COST-EFFECTIVENESS ANALYSIS OF AI-ASSISTED CHEST X-RAY (DEEPCXR) FOR PULMONARY TUBERCULOSIS SCREENING IN INDIA...
Author(s)
Devang R. Raval, MPH, M.A Economic1, Dhaval Parmar, PhD1, Somen Saha, PhD1, Harsh Shah, PhD1, Raju Sarkar, PhD1, Jigna Gohil, MD1, Archana Thakur, MD1, Junaid KP, PhD1, Krushnachandra Sahoo, PhD2, Kavitha Rajsekar, PhD2, Deepak B. Saxena, PhD1.
1Indian Institute of Public Health Gandhinagar, Gandhinagar, India, 2DHR, HTAIn, New Delhi, India.
1Indian Institute of Public Health Gandhinagar, Gandhinagar, India, 2DHR, HTAIn, New Delhi, India.
OBJECTIVES: To evaluate the cost-effectiveness of DeepCxR, an AI-assisted chest X-ray interpretation tool, compared with radiologist-interpreted digital chest X-ray for pulmonary TB screening in India.
METHODS: A decision-analytic model was developed to simulate pulmonary TB screening outcomes in a hypothetical cohort of 1,000 individuals undergoing facility-based screening, assuming 15% TB prevalence. Diagnostic accuracy inputs for DeepCxR were derived from Unpublished Indian Council of Medical Research validation data and adjusted using Mycobacterial Growth Indicator Tube (MGIT) culture as the microbiological reference standard. Cost inputs were estimated using a micro-costing approach from the healthcare system perspective. Outcomes included true positives, false positives, false negatives, correctly classified cases, total costs, and incremental cost-effectiveness.
RESULTS: DeepCxR demonstrated sensitivity of 85.9% and specificity of 62.6%, compared with 87.0% sensitivity and 89.0% specificity for radiologist-interpreted digital chest X-ray. In a cohort of 1,000 screened individuals, DeepCxR identified 129 true-positive and 532 true-negative cases, compared with 131 true-positive and 757 true-negative cases using conventional digital chest X-ray. DeepCxR generated substantially more false positives (318 vs 94) and 226 fewer correctly classified cases overall. Total screening cost per 1,000 individuals was lower for DeepCxR (INR 247,747) than conventional digital chest X-ray (INR 299,077), resulting in cost savings of INR 51,330.
CONCLUSIONS: DeepCxR reduced screening costs within India’s public health system but resulted in lower diagnostic accuracy compared with radiologist-interpreted digital chest X-ray, primarily due to substantially lower specificity. The savings were associated with 1.7 additional missed TB cases. AI-assisted screening offers operational advantages in settings with limited radiology capacity. Its economic value depends on implementation context, confirmatory diagnostic pathways, and acceptable trade-offs between cost savings and diagnostic performance.
METHODS: A decision-analytic model was developed to simulate pulmonary TB screening outcomes in a hypothetical cohort of 1,000 individuals undergoing facility-based screening, assuming 15% TB prevalence. Diagnostic accuracy inputs for DeepCxR were derived from Unpublished Indian Council of Medical Research validation data and adjusted using Mycobacterial Growth Indicator Tube (MGIT) culture as the microbiological reference standard. Cost inputs were estimated using a micro-costing approach from the healthcare system perspective. Outcomes included true positives, false positives, false negatives, correctly classified cases, total costs, and incremental cost-effectiveness.
RESULTS: DeepCxR demonstrated sensitivity of 85.9% and specificity of 62.6%, compared with 87.0% sensitivity and 89.0% specificity for radiologist-interpreted digital chest X-ray. In a cohort of 1,000 screened individuals, DeepCxR identified 129 true-positive and 532 true-negative cases, compared with 131 true-positive and 757 true-negative cases using conventional digital chest X-ray. DeepCxR generated substantially more false positives (318 vs 94) and 226 fewer correctly classified cases overall. Total screening cost per 1,000 individuals was lower for DeepCxR (INR 247,747) than conventional digital chest X-ray (INR 299,077), resulting in cost savings of INR 51,330.
CONCLUSIONS: DeepCxR reduced screening costs within India’s public health system but resulted in lower diagnostic accuracy compared with radiologist-interpreted digital chest X-ray, primarily due to substantially lower specificity. The savings were associated with 1.7 additional missed TB cases. AI-assisted screening offers operational advantages in settings with limited radiology capacity. Its economic value depends on implementation context, confirmatory diagnostic pathways, and acceptable trade-offs between cost savings and diagnostic performance.
Conference/Value in Health Info
2026-09, ISPOR Asia Pacific 2026, Bangkok, Thailand
Value in Health, Volume 55, Issue S1
Code
HTA2
Topic
Health Technology Assessment
Disease
SDC: Infectious Disease (non-vaccine)