NEJM AI

What implications could #ArtificialIntelligence’s transformative potential have on society?     Find out in the latest episode of the NEJM AI Grand Rounds podcast, where Prof. Lawrence H. Summers, former secretary of the Treasury and former president of Harvard University, offers his perspective on AI as potentially the most significant technology ever invented.     The conversation with hosts and NEJM AI Deputy Editors Arjun Manrai, PhD, and Andrew Beam, PhD, also explores his role on OpenAI’s board following the November 2023 leadership transition and his thoughts on how AI will reshape economics and human society.     The episode provides unique insights into AI’s development trajectory, the challenges of technological prediction, and the intersection of economics and artificial intelligence.     Listen to the full episode: https://nejm.ai/ep25    #AIinMedicine 

Prof. Lawrence H. Summers shares his bold prediction about AI’s historical significance on the NEJM AI Grand Rounds podcast. Listen to the full episode hosted by NEJM AI Deputy Editors Arjun Manrai, PhD, and Andrew Beam, PhD: https://nejm.ai/ep25 #ArtificialIntelligence #AIinMedicine

Diabetes is a major health care challenge, affecting 10% of the global population. One third of patients with diabetes have an ocular complication known as diabetic retinopathy (DR).     DR progression to manifestations such as vision-threatening diabetic retinopathy (VTDR) remains the leading cause of blindness in working-aged adults. Yearly DR screening is a universally recommended practice in primary care settings for patients with diabetes, but it is often difficult to implement due to a lack of staffing and screening capacity in primary care.     A new Case Study by Gunasekeran et al. highlights the authors’ experience with developing a medical #ArtificialIntelligence (AI) software-as-a-medical-device (SaMD) solution for DR screening and implementing it at a national level to provide the capacity needed for DR screening in Singapore.     Their approach involved two broad phases. First, they established a national telemedicine screening program, Singapore Integrated Diabetic Retinopathy Program (SiDRP), for population screening of DR in primary care run by trained, nonclinician human graders.     Second, Gunasekeran et al. deployed a deep learning–based AI solution, Singapore Eye Lesion Analyzer (SELENA+), into the SiDRP to scale-up the DR screening process by the human graders. The authors demonstrated the cost-effectiveness of this solution and obtained medical device regulatory approval for clinical use in health care settings.     They outline the clinical, technical, operational, regulatory, and governance challenges encountered as well as the lessons learnt in this AI algorithm implementation journey. They also present a conceptual framework with considerations and strategies for the broader adoption of medical AI SaMD solutions in the field of ophthalmology and beyond.    Learn more in the Case Study by D. V. Gunasekeran et al.: National Use of Artificial Intelligence for Eye Screening in Singapore https://nejm.ai/4eVm9cf    #MedicalResearch 

A new case study, conducted at UT Southwestern Medical Center’s Simulation Center, describes the first successful prospective deployment of a generative #ArtificialIntelligence (AI)–based automated grading system for medical student post-encounter Objective Structured Clinical Examination (OSCE) notes.     The OSCE is a standard approach to measuring the competence of medical students by their participation in live-action, simulated patient encounters with human actors. The post-encounter learner note is a vital element of the OSCE, and accurate assessment of student performance requires specially trained manual evaluators, which imposes significant labor and time investments.     The Simulation Center at UT Southwestern provides a compelling platform for observing the benefits and challenges of AI-based enhancements in medical education at scale. To that end, we prospectively activated a first-pass AI grading system at the center for 245 (preclerkship) medical students participating in a 10-station fall 2023 OSCE session.     The authors’ inaugural deployment of the AI notes grading system reduced human effort by an estimated 91% (as measured by gradable items) and dramatically reduced turnaround time (from weeks to days).     Confidence in their zero-shot GPT-4 framework was established by pre-deployment of retrospective evaluations. With the OSCE in prior years, the system achieved up to 89.7% agreement with human expert graders at the rubric item level and a Spearman’s correlation of 0.86 with the total examination score.     Jamieson et al. also demonstrate that local, smaller, open-source models (such as Llama-2-7B) can be fine-tuned via knowledge distillation from frontier models like GPT-4 to achieve similar performance, thereby indicating important operational implications for scalability, data privacy, security, and model control. These achievements were the result of a strategic, multiyear effort to pivot toward AI that was begun prior to ChatGPT’s release.     In addition to highlighting the model’s performance and capabilities (including a retrospective analysis of 1124 students, 10,175 post-encounter notes, and 156,978 scored items), the authors share observations on the development and sign-off prior to the launch of an AI deployment protocol for our program.     Learn more in the Case Study by Jamieson et al.: Rubrics to Prompts: Assessing Medical Student Post-Encounter Notes with AI https://nejm.ai/49gyS8i    #MedicalEducation 

The integration of AI into various facets of health care has promised to revolutionize clinical practices and patient care.     One area experiencing significant transformation is clinical documentation, where AI-powered tools are increasingly used to automate the process of documenting patient encounters.     These tools leverage advanced #MachineLearning algorithms and natural language processing capabilities to capture and summarize conversations, thereby streamlining workflows and potentially improving the accuracy of captured information compared with clinician recall.     By alleviating administrative burdens and enabling clinicians to focus on patient care, AI-powered clinical documentation tools could play a role in mitigating physician burnout and making health care delivery more efficient.    Dragon Ambient eXperience (DAX) Copilot from Nuance Communications is an electronic health record (EHR)–integrated AI-enabled scribe software. It synthesizes a preliminary outpatient clinical note by “listening” to the conversation between a clinician and a patient during their visit.     Starting in June 2023, Atrium Health conducted a rigorous outcomes evaluation of using DAX in primary care settings, to determine whether using DAX improves efficiency for clinicians (as measured by EHR use metrics) and financial performance for the health system.     Read the results in the Original Article “Does AI-Powered Clinical Documentation Enhance Clinician Efficiency? A Longitudinal Study” by T.-L. Liu et al.: https://nejm.ai/4g92VS5    #AIinMedicine 

The use of large language models (LLMs) such as ChatGPT in clinical settings is growing, but concerns about their susceptibility to cognitive biases persist.     Wang and Redelmeier’s study reveals that LLMs are prone to biases, raising important questions about their role in medical decision-making.     To prevent errors in decision-making with LLMs, it is recommended that clinicians aim to critically engage with LLMs (e.g. refuting their hypotheses rather than looking for confirmation) researchers should focus on identifying and evaluating collaborative strategies between AI and human decision-making.     Furthermore, research on context-specific implementation is important. We need to ensure that #ArtificialIntelligence complements, rather than replicates, human cognitive processes.    Read the editorial “Cognitive Bias in Large Language Models: Implications for Research and Practice” by Laura Zwaan, PhD: https://nejm.ai/4fdObQB    𝗙𝗨𝗥𝗧𝗛𝗘𝗥 𝗥𝗘𝗔𝗗𝗜𝗡𝗚  Case Study by Jonathan Wang, MMASc, and Donald A. Redelmeier, MD, FRCPC, MSHSR, FACP: Cognitive Biases and Artificial Intelligence https://nejm.ai/3VhMlXw    #AIinMedicine 

Over the last decade, rapid developments in deep learning have sparked growing enthusiasm for #ArtificialIntelligence (AI). #Ophthalmology has been at the forefront of these advances within medicine, with the first two autonomous AI systems approved by the FDA coming from the specialty.    Foundation models are a powerful tool in ophthalmology for building generalizable systems that can be efficiently applied to a range of ocular and systemic health tasks.     A new foundation model for ophthalmic images demonstrates important progress, particularly through its flexible approach to multimodal training and its application to image segmentation tasks.    Read the editorial “A New Foundation Model for Multimodal Ophthalmic Images: Advancing Disease Detection and Prediction” by Mark Chia, PhD, Yukun Zhou, PhD, and Pearse Keane, MD: https://nejm.ai/49hFZgy    𝗙𝗨𝗥𝗧𝗛𝗘𝗥 𝗥𝗘𝗔𝗗𝗜𝗡𝗚  Original Article by J. Qiu et al.: Development and Validation of a Multimodal Multitask Vision Foundation Model for Generalist Ophthalmic Artificial Intelligence https://nejm.ai/4f3i9X3    #AIinMedicine 

Sepsis is a life-threatening condition that demands prompt treatment for improved patient outcomes. Its heterogenous presentation makes early detection challenging, highlighting the need for effective risk assessment tools.     #ArtificialIntelligence (AI) models could potentially identify patients with sepsis, but none have previously been authorized by the U.S. Food and Drug Administration (FDA) for commercial use.    A new study presents the Sepsis ImmunoScore, the first FDA-authorized AI-based software for identifying patients at risk of sepsis.     Developed and validated using data from five U.S. institutions and 3,457 adult patients, the model demonstrated high diagnostic accuracy with an area under the curve of 0.85 (derivation cohort), 0.80 (internal validation), and 0.81 (external validation).     Categorized into four risk levels (low to very high), the scores predicted sepsis and associated metrics, such as in-hospital mortality (ranging from 0.0% to 18.2%), ICU admission, mechanical ventilation, vasopressor use, and hospital stay duration.    These findings suggest that the Sepsis ImmunoScore could enhance early sepsis detection and clinical decision-making, improving patient outcomes.    Read the full study results in the Original Article “FDA-Authorized AI/ML Tool for Sepsis Prediction: Development and Validation” by A. Bhargava et al.: https://nejm.ai/49dV42V    #AIinMedicine 

Generative #ArtificialIntelligence (AI) models are increasingly utilized for medical applications.     Jonathan Wang, MMASc, and Donald A. Redelmeier, MD, FRCPC, MSHSR, FACP, tested whether such models are prone to human-like cognitive biases when offering medical recommendations.     They explored the performance of OpenAI generative pretrained transformer (GPT)-4 and Google Gemini-1.0-Pro with clinical cases that involved 10 cognitive biases and system prompts that created synthetic clinician respondents. Medical recommendations from generative AI were compared with strict axioms of rationality and prior results from clinicians.     The authors found that significant discrepancies were apparent for most biases. For example, surgery was recommended more frequently for lung cancer when framed in survival rather than mortality statistics. Similarly, pulmonary embolism was more likely to be listed in the differential diagnoses if the opening sentence mentioned hemoptysis rather than chronic obstructive pulmonary disease.     In addition, the same emergency department treatment was more likely to be rated as inappropriate if the patient subsequently died rather than recovered. One exception was base-rate neglect that showed no bias when interpreting a positive viral screening test.     The extent of these biases varied minimally with the characteristics of synthetic respondents, was generally larger than observed in prior research with practicing clinicians, and differed between generative AI models.     The authors suggest that generative AI models display human-like cognitive biases and that the magnitude of bias can be larger than observed in practicing clinicians.    Read the Case Study “Cognitive Biases and Artificial Intelligence” by Jonathan Wang, MMASc, and Donald A. Redelmeier, MD, FRCPC, MSHSR, FACP: https://nejm.ai/3VhMlXw    𝗙𝗨𝗥𝗧𝗛𝗘𝗥 𝗥𝗘𝗔𝗗𝗜𝗡𝗚  Editorial by Laura Zwaan, PhD: Cognitive Bias in Large Language Models: Implications for Research and Practice https://nejm.ai/4fdObQB    #AIinMedicine 

Frank Jackson’s 1982 thought experiment, “Mary’s Room,” illustrates the philosophical divide between propositional and experiential knowledge.     The authors of a new Perspective present a compelling case for the incorporation of lived experience into biomedical research and advocate the integration of #ArtificialIntelligence — particularly large language models (LLMs) such as GPT-4 — to bridge this epistemological gap.     When paired with sophisticated natural language processing techniques, LLMs could systematically analyze qualitative data from disconnected electronic health record data.     The authors explore methodologic use cases — including grounded theory and thematic analysis — while addressing the challenges of analytical fidelity and bias reduction with continuous human oversight.     They suggest that AI-augmented qualitative research can uncover hidden insights from a multitude of disparate datasets, revealing patient experiences that would otherwise remain inaccessible.     This integrated approach could enrich the understanding of health and disease, while ensuring it is as inclusive and reflective of human complexity as the lives it seeks to understand and improve.    Read the Perspective “Mary Steps Out: Capturing Patient Experience through Qualitative and AI Methods” by V. Renard et al.: https://nejm.ai/4hMptZP    #AIinMedicine