Presentations

Abstract

Introduction: Despite the many studies deriving, validating, and/or testing artificial intelligence/machine learning (AI/ML) models in healthcare (HC), understanding how the emergence of AI/ML in medicine is being interpreted by clinicians and developers is integral to its proper stewardship and safe implementation.

Methods: A mixed method, convergent parallel design implementing both a survey and semi-structured interview was used to understand the attitudes and beliefs of obstetric and pediatric providers (HCP) regarding the use of ML/AI in HC and compare them with those of data scientists (DS) and clinical informaticists (CI).

Results: In the inferential analysis due to the small sample size of DS (N =5) and CI (N =3) groups in contrast to HCP (N= 24) statical significances could not be accurately assessed. However, descriptive survey analysis found that less HCP reported having a clear understanding or general familiarity with AI/ML in contrast to DS/CI (46% vs. 100%). HCP were less aware of using AI daily professionally than DS (52% vs. 100%). HCP also perceived AI/ML to be less professionally useful than DS (61% vs 100%). However, all groups reported that data source representation and size were important factors in trusting AI, which was reflected in the themes. HCP and DS had similar ideas about validation, but HCPs deemed professional society endorsement less important and requirement by regulatory agency more important than DS. Thematic analysis of interviews N = 17 highlighted many themes that went into multiple areas but included hopes that AI could alleviate provider task load, offer improvements in direct and indirect patient care activities, and aid in quality improvement/public health efforts. On the other hand, common concerns were societal obstacles, data/AI model validity, improper use, and the potential to undermine provider-patient relationships or humanity in medicine

Conclusion: With some notable differences between them, both clinical providers and data scientists report a positive outlook for improving patient care, efficiency of practice and education amongst other uses while also reporting concerns regarding accuracy/validity, proper usage, and societal obstacles. These findings should be incorporated into the design, implementation, and dissemination of ML/AI in health care.