Cost-effectiveness analysis of a prediction model for community-based screening of active tuberculosis

A study investigated whether a newly developed prediction model, transformed into a practical scoring system, could enhance the cost-effectiveness of active TB case finding in community settings compared to existing WHO symptom-based screening tools. The central question addressed the potential for improved TB detection and resource optimization using a data-driven, stratified approach rather than relying solely on broad symptom checklists.

To answer this, the researchers used a robust methodological design combining prediction model development and external validation with a cost-effectiveness analysis (CEA). The model was built using data from the ZAMSTAR trial in South Africa and validated using an independent dataset from Zambia. This separation between development and validation populations supports the model’s external validity. Stratifying analyses by HIV status was also appropriate, reflecting how TB symptoms can differ significantly between HIV-positive and HIV-negative individuals. Moreover, the inclusion of CEA is a strong match to the study objective, which centers on efficient use of limited public health resources.

Model performance was consistently better than existing WHO tools. Among HIV-positive individuals, the area under the curve (AUC) was significantly higher for the new model compared to the WHO-recommended four-symptom screen (W4SS), both in South Africa (0.652 vs. 0.568) and Zambia (0.778 vs. 0.725). Similar improvements were observed in HIV-negative or unknown-status populations, where the model outperformed standard symptom-based tools like “any TB symptom” or “prolonged cough.” These results suggest stronger predictive power and a higher potential to identify true TB cases.

Cost-effectiveness was also a key outcome. Across different cut-off strategies in the scoring system, 17 approaches consistently outperformed the WHO tools in both countries in terms of average cost-effectiveness ratio (ACER), which ranged from USD 246 to 1670 in South Africa and USD 164 to 7074 in Zambia. The model identified flexible screening thresholds that allowed programs to balance case detection efficiency with available budgets, making it a pragmatic solution for real-world implementation.

In conclusion, the study demonstrated that a tailored TB scoring system, based on predictive modeling and stratified by HIV status, is not only more accurate than current WHO symptom-based tools but also more cost-effective. Its practicality in field settings through simple score sheets enhances its implementation potential, especially in resource-limited areas. The findings support broader use of data-driven tools to optimize TB screening programs and improve public health outcomes.

Source: Yang, C.C., Shih, Y.J., Ayles, H., Godfrey-Faussett, P., Claassens, M. and Lin, H.H., 2024. Cost-effectiveness analysis of a prediction model for community-based screening of active tuberculosis. Journal of Global Health, 14, p.04226.