AI-Based TB Triage Cuts Costs and Improves Outcomes in Ghana
Cost-effectiveness analysis of computer aided detection for tuberculosis in Ghana: A decision-analytic modeling study
PloS one, 2026
Abstract
Introduction: Sustaining universal molecular testing for tuberculosis (TB) is a significant financial burden in high-prevalence settings. In Ghana, GeneXpert MTB/RIF is the standard confirmation tool, but its high unit cost demands efficient upstream triage. This study examines whether AI-powered Computer-Aided Detection (CAD) can serve as a cost-effective gatekeeper to molecular testing, reducing unnecessary expenditure without sacrificing diagnostic accuracy.
Methods: A prospective cross-sectional diagnostic accuracy study with an embedded decision-analytic economic evaluation was conducted using primary data from 701 symptomatic outpatients across 23 health facilities in Ghana. Three CAD4TB scoring thresholds were compared: the vendor-recommended threshold of 51, and two locally calibrated thresholds of 56 (weighted Youden index) and 68 (standard Youden index). These were benchmarked against the current Standard of Care (SoC), which comprises symptom screening and human-interpreted chest X-ray. Health outcomes were quantified using Quality-Adjusted Life Years (QALYs) lost, with a morbidity penalty applied to missed TB cases, and the primary economic outcome was the Incremental Cost-Effectiveness Ratio (ICER).
Results: The CAD4TB threshold of 56, calibrated using the weighted Youden index, emerged as the dominant clinical strategy — it was simultaneously more effective and less costly than the current SoC. Total pathway costs for the CAD-56 strategy and SoC were US$23,449 and US$33,239 respectively across 701 patients screened, representing a health system saving of approximately US$9,790. The CAD-56 strategy also delivered an incremental gain of 10.08 QALYs compared to standard practice.
Conclusion: Deploying CAD4TB version 7 at a locally optimised threshold of 56 as a triage filter before molecular testing represents a dominant economic strategy for TB control in Ghana. The authors recommend this approach be evaluated with nationwide data prior to broader national rollout of AI-based screening.