Publications

Abstract

Background: Safe implementation of autonomous AI in medicine requires rigorous evaluation through clinical trials. The 7 guiding principles for ethical clinical research endorsed by the National Institutes of Health (NIH) provide an established framework for promoting scientific rigor and protecting the safety of human participants. However, clinical trials of autonomous AI raise novel ethical issues that require adaptation of these principles to account for effects that can vary across stakeholders and implementation contexts, including model performance across clinical settings. Incorporating expert perspectives on such challenges is critical to developing effective and ethically robust guidelines for autonomous AI clinical trials.Objective: This Delphi study aimed to generate expert consensus on how the National Institutes of Health's 7 principles of ethical clinical research should be applied to clinical trials of autonomous AI.Methods: We conducted 2 rounds of surveys followed by a final virtual meeting using a modified Delphi approach with a multidisciplinary expert panel. Participants were purposively identified through PubMed literature searches and selected for expertise in AI, data science, ophthalmology, public policy, law, bioethics, and patient advocacy. Round 1 used open-ended survey questions based on a vignette describing an autonomous AI tool. Round 1 survey responses were analyzed qualitatively using thematic coding, and were used to generate representative statements for Round 2. In round 2, panelists rated statements on a 5-point Likert scale. In accordance with Delphi methodology, statements of consensus within the surveys (at least 80% rating agreement) and moderate agreement (60%-80% rating agreement) were identified for further discussion and iteration. Findings from a final virtual meeting were then analyzed thematically and synthesized into actionable recommendations.Results: Fourteen expert panelists participated in the Delphi study over a 6-month period. Participation was 12 (85.7%) of 14 experts in round 1, 10 (71.4%) of 14 experts in round 2, and 13 (92.9%) of 14 experts in the final virtual meeting. Round 2 survey results yielded 9 strong agreement statements, 2 moderate agreement statements, and 4 divisive statements for participants to explore further in developing recommendations. Final recommendations from the virtual meeting addressed transparency regarding training and validation data, bias assessment before deployment, performance across clinical settings, health inequities before implementation, stakeholder engagement, informed consent, comparison of AI tools with existing standards of care, downstream access to care after AI-generated recommendations, and cost and access implications.Conclusions: Ethical evaluation of autonomous AI clinical trials should extend beyond technical accuracy to help mitigate potential harms to patients. This study highlights key ethical considerations for clinical trials of autonomous AI and provides consensus recommendations from a multidisciplinary Delphi panel. These recommendations can inform future research, policy, and guidance for the ethical development and implementation of autonomous AI clinical trials.

View details for DOI 10.2196/91472

View details for PubMedID 42627957

Abstract

This study assessed over 2000 patient perspectives on the use of ambient AI scribes in outpatient visits.This prospective quality improvement study was conducted at Stanford Health Care between May and July 2025. Outcome measures included patient perceived helpfulness of ambient AI scribes and patient interest in future use.Among 2202 survey respondents, 70.1% patients found the ambient AI scribe helpful and 73.6% preferred future use of ambient AI scribes. Small but statistically significant differences were observed across gender, age, and race.In an evaluation of patient perceptions of an ambient AI scribe integrated into clinical practice, the majority of patients who had experienced the technology found it acceptable, and most found the tool helpful and desired use with future visits.These results suggest that ambient AI scribes may be acceptable to many patients in clinical practice. Further research is needed to inform patient-centered design and workflow improvements.

View details for DOI 10.1093/jamiaopen/ooag104

View details for PubMedID 42290933

View details for PubMedCentralID PMC13264434

Abstract

As healthcare organizations' use of generative AI moves from initial experimentation to scaled deployments, the need to build oversight systems to identify and address ethical challenges assumes greater urgency. Among the AI applications attracting the strongest early interest are clinical summarization tools, which use large language models (LLMs). To assist healthcare organizations weighing adoption of LLMs, we describe an ethical assessment process employed at our healthcare system to identify problems that may affect patient care so problems can be addressed prior to deployment or monitored over time to detect harms. The process uses stakeholder interviewing to explore risks and other concerns arising from integration of AI tools into clinical workflow and identify areas where values and priorities of different stakeholder groups do not align. We describe ethical issues identified in assessments of tools that (1) draft end-of-shift nursing notes and (2) generate clinical notes from conversations between clinicians and patients.

View details for DOI 10.1080/15265161.2026.2657875

View details for PubMedID 42089762

Abstract

PURPOSE: In recent years, researchers have brought attention to the underrepresentation of people with disabilities in biomedical research, including genomics research. However, little is known about how disability-related experiences influence participation in rare disease research. This omission is striking, because rare diseases are associated with disabling phenotypes that affect multiple body systems. As part of a study interrogating the relationship between rare disease status and disability identity, we conducted mixed-methods research to address this knowledge gap.METHODS: Parents of children enrolled in the UDN (n=25) completed semi-structured interviews to assess disability-related experiences in research participation. Directed content analysis was used to identify common themes.RESULTS: Participants' disability-related research experiences were characterized by: 1) disability-related facilitators to research participation, including benefits of research participation and disability-conscious approaches; 2) disability-related logistical barriers to research participation; and 3) research procedures, and the perception of research as minimally burdensome relative to clinical encounters.CONCLUSION: Parents of children in the UDN make considerable investments of time and resources to accommodate their children's disabilities to facilitate their participation. Future research should explore these issues in other genomic research studies and practical approaches to mitigating barriers and employing facilitators to disability-related research participation.

View details for DOI 10.1016/j.gim.2026.102663

View details for PubMedID 42439107

Abstract

Machine learning predictive analytics (MLPA) are utilized increasingly in health care, but can pose harms to patients, clinicians, health systems, and the public. The dynamic nature of this technology creates unique challenges to evaluating safety and efficacy and minimizing harms. In response, regulators have proposed an approach that would shift more responsibility to MLPA developers for mitigating potential harms. To be effective, this approach requires MLPA developers to recognize, accept, and act on responsibility for mitigating harms. In interviews of 40 MLPA developers of health care applications in the United States, we found that a subset of ML developers made statements reflecting moral disengagement, representing several different potential rationales that could create distance between personal accountability and harms. However, we also found a different subset of ML developers who expressed recognition of their role in creating potential hazards, the moral weight of their design decisions, and a sense of responsibility for mitigating harms. We also found evidence of moral conflict and uncertainty about responsibility for averting harms as an individual developer working in a company. These findings suggest possible facilitators and barriers to the development of ethical ML that could act through encouragement of moral engagement or discouragement of moral disengagement. Regulatory approaches that depend on the ability of ML developers to recognize, accept, and act on responsibility for mitigating harms might have limited success without education and guidance for ML developers about the extent of their responsibilities and how to implement them.

View details for PubMedID 36541003

Abstract

Regulatory agencies need to ensure the safety and equity of AI in biomedicine, and the time to do so is now.

View details for DOI 10.1126/scitranslmed.adi0336

View details for PubMedID 37703349

Abstract

Stepped-wedge cluster randomized trial (SW-CRT) designs are increasingly employed in pragmatic research; they differ from traditional parallel cluster randomized trials in which an intervention is delivered to a subset of clusters, but not to all. In a SW-CRT, all clusters receive the intervention under investigation by the end of the study. This approach is thought to avoid ethical concerns about the denial of a desired intervention to participants in control groups. Such concerns have been cited in the literature as a primary motivation for choosing SW-CRT design, however SW-CRTs raise additional ethical concerns related to the delayed implementation of an intervention and consent. Yet, PCT investigators may choose SW-CRT designs simply because they are concerned that other study designs are infeasible. In this paper, we examine justifications for the use of SW-CRT study design, over other designs, by drawing on the experience of the National Institutes of Health's Health Care Systems Research Collaboratory (NIH Collaboratory) with five pragmatic SW-CRTs. We found that decisions to use SW-CRT design were justified by practical and epistemic reasons rather than ethical ones. These include concerns about feasibility, the heterogeneity of cluster characteristics, and the desire for simultaneous clinical evaluation and implementation. In this paper we compare the potential benefits of SW-CRTs against the ethical and epistemic challenges brought forth by the design and suggest that the choice of SW-CRT design must balance epistemic, feasibility and ethical justifications. Moreover, given their complexity, such studies need rigorous and informed ethical oversight.

View details for DOI 10.1016/j.cct.2022.106703

View details for PubMedID 35176501

View details for DOI 10.1038/d41586-026-00090-1

View details for Web of Science ID 001666689000013

View details for PubMedID 41559225

Abstract

The rapid advancement of artificial intelligence (AI) in biomedical research presents considerable potential for misuse, including authoritarian surveillance, data misuse, bioweapon development, increase in inequity and abuse of privacy. We propose a multi-pronged framework for researchers to mitigate these risks, looking first to existing ethical frameworks and regulatory measures researchers can adapt to their own work, next to off-the-shelf AI solutions, then to design-specific solutions researchers can build into their AI to mitigate misuse. When researchers remain unable to address the potential for harmful misuse, and the risks outweigh potential benefits, we recommend researchers consider a different approach to answering their research question, or a new research question if the risks remain too great. We apply this framework to three different domains of AI research where misuse is likely to be problematic: (1) AI for drug and chemical discovery; (2) generative models for synthetic data; (3) ambient intelligence.

View details for DOI 10.1038/s42256-024-00926-3

View details for PubMedID 40994707

View details for PubMedCentralID PMC12456743

View details for Web of Science ID 001376888000001

View details for PubMedID 39587290

Abstract

PURPOSE: In recent years, researchers have brought attention to the underrepresentation of people with disabilities in biomedical research, including genomics research. However, little is known about how disability-related experiences influence participation in rare disease research. This omission is striking, because rare diseases are associated with disabling phenotypes that affect multiple body systems. As part of a study interrogating the relationship between rare disease status and disability identity, we conducted mixed-methods research to address this knowledge gap.METHODS: Parents of children enrolled in the UDN (n=25) completed semi-structured interviews to assess disability-related experiences in research participation. Directed content analysis was used to identify common themes.RESULTS: Participants' disability-related research experiences were characterized by: 1) disability-related facilitators to research participation, including benefits of research participation and disability-conscious approaches; 2) disability-related logistical barriers to research participation; and 3) research procedures, and the perception of research as minimally burdensome relative to clinical encounters.CONCLUSION: Parents of children in the UDN make considerable investments of time and resources to accommodate their children's disabilities to facilitate their participation. Future research should explore these issues in other genomic research studies and practical approaches to mitigating barriers and employing facilitators to disability-related research participation.

View details for DOI 10.1016/j.gim.2026.102663

View details for PubMedID 42439107

Abstract

Realizing the promise of precision medicine will require the highest standards of accuracy in genome sequencing and analysis. Here we describe challenges and opportunities for the field through the lens of genome data quality. We present recommendations in the context of specific areas of application for genomic sequencing in which isolated standards have arisen: germline sequencing, tumour sequencing, cell-free DNA testing, and sequencing for quality control in genetic therapy. Despite these distinct clinical contexts, technical challenges are often similar; for example, accurately detecting low-frequency genetic variants in tumour sequencing or gene-edited cells. We call for increased synchronization among these communities to establish new medical genome standards that promote confidence in genomic diagnostics and genetic therapies in a time of rapid technology-driven change. We suggest practical approaches for implementing these genome standards across contexts, and identify key areas that require further development.

View details for DOI 10.1038/s41586-026-10621-5

View details for PubMedID 42387168

View details for PubMedCentralID 10373632

Abstract

Genomics is transforming health care but its implementation raises challenges. This paper reports a 2025 workshop on justice in the implementation of genomics for rare disorders. The workshop goals were to develop a consensus understanding of the problems faced by rare disease patients and families where justice is at stake, to achieve a shared perspective on support for rare disease patients, and to consider the implications for justice in several areas of rare disease genomics, in both research and healthcare. We heard about the diverse experiences and needs of patients. Inequity between different rare diseases is marked. The need for coordination of care for rare disease patients is under-recognized but good models of rare disease care exist. The value of conscientious professionalism to nurture a rare disease mindset needs to be emphasized in the training of each new generation of healthcare students//trainees. The circumstances of different population groups differ systematically. The needs of indigenous and other historically marginalised groups must also be addressed. However, the subordination of individuals to the benefit of the population (i.e. eugenics) must be resisted. Those engaged in genomics projects or diagnostics may need protection from hype and misuse of their personal data, There are different perspectives on the fair allocation of resources to healthcare and research for rare conditions. Health economics and health technology assessment can be practised equitably, so as to meet the challenges of rare disease clinical trials and address the needs of patients and communities.

View details for DOI 10.1007/s12687-026-00911-w

View details for PubMedID 42384115

View details for PubMedCentralID 9160156

View details for Web of Science ID 001282805300174

View details for DOI 10.1001/jamanetworkopen.2023.50821

View details for PubMedID 38190187

Abstract

The rise of genomic technologies has catalyzed shifts in the health care landscape through the commercialization of genome sequencing and testing services in the genomics marketplace. The development of consumer genomics into a growing array of information technologies aimed at collecting, curating, and broadly sharing personal data and biological materials reconstitutes the meaning of health and reframes patients into biocitizens. In this context, the good biocitizen is expected to assume personal responsibility for health through consumption of genomic information and acquiescence to public and private efforts at data surveillance and aggregation. These shifts raise fundamental questions about how competing interests of the public, the state, and corporate entities will be reconciled and what trade-offs are demanded for the promise of precision health.

View details for DOI 10.1002/hast.1156

View details for Web of Science ID 000543918300008

View details for PubMedID 32597530

Abstract

This study describes challenges faced while incorporating sometimes conflicting stakeholder feedback into study design and development of patient-facing materials for a translational genomics study aiming to reduce health disparities among diverse populations.We conducted an ethnographic analysis of study documents including summaries of patient advisory committee meetings and interviews, reflective field notes written by study team members, and correspondence with our institutional review board (IRB). Through this analysis, we identified cross-cutting challenges for incorporating stakeholder feedback into development of our recruitment, risk assessment, and informed consent processes and materials.Our analysis revealed three key challenges: (1) balancing precision and simplicity in the design of study materials, (2) providing clinical care within the research context, and (3) emphasizing potential study benefits versus risks and limitations.While involving patient stakeholders in study design and materials development can increase inclusivity and responsiveness to patient needs, patient feedback may conflict with that of content area experts on the research team and IRBs who are tasked with overseeing the research. Our analysis highlights the need for further empirical research about ethical challenges when incorporating patient feedback into study design, and for dialogue with genomic researchers and IRB representatives about these issues.

View details for DOI 10.1038/s41436-020-0763-z

View details for PubMedID 32089547

View details for DOI 10.4088/PCC.26cr04204

View details for PubMedID 42284566

Abstract

Melatonin is an easily accessible, widely used drug for sleep issues, disrupted sleep-wake cycles, and jet lag, available in a variety of forms and dosages. Melatonin is also used in hospital settings to promote sleep onset, particularly in elderly patients, as a circadian rhythm regulator. Despite the popularity of melatonin, it is not approved by the US Food and Drug Administration (FDA). This creates ambiguity surrounding its proper usage for optimum results, including dosage and time of administration. The objective of this article is to shed light on the best timing to administer melatonin. Melatonin is a hormone that our body naturally produces to regulate our biological clock. Even though our body has a built-in "sleep system," many people still suffer from chronic sleep disorders such as insomnia. Melatonin has also proved to help prevent delirium in hospitalized patients due to its circadian rhythm regulatory effects. The elderly are at risk of developing insomnia because as one ages, melatonin production decreases. The most convenient solution for insomnia is to take melatonin supplements. To optimize the effects of melatonin supplements, proper dosage and timing must be considered. Additionally, patients who are oppositional to bedtime, which is known as bedtime resistance, are typically more willing to go to bed following melatonin administration. Melatonin administration at around 6PM (1-2hours before bedtime) is optimal to regulate sleep cycles of patients, and it can help with bedtime resistance. This should be the standard of care in all hospitals, nursing homes, and at home.

View details for DOI 10.1017/S109285292510062X

View details for PubMedID 41126740

View details for DOI 10.1080/15265161.2024.2308128

View details for PubMedID 38529987

Abstract

The study provides a comprehensive review of OpenAI's Generative Pre-trained Transformer 4 (GPT-4) technical report, with an emphasis on applications in high-risk settings like healthcare. A diverse team, including experts in artificial intelligence (AI), natural language processing, public health, law, policy, social science, healthcare research, and bioethics, analyzed the report against established peer review guidelines. The GPT-4 report shows a significant commitment to transparent AI research, particularly in creating a systems card for risk assessment and mitigation. However, it reveals limitations such as restricted access to training data, inadequate confidence and uncertainty estimations, and concerns over privacy and intellectual property rights. Key strengths identified include the considerable time and economic investment in transparent AI research and the creation of a comprehensive systems card. On the other hand, the lack of clarity in training processes and data raises concerns about encoded biases and interests in GPT-4. The report also lacks confidence and uncertainty estimations, crucial in high-risk areas like healthcare, and fails to address potential privacy and intellectual property issues. Furthermore, this study emphasizes the need for diverse, global involvement in developing and evaluating large language models (LLMs) to ensure broad societal benefits and mitigate risks. The paper presents recommendations such as improving data transparency, developing accountability frameworks, establishing confidence standards for LLM outputs in high-risk settings, and enhancing industry research review processes. It concludes that while GPT-4's report is a step towards open discussions on LLMs, more extensive interdisciplinary reviews are essential for addressing bias, harm, and risk concerns, especially in high-risk domains. The review aims to expand the understanding of LLMs in general and highlights the need for new reflection forms on how LLMs are reviewed, the data required for effective evaluation, and addressing critical issues like bias and risk.

View details for DOI 10.1371/journal.pdig.0000417

View details for PubMedID 38236824

View details for PubMedCentralID PMC10795998

Abstract

Self-referred imaging is one of the latest health care services to be marketed directly to consumers. Most aspects of these services are unregulated, and little is known about the messages in advertising used to attract potential consumers. We conducted a detailed analysis of print advertisements and informational brochures for self-referred imaging with respect to themes, content, accuracy, and emotional valence.Forty print advertisements from US newspapers around the country and 20 informational brochures were analyzed by 2 independent raters according to 7 major themes: health care technology; emotion, empowerment, and assurance; incentives; limited supporting evidence; popular appeal; statistics; and images. The Fisher exact test was used to identify significant differences in information content.Both the advertisements and the brochures emphasized health care and technology information and provided assurances of good health and incentives to self-refer. These materials also encouraged consumers to seek further information from company resources; virtually none referred to noncomplying sources of information or to the risks of having a scan. Images of people commonly portrayed European Americans. We found statistical differences between newspaper advertisements and mailed brochures for references to "prevalence of disease" (P<.001), "death" (P<.003), and "radiation" (P<.001). Statements lacking clear scientific evidence were identified in 38% of the advertisements (n = 15) and 25% of the brochures (n = 5).Direct-to-consumer marketing of self-referred imaging services, in both print advertisements and informational brochures, fails to provide prospective consumers with comprehensive balanced information vital to informed autonomous decision making. Professional guidelines and oversight for advertising and promotion of these services are needed.

View details for Web of Science ID 000225701900003

View details for PubMedID 15596630

Abstract

To examine different protocols for handling incidental findings on brain research MRIs, and provide a platform for establishing formal discussions of related ethical and policy issues.Corresponding authors identified from a database of peer-reviewed publications in 1991-2002 involving functional MRI (fMRI), alone or in combination with other imaging modalities, were invited to participate in this web-based survey. The survey asked questions regarding knowledge and handling of incidental findings, as well as characteristics of the scanning environment, training required, IRB protocol requirements, and neuroradiologist involvement.Seventy-four investigators who conduct MRI studies in the United States and abroad responded. Eighty-two percent (54/66) reported discovering incidental findings in their studies, such as arteriovenous malformations, brain tumors, and developmental abnormalities. Substantial variability was found in the procedures for handling and communicating findings to subjects, neuroradiologist involvement, personnel permitted to operate equipment, and training.Guidelines for minimum and optimum standards for detecting and communicating incidental findings on brain MRI research are needed.

View details for DOI 10.1002/jmri.20180

View details for Web of Science ID 000224762700001

View details for PubMedID 15503329

View details for PubMedCentralID PMC1506385

Abstract

Chylothorax is a potentially serious complication of lung and heart-lung transplantation. This article describes the clinical course of chylothorax in 3 heart-lung allograft recipients. We discuss management options, including dietary modifications, octreotide infusion, thoracic duct ligation and embolization, and surgical pleurodesis. In addition, we describe the novel use of aminocaproic acid to reduce lymph flow. We propose a multidisciplinary approach for the management of chylothorax that includes both medical and surgical options.

View details for Web of Science ID 000221393700018

View details for PubMedID 15135382

Abstract

Sharing human brain data can yield scientific benefits, but because of various disincentives, only a fraction of these data is currently shared. We profile three successful data-sharing experiences from the NIH BRAIN Initiative Research Opportunities in Humans (ROH) Consortium and demonstrate benefits to data producers and to users.

View details for DOI 10.1016/j.neuron.2023.09.029

View details for Web of Science ID 001135333000001

View details for PubMedID 37944519

View details for DOI 10.1080/15265161.2023.2237443

View details for Web of Science ID 001058425800017

View details for PubMedID 37647482

View details for PubMedCentralID PMC10502902

View details for DOI 10.1001/jama.2024.9450

View details for PubMedID 39141373

View details for DOI 10.1213/ANE.0000000000006061

View details for PubMedID 36806237

View details for DOI 10.1213/ANE.0000000000005947

View details for PubMedID 35709451

View details for DOI 10.1001/jamanetworkopen.2026.29872

View details for PubMedID 42616504

View details for PubMedCentralID PMC13491107

Abstract

Adults with intellectual and developmental disabilities (AIDD) experience persistent barriers to accessing and participating in health care yet are rarely engaged as partners in efforts to improve accessibility. This study used the evidence-based Our Voice (OV) citizen science approach to engage AIDD in identifying barriers and facilitators to primary care access and generating actionable solutions. Twelve participants (six AIDD and six adults without IDD) documented accessibility conditions across three medical clinics using photographs and narratives. Analysis of 170 photographs and 154 narratives identified five themes: (1) sensory considerations, (2) informational materials, (3) signage and wayfinding, (4) accommodations, and (5) physical accessibility. Findings demonstrate the feasibility of engaging AIDD in structured, clinic-based accessibility assessment and suggest a participatory framework that may inform research and quality improvement initiatives aimed at promoting disability-inclusive health care.

View details for DOI 10.1080/19349491.2026.2682771

View details for PubMedID 42584858