 12,000 × g for 10 min. The supernatant was aspirated and 100 µl of lysis reagent was added to the sediment. The suspension was incubated for 45 min at 60° C and then neutralized by the addition of 100 µl of neutralization reagent. Fifty microliters of each prepared specimen was transferred to tubes containing 50 µl of master mix. Specimens and controls (three negative and one positive) were amplified in a Perkin-Elmer GeneAmp PCR System 9600 thermal cycler. In addition a sequence of plasmid DNA with primer binding regions identical to those of the M. tuberculosis sequence was introduced into each reaction mixture and coamplified with the target DNA to provide an internal control of the PCR reaction. Carryover contamination was prevented by incorporation of dUTP in place of dTTP in the amplification reaction, and utilization of uracil-N-glycosylase (Amperase) to cleave any amplicon carried over from previous reactions.

Detection of mycobacteria of the M. tuberculosis complex was accomplished by hybridization of the biotin-labeled amplicon to probe-coated microwell plates, and addition of an avidine horseradish peroxidase conjugate and a tetramethylbenzidine substrate. The optical density at 450 nm was measured using an automated microwell plate reader. Specimens with an absorbance greater than 0.35, were interpreted as positive, irregardless of the internal control result. Specimens with an absorbance less than 0.35, and an internal control absorbance greater than 0.35, were classified as negative. Specimens with an absorbance less than 0.35, and an internal control absorbance less than 0.35, were classified as uninterpretable.

Serologic Testing (ELISA)

Ten milliliters of blood was obtained and centrifuged at 3500 rpm for 10 min, following which the supernatant was separated into 1-ml aliquots and frozen at 70° C. Assays (one per patient) were conducted using the Detect-TB test (Biochem ImmunoSystems, Montreal, PQ, Canada). In brief, this test was performed using microtiter plates, which contained one blank well, three negative control wells, two positive control wells, and the test wells. All were coated with M. tuberculosis antigens. Two hundred microliters of sample diluent was added into the blank well and all test wells, whereas 200 µl of positive control sample was added into the two positive control wells.

Ten microliters of each patient's serum sample was then added to all test wells. If specific M. tuberculosis antibodies were present in the patient's serum, these formed stable complexes with the M. tuberculosis antigens in the test well. Then 200 µl of peroxidase conjugation solution (goat anti-human immunoglobulin G [IgG] labeled with horseradish) was added into each well and incubated for 30 min at room temperature. Detection of the bound conjugate was performed by adding 200 µl of substrate solution, incubating for 30 min at 20° C, then adding 100 µl of stop solution. Within 30 min of addition of the stop solution, the optical density at 450 nm was measured with a plate reader.

If the average optical density of the positive controls was less than 1.4, the results were considered invalid and the run was repeated. The average optical density of the three negative control wells was defined as NCx, and the cut-off for a positive test as NCx + 0.15. A patient's test sample was considered positive if the sample optic density was equal to or greater than [NCx + 0.15].

All technicians who carried out the PCR and ELISA were blind to the final diagnosis.

Diagnoses and Definitions

Active TB. Confirmed if any culture of respiratory secretions (sputum or bronchoalveolar lavage) was positive for M. tuberculosis. Clinical (culture negative) if there was a compatible abnormal chest X-ray, which improved after treatment for 2-3 mon with three or four antituberculous drugs, as judged by the independent review of the chest X-ray.

Inactive TB. All cultures were negative for M. tuberculosis, compatible abnormal chest X-ray, stable during the follow-up period, as judged by the independent review, reaction to tuberculin skin test of 5 mm or greater, and absence of an alternative pulmonary diagnosis.

Nontuberculous mycobacterium (NTM). Potentially pathogenic NTM isolated from the respiratory secretions (sputum or bronchoalveolar lavage) and abnormal chest X-ray.

Stable radiographic scarring. All mycobacterial cultures of respiratory secretions were negative, abnormal chest X-ray that remained stable during the follow-up period, and tuberculin reaction less than 5 mm.

Other pulmonary diagnoses. Lung infections, cancer, bronchiectasis, other bronchial diseases, etc, were based on compatible chest X-rays, diagnoses by the treating physicians, and all other clinical information.

Statistical Analysis

All analysis was conducted using SAS (SAS Institute Inc., Cary, NC). Associations of demographic, clinical, and laboratory variables with clinical diagnoses and results of diagnostic tests were tested with t tests or ANOVA for continuous and chi square tests for categorical variables (18). An ROC curve was constructed to identify the criterion for a positive ELISA that would have best sensitivity and specificity (19). Sensitivity, specificity, positive and negative predictive values, and accuracy were calculated as suggested by Sackett and coworkers (19). Multivariate logistic regression was used to obtain adjusted parameter estimates for the clinical characteristics significantly associated with active disease and false-positive serologic results. In this regression continuous variables such as age were dichotomized at their median values. The final models included only factors remaining significant. The parameter estimates from logistic regression were used to estimate the probabilities of active disease, and corresponding positive and negative predictive values for serology and PCR, alone and together, in different clinical situations (20).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Over the 3 yr of the study, 510 patients were referred for sputum induction for diagnosis of suspected active tuberculosis, and initially agreed to participate. Three patients were excluded because no sputum was produced with induction, and a fourth because investigations were incomplete. Four patients had initially abnormal chest X-ray compatible with inactive TB, positive tuberculin skin test, and all sputum cultures AFB smear and culture negative, but the clinical diagnosis could not be finalized because of inadequate follow-up. These patients were also excluded. Two patients with extrapulmonary TB (genitourinary, and lymphadenitis) without active pulmonary disease were also excluded. In these patients their chest X-ray, sputum smear, culture, and PCR results accurately reflected their pulmonary status, but their serologic results may have reflected the extrapulmonary disease, making these results difficult to classify.

For the remaining 500 participants, results were available for AFB smear and culture, review of medical records, and chest X-ray. PCR results were missing in 13 (2%) because the quantity of sputum collected was sufficient only to perform AFB smear and culture. Serum for serologic testing was not available for 79 patients (16%), primarily because they refused to have a separate blood drawing for this, and 29% did not undergo tuberculin skin testing, mostly because their treating physician felt this was not necessary. Bronchoscopy was performed in 58 patients, including 10 of the 44 with culture-confirmed active TB. Of these 10 patients, mycobacterial cultures of bronchoscopy specimens were positive in only six.

As shown in Table 1, active tuberculosis was diagnosed in 60 patients (12%). Of these 44 were culture confirmed, among whom 10 had at least one positive smear. Only one patient was consistently smear positive on multiple specimens examined. In total 40 patients in whom all cultures were negative were treated with full anti-TB therapy (i.e., three or four drug regimens) for at least 2 mo; 16 had a radiographic response and were classified as active cases, whereas 24 did not and were classified as inactive TB. Two patients with active TB were human immunodeficiency virus (HIV) positive. Patients with active TB were younger, more often symptomatic, and more likely to have a past history of TB contact than the 278 patients with inactive TB (Table 2). The 14 patients with NTM were older, and more often female, but were otherwise similar clinically to those with active TB.

When mycobacterial culture was used as a standard, as shown in Table 3, PCR had a sensitivity of 57%, similar to several of the chest X-ray findings. Inclusion of clinically diagnosed cases reduced the sensitivity of PCR, but did not alter the sensitivity of the other tests. The specificity of PCR was excellentthere being no false positives among all those tested, so overall accuracy was 94%. Specificities of the tuberculin test and chest X-ray were much lower. Examination of the ROC curve for the serologic test (not shown) demonstrated that defining the test as positive if the specimen to cut-off ratio exceeded 1.6 yielded the best combination of sensitivity (38%), specificity (87%), and accuracy (80%), even though the manufacturer's recommendation is to use a ratio of 1.0 (sensitivity 45%, specificity 77%, accuracy 74%) for culture-confirmed cases.

Using multivariate logistic regression, younger age, symptoms of cough, fever, or weight loss, and an infiltrate/cavity affecting at least 5% of the lung parenchyma were independently associated with active TB (Table 4). After adjustment for these factors, sex, country of origin, history of bacillus Calmette-Guérin (BCG), TB contact, and tuberculin results were not significantly associated with disease, but serologic results were, even among the PCR-negative subjects only. PCR results were not entered in multivariate analyses because there were no false-positive PCR results, so the estimated odds of disease when PCR was positive was infinite. As shown in Table 5, false-positive serologic tests were associated with a history of past TB treatment at all cut-points, and with suspicious chest X-ray at higher cut-points. Interestingly, patients with false-positive tests at higher titers were less likely to have BCG vaccination or positive tuberculin tests.

As shown in Figure 1, the predictive value of a negative ELISA was high when the prevalence of active TB was low, regardless of the cut-point used. However, as shown in Figure 2, the predictive value of a positive test was much lower and more strongly affected by the cut-point selected. In Figure 2, the specificity of PCR was assumed to be 98%the average of several series (11) rather than 100% as was found here.

View larger version (11K): [in this window] [in a new window]   Figure 1.   Effect, at different disease prevalence, of changing the cut-point to define a positive serologic test on the negative predictive values of serologic and PCR tests. Solid line with diamonds: PCR. Dotted line with small squares: ELISA with cut-point ratio of 2.0. Solid line with large circles: ELISA with cut-point ratio of 1.6. Dashed line with triangles: ELISA with cut-point ratio of 1.0. Dashed line with inverted triangles: ELISA with cut-point ratio of 0.6.

View larger version (13K): [in this window] [in a new window]   Figure 2.   Effect, at different disease prevalence, of changing the cut-point to define a positive serologic test on the positive predictive values of serologic and PCR tests. Solid line with diamonds: PCR. Dotted line with small squares: ELISA with cut-point ratio of 2.0. Solid line with large circles: ELISA with cut-point ratio of 1.6. Dashed line with triangles: ELISA with cut-point ratio of 1.0. Dashed line with inverted triangles: ELISA with cut-point ratio of 0.6.

Parameter estimates from the final logistic regression models were used to calculate the probability of active TB in different clinical situations (Table 6). In this analysis probabilities of active disease were estimated when the ELISA was used alone (to simulate a setting in which PCR was not available, or not routinely performed) or in combination with PCR. For example, if the ELISA was used alone, and was positive in a young symptomatic patient with negative AFB smear but X-ray abnormality suspicious for TB, the probability of active TB would be 66% (95% CI 61-72%). If ELISA and PCR were both performed, and PCR results were negative, the likelihood of active TB would be 42% if ELISA was positive (36- 48%). Given the very high specificity of PCR (100% in this study) the predictive value of a positive test exceeded 99% in all clinical situations, so performing ELISA as well would not add anything. However, in many situations, the predictive value of a negative PCR results was equivalent to that of a negative ELISA. If PCR was already performed, then performance of ELISA offered additional information in situations in which the pretest probability was high.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study demonstrates that accurate and rapid diagnosis of minimal pulmonary tuberculosis remains difficult, and there is still no single ideal test. Nucleic acid amplification techniques had excellent specificity, but only moderate sensitivity, particularly for diagnosis of clinical and culture-positive cases. Chest X-ray and serology, which are or could be available in many settings, had similar or somewhat lower sensitivity but much lower specificity. Use of a combination of tests was superior to any single test. A positive PCR would indicate disease with a very high degree of certainty, but a negative result would be less helpful in situations with greater likelihood of disease (higher pretest probability). In these situations results of ELISA may provide additional clinical information to guide management in patients with possible minimal active tuberculosis.

The strengths of the study included the large number of subjects, allowing use of multivariate logistic regression to estimate with reasonable precision the likelihood of active disease in a wide variety of clinical situations. As well, this was a prospective series that included all patients suspected of possible active TB, but without spontaneous sputum or with negative AFB smearsthe group in which a new diagnostic test is most needed (21). Compared with studies utilizing healthy volunteers as control subjects (15, 16, 22, 23), these results provide a more realistic assessment of the utility of these diagnostic tests in clinical practice.

Weaknesses of the study include missing serologic tests results and inclusion of clinically diagnosed active cases. The missing serologic results should not have biased the results, because the clinical or radiographic characteristics of patients without serologic results were similar to patients for whom results were available. A relatively high proportion (27%) of the active cases were clinically diagnosed. However clinically diagnosed cases are common, accounting for 18% all cases reported in the United States in 1998 (24). In a recent diagnostic study of 72 patients with active disease, 60% were smear and culture positive, 30% were smear negative but culture positive, and 10% were smear and culture negative25% of all smear-negative patients were also culture negative (25). Patients were carefully characterized40 were placed on full anti-TB therapy, of whom only 16 (40%) were judged by an independent reviewer to have improved radiographically; the remainder were classified to have inactive TB. Only five patients dropped out: one failed to complete initial investigations, and four with negative mycobacterial cultures had insufficient follow-up to ensure stability of radiographic abnormalities.

Accuracy of radiographic diagnosis in this study was similar to previous studies (7). The finding of high specificity, but suboptimal sensitivity of PCR for diagnosis of paucibacillary active TB, was similar to that reported elsewhere (12, 26, 27). Similar high specificity, and moderate sensitivity for diagnosis of minimal active disease have also been reported for amplification tests of mycobacterial RNA (MTD; Gen-Probe, San Diego, CA) (14, 28), although one laboratory reported sensitivity of 77% in such patients using an in-house modification of the MTD test (29). In a recent report, a new commercially available second generation RNA amplification test was positive in 15 (52%) of 29 patients with smear-negative active disease, or 15 (68%) of 22 smear-negative culture-positive patients (25).

The serologic test evaluated is composed of a combination of antigens, and had sensitivity of 76% among patients with smear-positive active TB (22). Sensitivity was lower in this study because patients with less extensive disease may have lower antibody response (1), or react to different MTB antigens than patients with smear-positive disease (17). Multiantigen serologic tests may need to include different antigens for the diagnosis of extensive, smear-positive disease than for paucibacillary, smear-negative disease. False-positive serologic reactions were associated with a history of prior TB treatment, and more extensive radiographic abnormalities of inactive tuberculosis, as described elsewhere (16). Patients with active, inactive, and previously treated TB represent different stages of the same infection, so it is not surprising that an immunological assay did not distinguish these patients well. This emphasizes the importance of including such patients in any evaluation of a diagnostic test for active TB.

For the moment, it appears that mycobacterial cultures using both BACTEC and solid media are the most sensitive of currently available tests for the diagnosis of minimal active pulmonary TB. However, even cultures may be false negative, and a positive culture will be detected after an average of 2 wk, and longer in paucibacillary disease (30, 31). In this study, because of its excellent specificity the predictive value of a positive amplification test was very high. However, because of lower sensitivity the predictive value of a negative PCR was lower, especially when the presence of certain clinical and radiographic characteristics indicated a higher pretest probability, as has been noted elsewhere (25). When the pretest probability was low (atypical chest X-ray findings, asymptomatic, older age) negative predictive value was high. In a setting in which PCR was performed routinely on all smear-negative clinical specimens, serologic testing would be useful in high-risk situations, when the PCR was negative (see Table 6). However, given their high cost and complexity, it is unlikely that many centers will use amplification tests routinely in smear-negative clinical specimens, and they are very unlikely to be used in developing countries where the burden of disease, and need, is greatest (21).

In settings in which amplification tests are not in routine use, serologic testing may play a role in refining estimates of disease. For example, when the pretest probability was low, the predictive value of a negative serologic result was very high (similar to the results of amplification tests); in this situation a clinician could withhold treatment and await results of cultures. On the other hand, in patients with a higher pretest probability (younger age, and/or symptoms, typical chest X-ray findings) then a positive ELISA (defined at the cut-point of 1.6) increased the likelihood of active TB enough to consider initiation of anti-TB therapy.

We conclude that no currently available diagnostic test is reliable enough for the rapid and accurate diagnosis of minimal pulmonary TB. Amplification tests have excellent specificity but suboptimal sensitivity, and are costly and complex. The serologic test used in this study had suboptimal sensitivity and specificity. However, since ELISA tests are relatively simple and inexpensive, they could be introduced more widely, and used, together with clinical and radiographic characteristics, to refine estimates of likelihood of disease in patients suspected of having minimal active TB.