Conspicuous by Its Absence: Diagnostic Expert Testing Under Uncertainty

We study the problem a diagnostic expert (e.g., a physician) faces when offering a diagnosis to a client (e.g., a patient) that may be based only on the expert’s own diagnostic ability or supplemented by a diagnostic test—conventional and artificial intelligence (AI) tools alike—revealing the client’s true condition. The expert’s diagnostic ability (or type) is private information. The expert is impurely altruistic in that the expert cares about both the client’s utility and the expert’s own reputational payoff that depends on the peer perception of the expert’s diagnostic ability. The decision of whether to perform the test, which is costly for the client, provides the expert with an opportunity to influence that perception. We show a unique separating equilibrium exists in which the high-type expert does not resort to diagnostic testing and offers a diagnosis based only on the expert’s own diagnostic ability, whereas the low-type expert performs the test. Furthermore, we establish that high-type expert may skip necessary diagnostic tests to separate them from the low-type expert. Interestingly, the effect of reputational payoff on undertesting is nonmonotonic, and the desire to appear of high type leads to undertesting only when the reputational payoff is intermediate. Our results also suggest a more altruistic expert may be more likely to engage in undertesting. Furthermore, efforts to encourage testing by providing financial incentives or by raising malpractice lawsuit concerns may, surprisingly, help fuel undertesting in the equilibrium. Our paper sheds new light on barriers to the adoption of AI tools aimed at enhancing physicians’ diagnostic decision making.