INTRODUCTION: AI is big and moving fast into healthcare, creating opportunities and risks. However, current approaches to governance focus on high-level principles rather than tailored recommendations for specific domains like consumer health. This gap risks unintended consequences from generic guidelines misapplied across contexts and from providing answers before agreeing on the questions.

OBJECTIVE: Our objective is to explore pragmatic multi-stakeholder approaches to govern consumer-facing health AI. The aims are to (1) establish an approach tailored for consumer health AI governance and (2) identify key constraints and desirable model characteristics.

METHODS: This paper synthesizes insights informed by a 4-month multidisciplinary expert consensus process with nearly 200 participants. The deliberations provided guidance for the development of the proposed governance models in consumer health AI.

RESULTS: (1) A Shared View of Consensus: A process for consumer health AI governance should limit the scope and incorporate multi-stakeholder perspectives centered on patient needs. Desirable model characteristics include adaptability, patient empowerment, and transparency. (2) Recommended Collaborative Process: A pathway for effective governance should begin by forming a Health AI Consumer Consortium (HAIC2) representing patients and aligning incentives across stakeholders.

CONCLUSIONS: While examples focus on the United States healthcare system, core themes around incorporating consumer voices, enabling transparency, and balancing innovation with thoughtful oversight while avoiding overambitious scope will have relevance globally. As consumer AI spreads worldwide, the multi-stakeholder alignment and patient empowerment principles proposed here may offer productive ways to ensure AI for consumers is safe, effective, equitable, and trustworthy (SEET).

Digital health interventions (DHIs) have the potential to improve health care and health promotion. However, there is a lack of guidance in the literature for the development, refinement, and prioritization of key performance indicators (KPIs) for the evaluation of DHIs. This paper presents a 4-stage process used in the Gravitate Health project based on stakeholder consultation and consensus for this purpose. The Gravitate Health consortium, which comprises private and public partners from across Europe and the United States, is developing innovative digital health solutions in the form of Federated Open-Source Platform and G-lens to present users with individualized digital information about their medicines. The first stage of this was the consultative process for the development of KPIs involving stakeholder (Gravitate Health project leads) consultations at the planning stages of the project. This resulted in the formation of an extensive list of KPIs organized into 7 categories. The second stage was conducting a scoping review, which confirmed the need for extensive stakeholder consultation in all stages of the KPI development, refinement, and prioritization process. The third stage was a period of further consultation with all consortium members, which resulted in the elimination of 1 category of KPIs. The fourth stage involved using the Delphi technique for refining and prioritizing the remaining 6 categories of KPIs. It is unusual to use this methodology in a nonresearch exercise, but it provided a clear consultative framework and structure that facilitated the achievement of consensus within a large consortium of 250 members on a substantial list of KPIs for the project. Consortium members ranked the relevance and importance of each KPI. The final list of KPIs provides substantial indicators sensitive to the needs of a broad group of stakeholders that are being used to capture real-world data in developing and evaluating DHIs.

OBJECTIVE: To map how social, commercial, political and digital determinants of health have changed or emerged during the recent digital transformation of society and to identify priority areas for policy action.

METHODS: We systematically searched MEDLINE, Embase and Web of Science on 24 September 2023, to identify eligible reviews published in 2018 and later. To ensure we included the most recent literature, we supplemented our review with non-systematic searches in PubMed® and Google Scholar, along with records identified by subject matter experts. Using thematic analysis, we clustered the extracted data into five societal domains affected by digitalization. The clustering also informed a novel framework, which the authors and contributors reviewed for comprehensiveness and accuracy. Using a two-round consensus process, we rated the identified determinants into high, moderate and low urgency for policy actions.

FINDINGS: We identified 13 804 records, of which 204 met the inclusion criteria. A total of 127 health determinants were found to have emerged or changed during the digital transformation of society (37 digital, 33 social, 33 commercial and economic and 24 political determinants). Of these, 30 determinants (23.6%) were considered particularly urgent for policy action.

CONCLUSION: This review offers a comprehensive overview of health determinants across digital, social, commercial and economic, and political domains, highlighting how policy decisions, individual behaviours and broader factors influence health by digitalization. The findings deepen our understanding of how health outcomes manifest within a digital ecosystem and inform strategies for addressing the complex and evolving networks of health determinants.

BACKGROUND: TikTok is a global social media platform with over 1 billion active users. Presently, there are few data on how TikTok users navigate the platform for mental health purposes and the content they view.

OBJECTIVE: This study aims to understand the patterns of mental health-related content on TikTok and assesses the accuracy and quality of the advice and information provided.

METHODS: We performed a summative content analysis on the top 1000 TikTok videos with the hashtag #mentalhealth between October 12 and 16, 2021. Six content themes were developed to code the data: (1) a personal story, perspective, or confessional, (2) advice and information, (3) emoting, (4) references to death, (5) references to science or research, and (6) a product or service for sale. Advice and information were further assessed by clinical experts.

RESULTS: A total of 970 mental health-related videos were pulled for our analysis (n = 30 removed due to non-English content). The most prevalent content themes included a personal story, perspective, or confessional (n = 574), advice and information (n = 319), emoting (n = 198), references to death (n = 128). Advice and information were considered misleading in 33.0% of videos (n = 106), with misleading content performing better. Few videos included references to scientific evidence or research (n = 37).

CONCLUSION: Healthcare practitioners and researchers may consider increasing their presence on the platform to promote the dissemination of evidence-based information to a wider and more youth-targeted population. Interventions to reduce the amount of misinformation on the platform and increase people's ability to discern between anecdotal and evidence-based information are also warranted.

BACKGROUND: Social determinants of health (SDoH) contribute to disparities in SARS-CoV-2 infection, but their associations with long COVID are unknown.

OBJECTIVE: To determine associations between SDoH at the time of SARS-CoV-2 infection and risk for long COVID.

DESIGN: Prospective observational cohort study.

SETTING: 33 states plus Washington, DC, and Puerto Rico.

PARTICIPANTS: Adults (aged ≥18 years) enrolled in RECOVER-Adult (Researching COVID to Enhance Recovery) between October 2021 and November 2023 who were within 30 days of SARS-CoV-2 infection; completed baseline SDoH, comorbidity, and pregnancy questionnaires; and were followed prospectively.

MEASUREMENTS: Social risk factors from SDoH baseline questionnaires, ZIP code poverty and household crowding measures, and a weighted score of 11 or higher on the Long COVID Research Index 6 months after infection.

RESULTS: Among 3787 participants, 418 (11%) developed long COVID. After adjustment for demographic characteristics, pregnancy, disability, comorbidities, SARS-CoV-2 severity, and vaccinations, financial hardship (adjusted marginal risk ratio [ARR], 2.36 [95% CI, 1.97 to 2.91]), food insecurity (ARR, 2.36 [CI, 1.83 to 2.98]), less than a college education (ARR, 1.60 [CI, 1.30 to 1.97]), experiences of medical discrimination (ARR, 2.37 [CI, 1.94 to 2.83]), skipped medical care due to cost (ARR, 2.87 [CI, 2.22 to 3.70]), and lack of social support (ARR, 1.79 [CI, 1.50 to 2.17]) were associated with increased risk for long COVID. Living in ZIP codes with the highest (vs. lowest) household crowding was also associated with greater risk (ARR, 1.36 [CI, 1.05 to 1.71]).

LIMITATION: Selection bias may influence observed associations and generalizability.

CONCLUSION: Participants with social risk factors at the time of SARS-CoV-2 infection had greater risk for subsequent long COVID than those without. Future studies should determine whether social risk factor interventions mitigate long-term effects of SARS-CoV-2 infection.

PRIMARY FUNDING SOURCE: National Institutes of Health.

INTRODUCTION: The rapid integration of artificial intelligence (AI) into healthcare presents significant governance challenges, requiring balanced approaches that safeguard safety, efficacy, equity, and trust (SEET). This study proposes a cognitive framework to guide AI governance, addressing tradeoffs between speed, scope, and capability.

OBJECTIVE: To develop a structured governance model that harmonizes stakeholder perspectives, focusing on multi-dimensional challenges and ethical principles essential for AI in healthcare.

METHODS: A multidisciplinary team convened at the Blueprints for Trust conference, organized by the American Medical Informatics Association (AMIA), and the Division of Clinical Informatics at Beth Israel Deaconess Medical Center. Following extensive discussions with 190 participants across sectors, three governance models were identified to address specific domains: (1) Clinical Decision Support (CDS), (2) Real-World Evidence (RWE), (3) Consumer Health (CH).

RESULTS: Three governance models emerged, tailored to CDS, RWE, and CH domains. Key recommendations include establishing a Health AI Consumer Consortium for patient-centered oversight, initiating voluntary accreditation and certification frameworks, and piloting risk-level-based standards. These models balance rapid adaptation with SEET-focused safeguards through transparency, inclusivity, and ongoing learning.

CONCLUSION: A proactive, constraint-based governance framework is critical for responsible AI integration in healthcare. This structured, multi-stakeholder approach provides a roadmap for ethical, transparent governance that can evolve with technological advancements, enhancing trust and safety in healthcare AI applications.

BACKGROUND: The use of real-world data (RWD) in artificial intelligence (AI) applications for healthcare offers unique opportunities but also poses complex challenges related to interpretability, transparency, safety, efficacy, bias, equity, privacy, ethics, accountability, and stakeholder engagement.

METHODS: A multi-stakeholder expert panel comprising healthcare professionals, AI developers, policymakers, and other stakeholders was assembled. Their task was to identify critical issues and formulate consensus recommendations, focusing on the responsible use of RWD in healthcare AI. The panel's work involved an in-person conference and workshop and extensive deliberations over several months.

RESULTS: The panel's findings revealed several critical challenges, including the necessity for data literacy and documentation, the identification and mitigation of bias, privacy and ethics considerations, and the absence of an accountability structure for stakeholder management. To address these, the panel proposed a series of recommendations, such as the adoption of metadata standards for RWD sources, the development of transparency frameworks and instructional labels likened to "nutrition labels" for AI applications, the provision of cross-disciplinary training materials, the implementation of bias detection and mitigation strategies, and the establishment of ongoing monitoring and update processes.

CONCLUSION: Guidelines and resources focused on the responsible use of RWD in healthcare AI are essential for developing safe, effective, equitable, and trustworthy applications. The proposed recommendations provide a foundation for a comprehensive framework addressing the entire lifecycle of healthcare AI, emphasizing the importance of documentation, training, transparency, accountability, and multi-stakeholder engagement.

INTRODUCTION: Patient misunderstanding of instructions on medication labels is a common cause of medication errors and can result in ineffective treatment. One way to better improve patient comprehension of medication labels is by optimizing the content and display of the information.

OBJECTIVES: To review comparative studies that have evaluated the design of a medication label to improve patient knowledge or safety.

METHODS: Studies were selected from systematic computerized literature searches performed in PubMed, Embase (Elsevier), Cochrane Central (EBSCO), Cumulative Index to Nursing and Allied Health Literature-CINAHL (EBSCO), and Web of Science (Thomson Reuters). Eligible studies included comparative studies that evaluated the design of a medication label to improve patient knowledge or safety.

RESULTS: Of the 246 articles identified in the primary literature search, 14 studies were selected for data abstraction. Thirteen of these studies significantly impacted the patient understanding of medication labels. Three studies included a measure of patient safety in terms of medication adherence and dosing errors. The utilization of patient-centered language, pictograms/graphics, color/white space, or font optimization was seen to have the most impact on patient comprehension.

CONCLUSION: It is essential to present medication information in an optimal manner for patients. This can be done by standardizing the content, display, and format of medication labels to improve understanding and medication usage. Evidence-based design principles can, therefore, be used to facilitate the standardization of the structure of label content for both print and electronic devices. However, more research needs to be done on validating the implications of label content display to measure its impact on patient safety.

SYSTEMIC REVIEW REGISTRATION: PROSPERO CRD42022347510 ( http://www.crd.york.ac.uk/prospero/ ).

INTRODUCTION: Mental health Applications (MH Apps) can potentially improve access to high-quality mental health care. However, the recent rapid expansion of MH Apps has created growing concern regarding their safety and effectiveness, leading to the development of AETs (Assessment and Evaluation Tools) to help guide users. This article provides a critical, mixed methods analysis of existing AETs for MH Apps by reviewing the criteria used to evaluate MH Apps and assessing their effectiveness as evaluation tools.

METHODS: To identify relevant AETs, gray and scholarly literature were located through stakeholder consultation, Internet searching via Google and a literature search of bibliographic databases Medline, APA PsycInfo, and LISTA. Materials in English that provided a tool or method to evaluate MH Apps and were published from January 1, 2000, to January 26, 2021 were considered for inclusion.

RESULTS: Thirteen relevant AETs targeted for MH Apps met the inclusion criteria. The qualitative analysis of AETs and their evaluation criteria revealed that despite purporting to focus on MH Apps, the included AETs did not contain criteria that made them more specific to MH Apps than general health applications. There appeared to be very little agreed-upon terminology in this field, and the focus of selection criteria in AETs is often IT-related, with a lesser focus on clinical issues, equity, and scientific evidence. The quality of AETs was quantitatively assessed using the AGREE II, a standardized tool for evaluating assessment guidelines. Three out of 13 AETs were deemed 'recommended' using the AGREE II.

DISCUSSION: There is a need for further improvements to existing AETs. To realize the full potential of MH Apps and reduce stakeholders' concerns, AETs must be developed within the current laws and governmental health policies, be specific to mental health, be feasible to implement and be supported by rigorous research methodology, medical education, and public awareness.

OBJECTIVE: The increasing demands for curated, high-quality research data are driving the emergence of a novel registry type. The need to assemble, curate, and export this data grows, and the conventional simplicity of registry models is driving the need for advanced, multimodal data registries-the dawn of the next-generation registry.

MATERIALS AND METHODS: The article provides an outline of the technology roles and responsibilities needed for successful implementations of next-generation registries.

RESULTS: We propose a framework for the planning, construction, maintenance, and sustainability of this new registry type.

DISCUSSION: A rubric of organizational, computational, and human resource needs is discussed in detail, backed by over 40 years of combined in-the-field experiences by the authors.

CONCLUSIONS: A novel field, registry science, within the clinical research informatics domain, has arisen to offer its insights into conceiving, structuring, and sustaining this new breed of tools.