Prediction Models for Active Pulmonary Tuberculosis

A study in South Korea investigated the prevalence of diabetes mellitus (DM) among patients with active pulmonary tuberculosis (PTB) and found it to be significantly higher than in the general population. Specifically, 24.2% of patients with PTB had DM, compared to 11.6% in the general population. This higher prevalence of DM was consistent across all age groups and BMI categories. Furthermore, patients with both TB and DM (TB-DM) were generally older, with a median age of 65 years compared to 53 years in TB patients without DM (TB-NDM). They also had a higher BMI (22.18 vs. 21.06 kg/m²). Multidrug-resistant TB (MDR-TB) was more frequently observed in TB-DM patients (5.9%) than in TB-NDM patients (3.1%, P = 0.035), and TB-DM patients had a higher rate of AFB smear positivity (30.8% vs. 14.8%, P < 0.001).[3]

The study also examined treatment outcomes and found that DM negatively impacted TB treatment success. The overall treatment success rate was significantly lower in TB-DM patients (74.4%) compared to TB-NDM patients (84.9%, P < 0.001). Even after excluding patients categorized as ‘not evaluated,’ TB-DM patients continued to show lower success rates (85.1% vs. 91.4%, P = 0.008). DM was identified as an independent risk factor for persistent 2-month sputum culture positivity (OR 2.56, 95% CI 1.56–4.21) and for unsuccessful treatment outcomes (OR 1.67, 95% CI 1.03–2.70, P = 0.039). The study concluded that DM is not only more prevalent among PTB patients but also adversely affects their TB treatment outcomes by delaying sputum conversion and reducing the likelihood of successful treatment completion.[3]

Rapid treatment of tuberculosis (TB) remains challenging in settings with limited access to on-site diagnostic tools, such as microbiological testing, radiography, or specialized staff. In such environments, empirical (or clinical) TB diagnosis—diagnosis without microbiological confirmation—becomes critical. However, empirical diagnosis is often inconsistent, lacks standardization, and may not be routinely performed by midlevel clinicians who frequently staff these clinics. Furthermore, it often poorly correlates with microbiological results, leading to inappropriate treatment.[1]

A common clinical scenario in these settings involves patients with presumptive TB who are unlikely to receive same-day radiological or microbiological test results. Delaying treatment in these cases risks losing patients to follow-up, highlighting the need for immediate yet accurate decision-making tools. Unfortunately, most predictive models for active TB rely on resources—such as radiology, laboratory tests, or computerized calculations—that are unavailable in such clinics.[1]

To address this gap, a simple clinical risk score has been developed using easily accessible patient data. This score incorporates six predictors strongly associated with TB:[1]

1. Male sex
2. Age between 25 and 44 years
3. HIV positivity (based on clinical registers or self-reporting)
4. Presence of specific WHO-defined TB symptoms (cough, fever, night sweats, or weight loss exceeding 5 kg)
5. Duration of TB symptoms lasting more than two weeks
6. Self-reported history of diabetes

The score ranges from 1 to 10, is easy to calculate manually, and relies on information readily available during routine clinical visits. Despite its simplicity, the score demonstrated reasonable predictive accuracy, including in external validation studies. Its use adds clinical utility by enabling immediate diagnosis in patients where the benefits of initiating treatment outweigh the risks of a false-positive diagnosis. This straightforward approach offers a practical solution for settings where delays in diagnosis could lead to missed opportunities for treatment, ultimately improving outcomes in high-risk populations.[1]

Prompt identification of presumptive tuberculosis (TB) cases remains a significant challenge, often leading to delays in diagnosis and treatment. Research efforts have aimed to develop models that utilize clinical predictors such as patient history, physical examination findings, and chest radiography (CXR) to estimate the probability of pulmonary TB (PTB). Advanced imaging technologies were intentionally excluded from these studies due to their limited availability in high TB burden, low-resource settings, where such innovations are not practical.[2]

To ensure consistency, studies focused on specific outpatient populations, excluding settings such as inpatient care and certain subgroups like TB contacts, pregnant women, and drug users, to reduce variability. Out of numerous investigations, only six met the stringent criteria for developing and validating models that improve PTB detection. These models incorporated additional clinical factors such as CD4 count, body mass index (BMI), and the duration of antiretroviral therapy (ART), highlighting their relevance in co-infected populations.[2]

One significant finding was the enhanced sensitivity of TB detection when the tuberculin skin test (TST) was added to the WHO-recommended symptom screen. This underscores the potential of augmenting existing screening protocols with additional tools. Furthermore, these studies underscored the importance of creating clinical scores using various predictors to facilitate effective screening in routine practice, particularly in resource-limited settings.[2]

By integrating supplementary clinical data into the diagnostic process, these models showed promise in improving diagnostic accuracy over the standard WHO symptom screen. This approach also highlighted the pressing need for a low-cost, easy-to-use TB risk score tailored to the realities of high-burden, resource-constrained settings. However, due to the absence of superior alternatives, reliance on the WHO symptom screen remains the norm in many such environments. These findings emphasize the critical role of innovative, accessible tools in transforming TB screening and reducing delays in care.[2]

References:

[1] Baik, Y., Rickman, H.M., Hanrahan, C.F., Mmolawa, L., Kitonsa, P.J., Sewelana, T., Nalutaaya, A., Kendall, E.A., Lebina, L., Martinson, N. and Katamba, A., 2020. A clinical score for identifying active tuberculosis while awaiting microbiological results: development and validation of a multivariable prediction model in sub-Saharan Africa. PLoS medicine, 17(11), p.e1003420.

[2] Van Wyk, S.S., Lin, H.H. and Claassens, M.M., 2017. A systematic review of prediction models for prevalent pulmonary tuberculosis in adults. The International Journal of Tuberculosis and Lung Disease, 21(4), pp.405-411.

[3] Lee, E.H., Lee, J.M., Kang, Y.A., Leem, A.Y., Kim, E.Y., Jung, J.Y., Park, M.S., Kim, Y.S., Kim, S.K., Chang, J. and Kim, S.Y., 2017. Prevalence and impact of diabetes mellitus among patients with active pulmonary tuberculosis in South Korea. Lung, 195, pp.209-215.