Economic evaluation for medical artificial intelligence: accuracy vs. cost-effectiveness in a diabetic retinopathy screening case

Artificial intelligence (AI) models have shown high accuracy in health screening, but high accuracy may not guarantee cost-effectiveness. This study conducted a cost-effectiveness analysis of an AI model in a Chinese diabetic retinopathy (DR) screening program. The analysis considered 1100 different AI diagnostic performances (sensitivity/specificity pairs). Six scenarios were cost-saving and seven were cost-effective compared to the most accurate model. Cost-saving or cost-effective AI models needed a minimum sensitivity of 88.2% and specificity of 80.4%. Higher DR prevalence and willingness-to-pay levels required higher AI sensitivity for optimal cost-effectiveness. Urban regions and younger patients also needed higher sensitivity.

Yueye Wang, Chi Liu, Wenyi Hu, Lixia Luo, Danli Shi, Jian Zhang, Qiuxia Yin, Lei Zhang, Xiaotong Han, Mingguang He