Is crafty ‘Dr. AI’ already siphoning mindshare from actual doctors?

Patients can be persuaded to place almost as much confidence in diagnostic AI as they have in primary care physicians.

The conclusion comes from a study in which PCPs defeated machines at earning patient partiality by just six percentage points, 53% to 47%. Researchers at the University of Arizona conducted the experiment. PLOS Digital Health published the resulting study report May 19.

Senior author Marvin Slepian, MD, and colleagues ran the project in two phases. The first was a pretest of the materials in which the team interviewed 24 demographically diverse patients using a structured format. The second was a randomized, blinded survey of more than 2,600 healthcare consumers whose number was oversampled for minority representation.  

For the survey, the team presented participants with clinical scenarios randomly scrambling several variables. These included disease severity, whether AI is proven more accurate than human specialists, whether the AI diagnostician is personalized to the patient through listening and/or tailoring, whether the AI diagnostician avoids racial and/or financial biases, whether the PCP promises to explain and incorporate the advice, and whether the PCP nudges the patient toward AI as the established, recommended and easy choice.

Asking participants to choose one or the other—the doctor or the AI—the researchers found the aforementioned 53% to 47% score. Importantly, responses were weighted to represent the ethnic, racial and demographic makeup of the U.S. adult population, Slepian and co-authors report.

In unweighted experimental contrasts of qualified respondents, acceptance of AI—or AI “uptake”—significantly rose under three conditions:

1. a PCP’s explanation that AI has proven superior accuracy,
2. a PCP’s nudge toward AI as the established choice, and
3. reassurance that the AI clinic had trained counselors to listen to the patient’s unique perspectives.

Interestingly, patients weren’t swayed to choose a human doctor over AI when the hypothesized diagnosis was a dreaded disease such as leukemia (versus a comparatively benign condition like sleep apnea). Also noteworthy: Compared to White participants, Black participants selected AI less often, and Native Americans selected it more often. Also, older respondents were less likely to choose AI, as were those who identified as politically conservative or viewed religion as important.  

In general, higher educational levels corresponded with greater acceptance of AI for medical diagnostics. “To ensure that the benefits of AI are secured in clinical practice, future research on best methods of physician incorporation and patient decision making is required,” the authors comment.

The study is available in full for free, and the university has posted its own news coverage.