Primary Care Physician Perspectives on Artificial Intelligence for Dementia Screening: A Qualitative Study

Automated approaches to cognitive impairment screening may soon achieve sufficient levels of accuracy for clinical implementation but they present potentially serious ethical challenges. Overcoming such challenges for successful implementation of automated screening may depend on the perspective of the clinicians who are its intended users.

We conducted a qualitative study of internal medicine, geriatrics and family medicine physicians recruited from ambulatory practices in New York City, NY to identify ethical challenges of implementing automated CI screening (n = 22). In four focus groups, we explored attitudes and beliefs about routine manual screening for cognitive impairment and automated screening based in machine learning models of data from electronic medical records (EMR) or patient audio recordings, using hypothetical scenarios. Focus group recordings were transcribed and analyzed using grounded theory.

Participants reported routine screening for cognitive impairment only in the context of the Medicare annual wellness visit and generally avoided routine screening because of limited treatment options to support patients, such as poor access to geriatrics/neuropsychiatric services. Thematic analysis revealed several perceived benefits: enhanced clinical efficiency, engagement in cognitive healthcare of their patients, and expanded access to cognitive care. They identified several potential challenges with ethics implications: concerns about accuracy of the technology, bias, difficulty communicating the technology and results to patients and caregivers that could impact their ability to provide informed consent, and risks to patients (privacy, stigmatization and insurability). They noted that without more health system infrastructure to support patients with dementia, the benefits of automated screening would be limited.

Physicians identified several critical ethical challenges to automated CI screening. Health systems will need to address these challenges to ensure benefit for and the safety, autonomy and privacy of patients and ultimately, the successful implementation of such technology.