Ethics of AI in healthcare: a scoping review demonstrating applicability of a foundational framework

Aaron J. Gorelik, Mengyuan Li, Jessica Hahne, Junyi Wang, Yongqi Ren, Lei Yang, Xin Zhang, Xing Liu, Xiaomin Wang, Ryan Bogdan, Brian D. CarpenterFrontiers Digit. Health 2025 · DOI

TL;DR

This paper argues that the four established principles of biomedical ethics (Beneficence, Non-Maleficence, Respect for Autonomy, and Justice) provide a robust and universally applicable foundation for navigating the ethical complexities of Artificial Intelligence (AI) in healthcare. Through a comprehensive scoping review of 227 peer-reviewed articles, the authors demonstrate that existing ethical discussions surrounding healthcare AI map effectively onto these principles, offering a unified approach to guide responsible AI development and deployment.

What Problem Does This Paper Address?

The rapid integration of AI into healthcare promises significant advancements, from automating tasks to improving diagnoses. However, this integration also introduces novel and complex ethical risks, particularly concerning patient well-being and data privacy. The proliferation of numerous, often overlapping or contradictory, "Responsible AI" (RAI) frameworks can exacerbate these risks by leading to confusion in interpretation and prioritization. This paper addresses the critical need for a unified, broadly accepted ethical framework that can serve as a foundation for AI in healthcare, ensuring that its development and deployment are both innovative and ethically sound. The authors aim to demonstrate that the long-standing principles of biomedical ethics, already deeply embedded in healthcare practice, can fulfill this role.

Methods

The researchers conducted a scoping review of 227 peer-reviewed articles published on the use of AI in healthcare. Their methodology involved a semi-inductive thematic analysis using NVivo software. This approach combined deductive application of a predetermined framework (the four principles of biomedical ethics) with inductive identification of emerging themes from the data. The process included rigorous reviewer calibration to ensure consistency in data extraction and coding. The identified ethical issues were then categorized under the four core principles: Beneficence, Non-Maleficence, Respect for Autonomy, and Justice.

Datasets

Not discussed in the paper. The paper is a scoping review of existing literature and does not involve the creation or use of new datasets for training or evaluation.

Key Insights and Ideas

  • Biomedical Ethics as a Universal Foundation: The core insight is that the established four principles of biomedical ethics are not only relevant but also comprehensive enough to encompass the vast majority of ethical considerations surrounding AI in healthcare. This offers a much-needed unifying framework.
  • Mapping Existing Discussions: The review successfully mapped existing ethical discussions in the literature onto these four principles, demonstrating their practical applicability and showing that current debates are already implicitly or explicitly aligned with them.
  • Bridging the Gap: The four principles can serve as a bridge between general RAI frameworks and the specific, high-stakes ethical landscape of healthcare, providing a common language and set of priorities.
  • Facilitating Regulatory Governance: By grounding AI ethics in these well-understood principles, it becomes easier to develop consistent, streamlined, and globally applicable regulatory guidelines.
  • Addressing Ethical Dilemmas: The framework, with its long history and established methods for resolving conflicts (like the rule of double effect), can help navigate the inherent tensions and dilemmas that arise when applying AI in healthcare.

Key Results

The scoping review identified 227 relevant articles, with the majority published between 2020 and 2024, originating from 33 countries. The analysis revealed that discussions of ethical issues in healthcare AI spanned all four principles of biomedical ethics, with a comparable number of articles addressing each: Beneficence (n=136), Non-Maleficence (n=148), Respect for Autonomy (n=130), and Justice (n=136).

A total of 29 specific ethical issues were identified and categorized. Prominent themes included:

  • Privacy and data protection (under Non-Maleficence, n=107)
  • Transparency and understandability (under Respect for Autonomy, n=93)
  • Bias and discrimination (under Justice, n=80)
  • Accuracy and efficiency (under Beneficence, n=75)
  • Equity and fairness (under Justice, n=54)

The review found that while most identified themes mapped directly to general RAI principles, two themes under Beneficence—patient experience/doctor-patient relationship and social/humanistic dimensions of health services—highlighted aspects particularly unique to healthcare that are well-captured by the principle of Beneficence.

Limitations and Future Work

The authors acknowledge several limitations:

  • Search Scope: The review was limited to two databases (PubMed and EMBASE) and specific search terms, potentially missing relevant literature.
  • Quality Assessment: The focus was on mapping ethical discussions, not on evaluating the quality of the empirical research within the included articles.
  • Rapid Evolution of AI: The fast-paced nature of AI development means that any literature review is a snapshot in time, and future studies will need to address more recent advancements.

The authors suggest that future work should continue to explore how these foundational principles can be applied to specific AI technologies and applications in healthcare. They also emphasize the need for ongoing dialogue to adapt these principles as AI continues to evolve, ensuring that regulatory mechanisms remain robust and ethically grounded. The paper advocates for building upon these established principles to create a comprehensive governance system for responsible AI in healthcare.

Summary updated

BibTeX
@article{gorelik2025ethics,
  author = {Aaron J. Gorelik and Mengyuan Li and Jessica Hahne and Junyi Wang and Yongqi Ren and Lei Yang and Xin Zhang and Xing Liu and Xiaomin Wang and Ryan Bogdan and Brian D. Carpenter},
  title = {{Ethics of AI in healthcare: a scoping review demonstrating applicability of a foundational framework}},
  journal = {Frontiers Digit. Health},
  year = {2025},
  doi = {10.3389/fdgth.2025.1662642},
}
Figure