Thursday, July 27, 2023

Journal Club: NEJM Big Review Article on AI in Healthcare

Update:  NEJM the next week published a long article on AI in Medical Education - here.


On July 27, 2023, New England Journal publishes a masterful review article on AI in healthcare - here.

The article is by Sahni and Carrus, at Harvard and at the McKinsey Center for Healthcare Improvement.

Do you want to read it?  NEJM reviews generally don't have abstracts, making an AI summary particularly helpful.  Here's the GPT3.5 summary:


Can you provide a 50 word and a 100 word summary?

50-Word Summary:

The article discusses the slow adoption of artificial intelligence (AI) in U.S. health care delivery and identifies six key challenges hindering its implementation, including proper cost-benefit definition, data management, change management, workflow integration, patient confidence, and regulatory concerns. Overcoming these challenges is crucial to leverage AI's potential for improving patient care and safety.

100-Word Summary:

The article highlights the slow adoption of artificial intelligence (AI) in U.S. health care delivery due to various challenges. These include the need to properly define costs and benefits, data management complexities, change management considerations, seamless workflow integration, addressing patient confidence issues, and navigating regulatory and methodological concerns. Overcoming these challenges is vital for realizing AI's potential to revolutionize health care by improving patient care and safety. Organizations should focus on rigorous evaluation, increased investment, and conducting demonstration projects to identify effective AI use cases and ensure a smooth transition towards AI-driven health care delivery.



The article discusses the adoption of artificial intelligence (AI) in U.S. health care delivery and the challenges associated with its implementation. It is authored by Nikhil R. Sahni and Brandon Carrus and published in the New England Journal of Medicine.

Key Points:

Slow Adoption of AI in Health Care Delivery: Compared to other business sectors, the adoption of AI in health care delivery has been slow. Various factors contribute to this, including the heterogeneity and complexity of health care data, fee-for-service payment models, lack of patient confidence, regulatory issues, and methodological concerns.

Emerging High-Value Uses of AI: AI is being utilized in various domains of health care delivery, including reimbursement, clinical operations, and quality and safety. Examples include using AI to predict and manage sepsis, optimize operating room scheduling, and reduce clinician burnout by automating electronic health record (EHR) updates.

Challenges in AI Adoption: Several challenges need to be addressed for successful AI adoption in health care organizations. These challenges fall into six categories: strategic vision, key enabling factors, data management, change management, workflow integration, and implementation.

Promising Applications: AI shows potential in improving patient safety by predicting adverse events and enabling early intervention. It can also enhance patient experience by analyzing data to identify dissatisfied members and address their issues.

Need for Expanded Evaluation: Despite the potential benefits, the evidence supporting AI applications in health care is still limited. More research and rigorous evaluation, such as randomized controlled trials, are needed to assess the impact of AI on patient outcomes and safety.

Action Points and Next Steps:

Define Costs and Benefits: Health care organizations should consider the nonfinancial factors, such as improved patient safety and experience, when evaluating the costs and benefits of AI adoption.

Address Data Management: Proper data management is critical for successful AI implementation. Organizations should focus on preparing data, addressing biases, and ensuring continuous testing and validation.

Overcome Implementation Challenges: Organizations must address change management and workflow integration to ensure smooth AI adoption. AI should be seen as a supportive tool that enhances clinical judgment rather than a substitute for it.

Conduct Demonstration Projects: Testing AI applications in a few domains through demonstration projects can provide valuable insights and help identify the most effective use cases.

Investment in AI: Higher-performing organizations in AI adoption spend more on AI than lower performers. Continued investment in AI adoption is essential for reaping its benefits fully.

Rigorous Evaluation: As AI is deployed in health care, rigorous evaluation methods, including randomized controlled trials, should be utilized to measure its impact on patient outcomes and safety.

In conclusion, the potential of AI to revolutionize health care delivery is promising, but there are challenges to overcome. By addressing the key factors hindering AI adoption and investing in its implementation, health care organizations can leverage AI to improve patient care, safety, and overall health care delivery. Rigorous evaluation will be essential to understand the true impact of AI in the health care sector.


Can you specifically call out the 6 most important challenges.

The article identifies six important challenges in the adoption of artificial intelligence (AI) in health care delivery:

Properly Defining Costs and Benefits: Organizations need to accurately assess the costs and benefits of deploying AI. Traditionally, the decision to invest in AI has been based on financial return, but the calculation should also include nonfinancial factors, such as improved patient safety and experience.

Data Management: Health care data is heterogeneous and variable, making it more challenging to apply AI compared to other business sectors. Proper data management is crucial, including data preparation, addressing biases, and ensuring continuous testing and validation.

Change Management: Successfully adopting AI within an organization requires addressing change management. Different stakeholders, from hospital administrators to clinicians, may have different perspectives and implications on how AI impacts their workflows and responsibilities.

Workflow Integration: AI in health care delivery is more effective when seamlessly integrated into the clinical workflow. AI should be viewed as a supportive tool to enhance clinical judgment rather than a replacement for it.

Lack of Patient Confidence: Concerns about privacy, trust in AI output, and the potential consequences of AI-based decisions contribute to patient apprehension. Building patient confidence in AI applications is essential for widespread adoption.

Regulatory and Methodological Concerns: Regulatory issues, such as Food and Drug Administration (FDA) approval and reimbursement, need to be addressed. Additionally, concerns about validation and communication of AI-based recommendations or decisions, as well as reporting difficulties, must be overcome for successful AI adoption.

Addressing these challenges is crucial to the successful implementation of AI in health care delivery and realizing the potential benefits it offers for patient care and overall health care outcomes.


I use OpenAI ChatGPT, both the "free" 3.5 and "paid" 4.0.  I find the 3.5 often does better at article summaries.


In 3 Haiku:

AI's promise vast,
Health care's slow embrace revealed,
Challenges must pass.

Data's complexity,
Patient trust and workflow merge,
AI's path to forge.

Six challenges rise,
Adoption's key hurdles faced,
Health care's AI quest.

See 2 minute and 4 minute video scripts here.