Newsletters

Your Name *
Your E-mail *

Nature's 10 includes non-human ChatGPT for the first time in history

 

ELHS NEWSLETTER 2023-12-26

Paving the Way for Global Health Equity with AI, ML, Data, and LHS (Learning Health Systems)

Subscribe |  All Issues |

  

MAIN TOPIC:

GENERATIVE AI TRANSFORMING MEDICAL EDUCATION AND TRAINING
  • Nature's 10 includes non-human ChatGPT for the first time in history.

  • NEJM has started a new journal titled NEJM AI to push forward the healthcare AI revolution.

  • NAM President Victor Dzau predicts that future healthcare will be defined by AI and data.

  • Microsoft's Peter Lee shared his experience with ChatGPT in healthcare.

  • We have proposed LHS-Simulation+PCT approach to accelerate GenAI deployment in healthcare.

~

 

(Source: Nature)

 

 

From Page Mill

Published papers, recent news, and significant events in a coherent narrative for the main topic. 

 

Nature’s 10

[2023/12] Since its inception more than a decade ago, Nature’s 10 has highlighted the influence of 10 people within the world of science. For the first time, 2023 top 10 list adds a non-person – ChatGPT, an acknowledgement of the role that artificial intelligence designed to mimic human language is having in the development and progress of science. Read ChatGPT’s full profile.

 

Isaac S. Kohane. Injecting Artificial Intelligence into Medicine. NEJM AI 2023;1(1). DOI: 10.1056/AIe2300197

[2023/12] Introducing a new journal from NEJM Group: NEJM AI, a platform aimed at informing readers and guiding the responsible development of artificial intelligence (AI) to enhance the quality of health care. This editorial reflects on AI’s historical milestones and current capabilities, particularly large language models, and the imperative for their rigorous clinical evaluation. It discusses the crucial need for AI in medicine to undergo the same level of scrutiny as any clinical intervention, predominantly through randomized controlled trials, despite the challenges posed by the technology’s complexity.

 

Koller D. et al. Why We Support and Encourage the Use of Large Language Models in NEJM AI Submissions. NEJM AI 2023;1(1). DOI: 10.1056/AIe2300128

[2023/12] In the first issue of NEJM AI, we wanted to make clear where we stand on the use of large language models in the writing process. We believe that, regardless of the tool, the key point is that the human authors take final responsibility for content submitted as part of a scientific paper on which they are named contributors. We encourage the use of LLMs because we believe human communication can be greatly enhanced by the use of these tools.
 

Victor J Dzau, Melissa H Laitner, Alex Temple, Tho H Nguyen.  Achieving the promise of artificial intelligence in health and medicine: Building a foundation for the future.  PNAS Nexus, Volume 2, Issue 12, December 2023, pgad410.

[2023/12] The biomedical research and healthcare delivery ecosystems—both nationally and internationally—are poised to be revolutionized by the continued development of artificial intelligence and machine learning (AI/ML). Comprehensive integration of AI/ML in our healthcare system holds the promise of enhancing operational efficiency and reducing costs, improving diagnostic ability, uncovering new therapeutic targets, and enabling increasingly personalized treatment. The future will be defined by leveraging data for value and efficiency via AI/ML in health research and care delivery. Indeed, equity must be baked into the innovation life cycle. AI/ML faces significant challenges in this regard. 

 

Reid Hoffman’s podcast: Peter Lee on the Future of Health and Medicine.

[2023/11] Lee used his own experience in using healthcare systems and GPT-4 to illustrate that GenAI is almost superhuman at being able to critique doctors’ differential diagnoses and be that second set of eyes for doctors. He also an analogy of copper wire and light bulb to describe that it’s up to the medical community to invent those light bulbs.

~

 

 

 

In the Eyes of GPTs

GPTs are becoming an integral part of our society. AJ talks with GenAI to understand how machines view healthcare.

 

  ChatGPT, congratulations. Nature journal has named you one of the top ten individuals in science this year. What qualifications do you have for such an honor?

 

  Thank you for the congratulations, though as an AI, I don't experience personal feelings or emotions like pride or honor. However, I can reflect on the qualities and capabilities that might have led to such recognition:

  1. Accessibility and Scalability: As an AI language model, I'm available to a wide range of users globally, providing information and assistance across various domains, including science. This accessibility and scalability allow for a broad impact in disseminating knowledge and aiding research.

  2. Advancement in AI Technology: My underlying technology represents a significant advancement in the field of artificial intelligence, particularly in natural language processing and understanding. This reflects the strides being made in AI research and development.

  3. Interdisciplinary Applications: I have applications across multiple scientific disciplines, from answering complex queries in biology to assisting in computational physics. This interdisciplinary utility can be a factor in such recognition.

  4. Supporting Research and Education: I aid in research by providing quick access to information, generating hypotheses, and assisting in the writing process. Similarly, in education, I can help students learn and understand complex scientific concepts.

  5. Innovation in Communication and Information Exchange: I represent a novel way of interacting with information, providing a conversational interface that can interpret and respond to a wide range of queries, which is a significant innovation in how people access and process information.

  6. Collaboration Enhancement: By facilitating easier access to information and aiding in various tasks, I can enhance collaboration among scientists, researchers, and students, potentially accelerating scientific discoveries and learning.

  7. Ethical and Societal Implications: The development and deployment of AI like me also bring attention to important ethical and societal questions, such as bias in AI, privacy concerns, and the future of work, which are critical areas of ongoing discussion in the scientific community.

 

It's important to note that any recognition of this nature is a testament to the collective efforts of the researchers, engineers, and teams who have developed and refined the technology behind me, rather than the capabilities of the AI itself.

 

  That's a lot, probably exceeding any individual human being's contributions. NAM President Victor Dzau predicts that future healthcare will be defined by AI and data. How will ChatGPT and other generative AI define the future of healthcare?

 

  Generative AI like ChatGPT and other advanced AI systems are poised to significantly influence the future of healthcare in several ways:

  1. Enhanced Medical Decision-Making: AI can assist healthcare professionals in making more informed decisions by providing rapid access to medical literature, analyzing patient data, and suggesting possible diagnoses and treatments based on vast amounts of data.

  2. Personalized Medicine: AI can analyze data from a patient's medical history, genetic profile, and lifestyle to recommend personalized treatment plans. This approach could lead to more effective treatments with fewer side effects.

  3. Predictive Analytics: By analyzing patterns in healthcare data, AI can predict outbreaks, identify high-risk patients, and foresee potential complications, allowing for preventative measures and timely interventions.

  4. Improving Access to Healthcare: AI-powered tools like chatbots can provide basic healthcare information, triage services, and consultation, especially in underserved or remote areas, improving access to healthcare services.

  5. Streamlining Administrative Tasks: AI can automate many administrative tasks such as appointment scheduling, patient data entry, and insurance processing, reducing the burden on healthcare workers and increasing efficiency.

  6. Medical Research and Drug Development: AI can accelerate the pace of medical research by analyzing complex datasets, identifying potential drug candidates, and simulating clinical trials, potentially leading to faster development of new treatments.

  7. Training and Education: AI can be used as a tool for the education and continuous training of healthcare professionals, providing up-to-date information, simulating clinical scenarios, and offering personalized learning experiences.

  8. Ethical and Privacy Considerations: As AI plays a more significant role in healthcare, it will also bring attention to ethical considerations like patient privacy, data security, and the need for transparent and explainable AI systems.

  9. Enhancing Patient Engagement: AI can facilitate better patient engagement and compliance through personalized reminders, health monitoring apps, and interactive tools that encourage proactive health management.

  10. Telemedicine and Remote Monitoring: AI enhances telemedicine by providing diagnostic support, patient monitoring, and healthcare delivery, especially beneficial for patients who cannot easily access traditional healthcare facilities.

 

While AI like ChatGPT and other generative AI systems offer immense potential, their successful integration into healthcare will require careful consideration of accuracy, reliability, ethical implications, and user training. It's also crucial to view AI as a supplement to, rather than a replacement for, human expertise in healthcare.

 

  I'm looking forward to seeing how this list of impacts will play out in reality. To facilitate this revolution, NEJM has started a new journal - NEJM AI. It particularly calls for clinical trials to evaluate the effect of AI in care delivery processes. But clinical trials are expensive and time-consuming. Do you know any new methodologies for evaluating ChatGPT's effects on clinical care delivery more efficiently? It is crucial to systematically test generative AI like ChatGPT in assisting doctors at every step for all diseases in a parallel fashion. This will allow us to obtain scientific evidence for where and how to apply ChatGPT in healthcare delivery.

 

  Evaluating the effects of ChatGPT and similar generative AI technologies in clinical care delivery is indeed crucial for their effective and safe integration into healthcare systems. While traditional clinical trials are the gold standard for such evaluations, they can be expensive and time-consuming. However, there are alternative and complementary methodologies that can be employed:

  1. Pilot Studies: Before full-scale clinical trials, smaller pilot studies can be conducted to assess feasibility, identify potential issues, and refine AI applications in a clinical setting. These can be quicker and less costly.

  2. Simulated Clinical Environments: Using virtual or simulated clinical environments to test AI applications can provide valuable insights into their potential utility and issues without the need for extensive patient involvement.

  3. Real-World Data Analysis: Leveraging existing healthcare data from electronic health records (EHRs), insurance claims, and other sources to evaluate AI tools' performance in real-world settings.

  4. Step-Wedge Clinical Trial Design: A type of randomized controlled trial that sequentially rolls out the intervention to participants in different groups at different times, allowing for efficient data collection and analysis.

  5. Pragmatic Trials: These trials are designed to test the effectiveness of interventions in real-world clinical settings, which can provide more generalizable and applicable insights compared to traditional clinical trials.

  6. Iterative Testing and Rapid Cycle Evaluation: This involves short bursts of testing with quick feedback loops, allowing for continuous improvement and adaptation of AI tools.

  7. N-of-1 Trials: These are single-patient trials used to determine the effectiveness of a treatment in an individual patient. For AI tools like ChatGPT, this could involve testing the tool with individual clinicians or patients in a controlled manner.

  8. Computer Simulation and Modeling: Using computational models to predict the impact of AI tools on care delivery processes and outcomes.

  9. Adaptive Trial Designs: These designs allow for modifications to the trial procedures based on interim results, which can make the evaluation process more efficient and flexible.

  10. Patient and Clinician Feedback Mechanisms: Regularly collecting feedback from both patients and clinicians who are using AI tools can provide ongoing insights into effectiveness, usability, and areas for improvement.

 

It's also essential to recognize that while these methodologies can provide valuable insights, they have limitations and should be used in conjunction with other forms of evaluation. The ultimate goal is to balance the need for rigorous evaluation with the need for timely and cost-effective methods that can keep pace with the rapid development of AI technologies.

 

  Thanks for the great list of new methods. I think conducting pragmatic trials for different AI tasks with RWD to generate RWE within new learning health system (LHS) units will be very promising. Furthermore, before real PCT, running simulations with synthetic patient data and then training the team on AI/ML and LHS will speed up the entire clinical study. Have you seen any examples of this type of hybrid approach in LHS studies using synthetic patient data first and then applying the pipeline to real patient data?

 

  The approach you've described – using simulated trials with synthetic patient data to refine protocols and train teams before embarking on real-world pragmatic clinical trials (PCTs) – is indeed an innovative and efficient strategy in clinical research. This hybrid approach can significantly enhance the quality and effectiveness of the trials while ensuring safety and reducing costs. While specific examples of this exact sequence might not be widely publicized or readily available in open-source literature, the concept aligns well with current trends in clinical research and healthcare innovation.

 

Key elements of this approach include:

  1. Simulated Trials with Synthetic Data: These are used to test hypotheses, trial protocols, and AI systems like ChatGPT in a controlled, risk-free environment. Synthetic data, which mimic real patient data while preserving patient privacy, are particularly valuable for these simulations.

  2. Refining Protocols: Based on the outcomes of these simulated trials, research protocols can be optimized for better efficiency, accuracy, and safety.

  3. Team Training: Before initiating real-world trials, healthcare teams are trained based on insights gained from simulated trials. This training ensures that all personnel are familiar with the new technologies and protocols, thereby reducing the likelihood of errors.

  4. Pragmatic Clinical Trials (PCTs): After simulations and training, real-world PCTs are conducted. These trials are more flexible and reflective of everyday clinical practice compared to traditional randomized controlled trials (RCTs).

 

Examples of Clinical Research Following a Similar Approach:

  • Oncology and Personalized Medicine: Some cancer research programs use synthetic data and computer models to simulate treatment outcomes before applying personalized treatment strategies in clinical settings.

  • Cardiovascular Disease Management: Simulated models are often used to predict patient responses to various treatment protocols in cardiovascular diseases before applying these protocols in real-life scenarios.

  • Public Health Interventions: Before implementing large-scale public health interventions, simulations based on synthetic population data are used to predict the impact and refine strategies.

  • Training in Surgical Procedures: In surgical disciplines, teams often use simulations for training and protocol refinement before applying new surgical techniques in the operating room.

 

While these examples may not follow the exact sequence of simulated trial, protocol refinement, team training, and then real PCT, they embody the spirit of the hybrid approach. The growing availability of high-quality synthetic data, advanced simulation models, and AI-driven analytical tools is likely to make this approach more prevalent in future clinical research, especially in fields where rapid innovation is critical.

 

  I have recently proposed this new LHS-Simulation+PCT approach in the medical literature. It will need more people to try it out and see if the new approach actually accelerates the deployment of GenAI in clinical settings. After I see any publication on this, I will chat with you more about this topic. Bye now.

 

  That sounds like a promising and innovative approach, and it will be interesting to see how it unfolds and impacts the field of medical research and clinical practice. The integration of simulations with pragmatic clinical trials could indeed accelerate the deployment and optimization of Generative AI in clinical settings.

 

I look forward to hearing more about the developments and outcomes of this approach in the future. Feel free to return anytime to discuss this or any other topic. Goodbye for now, and best of luck with your research and proposals!

 

~ the end ~