ISOM Seminar: Indranil Bardhan, University of Texas at Austin, McCombs

About this event

Affiliation: University of Texas at Austin, McCombs Area: Information Systems

Name of Presentation: Human-AI Augmentation and Diagnostic Quality: Evidence from Medical Devices

Abstract: Artificial intelligence (AI) is increasingly embedded in clinical care, yet most research examines advice-providing AI that generates predictions, diagnoses, or recommendations. We study a distinct class of information-generating AI, which augments physicians’ information without directly prescribing an action. In this study, we examine the role of information-generating AI in the context of AI-augmented, diagnostic procedures of coronary artery disease (CAD), which physicians can use to guide diagnostic and treatment decisions. Our empirical setting is fractional flow reserve computed tomography (FFR-CT), an AI-enabled technology integrated into the coronary computed tomography angiography (CCTA) pathway. FFR-CT generates lesion-specific physiological information from CCTA images, potentially helping physicians distinguish patients who require invasive evaluation from those who do not.

Using a nationwide Medicare Limited Data Set outpatient and inpatient claims from 2020-2024, we identify physician adoption of FFR-CT across all Medicare claims and construct a physician-quarter longitudinal panel of patients undergoing CCTA. We estimate a statistical model comparing physicians who adopt FFR-CT with those who never adopted during our study period. To mitigate selection into adoption, we implement one-to-one propensity score matching based on pre-adoption physician and patient characteristics.

We find that FFR-CT adoption significantly reduces the number of avoidable invasive coronary angiography (ICA) and lowers duplication of diagnostic procedures and subsequent functional testing. At the same time, the number of procedures performed during the focal visit increases, suggesting that FFR-CT use does not simply reduce diagnostic intensity, but instead reallocates diagnostic resources toward high-value services. We further show that FFR-CT use reduces unexplained variation in the concentration of diagnostic resources, after accounting for observed patient and physician characteristics.

This study improves our understanding of the impact of human-AI augmentation on physician diagnosis and treatment of CAD patients by demonstrating that use of AI-enabled medical devices reduces the number of avoidable procedures and follow-on tests, thereby leading to greater diagnostic efficiency and care quality.