Objective:

To review the role of artificial intelligence (AI) in improving healthcare quality and patient safety through diagnostic, predictive, and decision support technologies.

Approach:

• Background: AI has gained prominence in modern medicine, particularly in medical image interpretation and patient deterioration prediction, driven by increased healthcare data, advancements in machine learning, and enhanced computational resources.
• Challenges: Despite promising individual studies, there is a significant gap in translating AI from development to routine clinical use, often referred to as the 'AI chasm'.
• Evidence Base: Many AI systems report strong performance but lack prospective validation, leading to questions about their reliability in real-world settings.
• Economic Assessments: Economic evaluations often forecast savings without accounting for implementation costs, potentially overestimating benefits.
• Equity Concerns: There is limited evidence on AI's impact across diverse patient populations, raising concerns about exacerbating existing disparities.

Key Findings:

• AI systems show strong performance in clinical applications but face challenges in validation and integration into clinical workflows.
• The economic impact of AI in healthcare is often overestimated due to inadequate consideration of implementation costs.
• There is a lack of comprehensive evidence regarding AI's effects on health equity and potential biases in algorithms.

Interpretation:

The gap between AI's potential and its actual clinical utility is influenced by validation methods, integration challenges, and governance issues rather than algorithmic performance alone.

Limitations:

• Many studies on AI use retrospective designs, which carry a risk of bias.
• A limited number of AI systems have undergone prospective external validation.
• Economic assessments often rely on static assumptions that may not reflect real-world complexities.

Conclusion:

Addressing the identified challenges is crucial for stakeholders, including healthcare administrators, clinicians, policymakers, and patients, to ensure effective AI integration in healthcare.

Sources:

• Hyland et al. (2020)
• Acosta et al. (2022)
• Topol (2019)
• Kelly et al. (2019)
• Schwalbe & Wahl (2020)
• Davenport & Kalakota (2019)
• Vaishya et al. (2021)
• Liu et al. (2019)
• Nagendran et al. (2020)
• Fleuren et al. (2020)
• Wolff et al. (2020)
• Gianfrancesco et al. (2018)
• Obermeyer et al. (2019)