Cost-Consequences Analysis (CCA) of Artificially Intelligent Clinician-Friendly Interpretable Computer-Aided Diagnosis (ICADX) Tool for Coronary Artery Disease (CAD) Developed at Hosmartai (Horizon 2020 Funded)

OBJECTIVES: Coronary artery disease (CAD) has been regarded as one of the most dangerous and life-threatening chronic diseases. The clinical recommendations are to use contrast enhanced Coronary Computed Tomography Angiography (CCTA) as a first-line diagnostic option in obstructive CAD Data-driven decision-making using AI algorithms has been increasingly common in the CAD field by providing improved diagnostic accuracy and automated, standardized interpretation and inference processes. CADXpert-CARDIO) uses clinically validated models that can efficiently predict the Pre-test probability of stable CAD (Stenosis>50%) and hence do not require additional diagnostic testing. The study evaluated the economic and clinical performance of the CCTA tool on the accurate detection of CAD emerging from an automatic AI-based tool.

METHODS: A micro-costing analysis, based on costs of the Greek healthcare system, was performed to identify the costs of development of the new AI technology and an array of the most important indicators to enable performance identification. The most important KPIs were i) clinical performance, ii) User satisfaction iii) Duration of diagnosis by experienced/less experienced physician. The analysis of costs and consequences was performed incrementally between the current practice and the new AI technology.

RESULTS: The annual cost of the new AI technology costs presents savings to the Greek Healthcare system since the annual cost per year has been estimated at €41.860 vs. €67.210 at current practice presenting savings of €25.350. The clinical performance of Hosmartai intervention has a ROC-AUC of 0.84 and a recall of 0.96 meaning that it is very sensitive to all positive cases in clinical practice. The system usability score of the new technology was 72.08%. The new AI technology required fewer human resources for the review of the examination.

CONCLUSIONS: AI could automate parts of the process providing improved accuracy, with less human resources, resulting in savings for the healthcare systems.