Abstract Body: Background:
Machine learning and artificial intelligence (ML/AI) are rapidly spreading in clinical medicine. Few data describe the perspectives of patients, proxies (e.g., patients’ spouses), and clinicians. In this mixed-methods study, we qualitatively characterized perspectives regarding ML/AI use and quantitatively explore sentiment towards ML/AI from acute neurology patients, proxies, and clinicians.
Methods:
We conducted semi-structured interviews with survivors of intracranial hemorrhage, proxies, and clinicians. We analyzed interview transcripts using framework analysis, organizing data within the domains of the Theoretical Framework of Acceptability, adding domains identified with input from all co-authors. We quantitatively analyzed the sentiment scores of responses from positive to negative using a transformer-based model, the same technology that underlies large language models. Sentiment scores were compared with Kruskal-Wallis H, and multiple comparisons adjusted using Dunn’s test.
Results:
We analyzed 21 interviews (14 patients, 1 proxy, and 6 clinicians), by which point there was thematic saturation. Help with clinical decision-making was cited as the key potential advantage of ML/AI. Participants noted the importance of considering ML/AI as an adjunct to clinical care, not as a replacement for clinicians. Over-reliance on recommendations potentially leading to diminution of clinician skill, incorrect ML/AI recommendations, potential liability, and bias were cited as challenges. Clinician and patient education were noted as potential burdens that impose opportunity costs, but are important for self-efficacy. Median sentiment scores ranged from 0.0 (neutral) to 0.3 (positive). Sentiment varied with question type (P < 0.001). Questions about clinicians’ using ML/AI for patient care had the highest sentiment score.
Conclusion:
Patients, caregivers, and clinicians expressed mixed views about ML/AI. Concerns related to potential burdens and opportunity costs were noted and should be considered as ML/AI is introduced. Future directions include how best to incorporate ML/AI into education and obviate potential burdens as ML/AI is integrated into clinical care.