Skip navigation

Tuberculosis diagnosis


Bandolier has become increasingly harsh on studies of diagnostic tests. There are several reasons for this, including designs that are almost always biased. The bias is so great that most studies of diagnostic tests can be forgotten ( Bandolier 70 ).

The other reason for forgetting about any studies of diagnostic tests that slip through the net is that they always discuss only the one test. Yet a diagnosis is made using many criteria - including knowledge of the patient, history, physical examination and clinical suspicion. The test result is just one piece of information that integrates to form the whole, where the whole is the diagnosis.

What a relief, therefore, when someone comes along and shows how diagnostic test evaluation should be done [1]. This study evaluated a new genetic test for tuberculosis, integrating the information with clinical suspicion and evaluating the test against the best available gold standard.


Consecutive patients at seven sites in the USA and Europe suspected of having TB were involved. Attending physicians classed them as being of low risk (less than 25%), intermediate risk (26% to 75%) or high risk (more than 75%) based on standard workup including history, clinical examination and risk factors.

Patients gave sputum specimens for six consecutive days for culture and sensitivity. The Enhanced Mycobacterium tuberculosis Direct (E-MTD) test, a nucleic acid amplification test, was done at the same time but not reported.

Clinical diagnosis of TB was established using criteria established by an expert panel in TB diagnosis and treatment. The combination of high clinical suspicion (more than 80%) and at least two positive cultures from separate specimens was considered definite evidence of active TB. In the absence of these conditions, cases were reviewed by the independent expert panel for adjudication.


At the end of the study the results were pooled and the E-MTD test and acid-fast bacilli (AFB) test were analysed according to the physician's assessment of risk, as well as overall (Table). In the low risk group, 12 of 244 patients (5%) had TB. In the intermediate risk group it was 20 of 68 patients (29%) and in the high risk group it was 40 of 46 patients (87%). Sensitivity and specificity of the E-MTD test were high at all levels of clinical suspicion, and higher than the AFB test.

Clinical suspicion TB (%) Positive likelihood ratio Negative likelihood ratio
Low 5 25.2 0.17
Medium 29 72.0 0.25
High 87 10.5 0.15
Overall 21 31.7 0.17
Low 5 17.7 0.6
Medium 29 1.1 1.0
High 87 2.5 0.3
Overall 21 7.2 0.44
Shaded block: indicates a rule-in decision for post-test probability of 50% or greater for positive likelihood ratio. A negative likelihood ratio indicates a rule-out decision where post-test probability less than 1%.


Fortunately the data used for calculating the sensitivity and specificity were available, because likelihood ratios were not given. Before doing this, Bandolier read the paper at least four times. Clearly it was a good paper, but what did it mean?

The likelihood ratios and post test probabilities helped, using the calculations from that wonderful book, "Evidence-based Medicine" [2]. The likelihood ratios and the TB frequency in the low, intermediate and high risk populations to calculate post-test probabilities gave some idea as to how the results of the test might be used.

Because TB is such a nasty disease, Bandolier guessed that any probability of having TB of over 50% might well result in antimicrobial therapy. All high risk patients would have been treated without a test, but a positive E-MTD test in low and intermediate risk groups would also result in treatment. The AFB test, by contrast, was unhelpful.

Bandolier also guessed that only when the post-test probability of having TB was below 1% might the diagnosis be dismissed. A negative E-MTD test did this for the low, but not intermediate or high risk groups. The AFB test was useless for ruling out the diagnosis.

It is entirely likely that the post-test probability for risk of having or not having TB chosen by Bandolier for illustrative purposes were wrong. It is possible, probable, even, that there is no right answer. They serve, though, to illustrate how to use excellent information about a laboratory test, how to integrate it with clinical factors in making a diagnosis and how to integrate it into clinical practice. It is a superb example of how to evaluate a diagnostic test, and worth reading for all our friends in laboratories and diagnostic companies, and those who teach others how to evaluate the literature.


  1. A Catanzaro et al. The role of clinical suspiciaon in evaluating a new diagnostic test for active tuberculosis. Results of a multicentre prospective trial. JAMA 2000 283: 639-45.
  2. D Sackett et al. Evidence-based medicine. Second Edition, Churchill Livingstone, 2000.
previous story in this issue