Paving the Way for Global Health Equity with AI, ML, Data, and LHS (Learning Health Systems)
Google's new chatbot takes patient history and suggests potential diagnoses.
University of California Health System transforms big data into better care for everyone.
The future of AI-augmented clinical trials, from RWD to RWE in action.
Published papers, recent news, and significant events in a coherent narrative for the main topic.
[2024/1] According to Nature news, Google Health’s new medical chatbot, Articulate Medical Intelligence Explorer (AMIE) trained to conduct medical interviews, matched, or even surpassed, human doctors’ performance at conversing with simulated patients and listing possible diagnoses on the basis of the patients’ medical history. “This in no way means that a language model is better than doctors in taking clinical history,” says a co-author. But researchers say their artificial-intelligence system could help to democratize medicine.
[2024/1] JAMA Chief Editor’s interview with Dr. Atul Butte of UCSF, chief data scientist over the entire University of California Health System. University of California Health System has 10 hospitals, over 9 million patients. Butte said: We’re trying to get from real-world data to real-world evidence. This is not the randomized controlled trial. This is what happens when we actually use these devices, these drugs, these biologics in real patients.
[2023/12] In this Q&A, JAMA Editor in Chief interviews Dr. John Ayers at UC San Diego to discuss how GenAI programs like ChatGPT can increase communication pathways and improve patient outcomes.
[2023/12] FDA issues a final guidance that provides industry, investigators and others recommendations on the use of digital health technologies (DHTs) to acquire data remotely from participants in clinical investigations that evaluate medical products.
Hernandez AF, Lindsell CJ. The Future of Clinical Trials: Artificial to Augmented to Applied Intelligence. JAMA. November 11, 2023. doi:10.1001/jama.2023.23822
[2023/11] This Editorial paper highlights the findings of a new article in JAMA Cardiology that examines the potential for natural language processing (NLP) models to accurately classify heart failure hospitalizations compared with the gold standard for clinical adjudication and places the findings from this study in context of the possibilities for improving clinical trials through AI.
GPTs are becoming an integral part of our society. AJ talks with GenAI to understand how machines view the main topics.
Google reported using a new GenAI chatbot trained for diagnostic dialogue to take clinical history from patient and suggest the possible diagnoses. This is a big step forward to realize the vision of AI-augmented virtual doctor. Could you summarize the existing research work in designing and testing AI-augmented virtual doctors?
The development of AI-augmented virtual doctors, especially in the context of diagnostic dialogue and clinical history-taking, is an area of active research and innovation. Here's a summary of the existing research work in this field:
Natural Language Processing (NLP) and Understanding: A key area of research involves using advanced NLP techniques to enable AI systems to understand and interpret patient language, including symptoms, medical history, and concerns. This involves training AI on vast datasets of medical dialogues and literature.
Diagnostic Algorithms: Research is focused on developing algorithms that can suggest possible diagnoses based on the clinical information provided by patients. These algorithms often employ machine learning techniques and are trained on large datasets of clinical cases and outcomes.
Clinical Decision Support Systems (CDSS): These systems are designed to assist healthcare professionals in making decisions. Research in AI-augmented CDSS includes integrating AI tools for diagnosis, treatment recommendations, and predicting patient outcomes.
User Interface and Experience: Research is also being conducted on how to design user interfaces for virtual doctor systems that are user-friendly, accessible, and efficient for both patients and healthcare providers. This includes voice recognition, chat interfaces, and integration with healthcare systems.
Ethical and Privacy Considerations: This area of research addresses the ethical implications of AI in healthcare, patient privacy, data security, and the need for transparent AI decision-making processes.
Clinical Trials and Validation: Several studies and trials are being conducted to validate the effectiveness and accuracy of AI-augmented virtual doctors. These studies compare AI-generated diagnoses and treatment recommendations with those of human doctors.
Regulatory Compliance: Research is also focused on ensuring that AI systems comply with healthcare regulations and standards, including FDA approvals where necessary.
Integration with Healthcare Systems: There is ongoing research on how to effectively integrate AI tools into existing healthcare infrastructures, including Electronic Health Records (EHRs) and telemedicine platforms.
Personalization and Adaptation: AI systems are being designed to adapt and personalize their interactions based on individual patient data, history, and preferences, to provide more tailored healthcare advice.
Overall, while significant progress has been made in the development of AI-augmented virtual doctors, the field is still evolving with ongoing research addressing various technological, ethical, and practical challenges.
Do you think GenAI can accelerate this field dramatically because LLM already has all the AI capabilities required? It is a matter of designing an end-to-end system and demonstrating its effectiveness in a specific clinical setting.
The integration of Generative AI (GenAI) models like large language models (LLMs) has the potential to significantly accelerate the field of AI-augmented virtual doctors. However, it's important to recognize the complexities and specific requirements of the medical field. Here are a few considerations:
Advanced Understanding of Medical Context: LLMs already possess sophisticated natural language processing capabilities, which can be crucial in understanding patient dialogues, medical histories, and symptom descriptions. This foundational ability can be a significant accelerator.
Data Sensitivity and Privacy: Medical applications require stringent adherence to data privacy and security standards like HIPAA. LLMs must be adapted to comply with these regulations, ensuring patient data is handled securely.
Clinical Accuracy and Reliability: While LLMs are powerful, ensuring clinical accuracy in diagnoses and recommendations is paramount. This involves not just the AI's understanding of medical knowledge but also its ability to apply this knowledge correctly in diverse clinical scenarios.
Integration with Healthcare Systems: Designing an end-to-end system involves seamless integration with existing healthcare infrastructures, like Electronic Health Records (EHRs), which can be complex. LLMs can facilitate this integration by processing and synthesizing information from these systems.
Customization for Clinical Settings: Different clinical settings have varied requirements. LLMs offer the flexibility to be customized for specific medical specialties or patient demographics.
Validation and Testing: Any AI system in healthcare must undergo rigorous testing and validation to ensure its effectiveness and safety. This includes clinical trials and validation against current diagnostic and treatment standards.
Ethical and Regulatory Considerations: The deployment of AI in healthcare raises ethical questions about decision-making, accountability, and patient consent. Regulatory approval processes, especially for diagnostic tools, are also stringent.
In conclusion, while GenAI and LLMs provide a strong foundation with their existing AI capabilities, the development of an effective AI-augmented virtual doctor system requires careful consideration of medical accuracy, ethical and regulatory compliance, integration challenges, and thorough validation. These aspects go beyond the AI's core capabilities and involve a holistic approach to system design and implementation in healthcare.
Please list the top research labs in this field?
As of my last update in April 2023, several leading research labs and institutions are at the forefront of AI-augmented virtual doctor and medical AI research. Here's a list of some of the top players in this field:
Stanford University's Artificial Intelligence Lab: Known for its cutting-edge research in AI, Stanford's AI Lab conducts extensive research in medical AI, including applications in diagnostics and patient care.
MIT Computer Science and Artificial Intelligence Laboratory (CSAIL): MIT CSAIL is a hub for AI research, including medical AI, focusing on areas like machine learning, natural language processing, and data science.
Harvard Medical School's Computational Health Informatics Program (CHIP): CHIP at Harvard is known for its research in applying AI and machine learning to healthcare data.
IBM Watson Health: Known for its AI capabilities, IBM Watson Health is involved in research and development of AI applications in healthcare, including virtual health assistants.
Google Health and DeepMind: Google Health and its sister company DeepMind are actively involved in healthcare-related AI research, including the development of AI systems for disease detection and diagnosis.
Allen Institute for AI (AI2): AI2, founded by Paul Allen, conducts research in various AI fields, including AI applications in medicine and healthcare.
Johns Hopkins University's Malone Center for Engineering in Healthcare: This center focuses on developing innovative healthcare solutions using AI and engineering principles.
University of Toronto's Vector Institute: The Vector Institute specializes in machine learning and AI research, with a focus on healthcare applications.
UC Berkeley Artificial Intelligence Research Lab (BAIR): BAIR conducts broad AI research, including healthcare applications, machine learning, and robotics.
Microsoft Research Healthcare: Microsoft's research division is involved in developing AI technologies for healthcare, focusing on areas like medical imaging and genomics.
These institutions are known for their pioneering research and often collaborate with hospitals, tech companies, and other research organizations. Additionally, many other universities and private research labs worldwide are making significant contributions to this field.
Big health systems like UC Health System have millions of patients and billions of clinical event data. These real-world data have potential to be converted to real-world evidence. Please summarize some major studies of this approach in literature.
Utilizing real-world data (RWD) from large health systems like the UC Health System to create real-world evidence (RWE) has become an increasingly important approach in medical research. Here are summaries of some major studies and approaches in this area:
Comparative Effectiveness Research: Studies have used RWD to compare the effectiveness of different treatments. For example, a study might use patient data from electronic health records (EHRs) to compare the outcomes of two cancer treatments, providing valuable insights into which treatment works better in real-world settings.
Drug Safety and Pharmacovigilance: Many studies have utilized RWD for monitoring drug safety post-marketing. By analyzing patient data, researchers can identify adverse drug reactions and interactions that might not have been apparent in clinical trials.
Patient-Centered Outcomes Research: Studies have focused on understanding patient experiences and outcomes using RWD. This includes analyzing patient-reported outcomes and health-related quality of life data to inform healthcare practices.
Healthcare Utilization and Cost Analysis: RWD from large health systems has been used to study patterns in healthcare utilization and costs. These studies help in identifying areas where healthcare delivery can be more efficient and cost-effective.
Predictive Modeling: RWD has been employed to develop predictive models for various diseases. For instance, using data from EHRs, researchers have developed models to predict the likelihood of patients developing conditions like diabetes or heart disease.
Personalized Medicine Studies: Leveraging RWD, researchers have been able to study how genetic and environmental factors influence drug responses and disease progression, aiding in the advancement of personalized medicine.
Chronic Disease Management: There have been studies using RWD to evaluate and improve chronic disease management strategies. For example, analyzing patient data to identify effective interventions for managing conditions like hypertension or asthma.
Epidemiological Studies: RWD is a valuable resource for conducting epidemiological studies, such as determining the incidence and prevalence of diseases within a population and identifying risk factors.
These studies highlight the potential of RWD in enhancing our understanding of diseases, treatments, and patient outcomes in real-world settings. However, it is crucial to address challenges related to data quality, privacy, and standardization to maximize the value of RWD in research.
Is there any observational study that was automatically designed by AI and conducted using EHR records?
As of my last update in April 2023, the field of AI-driven observational study design using electronic health records (EHRs) was still emerging, and there were limited instances of fully automated study designs. However, there were significant advancements and explorations in this area. Here are some key points regarding the use of AI in observational study design:
AI-Assisted Study Design: While there might not have been fully automated AI-designed studies, AI and machine learning tools have been increasingly used to assist researchers in designing observational studies. These tools can help identify relevant patient cohorts, suggest potential hypotheses based on data patterns, and assist in determining appropriate study parameters.
Data Mining and Pattern Recognition: AI algorithms, particularly in machine learning and data mining, have been used to analyze EHRs for identifying patterns and correlations that could inform observational studies. These AI tools can process vast amounts of data to uncover trends that might not be immediately apparent to human researchers.
Predictive Modeling: AI has played a significant role in developing predictive models using EHR data. These models often form the basis for observational studies, especially in identifying risk factors for diseases or predicting patient outcomes.
Natural Language Processing (NLP): NLP has been used to extract and analyze unstructured data from EHRs, such as doctor's notes and patient histories, which are crucial for observational study design. NLP can automate the process of turning this unstructured data into usable information for research.
Ethical and Privacy Considerations: Any AI-driven research using EHRs must adhere to strict privacy and ethical guidelines. This includes ensuring patient confidentiality and complying with regulations like HIPAA in the United States.
Integration Challenges: One of the challenges in fully automating observational study design is the integration of AI tools with diverse EHR systems, which often have varying formats and standards.
In summary, while fully automated AI-driven observational studies using EHRs were not commonplace as of early 2023, AI technologies were increasingly being integrated into the research process, enhancing the efficiency and capabilities of observational study design in healthcare. The field is rapidly evolving, and more automated approaches may emerge as AI technology and healthcare data systems continue to advance.
When I am putting together the latest studies into this newsletter, I see the possibility of collecting data and delivering medical decisions automatically by chatbot while EHR is automatically conducting observational studies and generating evidence. Connecting these two components can potentially result in an autonomous healthcare unit. Do you know what I envision?
Yes, I understand your vision. You're envisioning a highly integrated, autonomous healthcare system that leverages advanced technology to both deliver medical care and continuously improve its delivery methods. Here's a breakdown of the two key components in your vision:
Data Collection and Medical Decision-Making via Chatbots: In this component, AI-powered chatbots interact with patients to collect health data and provide medical decisions. These chatbots use sophisticated natural language processing (NLP) to understand patient queries, symptoms, and medical history. They apply medical knowledge, possibly augmented by machine learning algorithms, to offer diagnostic suggestions, treatment recommendations, or health advice. The chatbots could also play a role in ongoing patient monitoring and follow-up care.
EHR-Driven Observational Studies and Evidence Generation: Parallel to the chatbot interactions, an electronic health record (EHR) system continuously conducts observational studies. It analyzes the vast amounts of data generated through healthcare interactions, including those from the chatbots. Utilizing AI and machine learning, it identifies patterns, evaluates treatment outcomes, and generates real-world evidence. This evidence can then be used to update and refine medical guidelines, treatment protocols, and the AI algorithms themselves.
Connecting these two components would create a feedback loop where:
The chatbot provides personalized healthcare based on current medical evidence and patient-specific data.
The EHR system continuously analyzes outcomes and generates new insights.
These new insights are fed back into the chatbot, constantly improving its accuracy and effectiveness.
Such an autonomous healthcare unit could potentially revolutionize healthcare delivery, making it more efficient, evidence-based, and accessible. It would represent a significant leap in telemedicine and AI in healthcare, offering real-time, data-driven medical care and continuously evolving based on the latest clinical evidence.
However, it's important to note that while this vision is technologically feasible, its realization would require careful consideration of ethical, regulatory, and practical aspects, including patient privacy, data security, the accuracy of AI decisions, and the need for human oversight in healthcare.
You are good. I want to emphasize that, while the ultimate goal is autonomous health system, the mid-term goal should be autonomous units of some complete health systems. For instance, after my recent study of ML-enabled Learning Health System (LHS) Unit that was published in Nature Sci. Rep., I have proposed a hypothesis of ML-LHSU approach for building equitable learning health systems. Given the recent emergence of generative AI, it seems more feasible to develop autonomous ML-LHSU for certain clinical tasks. What do you think?
Your vision of developing autonomous Machine Learning-enabled Learning Health System Units (ML-LHSUs) for specific clinical tasks is both forward-thinking and timely, especially with the advancements in generative AI. Here are some considerations and potential impacts of this approach:
Targeted Clinical Tasks: Focusing on specific clinical tasks is a strategic approach. It allows for the development of highly specialized ML-LHSUs that can address distinct challenges or opportunities within a healthcare system. For example, an ML-LHSU might focus on chronic disease management, post-operative care, or early disease detection.
Generative AI Advancements: The emergence of generative AI can significantly enhance the capabilities of ML-LHSUs. Generative models can create realistic and diverse medical scenarios for training purposes, generate patient education materials, and even aid in simulating patient responses to various treatments. This can lead to more robust and adaptive learning systems.
Enhanced Personalization and Equity: One of the key goals of an LHS is to provide personalized care that adapts to individual patient needs. ML-LHSUs can leverage patient data to tailor treatments and interventions, potentially reducing healthcare disparities and promoting equity.
Real-Time Data Utilization: ML-LHSUs can continuously learn from real-time data streams, such as EHRs, wearable health devices, and patient-reported outcomes. This ongoing learning process can help healthcare systems rapidly adapt to new information and changing patient needs.
Scalability and Integration: Developing autonomous units allows for scalability. Each unit can serve as a building block for a larger, more comprehensive autonomous health system. Integration with existing healthcare infrastructures will be crucial for the success of ML-LHSUs.
Ethical and Regulatory Considerations: As with any AI application in healthcare, developing ML-LHSUs will require careful attention to ethics, privacy, and regulatory compliance. Ensuring transparency, accountability, and patient safety in these systems will be paramount.
Human-Centric Design: While aiming for autonomy, it's important to maintain a human-centric approach. This includes ensuring that healthcare professionals are involved in the oversight of ML-LHSUs and that patient preferences and values are respected.
Your hypothesis about building equitable learning health systems through ML-LHSUs is not only feasible with current technological advancements but also necessary for the evolution of healthcare. It represents a significant step towards more adaptive, efficient, and patient-centered healthcare delivery.
~ the end ~
Join us in building Learning Health System units empowered by AI/ML to move towards global health equity.
Palo Alto, California, USA