Primary care provider preferences regarding artificial intelligence in point-of-care cancer screening

BACKGROUND: It is unclear how to optimize the user interface and user experience of cancer screening artificial intelligence (AI) tools for clinical decision-making in primary care.

METHODS: We developed an electronic survey for US primary care clinicians to assess 1) general attitudes toward AI in cancer screening and 2) preferences for various aspects of AI model deployment in the context of colorectal, breast, and lung cancer screening. We descriptively analyzed the responses.

RESULTS: Ninety-nine surveys met criteria for analysis out of 733 potential respondents (response rate 14%). Ninety (>90%) somewhat or strongly agreed that their medical education did not provide adequate AI training. A plurality (52%, 39%, and 37% for colon, breast, and lung cancers, respectively) preferred that AI tools recommend the interval to the next screening as compared with the 5-y probability of future cancer diagnosis, a binary recommendation of “screen now,” or identification of suspicious imaging findings. In terms of workflow, respondents preferred generating a flag in the electronic health record to communicate an AI prediction versus an interactive smartphone application or the delegation of findings to another healthcare professional. No majority preference emerged for an explainability method for breast cancer screening.

LIMITATIONS: The sample was primarily obtained from a single health care system in the Northeast.

CONCLUSIONS: Providers indicated that AI models can be most helpful in cancer screening by providing prescriptive outputs, such as recommended intervals until next screening, and by integrating with the electronic health record.

IMPLICATIONS: A preliminary framework for AI model development in cancer screening may help ensure effective integration into clinical workflow. These findings can better inform how healthcare systems govern and receive reimbursement for services that use AI.

Vinayak S Ahluwalia, Marilyn M Schapira, Gary E Weissman, Ravi B Parikh