INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Although no longer the scourge it once was, pulmonary tuberculosis remains a significant global health problem, especially in developing countries. There are 1.6 billion people carrying the bacteria; 10 million new cases of tuberculosis occur worldwide every year, with about 3 million deaths annually (1).

The identification of the microorganism in secretions or tissues from the patient is the mainstay of the diagnosis of tuberculosis. However, the process is quite difficult and has some limitations. The sensitivity of sputum acid-fast stain is low (especially in noncavitary tuberculosis), and the time needed for culture results is no shorter than 6 to 8 wk. In addition, the clinical features are nonspecific. Chest roentgenographic findings have been estimated as atypical in more than 30% of patients with tuberculosis in developed countries (2). Therefore, a rapid diagnostic tool with both high sensitivity and specificity is needed to improve the conventional diagnostic methods.

Several new techniques have been developed to improve the diagnosis of tuberculosis, including newer radiometric methods, DNA probes, chromatography of mycolic acid, polymerase chain reaction, and serologic tests. These diagnostic approaches have had a dramatic effect on the ability to diagnose disease accurately and expeditiously (3).

A serologic diagnostic method was first introduced by Arloing (4) as a technique of hemagglutination in 1898. However, it did not fulfill the clinical requirement with acceptable sensitivity and specificity in the diagnostic field of tuberculosis until 1972, when Engvall and Perlmann (5) described a simple, highly sensitive, reproducible, and inexpensive technique of enzyme-linked immunosorbent assay (ELISA).

Many antigenic materials have been subsequently employed in the ELISA method in an attempt to improve both the sensitivity and specificity. These have included complex antigen from Mycobacterium tuberculosis (6), bacille Calmette-Guérin (BCG) socinate (7), purified protein derivative (PPD) antigen (8, 9), Antigen 5,6, and Mycobacterium glycolipids. The overall sensitivity of these methods ranges from 45 to 95%, and their specificity ranges from 90 to 100%, respectively (10). Antigen 60 is one of the best-known antigens used in the TB ELISA method. According to the latest literature reviewed, both sensitivity and specificity are approximately 80% (16). Newer antigens such as 38kda (with identity with antigen 5) (17) and Kp90 (Kreatech Diagnostics, Madrid, Spain) have recently been introduced as commercially available kits, claiming even higher specificity (18, 19). However, the comparison between different antigens in the literature is difficult since the patient populations in the studies differed significantly. Furthermore, the performance of the diagnostic kits from a clinical point of view was not satisfactory (20). The positive predictive value was as low as 24% according to calculations based on a sensitivity of 84%, a specificity of 88%, and disease probability of 0.05 (21). Clinical trials of antituberculosis drugs according to ELISA data would be inappropriate. Different antigens, combined in ways to improve sensitivity and specificity, may be the answer until the so-called ideal antigen is discovered.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Subjects

Serum samples were obtained from 594 adult subjects (> 15 yr of age), including 312 of a tuberculosis group and 282 of a nontuberculosis group. All the subjects were prospectively and consecutively enlisted from among the patients presenting at Taiwan Provincial Chronic Disease Control Bureau during an 8-mo period, from September 1994 to April 1995.

Tuberculosis group. This group consisted of 312 patients with active tuberculosis. The inclusion criteria required one of the following: (1) positive sputum acid-fast stain, (2) positive sputum culture, (3) histologic proof from tissue biopsy, and (4) suggestive chest roentgenographic findings with marked improvement after antituberculosis treatment.

Fibrocalcified tuberculosis group. This group was made up of 101 patients with inactive tuberculosis. The patients had been treated either with a complete course of antituberculosis drugs for their active tuberculosis or had presented with chest roentgenograms showing a fibrocalcified pulmonary tuberculous lesion. All chest roentgenograms of the patients were stable for at least 6 mo, and sputa, obtained on six different days, tested negative for TB smear and culture.

Healthy control group. This group consisted of 118 healthy adults documented by a general health examination. Their chest roentgenograms were normal, and no respiratory symptoms were indicated.

Nontuberculous pulmonary disease group. This group consisted of 63 patients admitted to our hospital with chest problems other than tuberculosis. The diseases included bacterial pneumonia, chronic obstructive pulmonary disease, bronchial asthma, bronchiectasis, pneumoconiosis, and lung cancer.

ELISA for Antibody Level Determination

Antigen 60 IgG measurement. The diagnostic kits were obtained from Anda Biologicals (Strasbourg, France). The antigen used in this kit was prepared from the cytoplasm of Mycobacterium bovis-BCG. In the standard procedure recommended by the manufacturer, 100 µl of diluted sera (1:100) was distributed in microtiter wells and incubated for 60 min at 37° C. The unbound components were eliminated by washing with a buffer solution. The wells were subsequently incubated with 100 µl of peroxidase-labeled antihuman gammaglobulin (IgG) conjugate at 37° C for 30 min. After the incubation period and a second washing cycle, the peroxidase substrate tetramethyl-benzidine (TMB) containing hydrogen peroxide was introduced into the wells, and the colorimetric reaction was prolonged for 15 min in the dark at 37° C. Stop agent (0.5N sulfuric acid) was added at the end of incubation. This yielded a yellow color, and the results were interpreted by measuring the optical density (OD) of the stained precipitates with a spectrophotometer (Emax Model 4845-02; Sunnyvale, CA) at 450 nm.

Six standards (0, 1, 2, 4, 8, and 16 serounits) were provided for a semilog reference curve. Because the samples had been diluted 1:100, they corresponded to 0, 100, 200, 400, 800, and 1,600 serounits, respectively. For those titers higher than 1,600 units, the data were analyzed by extrapolation.

Antigen 38kda IgG measurement. The serologic test of 38kda IgG (Omega Diagnostics, Alloa, Scotland, UK) consisted of microtiter wells coated with 38kda antigen from M. tuberculosis. All the procedures required for the assay were the same as described above in the section on Antigen 60, except that the sera were diluted by a factor of 1:50 rather than 1:100.

The cutoff level of the OD readings in this kit was determined by dividing the OD value of the low positive control by 1.5 (as recommended by the manufacturer). The Antibody Index derived from the ratio of the unknown sera to the cutoff level permits comparison of samples by different measurements.

Antigen Kp90 IgA measurement. Diagnostic kits were obtained from Kreatech Diagnostics. Serum IgA antibody to a mycobacterial Kp90 ImCRAC (immuno-cross-reactive antigenic compound) was measured. The procedures were also similar to those described in the section on Antigen 60, except (1) the sera were diluted by a factor of 1:500; (2) the second incubation time was 1 h rather than 30 min, and (3) the third incubation was performed for 30 min at room temperature.

The cutoff OD value was derived from the mean of the three cutoff control OD readings (as recommended by the manufacturer). The ratio of the unknown sera to the cutoff level provided the basis of comparison of data in different measurements.

All the sera were tested in duplicate wells to ensure greater precision of the results. If duplicate absorbency values varied so much that they represented different interpretation, the data were eliminated. The quality control procedures recommended by the manufacturers were strictly followed.

Data Analysis

The diagonstic value of the ELISA was evaluated in terms of sensitivity and specificity as well as positive and negative predictive values. The sensitivity of a diagnostic test was interpreted as the proportion of patients correctly identified by the test as abnormal; the specificity is the proportion of normal subjects correctly identified as normal by the test (22). Furthermore, the positive predictive value was interpreted as the proportion of patients with active tuberculosis among those with a positive test; and the negative predictive value is the proportion of subjects without tuberculosis among those testing negative.

For the calculation of a mean titer in Antigen 60 IgG, titers less than 100 units were calculated as 100 units. The ELISA units in Antigen 60, the Antibody Index in 38kda, and the Antibody ratio in Kp90 were all calibrated to log units (e.g., 100 units = 2), which seemed to normalize the distribution. The mean and standard deviation were calculated from the log units. The cutoff value was chosen according to a receiver operating characteristic (ROC) analysis (23) rather than that recommended by the manufacturer (in 38kda and Kp90, see description above).

For comparison of mean titers and antibody ratios, a t-test of the log units was employed for group comparison.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The demographic data of the patients are shown in Table 1. There were 14 patients who had positive acid-fast smears, 72 patients had positive sputum cultures for M. tuberculosis, 134 patients had both positive acid-fast stain and sputum TB culture, 13 patients had histologic evidence suggesting M. tuberculosis infection, and 79 patients had clinical diagnosis only. There were 101 patients with fibrocalcified lesions on their chest roentgenograms, 118 healthy control subjects, and 63 subjects with nontuberculous pulmonary disease.

Antigen 60 IgG Measurements

Antigen 60 IgG measurements are shown in Figure 1 as a function of the different groups. The mean and standard deviation of each group are listed in Table 2. The p values between the tuberculosis group and the fibrocalcified tuberculosis group, and the healthy control and nontuberculous disease group were p < 0.0001, p < 0.00001, and p < 0.00001, respectively. Using ROC analysis, a reasonable cutoff value was determined as 340 ELISA units, which defined the sensitivity (80.77%) and the specificity of the test (88.40%). Under this cutoff value, the false positive rate still remained as high as 31.68% in the fibrocalcified tuberculosis group and 11.60% in the joint group of healthy control and nontuberculous pulmonary disease. The positive and negative predictive values based on different estimated disease probability and the sensitivity/specificity determined above are shown in Table 3.

View larger version (20K): [in this window] [in a new window]   Figure 1.   The Antigen 60 (A60) IgG value in the different groups. Each patient is represented as a single diamond. Titers more than 5,000 are shown as 5,000 in the graph. Abbreviations: TB = tuberculosis group; Fibro = fibrocalcified tuberculosis group; Normal = healthy control group; Other = nontuberculous pulmonary disease group.

Antigen 38kda IgG Measurements

The IgG ELISA results against 38kda are presented in Figure 2. The mean and standard deviation of each group are shown in Table 2. The p values between the tuberculosis group, the fibrocalcified group, and the healthy control and nontuberculous group are p < 0.0001, p < 0.00001, and p < 0.00001, respectively. Using ROC analysis, a reasonable cutoff value was determined as 0.38 Antibody Index, which defined the sensitivity (64.12%) and the specificity of the test (80.74%). Under this cutoff value, the false positive rate still remained as high as 52.48% in the fibrocalcified tuberculosis group and 19.26% in the joint group of healthy control and nontuberculous pulmonary disease. The positive and negative predictive values based on different estimated disease probability and the sensitivity and specificity determined above are shown in Table 3.

View larger version (22K): [in this window] [in a new window]   Figure 2.   The 38kda IgG value in the different groups. Each patient is represented as a single diamond. The values larger than 14 are shown as 14 in the graph. For definition of abbreviations, see Figure 1.

Antigen Kp90 IgA Measurements

The IgA ELISA results against Kp90 are presented in Figure 3. The mean and standard deviation of each group are shown in Table 2. The p values between the tuberculosis group, the fibrocalcified group, and the healthy control and nontuberculous pulmonary disease group are p < 0.001, p < 0.00001, p < 0.001, respectively. A stepwise analysis was also used to establish a cutoff value for a positive antibody test. The trends of change in the sensitivity and specificity are illustrated in Figure 4. Using ROC analysis, a reasonable cutoff value was determined as 1.4 antibody ratio, which defined the sensitivity at 62.58% and the specificity of the test at 66.30%. Under this cutoff value, the false positive rate still remained as high as 48.51% in the fibrocalcified tuberculosis group and 33.70% in the joint group of healthy control and nontuberculous pulmonary disease. The positive and negative predictive values based on different estimated disease probability and the sensitivity and specificity determined above are shown in Table 3.

View larger version (25K): [in this window] [in a new window]   Figure 3.   The Kp90 IgA value in different groups. Each patient is represented as a single diamond. The values larger than 14 are shown as 14 in the graph. For definition of abbreviations, see Figure 1.

View larger version (19K): [in this window] [in a new window]   Figure 4.   Comparison of ROC curves of Antigent 60 IgG, 38kda IgG, and Kp90 IgA.

Combination of Different Antigens

Combinations of different antigens were attempted in order to improve the diagnostic yield. However, it could only improve the sensitivity and specificity by about 10%, with the sacrifice of the opposite parameter by no less than 20%, no matter which antigens chosen.

Direct Comparison of Different Antigens

We performed ROC analysis to provide direct comparisons among the specific antigens applied. The ROC curve is displayed in Figure 4, and it can be seen that Ag60 IgG was the most discriminative one all the way except in the strictest confidence threshold zone. The 38kda possessed the highest sensitivity (27.65%) while the specificity was set to 100%.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The data indicate that Antigen 60 was the more indicative among the tests studied from the point of view of sensitivity and specificity. Theoretically, combining different antigen ELISA tests may be another way to improve diagnostic yields. The sensitivity can be improved by adding patients for whom either, or any one, of the tests was positive. On the other hand, specificity can be improved by selecting the patients for whom both or all of the tests were positive. Such a proposal is rejected since the results show that the combination makes a change that can also be accomplished by just adjusting the cutoff value of a single antigen ELISA test.

Cutoff Value Determination

Cutoff value determination is the critical issue to define the role of a particular ELISA test in the correct diagnosis of active tuberculosis. The cutoff value may be different in different areas with different prevalence of tuberculous infection or history of BCG vaccination. In this study, an ROC analysis was used to determine the cutoff value. After the analysis, the patients with fibrocalcified tuberculosis were also checked by the cutoff value to determine its own false positive rate. It is believed that the inclusion of inactive cases of tuberculosis is unavoidable because, from a clinical point of view, it is much more difficult to differentiate active from inactive tuberculosis than it is to differentiate pulmonary tuberculosis from other chest diseases. There is more dependency on ELISA methods in this issue.

Because the subjects studied here were enlisted in a randomized manner during a specific period of time, it seems they could reasonably represent the patient population in this hospital. Thus, the ratio between the TB group and the non-TB control group in this study was valid for the determination of the cutoff value. The method for choosing a cutoff value recommended by the manufacturers of 38kda and Kp90 was abandoned because of unsatisfactory sensitivity and specificity deduced (sensitivity and specificity: of 38kda-antigen, 39.17%/ 89.79%; of Kp90-antigen, 80%/41.84%). The most commonly used method to compute the cutoff point as the mean + 2 SD measured in a control population was also not acceptable for the same reason.

Comparisons between Different Antigens Measured

Three different antigens, Antigen 60, 38kda, and Kp90, were tested in the same study subjects. Antigen 60 seemed to be superior to the others if a so-called "balanced" antigen is sought for the serodiagnosis (i.e., the cutoff value was justified by both the sensitivity and the specificity).

For the clinician, specificity is a more meaningful parameter. Therefore, the cutoff level can be increased to obtain a diagnostic tool with better specificity. From the trends shown in Figure 4, if there is an increase in specificity to 100% by increasing the cutoff values, the resulting test has a sensitivity of 21.79% (Antigen 60 cutoff, 2199 ELISA Unit), 27.65% (38kda cutoff Antibody Index, 4.91), and 15.63% (Kp90 cutoff Antibody ratio, 9.29), respectively. Thus, 38kda is superior to Antigen 60 by a small percentage. However, the false positive rate of the fibrocalcified tuberculosis group was still present at 2.97% (Antigen 60), 2.97% (38kda), and 2.97% (Kp90), respectively.

However, the Kp90 is a diagnostic kit based on IgA measurement that differs from the others, and it may have clinical significance that is vague in gross sensitivity/specificity measurements. An ELISA study based on different immunoglobulin of the same antigen is recommended to clarify this issue.

Combination of Different Antigens

The sensitivity/specificity of Antigen 60 can be increased by combining a second ELISA kit, and there is no difference whether 38kda or Kp90 is chosen. However, the opposite parameter (i.e., specificity and sensitivity) is also sacrificed.

It is not recommended to recheck the sera by a different diagnostic antigen. The cost is doubled and the same result can be obtained by simply adjusting the cutoff value. It is proposed that the cutoff value of different subgroups may also be different. Therefore, there is a need to analyze the data in greater detail in order to clarify and select an optimal cutoff titer of best diagnostic yields.

Predictive Values

Sensitivity and specificity underlie the quality of a diagnostic tool. However, the clinician in daily practice is much more concerned about the predictive values. The predictive values of the kits in different prevalence of patient population are shown in Table 3. The disease probability of the symptomatic patient in a medical center in Taiwan has been estimated to be around 0.26 (24). In a patient population with such a probability of disease, the antigens tested here would not provide a satisfactory positive predictive value for definitive diagnosis of tuberculosis, yet they would provide a good negative predictive value for excluding the disease. It is interesting that the difference in the ability of disease exclusion among the antigens is limited (negative predictive value of Antigen 60 versus 38kda and Kp90: 93% versus 86% and 83%). It is certain that the newer antigens do not provide better predictive values than that of Antigen 60. As described above, neither do different antigen combinations.

In conclusion, Antigen 60 is more effective than 38kda and Kp90 in gross sensitivity and specificity based on a study of 594 subjects. However, the difference in sensitivity and specificity does not imply a significant difference in predictive values (especially the negative) because of the relatively low disease probability in Taiwan. That is to say the clinical importance of the difference is limited. A combination of different antigens was implemented to improve the diagnostic yields. The effort, however, was ineffective since the improvement was negligible and can also be easily obtained by adjusting the cutoff value, which does not increase the cost. Further analysis of the detailed information of the study subjects is warranted to clarify the role of different immunoglobulins measured and the cutoff values of different subgroups of a patient population.