Maxim Topaz

References to nonexistent studies are well documented in biomedical research, but evidence outside biomedicine is limited. We analyzed a stratified random sample of 500,000 peer-reviewed journal papers in chemistry, physics, materials science, engineering, and computer science published from January 2023 through March 31, 2026, using OpenAlex, a large open index of scholarly works and citations. Our automated pipeline classified 1,135 references across 619 papers as fabricated, with 85.8% precision on masked expert review; fabricated is an operational label for a claimed title that could not be located in any of seven bibliographic and open-web sources. Affected-paper prevalence rose from 1 in 1,600 (6.3 per 10,000) in 2023 to 1 in 435 (23.0 per 10,000) in the first quarter of 2026, and full-period rates differed roughly sixfold across fields, from 25.6 per 10,000 papers in physics to 4.3 in chemistry. Most affected papers contained one or two fabricated references, but the 27 papers containing five or more accounted for more than a third of all fabricated references identified. Applying each field’s rate to the approximately 7 million OpenAlex papers indexed in these fields gives an estimated 8,500 papers carrying at least one fabricated reference. Routine automated reference verification at submission, together with retrospective screening of the published literature, can detect fabricated references before and after they enter the scientific record.

Aim: Artificial intelligence (AI) is reshaping healthcare delivery. This work aims to explore how nursing expertise can be fully integrated into the development of AI‐based health technologies and to assess the implications for clinical practice, education, leadership, industry, policy and research. Methods: This qualitative expert elicitation study drew on data generated during a structured three‐day international think tank hosted by the Brocher Foundation in November 2024. Interdisciplinary experts in nursing, informatics, ethics, philosophy, medicine, computer science and law participated in structured presentations, facilitated subgroup discussions, plenary synthesis activities and collaborative development of recommendations. Reflexive thematic analysis was used to analyse workshop outputs and identify cross‐cutting patterns, tensions and implications. Results: Six interrelated thematic findings were generated. Participants framed nursing expertise as a form of sociotechnical governance rather than simply end‐user input: (1) nurses were positioned as co‐developers who can shape AI problem definition, design and implementation; (2) AI education was viewed as building critical interpretive capacity, not only technical literacy; (3) structural barriers limited nurse participation in innovation; (4) legal and ethical concerns reflected a gap between accountability and authority; (5) nurses were positioned as frontline actors for identifying AI bias and inequity; and (6) competency frameworks were viewed as tools for clarifying role‐specific expectations in AI‐enabled care. Discussion: The findings suggest that embedding nursing expertise in healthcare AI is a matter of sociotechnical governance, requiring attention to organisational structures, education, legal accountability, professional roles and mechanisms for identifying safety, equity and workflow concerns. Implications for the Profession: This study provides an expert‐informed framework for strengthening nursing participation in healthcare AI development. While the recommendations require empirical validation, they indicate priority areas for action, including nurse‐led innovation, competence development, cross‐sector collaboration, and educational and policy frameworks that recognise nursing expertise in AI development. Reporting Method: The study adhered to the Standards for Reporting Qualitative Research (SRQR) guidelines.

Objective: To examine posthumous health data donation as a pathway toward more representative datasets used to develop healthcare artificial intelligence (AI). Perspective: Healthcare AI systems are increasingly deployed in clinical settings, yet the datasets underlying these systems remain narrow, fragmented, and systematically underrepresenting key patient populations. Current consent frameworks, designed for living patients, create structural barriers to assembling diverse training data. Death meaningfully shifts the ethical risk–benefit calculus, substantially reducing harms, employment discrimination, insurance denial, social stigma, that presuppose a living subject. This creates a unique opportunity for individuals to contribute their health data posthumously through informed, opt-in consent established during life. Conclusion: Posthumous health data donation warrants serious exploration by the informatics, policy, and governance communities as a complementary mechanism for reducing structural barriers to inclusion, one that does not guarantee representativeness but may address gaps that living-patient consent frameworks cannot reach.

Background: Natural language processing can extract meaningful information from clinical notes. However, human annotation is time-consuming and costly, and scarce data poses a challenge. Objective: This study aimed to explore a semantic similarity search approach to extract exemplars of stigmatizing and positive language in obstetric clinical notes. Methods: We used electronic health record data from labor and birth admissions at 2 hospitals in the United States from 2017 to 2019. We used a semantic similarity search approach, which used 200 randomly selected true exemplars, stratified by language categories, as queries to search for similar exemplar candidates. We extracted the top 5 candidates with the highest cosine similarities, which were assessed for accuracy. Results: We retrieved 1000 candidates. An average precision of 0.69 was achieved when candidates with cosine similarity thresholds of 0.75 or higher were included, at which point 68.8% (64/93) of exemplar candidates accurately represented true cases. At the 0.75 threshold, the proportion of true cases was higher for preferred language (41/56, 73.2%) and unilateral/authoritarian decisions (5/7, 71.4%). The proportion of true cases was lower for difficult patients (2/7, 28.6%) and marginalized identities (3/9, 33.3%). Conclusions: The semantic similarity search approach shows promise in efficiently extracting exemplars while reducing the annotation burden, laying the groundwork for future applications in other domains.

Was not provided for this article.

Fabricated references, citations to nonexistent sources, appeared at a rate of 56.9 per 10 000 biomedical papers indexed in early 2026, a more than 12-fold increase since 2023. Manuscript volume is rising faster than editorial capacity, and conventional peer review does not routinely verify references. This viewpoint argues that automated reference verification, integrated into submission workflows, is a practical and proportionate response. The technical readiness of current tools, the evidence gaps and economic barriers to adoption, and the structured editorial response triggered by fabricated references are examined. The core principle is ‘artificial intelligence should handle the verifiable; humans should judge the evaluable’.

As artificial intelligence (AI) tools are increasingly integrated into healthcare settings, their application to ethically complex care decisions for patients who lack decision-making capacity and identifiable surrogates remains largely unexplored. This study explored ethics consultants’ perspectives on the use of artificial intelligence (AI) in ethical and clinical decision-making for patients who are incapacitated with no evident advance directives or surrogates (INEADS). We conducted a qualitative study using in-depth semi-structured interviews with 19 ethics consultants across nine U.S. states, analyzed using thematic analysis. Three themes were generated. Theme 1, Human Accountability as a Non-Negotiable Boundary, captured ethical concerns including clinician accountability, AI’s inability to capture individual context, and algorithmic bias risk. Theme 2, Designing AI That Reflects the Complexity of INEADS Care, identified requirements for high-quality training data, interdisciplinary development teams, and explainability. Theme 3, A Conditional and Constructive Vision for AI as Supportive Tool, described applications encompassing surrogate identification, goals-of-care facilitation, longitudinal decisional pattern review, and individualized prognostication. Participants framed AI as a resource to support rather than replace human moral judgment, raised concerns about algorithmic bias and institutional variation, and articulated a constructive vision grounded in prior natural language processing work identifying INEADS patients across care settings. Ethics consultants expressed conditional acceptance of AI in INEADS care when used to support, not replace, accountable human judgment. Findings highlight the need for equity-centered design, transparent communication of data limitations, interdisciplinary development, and ethics consultant involvement in validation before clinical implementation.