Translating AI to the Bedside with Physician Buy-In: Recommendations from a Meta-Analysis and Systematic Review of the Literature

Grace Hwang, Seyyed Sina Hejazian, Alireza Vafaei Sadr, Jennifer K. Wagner, Peng-Jie Hao, Ajith Vemuri, Yuki I. Kawamura, Khalid Nawab, Shadi Hijjawi, Ramin Zand, Vida AbediBioengineering 2025 · DOI

TL;DR

This systematic review and meta-analysis reveals that while physicians are cautiously optimistic about AI, their buy-in is hindered by a lack of knowledge, concerns about workflow, and legal uncertainties. The paper proposes five actionable recommendations to foster physician adoption of AI in healthcare, emphasizing education, early clinician involvement in design, and clear communication.

What Problem Does This Paper Address?

The integration of Artificial Intelligence (AI) into healthcare holds immense promise for improving diagnostics, streamlining workflows, and augmenting medical knowledge. However, the successful translation of AI from research settings to actual clinical practice, particularly at the bedside, hinges critically on the acceptance and adoption by physicians and trainees. This paper addresses the significant gap between the potential of AI and its practical implementation by investigating the factors influencing physician buy-in. It aims to understand physicians' current knowledge and attitudes towards AI, identify the barriers to its adoption, and provide evidence-based recommendations to overcome these challenges, ultimately facilitating the seamless integration of AI into daily clinical practice.

Methods

The study employed a comprehensive systematic review and meta-analysis approach. Researchers searched major databases (PubMed, OVID, IEEE, Scopus, Web of Science) for studies published between 2013 and 2024, focusing on physician and trainee knowledge, attitudes, and perspectives on AI in healthcare. After an initial screening of over 11,000 records, 115 studies met the inclusion criteria. Of these, 53 studies provided quantitative data suitable for a meta-analysis on physician knowledge levels. Inductive thematic analysis was used to identify overarching themes related to physician buy-in, categorizing findings into knowledge, engagement, legal/regulatory uncertainties, workflow, and patient-centeredness. A scoring system was developed to categorize studies based on reported AI knowledge levels (low to high).

Datasets

This paper does not introduce new datasets. Instead, it synthesizes data from 115 existing studies that collected information from physicians and medical trainees regarding their knowledge, attitudes, and perceptions of AI in healthcare. These studies utilized various methodologies, primarily questionnaires, to gather data.

Key Insights and Ideas

  • Knowledge is a Major Hurdle: A significant portion of physicians and trainees possess only moderate conceptual knowledge of AI, with limited technical understanding. This knowledge gap is a primary barrier to buy-in.
  • Conceptual vs. Technical Knowledge: While conceptual understanding of AI's potential is linked to greater receptiveness, deeper technical knowledge about how AI works appears to foster even higher trust and willingness to adopt.
  • Engagement is Crucial, Early On: Involving physicians and trainees in the early stages of AI design is vital. They view AI as a support tool, not a replacement, and their input ensures AI aligns with clinical decision-making and workflows.
  • Legal and Regulatory Fog: Significant uncertainty surrounds legal aspects like liability and malpractice related to AI use, creating hesitancy among clinicians.
  • Workflow Integration is Key: AI tools need to seamlessly integrate into existing workflows, avoiding excessive clicks, alert fatigue, and technical glitches, which mirror past frustrations with EHR systems.
  • Patient-Centeredness Matters: Physicians are concerned that AI might negatively impact the physician-patient relationship by reducing direct interaction and empathy. AI tools must augment, not detract from, patient-centered care.

Key Results

  • Low AI Knowledge is Prevalent: The meta-analysis estimated that only 19.6% of physicians and trainees have high overall AI knowledge, while a substantial 36.3% have low knowledge.
  • Moderate Conceptual Understanding: The majority of studies indicated that physicians and trainees possess only moderate conceptual knowledge of AI, with limited technical expertise.
  • Positive but Cautious Outlook: Most studies (52%) concluded that physicians and trainees are "cautiously optimistic" about AI in healthcare.
  • Five Key Barriers Identified: The review identified five major barriers to AI adoption: knowledge, engagement, legal/regulatory uncertainties, workflow, and patient-centeredness.

Limitations and Future Work

The authors acknowledge several limitations. The heterogeneity in questionnaires used across studies makes direct comparisons of knowledge levels challenging. The meta-analysis can only suggest trends in overall knowledge due to the varied methodologies. Future work could benefit from the development of a validated study tool for more precise comparisons. The paper also highlights the need for further research into how medical trainees utilize AI and their preferences for learning about it, to inform the development of evidence-based educational guidelines. Additionally, more research is needed to distinguish between moral/professional responsibility for patient outcomes and malpractice standards in the context of AI.

Summary updated

Abstract

Although physicians and trainees are generally receptive to AI, many barriers hinder adoption. To address them, we recommend establishing standardized AI education and workforce training, involving clinicians early in AI design, clarifying legal and regulatory issues, leveraging insights from clinical decision support system implementation to reduce workflow challenges, and integrating patient-centered communication principles to enhance trust and transparency.
BibTeX
@inproceedings{hwang2025translating,
  author = {Grace Hwang and Seyyed Sina Hejazian and Alireza Vafaei Sadr and Jennifer K. Wagner and Peng-Jie Hao and Ajith Vemuri and Yuki I. Kawamura and Khalid Nawab and Shadi Hijjawi and Ramin Zand and Vida Abedi},
  title = {{Translating AI to the Bedside with Physician Buy-In: Recommendations from a Meta-Analysis and Systematic Review of the Literature}},
  booktitle = {Bioengineering},
  year = {2025},
  doi = {10.3390/bioengineering12121363},
}
Figure