Nathan Geffen, TAC

A simple, reliable and affordable algorithm to identify if a patient can safely be initiated onto isoniazid preventative therapy (IPT) would be very useful. It is important to reduce the risk of initiating patients with active TB on IPT so that they are not effectively given monotherapy with the prospect of developing resistance to isoniazid. Furthermore, it is important that patients with active TB are initiated onto TB treatment. Research published by Cain and colleagues in the NEJM examines this problem. Their large prospective study of HIV-positive antiretroviral-naive patients in Cambodia, Thailand and Vietnam examined 80 million unique combinations of one to five predictors of TB in order to identify the most effective TB screening algorithm. They also identified the most effective TB diagnostic algorithm for patients who were received positive screening results. [1]

Cohort

A total of 1,836 HIV-positive antiretroviral-naive people from eight out patient facilities were screened for eligibility for the study. Of these, 1,768 were eventually enrolled. Three samples of sputum and one each of urine, stool, blood, and lymph node aspirate for patients with lymphadenopathy were obtained for TB culture testing. Patients who were culture-positive for TB were assumed to have TB. Twenty patients could not have their TB status classified and were excluded from the analysis. TB was diagnosed in 267 (15%) of patients.

The sensitivity of screening and diagnostic algorithms was determined based on the results of the culture tests. Sputum smears from and chest radiographs of patients were also taken.

At baseline, the median CD4 count was 242 cells/mm3 (IQR: 82 to 396). Median age was 31 years (IQR: 7 to 72) and just over half were male.

Screening

The challenge for a screening algorithm is to find the best combination of signs, symptoms and history of exposure, easily determined at any level health facility, that maximises the number of negative patients correctly diagnosed as negative but also minimises the number of patients who are suspected of being positive and consequently require microscopy and culture. For example, an algorithm that simply assumes everyone does not have TB might result in too many patients with active TB being inadvertently prescribed isoniazid monotherapy resulting in unnecessarily high isoniazid resistance, as well as many patients with active TB going untreated. On the other hand an algorithm that assumes everyone needs diagnostic testing might be impractical and too expensive in most settings.

The study found that requiring more than one symptom present for a positive diagnosis resulted in too low sensitivity. Also, considering only one or two symptoms was too insensitive. On the other hand, considering four or more symptoms conferred no substantial additional benefit over three symptoms but added complexity to the algorithm. The best screening algorithms therefore were ones, which required the presence of one or more of three symptoms. These were:

• Cough or fever of any duration or drenching night sweats for three weeks out of the previous four weeks: This algorithm has a sensitivity of 93%, specificity of 36% and negative predictive value of 97%.
• Cough, drenching night sweats, or loss of appetite of any duration in previous four weeks: This has the same sensitivity and negative predictive value as above, and a specificity of 35%.
• Cough in previous 24 hours or fever of any duration or drenching night sweats for three weeks in previous 4 weeks: This has a sensitivity, specificity and negative predictive value of 90%, 43% and 96% respectively.
• Cough in previous 24 hours or drenching night sweats or loss of appetite of any duration in previous 4 weeks: This has a sensitivity, specificity and negative predictive value of 89%, 44% and 96% respectively.

The authors show the benefit of three symptoms over two by giving this example: As compared with two symptoms (cough or fever of any duration in the previous four weeks) a combination of three symptoms (the addition of night sweats for three weeks or more) reduced by five the number of patients with false negative screens but increased by 18 the number of patients who needed diagnostic evaluation.

Using the recommended three criteria screening algorithm, 1,199 of the 1,748 participants would have needed diagnostics, but there would have been 18 patients with false negatives not referred for diagnostics. By comparison, the WHO screening approach of only referring patients for diagnostics if they have had a cough for more than two or three weeks would have resulted in only 355 patients being referred for further diagnostics, but 179 false negatives.

Diagnostics

For patients who receive a positive screen, the researchers recommend the following algorithm:

• Start with AFB microscopy of two sputum smears. Patients with at least one positive sputum result should be initiated on TB therapy in most cases.
• For patients with two smear negative results, conduct a chest x-ray. If the x-ray is abnormal, clinical judgment should be used to determine if the patient should be treated, followed by confirmatory mycobacterial culture.
• For patients with a normal x-ray but a CD4 count below 350 cells/mm3, clinical judgment should be used to determine if the patient should be treated, followed by confirmatory mycobacterial culture.
• The authors state that for patients with a CD4 count ≥ 350 cells/mm3 it is unclear what to do. Remember, these are patients with at least one symptom associated with TB.

Applying the screening/diagnostics algorithm to this cohort, the results were as follows:

• Of the 1,199 patients with a positive screen for TB using the screening algorithm described by the authors, 113 had at least one sputum positive result from two sputum smear microscopy examinations. Of these, 98 (87%) had culture-confirmed TB.
• Of the 1,086 with two sputum-negative results, 250 had an abnormal chest x-ray, of whom 83 (33%) had TB.
• Of the 836 with normal chest x-rays, 558 had a CD4 count < 350 cells/mm3. Of these, 55 (10%) had TB.
• Of the 278 with a CD4 count ≥ 350 cells/mm3, 13 (5%) had TB.
• A total of 808 culture tests would need to be done (comprised of the 250 patients with abnormal chest x-rays and the 558 patients with CD4 counts below 350 cells/mm3)

Using this algorithm, the number of false negative results in the screening and diagnostic steps combined was 31, far fewer than the WHO methodology. The authors also compared their combined screening/diagnostic algorithm to one in which all patients underwent chest x-rays and microscopy of two sputum smears, without symptom screening or culture confirmation. This method would have yielded 75 false negatives; ie patients who had tuberculosis but nevertheless had normal x-rays and two negative smears.

The study also found that patients who had false negative results with the screening/diagnostic algorithm tended to have higher CD4+ cell counts than patients who had false negative results with the other two approaches.

The authors conclude by recommending validation of the algorithm in Africa and other parts of the world. They also argue that policy changes should be considered based on their findings.

comment

This excellent paper is the most detailed analysis of how to diagnose TB in HIV-positive antiretroviral-naive patients, a critically important challenge. The sample size is large and the methodology sound. Furthermore, although the cohort is Asian, the high rate of TB in this cohort means it has some applicability to southern African settings. However, the authors are right that validation of the algorithm is needed. This study could and should be carried out in southern Africa.

Even for well-resourced settings, this study is important. It takes a few weeks for the gold-standard TB diagnostic, a liquid medium culture test, to return a result. In the meanwhile clinicians need to decide whether or not to commence treatment or isoniazid preventative therapy.

However, the study also shows how costly, slow and difficult TB diagnosis is. Even the authors’ algorithm, albeit an improvement on the WHO’s current one, has a large number of false negatives (cf HIV testing). It is also more expensive to carry out than the WHO one. Many health facilities do not have the resources to carry out the diagnostic part of the algorithm, which involves far more sputum smears and chest x-rays than the WHO one. Indeed, the shortcomings of all current TB screening and diagnostic algorithms makes it difficult to implement isoniazid preventative therapy in many health facilities.

The findings of this study remind us again of the need for faster, cheaper and more sensitive and specific TB diagnostics.

Ref: Cain K et al. An algorithm for tuberculosis screening and diagnosis in people with HIV. NEJM 362;8 February 25, 2010.

http://content.nejm.org/cgi/content/full/362/8/707