Governance for safe and responsible AI in healthcare organisations: A scoping review of frameworks

Governance for safe and responsible AI in healthcare organisations: A scoping review of frameworks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Governance for safe and responsible AI in healthcare organisations: A scoping review of frameworks Amy Wang, Sam Freeman, Farah Magrabi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7209096/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 01 May, 2026 Read the published version in npj Digital Medicine → Version 1 posted 9 You are reading this latest preprint version Abstract This scoping review synthesises existing knowledge about AI governance in healthcare organisations by identifying key ethics and governance principles, and key components of AI governance frameworks. Following PRISMA-ScR guidelines, we searched MEDLINE, Embase, and Scopus (April 2024, updated March 2025) for AI governance frameworks in acute care. Seventy-seven frameworks were identified and examined for: 1) Guiding principles (ethics or governance-related); 2) Assessment methods; 3) AI life cycle stages; and 4) Oversight mechanisms. Most lacked real-world applicability and missed key principles or components such as an oversight mechanism. Only 10 frameworks (13.0%) included all four framework components, with oversight mechanisms (e.g. AI-specific governance committee) being the least common (n = 15, 19.5%). No framework had been evaluated for effectiveness in enabling safe and responsible AI. There is a need to move beyond principles to implementing AI governance frameworks in healthcare organisations and assessing their real-world impact. Health sciences/Health care Health sciences/Medical research Scientific community and society/Scientific community Figures Figure 1 INTRODUCTION Effective governance of artificial intelligence (AI) in healthcare organisations is critical to facilitating its safe and responsible implementation. 1 However, attempts to conceptualise AI governance are fragmented, and many proposed ethical and governance frameworks have not been operationalised into actionable practices or evaluated in real-world healthcare settings. 2 The lack of cohesion and real-world applicability has made it challenging for healthcare organisations to safely navigate the rapidly expanding technological capabilities of AI and associated ethical and regulatory concerns, 3 highlighting the need for more comprehensive and practical guidance on AI governance. 4 Current understanding lacks a structured, holistic approach that recognises AI tools as one part of a larger sociotechnical system, integrating concerns regarding the technological product, 5 ethical principles, organisational processes, 6 and the regulatory landscape. 7 This has hindered AI adoption despite rapid advancements in research and technology, 8 limiting the transformative potential of AI tools on patient care across both clinical and non-clinical tasks, such as predictive analysis, diagnostic assistance, or optimisation of administrative processes. 9 , 10 Previous reviews have examined AI governance principles and practices broadly across industries, 11 – 14 or in specific health disciplines. 15 , 16 Others have focused only on a specific aspect of governance such as ethics principles, 17 – 20 on specific AI technologies or applications, 21 , 22 or on digital health technology more broadly, which may or may not include AI. 23 – 25 To date, no review has examined the frameworks governing AI within healthcare organisations. Healthcare represents a distinctive context where sector-agnostic frameworks may not adequately consider the rigorous ethical and regulatory requirements unique to healthcare, while discipline- or technology-specific frameworks may not facilitate governance that covers the wide variety of AI use cases relevant to multidisciplinary healthcare organisations. To address this gap, we conducted a scoping review to provide an overview of the state of AI governance in healthcare, with a specific focus on governance at the organisational level. Scoping reviews are a suitable method for identifying key characteristics related to a concept, particularly for emerging areas such as AI governance. 26 We sought to identify the key ethics and governance principles, and the components of practical AI governance frameworks for safe and responsible AI in healthcare organisations. RESULTS Overview of frameworks We identified 77 frameworks for AI governance in healthcare organisations (Table 1; Supplement S2). Almost half were published in 2023 or 2024 (n=36, 46.8%), and the most common country of origin was the USA (n=22, 28.6%). Most frameworks (n=56, 72.7%) were designed for application across general healthcare settings (i.e. any health field), with a primary focus on evaluation (n=18, 23.4%) or overall governance (n=16, 20.8%). Frameworks were mostly derived from the literature (n=38, 49.4%) and typically structured by themes or principles (n=39, 50.6%). There was no evidence of systematic evaluation of any frameworks on organisational or health outcomes. Table 1. Characteristics of AI ethics and governance frameworks identified in the scoping review. Characteristic Frequency (n=77) % Type of publication Academic Government Industry 63 10 4 81.8 13.0 5.2 Year published 2019 2020 2021 2022 2023 2024 2025 4 9 13 12 17 19 3 5.2 11.7 16.9 15.6 22.1 24.7 3.9 Country of origin USA UK Canada Australia New Zealand Singapore Germany Italy Other International 22 12 9 9 3 2 2 2 11 5 28.6 15.6 11.7 11.7 3.9 2.6 2.6 2.6 14.3 6.5 Setting General healthcare Research Imaging/radiology Surgery-related* Emergency medicine Oncology Dermatology Ophthalmology Public health 56 9 4 3 1 1 1 1 1 72.7 11.7 5.2 3.9 1.3 1.3 1.3 1.3 1.3 Purpose Evaluation Governance Ethical principles Implementation Guidelines or standards Translation and integration 18 16 13 11 11 8 23.4 20.8 16.9 14.3 14.3 10.4 Method of derivation^ Literature review Research and consultation † Previous framework(s) Self-created Expert consensus Case studies N/A 38 13 13 13 7 4 10 49.4 16.9 16.9 16.9 9.1 5.2 13.0 Structure Theme-based Stage-based Theme x stage Other or N/A 39 13 21 4 50.6 16.6 27.3 5.2 Number of key components included One component Two components Three components Four components 19 31 17 10 24.7 40.3 22.1 13.0 Inclusion of key components Component 1: Guiding principles Component 2: Assessment method Component 3: AI lifecycle stages Component 4: Oversight mechanism 72 50 35 15 93.5 64.9 45.5 19.5 *Perioperative medicine, operating rooms, otolaryngology-head and neck surgery ^Some frameworks used multiple methods/sources † Includes expert consultation, interviews, focus groups, and workshops Key framework components Most frameworks only included one or two out of the four key components (Table 1, Supplement S5). Of the 10 with all four components, seven were operational or had been tested (Table 2). Each key component is presented in the subsequent sections. Component 1: Guiding principles The first key component was a set of guiding principles for safe and responsible practices, included in nearly all the frameworks reviewed (n=72, 93.5%). We identified a total of 25 key principles across two broad categories – ethics principles and governance principles (Table 3, Supplement S3, S4). The most common ethics principle was transparency (n=61, 79.2%), followed by fairness (n=50, 64.9%) and privacy and security (n=48, 62.3%). Governance principles were distinct from ethics concerns, relating to the organisational, regulatory, and technical processes and practices that ensure safe and responsible use of AI tools. The most common governance principles were data selection and management (n=57, 74.0%), accuracy and model performance (n=55, 71.4%), and ongoing monitoring and maintenance (n=54, 70.1%) (Table 3, Supplement S3, S4). Table 2. Summary of frameworks that include all four key components of AI governance (n=10). Study Year Implementation status Guiding principles AI lifecycle stages Oversight mechanism Apfelbacher et al. 27 2024 Status not provided Details/checklist not provided - describes principles in text. Principles include safety, reliability, integration, education and training, autonomy, and accountability. Before implementation During implementation After implementation Specific AI governance group – Suggests establishing an AI committee prior to implementation. Callahan et al. 28 2024 Operational Stage 1: Problem, need and use case definition Usefulness estimates by workflow simulation Financial projections Ethical considerations (Responsibility, Equity, Traceability, Reliability, Governance, Non-maleficence, Autonomy) Stage 2: Model formulation Model training and testing Deployment on SHC infrastructure Organisational integration Stage 3: Monitoring Prospective evaluation What & Why How Impact Existing committee, new remit - An existing data science team at Stanford Health Care. Coalition for Health AI 29 2024 Status not provided 5 core principles for trustworthy health AI: Usefulness, usability, and efficacy Fairness and equity Safety and reliability Transparency, intelligibility, and accountability Security and privacy Stage 1: Define the problem Stage 2: Design the AI system Stage 3: Engineer the AI solution Stage 4: Assess Stage 5: Pilot Stage 6: Deploy & monitor Independent review - Suggests a combination of a local leadership structure and regular independent third-party review. Economou-Zavlanos et al. 30 2024 Operational Clinical Value and Safety Fairness & Equity Usability & Adoption Regulatory Compliance Transparency and Accountability Model development Silent evaluation Effectiveness evaluation General deployment Specific AI governance group - ABCDS Oversight Committee and Review Committee. Liao et al. 31 2022 Operational Set of guiding principles endorsed by the AI Committee: Predictive model Model evaluation includes statistical measures and relevant operational health metrics Model output follows the five rights of CDSs and is associated with interventions Model monitoring Healthcare ethics incorporated in all stages of model evaluation and validation Value stream from initial presentation to the Committee to periodic review and update or decommissioning. Specific AI governance group - Clinical-AI-Predictive Analytics (CAIPA) Committee and case-based sub-committees. Morley et al. 32 2021 Not implemented Key questions at each stage: Is this AI system the right solution to the problem? Has the AI system been designed in the right way? Is the AI system working in the right way? Is the AI system having the right kind of impact? Preclinical stage (Theoretical Stage) Exploratory stage (Validation Stage) Definitive Stage (Real-world Evaluation Stage) Post-market surveillance Include AI governance in existing processes – Suggests identifying and mapping which aspects of evaluation are already covered by existing bodies and governing processes. Saenz et al. 33 2024 Tested I. Model Assessment Data provenance for training and testing Regulatory Compliance II. Pre-Deployment (Shadow Deployment) Testing phase Demographic Evaluation System Customization Model Output Security and Privacy Autonomy and Human Oversight Transparency Education III. Deployment and Continuous Monitoring Documentation/Communication Performance Metrics Risk Management Equity and Fairness I. Model Assessment II. Pre-Deployment (Shadow Deployment) III. Deployment and Continuous Monitoring Specific AI governance group – recommends presence of an AI monitoring committee. Singapore MOH, HSA, and IHiS 34 2021 Tested Guiding principles : Fairness, Responsibility, Transparency, Explainability, Patient-Centricity Guideline sections : Development Design: clinical inputs, end-user inputs, understanding the current clinical practice, data, cybersecurity, explainability Build: development standards, self-validation Test: Evaluation and monitoring of AI-MD, intended use and workflow Implementation Use: Clinical governance, operational workflows and processes, end-user communication Monitor & respond: Post-deployment monitoring Review: Review and tracking Development - design, build, test Implementation - use, monitor & respond, review Non-specific - suggests review by “Organisational Leadership” who then makes a decision to implement. van der Vegt et al. 35 2023 Tested Specification Component development Combination of components into systems Integration of system into environment Routine use Organisation engagement Policy domains Stage I - Definition Stage II - Retrospective study Stage III - Silent trial Stage IV - Pilot trial Stage V - Large trial/roll-out Non-specific – includes “Governance” as a key principle under “F. Organisation engagement” but does not provide specific structure. Whittaker et al. 36 2023 Operational Appropriateness Consumers/population perspectives Māori perspectives Equity and fairness Ethical principles Clinical perspectives Data availability, quality, appropriateness, and completeness Technical processes Contractual and legal issues Checklist for each domain across 3 stages: Concept development Access to data for pre-processing, labelling, or model development Validation or implementation of an existing AI model Specific AI governance group - National AI Expert Advisory Group (NAIAEAG) and AI Lab. Table 3. Summary of guiding ethics and governance principles for AI. Principle Description n (%) References Ethics principles Accountability “Those responsible for the different phases of the AI system lifecycle should be identifiable and accountable for the outcomes of the AI systems, and human oversight of AI systems should be enabled.” 37 41 (53.2) 2,15-17,20,22,25,27-30,33-36,38-63 Autonomy & human-centred values “Throughout their lifecycle, AI systems should respect human rights, diversity, and the autonomy of individuals.” 37 Includes the principles of consent (n=23, 29.9%). 31 (40.3) 2,15,17,20,22,25,27,28,31,33,36,38,40,43-45,48,49,53,54,56-58,61,63-69 Beneficence/ Human, societal & environmental wellbeing “Throughout their lifecycle, AI systems should benefit individuals, society and the environment.” 37 40 (51.9) 2,4,7,17,20,22,25,27-31,36,38-41,43-45,48-52,56,59-66,68-73 Contestability “When an AI system significantly impacts a person, community, group or environment, there should be a timely process to allow people to challenge the use or outcomes of the Ai system.” 37 10 (13.0) 2,27,29,34,42,44,56,61,63,66 Fairness & equity “Throughout their lifecycle, AI systems should be inclusive and accessible and should not involve or result in unfair discrimination against individuals, communities or groups.” 37 Involves ensuring AI systems do not create or perpetuate bias. 50 (64.9) 2,7,15-17,20,22,25,28-31,33-36,38,40-51,53,54,56-58,60,61,63-69,71-77 Non-maleficence “The injunction to “Do No Harm,” that is, that every reasonable effort shall be made to avoid, prevent, and minimize harm or damage to any stakeholder.” 56 36 (46.8) 2,4,20-22,28-31,34,36,38,40-42,44,49-52,56,57,59,61-65,68,69,72,74,75,77-79 Privacy & security “Throughout their lifecycle, AI systems should respect and uphold privacy rights and data protection and ensure the security of data”. 37 Includes the principle of confidentiality (n=19, 24.7%). 48 (62.3) 2,15-17,20-22,25,27,29,33-36,38,40,42-45,48-53,55,57-61,63-70,74,76,79-84 Reliability & safety “Throughout their lifecycle, AI systems should reliably operate in accordance with their intended purpose.” 37 43 (55.8) 2,15-17,20,21,25,27-32,34-36,38,40,42-50,52,53,55,56,58-61,63,66,68,70,71,79,81,85 Sustainability AI should be responsibly maintained and continue to produce benefit over time with minimal harm to the environment and health system. 2,20,61 16 (20.8) 2,20,40,43-45,48,50,61,63,66,68,71,74,81,86 Transparency “There should be transparency and responsible disclosure so people can understand when they are being significantly impacted by AI, and can find out when an AI system is engaging with them”. 37 Includes explainability, interpretability, and traceability/auditability. 61 (79.2) 2,4,15-17,20,22,25,27-30,32-36,38,40-45,47-58,60-72,74-79,84,86-90 Trust “References to trust include calls for trustworthy AI research and technology, trustworthy AI developers and organizations, trustworthy ‘design principles’, or underline the importance of customers’ trust” 19 . 23 (29.9) 2,17,20-22,29,32,36,40,46,48,49,54-56,60,63,74,82,86-88,91 Governance principles Accuracy & model performance “Accuracy describes evaluation based on how many right or wrong decisions a system makes. Example measurements are algorithmic accuracy, area under the curve (AUC) values, F1 scores, recall (sensitivity), precision, and specificity” 80 AI models should demonstrate accuracy and perform in line with its intended purpose. 55 (71.4) 2,7,15,16,20,25,27-36,38,41,43-48,51,52,55,57,59,60,62,63,65,67,69,71-75,77-88,90,92,93 Appropriateness & relevance AI should be the best solution to the identified problem, demonstrating fitness-for-purpose, superiority to non-AI solutions, and usefulness. 47 (61.0) 4,7,16,20,21,25,28-32,34,36,38,39,42,44-49,51,52,60-63,65-67,69,71-74,78-82,84,85,89,90,93,94 Awareness, education, & training System users should receive appropriate training and education to interact with the system as intended. 38 (49.4) 2,4,16,17,20,27-30,33-36,38,43,44,46,48-52,54,56,62-66,71,73,74,76,77,81,88,91,93 Data selection & management Data needs to be selected, handled, and governed safely and responsibly. Appropriate practices and protocols should be established and enforced. 57 (74.0) 2,15,16,20,27,29-36,38,41-48,51,52,54,55,57-63,65-69,71-81,83-90 Financial considerations Consideration of requirements for the AI system’s funding, cost-effectiveness, and commercialisation. 30 (39.0) 16,17,21,25,27-31,35,42,43,45,48,51,52,60,62,65,66,68,71,72,74,81,84-86,92,93 Interoperability “Data interoperability refers to the ability to accurately interpret data that is exchanged between different systems or organizations. It involves ensuring that the data has clear and unambiguous meaning, is correctly mapped, and is formatted in the required form” 95 . 16 (20.8) 16,21,25,29,35,43,48,49,52,57,59,60,66,68,70,83 Ongoing monitoring & maintenance AI systems should be monitored and maintained continuously across the life cycle. Includes risk management, appropriate reporting, and plans for decommissioning. 54 (70.1) 2,4,7,15,16,25,27-36,38,41-48,50-52,55-57,59-65,68,71-74,76-78,81-83,85,86,88-90 Organisational integration “The organisational aspects of deploying the AI model and workflow.” 28 Includes leadership, policy and strategy, and resources. 31 (39.0) 4,15,16,20,27-31,33-36,38,42,45-48,50,51,54,55,64,66,71,74,81,82,86,89,91,92 Regulatory compliance AI development and use should comply with legal, regulatory, contractual, or human rights laws and guidelines, e.g. privacy and confidentiality law, medical device regulations, clinical guidelines. 41 (53.2) 2,4,15,16,25,27,29-31,33-36,38,43-45,48,51-53,57,59-61,63,64,66,68-70,72,74,75,80-83,85,90,93 Stakeholder engagement The active involvement of stakeholders in decision making processes, e.g. inclusion of multidisciplinary stakeholder perspectives in the design, implementation, or evaluation of AI solutions. 35 (45.5) 2,15,16,20,29,32,34-36,38,42,47,48,50,51,53,55,59-61,63,64,66,72-75,77,80,81,83,84,90,92,93 Technical integration Integration of the AI tool into existing technical infrastructure (e.g. Electronic Health Records (EHR)). 29 (37.7) 27-30,35,36,43,46,48,52,54,57,59,60,62,65,66,72-75,78,79,85,88,89,92-94 Training & validation AI models should be trained and validated on appropriate and representative data, following responsible processes and practices. 45 (58.4) 2,16,17,25,28-36,38,41,45-49,51,52,57,60,62,63,65,67,68,71-73,77-80,83-87,89,90,92,93 Usability & adoption When integrating an AI solution, it should be designed to ensure it is intuitive, accessible, learnable, efficient, and user-centric. Usability has direct impacts on user experience, satisfaction, and uptake. 35 (45.5) 16,17,21,25,29,30,34-36,42,43,47,48,50,52,57,59,62,65,68,70,71,74,75,77,79-81,84,85,88,89,92-94 Workflow integration Integration of the AI tool into existing clinical workflows with minimal disruption. May also include workflow redesign to align with new needs of the AI tool and its users. 39 (50.6) 4,7,21,27-32,34-36,38,39,42,43,46-48,52,54,55,57,62,65-67,71,73-75,77,78,81,82,85,88,89,94 Component 2: Assessment method The second key component was a method to obtain information about AI tools and assess their alignment to predefined standards. This was included in 50 (64.9%) frameworks and typically comprised of a set of conceptual questions, or for more practical frameworks, a checklist or form to operationalise guiding principles into actionable practices or assessable criteria. For example, to assess fairness, the ABCDS framework required submission of a bias analysis and management plan with results of AI tool performance across subgroups. 30 Frameworks like the ABCDS framework, 30 FURM assessment, 28 and Clinical-AI-Predictive Analytics model review form 31 used formats similar to an ethics submission form, where developers or implementers must provide details about proposed AI tools. This method was often used in frameworks which were operational in organisations with an oversight committee who would use the form to review AI tools. Other frameworks 22,60,68,91,93 proposed a checklist structure, where the presence or absence of criteria are indicated in a checkbox or binary yes/no style. Some frameworks 66,79,84 included a scoring method where the AI tool’s consideration of each principle is given a numerical value (e.g. between 0-5, with 0 indicating no evidence of consideration and 5 indicating evidence of extensive consideration of a principle). A total score for the AI tool overall and/or multiple scores for different domains were then calculated. Checklists and scoring methods were often used for self-assessment by relevant stakeholders (e.g. researchers, developers, or implementers). Component 3: AI life cycle stages The third key component was a review timeline or consideration of the AI life cycle stages which was included in 35 (45.5.%) frameworks. Effective governance acknowledges the changing requirements and challenges across different stages of the AI life cycle. Frameworks assigned different recommendations to different stages of AI development (e.g. FUTURE-AI framework 48 ) or adopted a multi-stage review process requiring submission of new materials at different stages as AI tools were further developed and evaluated (e.g. ABCDS framework and FURM assessments 28,30 ). Five key life cycle stages are summarised in the following sections. Stage 1: Problem identification This involves identifying and defining the issue and planning an AI-driven solution. Problems may be identified by clinicians who raise pain points to organisational leaders or by leaders who identify opportunities for growth. 46 This typically involved: Identifying and engaging with stakeholders to define problems 29,34,42,46,48,55,56,60,82,83,90 Assessing whether AI is an appropriate solution 28,32,34,36,56,57,60,61,82 Defining use cases and potential impacts of solutions 28,29,48,55-57,59,72,82,85 Evaluating resource requirements and feasibility 28,29,48,55,60,82 Preliminary ethics and/or risk assessments 28,34,45,48,59-61,63 Once the problem has been defined, and an AI tool has been deemed an appropriate solution, a decision needs to be made to either procure an existing AI tool or develop a new one (in-house or in partnership with external developers). 29,46,60 Clearly defining and understanding problems can ensure AI tools that are procured or developed are purpose-driven and not merely guided by technological trends or availability. 29,42,82 Stage 2: Design and development This involves development or procurement of suitable AI tools. Though each method of obtaining an AI tool may involve case-specific activities, this stage typically involves the consideration of technical and workflow requirements, the selection and preparation of data, and ensuring tool design is tailored to its intended use. 29,35,45,60,83 Ethics principles such as privacy, security, autonomy, and fairness should be embedded during design. 29,63 Risk management, monitoring, and contestability plans should also be developed. 29 Stage 3: Training, validation, and evaluation This involves training, validation, and evaluation of AI tools, and ensures that ethics and governance principles are effectively applied prior to clinical or administrative use. For traditional machine learning, AI models need to be trained and validated in an iterative process using appropriate and representative data to ensure the model is refined according to its purpose, performs as intended, and does not create or perpetuate any existing bias. Evaluation often involves a series of studies from silent or small-scale evaluation to real-world or clinical studies. 29,30,62,67,85 Outcome measures may include statistical outcomes like accuracy and reliability, clinical outcomes, usability and adoption, cost-analyses, or ethics analyses (e.g. fairness evaluation). 21,29,30,32,41,48,52,57,67,72,85 Performance metrics should be carefully selected so they are relevant to the given task and models can be reliably evaluated. 2,32,52,72,79 AI tool performance and evaluation should be reported in line with established reporting standards such as MINIMAR, 87 DECIDE-AI, 77 CONSORT-AI, 78 TRIPOD-AI, STARD-AI, SPIRIT-AI. 62 Stage 4: Implementation and integration This is the stage at which AI tools are deployed, and often involves: Technical integration: integration of AI tools into existing technical infrastructure and workflows (e.g. with EHRs) and changes to technical infrastructure. 28,65,85,89 Clinical workflow integration: integration of AI tools into existing workflows, which can include workflow redesign and introduction or removal of key roles. 42,57 Effective integration often involves extensive stakeholder engagement and regular feedback. Organisational integration and governance: ensures organisations have the appropriate routines, competencies, and resources 4 to successfully implement AI tools, and that implementation is in line with organisational structures, systems, and values. 29,55,81,86,91 This includes the establishment of leadership and management structures, such as an oversight committee and local champions. This may also involve new policies and procedures that must be communicated to all stakeholders. 29,42,46,81 Training and education ensures that all stakeholders build literacy for safe and appropriate use. 2,16,48,50,61 Stage 5: Monitoring and maintenance This stage involves ongoing oversight of AI tool performance post-deployment, including processes for risk or error identification and management, and mechanisms for updating or decommissioning tools when appropriate. 46,47,56 An effective monitoring strategy should encompass a range of indicators, such as the monitoring of technical components, clinical outcomes, unintended consequences and harms, privacy and transparency concerns, and feedback on usability and acceptance. 35,46,48 Component 4: Oversight mechanism Mechanisms for implementing and managing AI governance were only included by 15 (19.5%) frameworks. While others provided guidance on their use by suggesting target audience/s or assigning stakeholders to different tasks, they did not provide guidance on management and organisational oversight. When included, the most common approach was a specific AI governance group which tested or used the framework. Three frameworks 30,31,36 reported establishing a specific AI governance group and six 27,33,42,50,54,89 recommended establishing one. All emphasised the need for diverse, multidisciplinary membership by ensuring the governance group was representative of key stakeholders (e.g. clinical, ethics, legal, IT, operations, research, consumers). One framework 32 suggested mapping AI governance requirements onto existing organisational governance bodies and processes, instead of creating a new group. Another 28 was being used by an existing team in the organisation who expanded their scope of responsibility and introduced new governance processes for AI. One framework suggested a combination of local oversight and an independent review process, where AI tools are reviewed by an external third party at regular intervals. 29 Three frameworks 34,35,51 included organisational-level leadership and oversight as key components, but did not specify how this was facilitated. DISCUSSION Despite rapid development and extensive research, guidance for healthcare organisations to navigate ethical and organisational challenges, and ensure safe and responsible AI adoption remains limited. To address this gap and integrate AI approaches with existing healthcare governance, we conducted a scoping review and identified ethical and governance principles, along with four key components for AI governance frameworks: guiding principles; an assessment method; consideration of the AI lifecycle stages; and an oversight mechanism. There was extensive literature on AI governance principles, however, depending on the purpose and focus of the frameworks, principles ranged from ethics principles to principles focused on safe and responsible organisational processes. We proposed a distinction between ethics and governance principles, where governance principles are a separate category related to the organisational, regulatory, and technical factors that impact AI at the organisational level, such as data selection and management, regulatory compliance, integration, and stakeholder engagement. Existing frameworks often had a stronger focus on one category while lacking depth in the other, such as ethics only frameworks, or frameworks including broad ethics principles without further elaboration. (e.g. 35,86 ) Our distinction provides clarity on governance needs at the organisational level and narrows the gap between conceptual ethics frameworks and practical AI governance requirements. The most common ethics principles such as transparency, fairness and equity, privacy and security, accountability, and beneficence aligned closely with established AI ethics frameworks. 19,96 However, subthemes specifically relevant to healthcare, such as consent, confidentiality, and medicolegal liability were not commonly included. Noticeably, some critical ethics principles such as contestability, sustainability, and cultural safety were absent from more than three-quarters of the frameworks. The principle of contestability, which is critical in ensuring public trust in AI and upholding the principles of justice and autonomy, 97 was only included in 10 frameworks (13.0%). 2,27,29,34,42,44,56,61,63,66 All stakeholders, including clinicians, patients, and caregivers require assurance that efficient and accessible processes exist to lodge complaints and receive appropriate and adequate redress. The ability for consumers to contest decisions or the use of AI in care delivery acknowledges the potential fallibility of AI tools, the importance of autonomy in healthcare, and the necessity for regular human oversight or, where necessary, intervention. 97 If the process for contesting decisions and seeking redress is too difficult, justice is compromised and there is little opportunity for similar harms to be avoided in the future. 63 Sustainability was a key principle in only 16 (20.8%) frameworks. References to sustainability included health system sustainability, environmental sustainability, economic sustainability, and social sustainability more broadly. 2,20,40,61,71 Sustainability is a crucial concern in healthcare, and related issues like climate change pose an existential threat to humans, health systems, and health infrastructure. 98,99 For healthcare AI to be sustainable, organisations should be adequately resourced with the ability to fully integrate, maintain, and update tools over time. 61 Without consideration of sustainability in AI design and monitoring, potential benefits and harms of AI tools may be missed, including harms to human and environmental wellbeing, impacts on intergenerational equity, and workplace disruptions like job changes or losses. 19,20,61 Patient trust in the organisation may also be compromised, especially if sustainability continues to become a public priority. Measures to improve sustainability include the use of energy efficient AI, renewable energy, and collaboration to reduce redundancy. 100,101 Various tools and frameworks could also be used to address sustainability. 102-104 Although most frameworks included themes or discussion surrounding fairness, equity, and data bias, only three frameworks, two from New Zealand 36,51 and one from the USA 33 incorporated specific criteria regarding culturally safe and appropriate practices. The New Zealand frameworks were the only ones to specifically address governance concerns for Indigenous populations. Culturally appropriate AI is a relevant concern for many healthcare organisations globally, and Indigenous and culturally and linguistically diverse groups have expressed the need for better inclusion and consideration of cultural safety in AI governance, beyond issues with data or algorithmic bias. 105,106 For example, in the Australian context, Aboriginal and or Torres Strait Islander peoples still experience disparities in health outcomes and access due to the ongoing effects of colonisation and continued discrimination and exclusion. There are specific guidelines on data governance and Data Sovereignty 107,108 and culturally safe healthcare 109 for Aboriginal and or Torres Strait Islander peoples that require consideration when designing and implementing safe and inclusive AI. Incorporating Indigenous principles and leadership into AI governance in healthcare organisations has the potential to address national priorities for digital inclusion and health equity (e.g. National Agreement on Closing the Gap, Priority Reform 4 and Targets 1, 2, 14, 17 110 ) and promote further benefit from AI for both Indigenous and non-Indigenous populations. The approach adopted by Whittaker et al. 36 to develop a framework specific to the Aotearoa New Zealand context represents a thoughtful approach that could inform future work, particularly as it highlights the insufficiencies of existing international frameworks in addressing Māori data governance and guardianship (kaitiakitangi) rights. Their resulting framework and governance group embeds Māori data sovereignty principles, ethics principles (tikanga), and philosophies (matauranga) to honour Te Tiriti (the Treaty of Waitangi) and prioritise consumer needs. Whittaker et al. demonstrate the value of going beyond inclusion and consultation to fully integrate and embed Indigenous principles in their framework to ensure value and safety for all stakeholders. We also identified governance considerations such as commercialisation that were not commonly included in frameworks, despite their importance. 20,70,72 Only seven frameworks (10.6%) 20,30,36,70,72,82,86 included discussion on the commercialisation of AI. This may have been due to the overall lack of consideration of practical AI governance needs in the literature. When commercialisation is an organisational goal, it is important that agreements recognise institutional contributions (e.g. the provision of patient data), benefits are fairly distributed, intellectual property rights are defined, and lines of accountability are clear. 72 This includes considerations around ethics principles such as privacy and fairness, which may be sidelined during commercialisation. 20 Commercialisation may also impact consent. For example, patients may be comfortable with the use of their data for internal quality improvement but may desire a more formal informed consent process if their data were to be provided to an external organisation. Institutional needs and pathways for commercialisation should be clarified in the planning stages. 20,72 Some frameworks (e.g. 7,28,47,66 ) identified the potential limitations in their application to diverse use cases and technologies including generative AI. Five (6.5%) frameworks 22,53,63,89,93 focused specifically on generative AI governance, and one 21 focused on conversational agents incorporating generative AI, but had limited applicability to AI technologies more broadly. Generative AI presents new challenges and risks with its ability to generate original material in a variety of mediums (e.g. text, images, audio) as opposed to traditional uses of AI for tasks like classification, analysis, or prediction. 111 These risks include a greater potential for misuse, errors, and inaccuracies (e.g. hallucinations), less transparency around its processes, and more complexity surrounding liability. 112,113 As AI continues to evolve, organisations and regulatory bodies are finding it increasingly difficult to envision the extent of emerging technological capability and associated ethical and regulatory requirements. 114 A responsive and sustainable framework would anticipate the unique risks and challenges of generative AI and other emerging technologies. An earlier review on clinical AI implementation frameworks by van der Vegt et al. 35 found that most existing frameworks only included one of either theme-based (determinant) or stage-based (process) elements, which presented a critical gap in the availability of comprehensive and actionable AI implementation frameworks. Many of the 77 frameworks we examined addressed this gap by including both theme-based and stage-based elements. However, we found that most were still missing components that could enable their real-world implementation, and identified the additional components of an assessment method and oversight mechanism. Only 10 out of 77 (13.0%) 27-36 included all four of the key components we identified, with the least common being an oversight mechanism. Organisational oversight dictates how human oversight will be maintained, who has the decision-making power to approve or deny the development and/or deployment of new AI tools in the organisation, and where stakeholders can obtain expert advice. Having a clear organisational apparatus for AI governance and embedding standardised processes and policies allows organisations and relevant stakeholders to be confident that deployed AI tools have met an established standard that aligns with the organisation’s values and purpose. 30 This transparency in processes may also encourage ethical and responsible practices from the early stages and can help promote trust in AI. Furthermore, an oversight structure that includes membership from diverse fields (e.g. clinical, operational, legal/ethical, technical, research, consumers) facilitates consideration of a variety of relevant concerns, builds AI literacy and expertise, and ensures alignment across the organisation. 31,54 The most common practice or recommendation for an oversight mechanism was for an AI-specific governance group to oversee all AI projects and tools within the organisation. The primary role of these committees is to identify, review, and monitor AI tools. 115 Liao et al. 31 and Economou et al. 30 each describe versions of this approach, where there is a broader AI-specific governance committee that oversees the review process, and a secondary group/s that assist with the ongoing development of AI tools. University of Wisconsin Health 31 has an institution-level Clinical AI and Predictive Analytics Committee from which separate subcommittees are created to assist with specific AI projects on a case-by case basis. Similarly, at Duke Health, 30 the broader ABCDS Oversight Committee has three subcommittees (ABCDS Regulatory Subcommittee, ABCDS Evaluation Subcommittee, and ABCDS Implementation and Monitoring Subcommittee) overseeing specific checkpoints at each AI lifecycle stage. The dual approach allows these organisations to provide specificity and appropriate expertise for the wide variety of technologies (e.g. traditional versus generative AI) and use cases (e.g. operational versus clinical, or different clinical areas) that a multidisciplinary setting may need to consider, while maintaining a centralised process to ensure appropriate oversight and consistency across the organisation. This approach also allows for more effective risk stratification and resource allocation so benefits can be maximised. We did not find any frameworks that were systematically evaluated for their feasibility and effectiveness in facilitating safe and responsible AI. Though preliminary learnings and measures of success were recorded for some (e.g. number of successful deployments, retirements, and non-deployments; 31 number of AI models under the organisation’s governance portfolio; 30,116 use cases advancing to implementation 28 ), systematic evaluation of AI governance frameworks is needed to identify the barriers, enablers, gaps, and unintended impacts of implementing different frameworks on organisational, health, and patient experience outcomes. Future research involving systematic evaluation, including engagement with relevant stakeholders such as clinicians and consumers will aid in consolidating our understanding of best practice for AI governance. We aimed to conduct a comprehensive review of AI governance frameworks for healthcare organisations, however, there were several limitations. First, we excluded primary and home care which are integral to care delivery. Their focus on accessibility and community-centred care may have provided additional insights and limits the applicability of our findings to these contexts, despite the need for governance in areas such as home care, which is being revolutionised by AI. Second, we were limited to frameworks available in English. The key components and principles represent a mostly Western perspective, with most frameworks originating from North America, the UK, and Australia. Future research inclusive of frameworks in other languages may provide a broader or more culturally responsive view. Lastly, this review will likely miss relevant published work that could influence findings, especially as AI evolves rapidly and publications grow exponentially. It is also likely that we have missed internally developed frameworks that are not publicly or freely available. We sought to mitigate these limitations by updating the search close to the time of submission, and including multiple bibliographic databases and search methods. This review highlights both ethics and governance principles that healthcare organisations should consider to support safe and responsible AI implementation. We identified four key components of AI governance frameworks including guiding principles, an assessment method, AI lifecycle stages, and an oversight mechanism. The findings provide a sound foundation for healthcare organisations to develop locally tailored AI governance frameworks that are both comprehensive and practical. METHODS We used the framework proposed by Arksey and O'Malley 117 and advanced by Levac et al. 118 which recommends five-stages for scoping reviews: identifying the research question; identifying relevant studies; study selection; data charting; and collating, summarising, and reporting results. We developed a protocol prior to commencement, and used it to guide eligibility criteria, search strategies, data extraction, and overall objectives. Reporting was guided by the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist. 119 Eligibility criteria The eligibility criteria were adapted from van der Vegt et al. 35 and adjusted to widen the scope from implementation to the broader concept of governance (Table 4). Table 4. Eligibility criteria. Component Inclusions Exclusions Concept - Studies reporting on either proposed or enforced AI governance. - Where a framework is not proposed. - Frameworks that are not specific to AI or do not include a specific AI component. - Frameworks that are not about governance or an aspect of governance. - Specific technology/IT ‘frameworks’ rather than generic. Context - Proposed AI governance frameworks must be aimed to be implemented in acute care settings. Non-acute settings: - Primary care. - Home care. Participants - Targeting any health condition or patient group. Type of evidence - Published peer-reviewed or grey literature. - Not available in English. Information sources and search Bibliographic databases including MEDLINE, Embase, and Scopus were searched in April 2024 and updated in March 2025. Databases were selected in consultation with a clinical librarian. Appropriate search strings were developed by linking concept clusters relating to each component of the question (artificial intelligence, governance, and a healthcare setting) with Boolean operators “AND” and “OR” (Supplement S1). Searches were also conducted on Google Scholar, where the first 200 results were included for review based on guidance for grey literature searching. 120 Reference lists of key papers were searched for additional peer-reviewed and grey literature. Figure 1. PRISMA flowchart of the stages of the review. Selection of sources of evidence Search results were imported into Covidence, where a two-stage review process was conducted by a team of independent reviewers (SF, TJ, AW). 6,391 studies were retrieved from the searches (Figure 1). After removal of duplicates, 5,194 studies remained. An initial title and abstract screening was conducted by two reviewers to exclude non-relevant studies. The remaining studies then underwent a second screening of the full text by two reviewers. An additional 23 studies from citation searching and grey literature were included at full text screening, resulting in 315 studies to review. After exclusions, 77 frameworks were analysed. Conflicts were resolved by two-way consensus. Further exclusions were based on researcher consensus during data charting and extraction. Data extraction and synthesis Included studies were exported from Covidence to Excel for data charting and extraction by AW, with verification by SF. A data charting table was developed to ensure extraction of relevant information (Supplement S2). For each included study descriptive information about the framework including name, setting, purpose, and method of derivation were extracted. The following data were also examined: Structure : Framework structure was examined using a previously published classification 35 based on two of Nielsen’s five implementation framework categories 121 which we adapted to the context of governance: Theme-based (determinant), where frameworks are structured around determinants that influence governance outcomes (e.g. 16,53,61 ) Stage-based (process), where frameworks are structured by steps or stages (e.g. 45,83,90 ) We added a third classification (theme x stage) for frameworks with both stage-based and theme-based elements (e.g. 30,35,79 ). Key components : Upon initial reading and synthesis, we identified four components that form a practical AI governance framework: Set of guiding principles. Method to review or assess AI tools (e.g. questions, checklist items, supporting materials). Review timeline or consideration of the AI life cycle stages. Organisational oversight mechanism (e.g. governance committee). Key principles : A content analysis was conducted to determine the key AI ethics and governance principles in each framework. An inductive approach was applied, where principles were added to the matrix throughout analysis (Supplement S3, S4). Principles were marked as present only when included as a discrete framework component. A narrative synthesis then integrated findings into descriptive summaries for each of the four key components. Declarations COMPETING INTERESTS The authors have no competing interests to declare. Author Contribution AW and SF conceived this review, designed, and conducted the analysis with advice and input from FM. AW and SF selected the studies. AW drafted the manuscript with input from all authors. All authors provided revisions for intellectual content. All authors have approved the final manuscript. Acknowledgement We wish to acknowledge the contributions of Tamasha Jayawardena (TJ) who assisted with the study screening. DATA AVAILABILITY All data relevant to the analysis are included in the article and online supplementary material. There are no new data associated with this article. References Baig, M. A., Almuhaizea, M. A., Alshehri, J., Bazarbashi, M. S. & Al-Shagathrh, F. Urgent Need for Developing a Framework for the Governance of AI in Healthcare. Stud Health Technol Inform. 272 , 253-256 (2020). https://doi.org/10.3233/shti200542 Solanki, P., Grundy, J. & Hussain, W. Operationalising ethics in artificial intelligence for healthcare: A framework for AI developers. AI and Ethics. 3 , 223-240 (2023). https://doi.org/10.1007/s43681-022-00195-z Carter, S. M. et al. How should artificial intelligence be used in Australian health care? Recommendations from a citizens’ jury. Med J Aust. 220 , 409-416 (2024). https://doi.org/10.5694/mja2.52283 Novak, L. L. et al. Clinical use of artificial intelligence requires AI-capable organizations. JAMIA Open. 6 , ooad028 (2023). https://doi.org/10.1093/jamiaopen/ooad028 Lyell, D., Wang, Y., Coiera, E. & Magrabi, F. More than algorithms: an analysis of safety events involving ML-enabled medical devices reported to the FDA. J Am Med Inform Assoc. 30 , 1227-1236 (2023). https://doi.org/10.1093/jamia/ocad065 Labkoff, S. et al. Toward a responsible future: recommendations for AI-enabled clinical decision support. J Am Med Inform Assoc. 31 , 2730-2739 (2024). https://doi.org/10.1093/jamia/ocae209 McCradden, M. D., Joshi, S., Anderson, J. A. & London, A. J. A normative framework for artificial intelligence as a sociotechnical system in healthcare. Patterns. 4 , 100864 (2023). https://doi.org/10.1016/j.patter.2023.100864 Davenport, T. H. & Glaser, J. P. Factors governing the adoption of artificial intelligence in healthcare providers. Discov Health Syst. 1 , 4 (2022). https://doi.org/10.1007/s44250-022-00004-8 Maleki Varnosfaderani, S. & Forouzanfar, M. The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century. Bioengineering. 11 (2024). https://doi.org/10.3390/bioengineering11040337 Magrabi, F., Lyell, D. & Coiera, E. Automation in Contemporary Clinical Information Systems: a Survey of AI in Healthcare Settings. Yearb Med Inform. 32 , 115-126 (2023). https://doi.org/10.1055/s-0043-1768733 Batool, A., Zowghi, D. & Bano, M. AI governance: a systematic literature review. AI and Ethics. (2025). https://doi.org/10.1007/s43681-024-00653-w Papagiannidis, E., Mikalef, P. & Conboy, K. Responsible artificial intelligence governance: A review and research framework. The Journal of Strategic Information Systems. 34 , 101885 (2025). https://doi.org/10.1016/j.jsis.2024.101885 Morley, J., Floridi, L., Kinsey, L. & Elhalal, A. From What to How: An Initial Review of Publicly Available AI Ethics Tools, Methods and Research to Translate Principles into Practices. Sci Eng Ethics. 26 , 2141-2168 (2020). https://doi.org/10.1007/s11948-019-00165-5 Birkstedt, T., Minkkinen, M., Tandon, A. & Mäntymäki, M. AI governance: themes, knowledge gaps and future agendas. Internet Research. 33 , 133-167 (2023). https://doi.org/10.1108/INTR-01-2022-0042 Khanna, S., Khanna, I., Srivastava, S. & Pandey, V. AI Governance Framework for Oncology: Ethical, Legal, and Practical Considerations. Quarterly Journal of Computational Technologies for Healthcare. 6 , 1-26 (2021). Stogiannos, N. et al. Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK. Br J Radiol. 96 , 20221157 (2023). https://doi.org/10.1259/bjr.20221157 Čartolovni, A., Tomičić, A. & Lazić Mosler, E. Ethical, legal, and social considerations of AI-based medical decision-support tools: A scoping review. Int J Med Inform. 161 , 104738 (2022). https://doi.org/10.1016/j.ijmedinf.2022.104738 Goirand, M., Austin, E. & Clay-Williams, R. Implementing Ethics in Healthcare AI-Based Applications: A Scoping Review. Sci Eng Ethics. 27 , 61 (2021). https://doi.org/10.1007/s11948-021-00336-3 Jobin, A., Ienca, M. & Vayena, E. The global landscape of AI ethics guidelines. Nature Machine Intelligence. 1 , 389-399 (2019). https://doi.org/10.1038/s42256-019-0088-2 Siala, H. & Wang, Y. SHIFTing artificial intelligence to be responsible in healthcare: A systematic review. Social Science & Medicine. 296 , 114782 (2022). https://doi.org/10.1016/j.socscimed.2022.114782 Ding, H., Simmich, J., Vaezipour, A., Andrews, N. & Russell, T. Evaluation framework for conversational agents with artificial intelligence in health interventions: a systematic scoping review. J Am Med Inform Assoc. 31 , 746-761 (2023). https://doi.org/10.1093/jamia/ocad222 Ning, Y. et al. Generative artificial intelligence and ethical considerations in health care: a scoping review and ethics checklist. Lancet Digit Health. 6 , e848-e856 (2024). https://doi.org/10.1016/S2589-7500(24)00143-2 Fernando, M. et al. Using Theories, Models, and Frameworks to Inform Implementation Cycles of Computerized Clinical Decision Support Systems in Tertiary Health Care Settings: Scoping Review. J Med Internet Res. 25 , e45163 (2023). https://doi.org/10.2196/45163 Lehoux, P., Rivard, L., de Oliveira, R. R., Mörch, C. M. & Alami, H. Tools to foster responsibility in digital solutions that operate with or without artificial intelligence: A scoping review for health and innovation policymakers. Int J Med Inform. 170 , 104933 (2023). https://doi.org/10.1016/j.ijmedinf.2022.104933 Segur-Ferrer, J., Molto-Puigmarti, C., Pastells-Peiro, R. & Vivanco-Hidalgo, R. M. Methodological Frameworks and Dimensions to Be Considered in Digital Health Technology Assessment: Scoping Review and Thematic Analysis. J Med Internet Res. 26 , e48694 (2024). https://dx.doi.org/10.2196/48694 Munn, Z. et al. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Med Res Methodol. 18 , 143 (2018). https://doi.org/10.1186/s12874-018-0611-x Apfelbacher, T., Kocman, S. E., Prokosch, H. U. & Christoph, J. A Governance Framework for the Implementation and Operation of AI Applications in a University Hospital. Stud Health Technol Inform. 316 , 776-780 (2024). https://dx.doi.org/10.3233/SHTI240527 Callahan, A. et al. Standing on FURM ground: A framework for evaluating Fair, Useful, and Reliable AI Models in healthcare systems. NEJM Catal Innov Care Deliv. 5 (2024). https://doi.org/10.1056/CAT.24.0131 Coalition for Health AI (CHAI). Assurance Standards Guide and Reporting Checklist , https://chai.org/assurance-standards-guide/ (2024). Economou-Zavlanos, N. J. et al. Translating ethical and quality principles for the effective, safe and fair development, deployment and use of artificial intelligence technologies in healthcare. J Am Med Inform Assoc. 31 , 705-713 (2024). https://dx.doi.org/10.1093/jamia/ocad221 Liao, F., Adelaine, S., Afshar, M. & Patterson, B. W. Governance of Clinical AI applications to facilitate safe and equitable deployment in a large health system: Key elements and early successes. Front Digit Health. 4 , 931439 (2022). https://doi.org/10.3389/fdgth.2022.931439 Morley, J., Morton, C., Karpathakis, K., Taddeo, M. & Floridi, L. Towards a framework for evaluating the safety, acceptability and efficacy of AI systems for health: an initial synthesis. ArXiv. (2021). https://doi.org/10.48550/arXiv.2104.06910 Saenz, A. D. et al. Establishing responsible use of AI guidelines: a comprehensive case study for healthcare institutions. NPJ Digit Med. 7 , 348 (2024). https://doi.org/10.1038/s41746-024-01300-8 Singapore Ministry of Health. Artificial Intelligence in Healthcare Guidelines. 2021. https://www.moh.gov.sg/licensing-and-regulation/artificial-intelligence-in-healthcare. van der Vegt, A. H. et al. Implementation frameworks for end-to-end clinical AI: derivation of the SALIENT framework. J Am Med Inform Assoc. 30 , 1503-1515 (2023). https://dx.doi.org/10.1093/jamia/ocad088 Whittaker, R. et al. An example of governance for AI in health services from Aotearoa New Zealand. NPJ Digit Med. 6 , 164 (2023). https://doi.org/10.1038/s41746-023-00882-z Australian Government Department of Industry & Science and Resources. Australia’s AI Ethics Principles. (2022). https://www.industry.gov.au/publications/australias-artificial-intelligence-ethics-framework/australias-ai-ethics-principles. Australian Medical Association. Position Statement - Artificial Intelligence in Healthcare. 2023. https://www.ama.com.au/sites/default/files/2023-08/Artificial%20Intelligence%20in%20Healthcare%20-%20AMA.pdf. Bellini, V., Cascella, M., Montomoli, J. & Bignami, E. From Big Data's 5Vs to clinical practice's 5Ws: enhancing data-driven decision making in healthcare. J Clin Monit Comput. 37 , 1423-1425 (2023). https://dx.doi.org/10.1007/s10877-023-01007-3 Cawthorne, D. & Robbins-van Wynsberghe, A. An Ethical Framework for the Design, Development, Implementation, and Assessment of Drones Used in Public Healthcare. Sci Eng Ethics. 26 , 2867-2891 (2020). https://dx.doi.org/10.1007/s11948-020-00233-1 Char, D. S., Abramoff, M. D. & Feudtner, C. Identifying Ethical Considerations for Machine Learning Healthcare Applications. Am J Bioeth. 20 , 7-17 (2020). https://dx.doi.org/10.1080/15265161.2020.1819469 Dixit, A., Quaglietta, J. & Gaulton, C. Preparing for the future: How organizations can prepare boards, leaders, and risk managers for artificial intelligence. Healthc Manage Forum. 34 , 346-352 (2021). https://dx.doi.org/10.1177/08404704211037995 Jacob, C. et al. AI for IMPACTS Framework for Evaluating the Long-Term Real-World Impacts of AI-Powered Clinician Tools: Systematic Review and Narrative Synthesis. J Med Internet Res. 27 , e67485 (2025). https://dx.doi.org/10.2196/67485 Katirai, A. The ethics of advancing artificial intelligence in healthcare: analyzing ethical considerations for Japan's innovative AI hospital system. Front Public Health. 11 , 1142062 (2023). https://dx.doi.org/10.3389/fpubh.2023.1142062 Kim, J. et al. Developing a Framework for Self-regulatory Governance in Healthcare AI Research: Insights from South Korea. Asian Bioeth Rev. 1-16 (2024). https://doi.org/10.1007/s41649-024-00281-w Kim, J. Y. et al. in Proceedings of the 2023 ACM conference on fairness, accountability, and transparency (FAccT '23). 1396-1417. Kim, J. Y. et al. Development and preliminary testing of Health Equity Across the AI Lifecycle (HEAAL): A framework for healthcare delivery organizations to mitigate the risk of AI solutions worsening health inequities. PLOS Digit Health. 3 , e0000390 (2024). https://doi.org/10.1371/journal.pdig.0000390 Lekadir, K. et al. FUTURE-AI: International consensus guideline for trustworthy and deployable artificial intelligence in healthcare. ArXiv. (2023). https://doi.org/10.48550/arXiv.2309.12325 Mirbabaie, M., Hofeditz, L., Frick, N. R. J. & Stieglitz, S. Artificial intelligence in hospitals: providing a status quo of ethical considerations in academia to guide future research. AI Soc. 37 , 1361-1382 (2022). https://doi.org/10.1007/s00146-021-01239-4 Nasir, S., Khan, R. A. & Bai, S. Ethical Framework for Harnessing the Power of AI in Healthcare and Beyond. IEEE Access. 12 , 31014-31035 (2024). https://doi.org/10.1109/ACCESS.2024.3369912 New Zealand Ministry of Health. Emerging Health Technology: Introductory Guidance. (2019). https://www.tewhatuora.govt.nz/assets/Uploads/introductory_guidance_-_algorithms_v0.4_-_web.pdf. Omoumi, P. et al. To buy or not to buy—evaluating commercial AI solutions in radiology (the ECLAIR guidelines). Eur Radiol. 31 , 3786-3796 (2021). https://doi.org/10.1007/s00330-020-07684-x Oniani, D. et al. Adopting and expanding ethical principles for generative artificial intelligence from military to healthcare. NPJ Digit Med. 6 , 225 (2023). https://dx.doi.org/10.1038/s41746-023-00965-x Reddy, S., Allan, S., Coghlan, S. & Cooper, P. A governance model for the application of AI in health care. J Am Med Inform Assoc. 27 , 491-497 (2020). https://doi.org/10.1093/jamia/ocz192 Roski, J., Maier, E. J., Vigilante, K., Kane, E. A. & Matheny, M. E. Enhancing trust in AI through industry self-governance. J Am Med Inform Assoc. 28 , 1582-1590 (2021). https://dx.doi.org/10.1093/jamia/ocab065 Solomonides, A. E. et al. Defining AMIA's artificial intelligence principles. J Am Med Inform Assoc. 29 , 585-591 (2022). https://dx.doi.org/10.1093/jamia/ocac006 Taribagil, P., Hogg, H. D. J., Balaskas, K. & Keane, P. A. Integrating artificial intelligence into an ophthalmologist’s workflow: obstacles and opportunities. Expert Rev Ophthalmol. 18 , 45-56 (2023). https://doi.org/10.1080/17469899.2023.2175672 The Royal Australian and New Zealand College of Radiologists. Ethical Principles for AI in Medicine. 2023. https://www.ranzcr.com/search/ethical-principles-for-ai-in-medicine. UK National Health Services (NHS). Digital Technology Assessment Criteria (DTAC). 2021. https://transform.england.nhs.uk/key-tools-and-info/digital-technology-assessment-criteria-dtac/. US Department of Health and Human Services (DHHS). Trustworthy AI Playbook. 2021. https://www.hhs.gov/sites/default/files/hhs-trustworthy-ai-playbook.pdf. World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance., (World Health Organization, Geneva, 2021). https://www.who.int/publications/i/item/9789240029200. World Health Organization. Generating evidence for artificial intelligence-based medical devices: a framework for training, validation and evaluation., (World Health Organization, Geneva, 2021). https://www.who.int/publications/i/item/9789240029200. World Health Organization. Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. (World Health Organization, Geneva, 2024). https://www.who.int/publications/i/item/9789240084759. Abujaber, A. A. & Nashwan, A. J. Ethical framework for artificial intelligence in healthcare research: A path to integrity. World J Methodol. 14 , 94071 (2024). https://doi.org/10.5662/wjm.v14.i3.94071 Brady, A. P. et al. Developing, purchasing, implementing and monitoring AI tools in radiology: Practical considerations. A multi-society statement from the ACR, CAR, ESR, RANZCR & RSNA. J Med Imaging Radiat Oncol. 68 , 7-26 (2024). https://doi.org/10.1111/1754-9485.13612 Lehoux, P. et al. A Comprehensive, Valid, and Reliable Tool to Assess the Degree of Responsibility of Digital Health Solutions That Operate With or Without Artificial Intelligence: 3-Phase Mixed Methods Study. J Med Internet Res. 25 , e48496 (2023). https://doi.org/10.2196/48496 McCradden, M. D. et al. A Research Ethics Framework for the Clinical Translation of Healthcare Machine Learning. Am J Bioeth. 22 , 8-22 (2022). https://doi.org/10.1080/15265161.2021.2013977 Nebeker, C., Torous, J. & Bartlett Ellis, R. J. Building the case for actionable ethics in digital health research supported by artificial intelligence. BMC Med. 17 , 137 (2019). https://doi.org/10.1186/s12916-019-1377-7 The Australasian College of Dermatologists. Position Statement - Use of Artificial Intelligence in Dermatology in Australia. 2022. https://www.dermcoll.edu.au/wp-content/uploads/2022/11/ACD-Position-Statement-Use-of-Artifical-Intelligence-in-Dermatology-in-Australia-Nov-2022.pdf. UK Government Department of Health and Social Care. A guide to good practice for digital and data-driven health technologies. 2021. https://www.gov.uk/government/publications/code-of-conduct-for-data-driven-health-and-care-technology/initial-code-of-conduct-for-data-driven-health-and-care-technology. UK Government National Institute for Health and Care Excellence (NICE). Evidence standards framework (ESF) for digital health technologies. 2022. https://www.nice.org.uk/corporate/ecd7/resources/evidence-standards-framework-for-digital-health-technologies-pdf-1124017457605. Vollmer, S. et al. Machine learning and artificial intelligence research for patient benefit: 20 critical questions on transparency, replicability, ethics, and effectiveness. BMJ. 368 , l6927 (2020). https://doi.org/10.1136/bmj.l6927 Yan, A. P. et al. A roadmap to implementing machine learning in healthcare: from concept to practice. Front Digit Health. 7 (2025). https://doi.org/10.3389/fdgth.2025.1462751 Hassan, M., Borycki, E. M. & Kushniruk, A. W. Artificial intelligence governance framework for healthcare. Healthc Manage Forum. 38 , 125-130 (2024). https://dx.doi.org/10.1177/08404704241291226 Kawamoto, K., Finkelstein, J. & Del Fiol, G. Implementing Machine Learning in the Electronic Health Record: Checklist of Essential Considerations. Mayo Clin Proc. 98 , 366-369 (2023). https://dx.doi.org/10.1016/j.mayocp.2023.01.013 Nwosu, O. I., Crowson, M. G. & Rameau, A. Artificial Intelligence Governance and Otolaryngology-Head and Neck Surgery. Laryngoscope. 133 , 2868-2870 (2023). https://doi.org/10.1002/lary.31013 Vasey, B. et al. Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. BMJ. 377 , e070904 (2022). https://doi.org/10.1136/bmj-2022-070904 Liu, X. et al. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Lancet Digit Health. 2 , e537-e548 (2020). https://doi.org/10.1016/S2589-7500(20)30218-1 Reddy, S. et al. Evaluation framework to guide implementation of AI systems into healthcare settings. BMJ Health Care Inform. 28 (2021). https://dx.doi.org/10.1136/bmjhci-2021-100444 Bertl, M., Ross, P. & Draheim, D. Systematic AI support for decision-making in the healthcare sector: Obstacles and success factors. Health Policy and Technology. 12 , 100748 (2023). https://doi.org/10.1016/j.hlpt.2023.100748 Cresswell, K., Williams, R. & Sheikh, A. Developing and applying a formative evaluation framework for health information technology implementations: Qualitative investigation. J Med Internet Res. 22 , 15068 (2020). https://dx.doi.org/10.2196/15068 UK National Health Services (NHS). A Buyer's Guide to AI in Health and Care. 2020. https://transform.england.nhs.uk/ai-lab/explore-all-resources/adopt-ai/a-buyers-guide-to-ai-in-health-and-care/. Van de Sande, D. et al. Developing, implementing and governing artificial intelligence in medicine: a step-by-step approach to prevent an artificial intelligence winter. BMJ Health Care Inform. 29 (2022). https://doi.org/10.1136/bmjhci-2021-100495 Marconi, L., Pirovano, E. & Cabitza, F. CLARITY AI: A Comprehensive Checklist Integrating Established Frameworks for Enhanced Research Quality in Medical AI Studies. Bolzano, Italy, 2025. Sendak, M. P. et al. A Path for Translation of Machine Learning Products into Healthcare Delivery. Euro Med J Innov. (2020). https://doi.org/10.33590/emjinnov/19-00172 Truong, T., Gilbank, P., Johnson-Cover, K. & Ieraci, A. A Framework for Applied AI in Healthcare. Stud Health Technol Inform. 264 , 1993-1994 (2019). https://dx.doi.org/10.3233/SHTI190751 Hernandez-Boussard, T., Bozkurt, S., Ioannidis, J. P. A. & Shah, N. H. MINIMAR (MINimum Information for Medical AI Reporting): Developing reporting standards for artificial intelligence in health care. J Am Med Inform Assoc. 27 , 2011-2015 (2020). https://dx.doi.org/10.1093/jamia/ocaa088 Kwong, J. C. C., Nickel, G. C., Wang, S. C. Y. & Kvedar, J. C. Integrating artificial intelligence into healthcare systems: more than just the algorithm. NPJ Digit Med. 7 (2024). https://doi.org/10.1038/s41746-024-01066-z Reddy, S. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. Implement Sci. 19 , 27 (2024). https://dx.doi.org/10.1186/s13012-024-01357-9 Wiens, J. et al. Do no harm: a roadmap for responsible machine learning for health care. Nat Med. 25 , 1337-1340 (2019). https://doi.org/10.1038/s41591-019-0548-6 Smith, T. G. et al. Creating a Practical Transformational Change Management Model for Novel Artificial Intelligence-Enabled Technology Implementation in the Operating Room. Mayo Clin Proc Innov Qual Outcomes. 6 , 584-596 (2022). https://dx.doi.org/10.1016/j.mayocpiqo.2022.09.004 Chae, A. et al. Strategies for Implementing Machine Learning Algorithms in the Clinical Practice of Radiology. Radiology. 310 (2024). https://doi.org/10.1148/radiol.223170 Raza, M. M., Venkatesh, K. P. & Kvedar, J. C. Generative AI and large language models in health care: pathways to implementation. NPJ Digit Med. 7 , 62 (2024). https://doi.org/10.1038/s41746-023-00988-4 Salwei, M. E. et al. Workflow integration analysis of a human factors-based clinical decision support in the emergency department. Applied Ergonomics. 97 , 103498 (2021). https://dx.doi.org/10.1016/j.apergo.2021.103498 Belete, G. F., Voinov, A. & Laniak, G. F. An overview of the model integration process: From pre-integration assessment to testing. Environmental Modelling & Software. 87 , 49-63 (2017). https://doi.org/10.1016/j.envsoft.2016.10.013 Beauchamp, T. L. & Childress, J. F. Principles of biomedical ethics . Eighth edn, (Oxford University Press, 2019). Alfrink, K., Keller, I., Kortuem, G. & Doorn, N. Contestable AI by Design: Towards a Framework. Minds and Machines. 33 , 613-639 (2023). https://doi.org/10.1007/s11023-022-09611-z World Health Organization. Climate change , https://www.who.int/news-room/fact-sheets/detail/climate-change-and-health (2023). Coiera, E. & Magrabi, F. What did you do to avoid the climate disaster? A call to arms for health informatics. J Am Med Inform Assoc. 29 , 1997-1999 (2022). https://doi.org/10.1093/jamia/ocac185 Doo, F. X. et al. Environmental Sustainability and AI in Radiology: A Double-Edged Sword. Radiology. 310 , e232030 (2024). https://doi.org/10.1148/radiol.232030 Braithwaite, J. et al. Strategies and tactics to reduce the impact of healthcare on climate change: systematic review. BMJ. 387 , e081284 (2024). https://doi.org/10.1136/bmj-2024-081284 Anthony, L. F. W., Kanding, B. & Selvan, R. Carbontracker: Tracking and predicting the carbon footprint of training deep learning models. ArXiv. (2020). https://doi.org/10.48550/arXiv.2007.03051 Thelisson, E., Mika, G., Schneiter, Q., Padh, K. & Verma, H. Toward Responsible AI Use: Considerations for Sustainability Impact Assessment. ArXiv. (2023). https://doi.org/10.48550/arXiv.2312.11996 Rohde, F. et al. Broadening the perspective for sustainable artificial intelligence: sustainability criteria and indicators for Artificial Intelligence systems. Current Opinion in Environmental Sustainability. 66 , 101411 (2024). https://doi.org/10.1016/j.cosust.2023.101411 Barrowcliffe, R. et al. Envisioning Aboriginal and Torres Strait Islander AI Futures Communique: March 2025. Journal of Global Indigeneity. 9 (2025). https://doi.org/10.54760/001c.133656 Lewis, J. E. Indigenous Protocol and Artificial Intelligence Position Paper. (The Initiative for Indigenous Futures and the Canadian Institute for Advanced Research (CIFAR). Honolulu, Hawai'i, 2020). Commonwealth of Australia. Framework for Governance of Indigenous Data. NIAA, (2024). https://www.niaa.gov.au/resource-centre/framework-governance-indigenous-data. Maiam nayri Wingara Indigenous Data Sovereignty Collective. Taking Control of Our Data: A Discussion Paper on Indigenous Data Governance for Aboriginal and Torres Strait Islander People and Communities. (Lowitja Institute, Melbourne, 2024). https://www.lowitja.org.au/resource/taking-control-of-our-data-a-discussion-paper-on-indigenous-data-governance-for-aboriginal-and-torres-strait-islander-people-and-communities/. De Zilva, S., Walker, T., Palermo, C. & Brimblecombe, J. Culturally safe health care practice for Indigenous Peoples in Australia: A systematic meta-ethnographic review. J Health Serv Res Policy. 27 , 74-84 (2022). https://doi.org/10.1177/13558196211041835 National Agreement on Closing the Gap , https://www.closingthegap.gov.au/national-agreement. Coiera, E. & Fraile-Navarro, D. AI as an Ecosystem — Ensuring Generative AI Is Safe and Effective. NEJM AI. 1 , AIp2400611 (2024). https://doi.org/10.1056/AIp2400611 Duffourc, M. & Gerke, S. Generative AI in Health Care and Liability Risks for Physicians and Safety Concerns for Patients. JAMA. 330 , 313-314 (2023). https://doi.org/10.1001/jama.2023.9630 Moulaei, K. et al. Generative artificial intelligence in healthcare: A scoping review on benefits, challenges and applications. Int J Med Inform. 188 , 105474 (2024). https://doi.org/10.1016/j.ijmedinf.2024.105474 The Lancet Regional Health –, E. Embracing generative AI in health care. The Lancet Regional Health – Europe. 30 (2023). https://doi.org/10.1016/j.lanepe.2023.100677 Parker, V. J., Economou, N. J. & Silcox, C. AI Governance in Health Systems: Aligning Innovation, Accountability, and Trust. Duke-Margolis Institute for Health Policy, 2024. https://healthpolicy.duke.edu/publications/ai-governance-health-systems-aligning-innovation-accountability-and-trust. Bedoya, A. D. et al. A framework for the oversight and local deployment of safe and high-quality prediction models. J Am Med Inform Assoc. 29 , 1631-1636 (2022). https://doi.org/10.1093/jamia/ocac078 Arksey, H. & and O'Malley, L. Scoping studies: towards a methodological framework. Int J Soc Res Methodol. 8 , 19-32 (2005). https://doi.org/10.1080/1364557032000119616 Levac, D., Colquhoun, H. & O'Brien, K. K. Scoping studies: advancing the methodology. Implement Sci. 5 , 69 (2010). https://doi.org/10.1186/1748-5908-5-69 Tricco, A. C. et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. 169 , 467-473 (2018). https://doi.org/10.7326/m18-0850 Haddaway, N. R., Collins, A. M., Coughlin, D. & Kirk, S. The Role of Google Scholar in Evidence Reviews and Its Applicability to Grey Literature Searching. PLoS One. 10 , e0138237 (2015). https://doi.org/10.1371/journal.pone.0138237 Nilsen, P. Making sense of implementation theories, models and frameworks. Implement Sci. 10 , 53 (2015). https://doi.org/10.1186/s13012-015-0242-0 Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformationFileGovernanceforsafeandreponsibleAIScRnpj.pdf SupplementaryDataS2DetailsofIncludedStudies.xlsx SupplementaryDataS3EthicsPrinciplesMapping.xlsx SupplementaryDataS4GovernancePrinciplesMapping.xlsx SupplementaryDataS5KeyComponentsMapping.xlsx Cite Share Download PDF Status: Published Journal Publication published 01 May, 2026 Read the published version in npj Digital Medicine → Version 1 posted Editorial decision: Revision requested 10 Sep, 2025 Reviews received at journal 07 Sep, 2025 Reviews received at journal 03 Sep, 2025 Reviewers agreed at journal 26 Aug, 2025 Reviewers agreed at journal 26 Aug, 2025 Reviewers invited by journal 11 Aug, 2025 Editor assigned by journal 01 Aug, 2025 Submission checks completed at journal 01 Aug, 2025 First submitted to journal 24 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7209096","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":500360493,"identity":"1739b400-721e-4e76-b473-e8218218abc5","order_by":0,"name":"Amy Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIiWNgGAWjYDADfjDJRooWyQaStRgcIFYL/7QzBsw8NXfsNl87vIHhQ9lhBv4ZCfi1SNzOAWo59ix52+20AsYZ5w4zSNwgoIUBpCWH7XCyGYjB23aYgYGQFnmwln+Hk41nAxl/gVrkCWkxAGnJbTtsZyANZDACtRgQ0mII9MLhv32HEySAjIM959J5DM88wK9F7nbyxoczvh2255+dvPHBjzJrObnjBGxhYOAAx0hiAzRqeAipBwJ2sDvsgdiACNWjYBSMglEwEgEA/YdHQvVyDUMAAAAASUVORK5CYII=","orcid":"","institution":"Australian Institute of Health Innovation","correspondingAuthor":true,"prefix":"","firstName":"Amy","middleName":"","lastName":"Wang","suffix":""},{"id":500360494,"identity":"ac67f2ef-564b-413b-8918-b548bdfb872a","order_by":1,"name":"Sam Freeman","email":"","orcid":"","institution":"Australian Institute of Health Innovation","correspondingAuthor":false,"prefix":"","firstName":"Sam","middleName":"","lastName":"Freeman","suffix":""},{"id":500360495,"identity":"77fe1bdc-25c0-4635-912a-e04b49f60e0f","order_by":2,"name":"Farah Magrabi","email":"","orcid":"","institution":"Australian Institute of Health Innovation","correspondingAuthor":false,"prefix":"","firstName":"Farah","middleName":"","lastName":"Magrabi","suffix":""}],"badges":[],"createdAt":"2025-07-25 00:08:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7209096/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7209096/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41746-026-02679-2","type":"published","date":"2026-05-01T15:57:27+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89404796,"identity":"a777334f-ff0b-4bdc-9723-3cb9d03ce328","added_by":"auto","created_at":"2025-08-19 14:59:19","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":164990,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA flowchart of the stages of the review.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/a947737f8be36d5d24c38052.png"},{"id":108437601,"identity":"2f1076e4-ab2a-45b2-85d9-80b312e374c4","added_by":"auto","created_at":"2026-05-04 16:00:00","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":864267,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/6940d3ec-60fe-4cf6-95d8-0289d07e466d.pdf"},{"id":89405079,"identity":"97b9e345-89a8-4a17-8e8d-c3abc9a028f8","added_by":"auto","created_at":"2025-08-19 15:07:19","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":93549,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformationFileGovernanceforsafeandreponsibleAIScRnpj.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/25faf940fa7f43d037977856.pdf"},{"id":89404797,"identity":"4e8f6a8d-0071-4cba-a508-3d22380b1daa","added_by":"auto","created_at":"2025-08-19 14:59:19","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":67617,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryDataS2DetailsofIncludedStudies.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/a01e07f6e1462e53ca788923.xlsx"},{"id":89404803,"identity":"7f460829-cdbb-45ef-9f55-75c59874bc99","added_by":"auto","created_at":"2025-08-19 14:59:19","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":37337,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryDataS3EthicsPrinciplesMapping.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/50d80f0f0633ba3b70733d0e.xlsx"},{"id":89404801,"identity":"e5a68a47-eae6-4586-9566-b864ea6a809f","added_by":"auto","created_at":"2025-08-19 14:59:19","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":33275,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryDataS4GovernancePrinciplesMapping.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/8f81989cd14b87172ea3b9fb.xlsx"},{"id":89406103,"identity":"efad3934-282c-4a7a-9c36-9febcf684205","added_by":"auto","created_at":"2025-08-19 15:15:19","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":32499,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryDataS5KeyComponentsMapping.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7209096/v1/bc995cc5b5ddee68f4a41ac9.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Governance for safe and responsible AI in healthcare organisations: A scoping review of frameworks","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eEffective governance of artificial intelligence (AI) in healthcare organisations is critical to facilitating its safe and responsible implementation.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e However, attempts to conceptualise AI governance are fragmented, and many proposed ethical and governance frameworks have not been operationalised into actionable practices or evaluated in real-world healthcare settings.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e The lack of cohesion and real-world applicability has made it challenging for healthcare organisations to safely navigate the rapidly expanding technological capabilities of AI and associated ethical and regulatory concerns,\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e highlighting the need for more comprehensive and practical guidance on AI governance.\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e Current understanding lacks a structured, holistic approach that recognises AI tools as one part of a larger sociotechnical system, integrating concerns regarding the technological product,\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e ethical principles, organisational processes,\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e and the regulatory landscape.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e This has hindered AI adoption despite rapid advancements in research and technology,\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e limiting the transformative potential of AI tools on patient care across both clinical and non-clinical tasks, such as predictive analysis, diagnostic assistance, or optimisation of administrative processes.\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003ePrevious reviews have examined AI governance principles and practices broadly across industries,\u003csup\u003e\u003cspan additionalcitationids=\"CR12 CR13\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e or in specific health disciplines.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e Others have focused only on a specific aspect of governance such as ethics principles,\u003csup\u003e\u003cspan additionalcitationids=\"CR18 CR19\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e on specific AI technologies or applications,\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e or on digital health technology more broadly, which may or may not include AI.\u003csup\u003e\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e To date, no review has examined the frameworks governing AI within healthcare organisations. Healthcare represents a distinctive context where sector-agnostic frameworks may not adequately consider the rigorous ethical and regulatory requirements unique to healthcare, while discipline- or technology-specific frameworks may not facilitate governance that covers the wide variety of AI use cases relevant to multidisciplinary healthcare organisations. To address this gap, we conducted a scoping review to provide an overview of the state of AI governance in healthcare, with a specific focus on governance at the organisational level. Scoping reviews are a suitable method for identifying key characteristics related to a concept, particularly for emerging areas such as AI governance.\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e We sought to identify the key ethics and governance principles, and the components of practical AI governance frameworks for safe and responsible AI in healthcare organisations.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003eOverview of frameworks\u003c/p\u003e\n\u003cp\u003eWe identified 77 frameworks for AI governance in healthcare organisations (Table 1; Supplement S2). Almost half were published in 2023 or 2024 (n=36, 46.8%), and the most common country of origin was the USA (n=22, 28.6%). Most frameworks (n=56, 72.7%) were designed for application across general healthcare settings (i.e. any health field), with a primary focus on evaluation (n=18, 23.4%) or overall governance (n=16, 20.8%). Frameworks were mostly derived from the literature (n=38, 49.4%) and typically structured by themes or principles (n=39, 50.6%). There was no evidence of systematic evaluation of any frameworks on organisational or health outcomes.\u003c/p\u003e\n\u003cp\u003eTable 1. Characteristics of AI ethics and governance frameworks identified in the scoping review.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristic\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFrequency (n=77)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eType of publication\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eAcademic\u003c/p\u003e\n \u003cp\u003eGovernment\u003c/p\u003e\n \u003cp\u003eIndustry\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e81.8\u003c/p\u003e\n \u003cp\u003e13.0\u003c/p\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYear published\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e2019\u003c/p\u003e\n \u003cp\u003e2020\u003c/p\u003e\n \u003cp\u003e2021\u003c/p\u003e\n \u003cp\u003e2022\u003c/p\u003e\n \u003cp\u003e2023\u003c/p\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003cp\u003e2025\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003cp\u003e11.7\u003c/p\u003e\n \u003cp\u003e16.9\u003c/p\u003e\n \u003cp\u003e15.6\u003c/p\u003e\n \u003cp\u003e22.1\u003c/p\u003e\n \u003cp\u003e24.7\u003c/p\u003e\n \u003cp\u003e3.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCountry of origin\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003cp\u003eUK\u003c/p\u003e\n \u003cp\u003eCanada\u003c/p\u003e\n \u003cp\u003eAustralia\u003c/p\u003e\n \u003cp\u003eNew Zealand\u003c/p\u003e\n \u003cp\u003eSingapore\u003c/p\u003e\n \u003cp\u003eGermany\u003c/p\u003e\n \u003cp\u003eItaly\u003c/p\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003cp\u003eInternational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e28.6\u003c/p\u003e\n \u003cp\u003e15.6\u003c/p\u003e\n \u003cp\u003e11.7\u003c/p\u003e\n \u003cp\u003e11.7\u003c/p\u003e\n \u003cp\u003e3.9\u003c/p\u003e\n \u003cp\u003e2.6\u003c/p\u003e\n \u003cp\u003e2.6\u003c/p\u003e\n \u003cp\u003e2.6\u003c/p\u003e\n \u003cp\u003e14.3\u003c/p\u003e\n \u003cp\u003e6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSetting\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eGeneral healthcare\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eResearch\u003c/p\u003e\n \u003cp\u003eImaging/radiology\u003c/p\u003e\n \u003cp\u003eSurgery-related*\u003c/p\u003e\n \u003cp\u003eEmergency medicine\u003c/p\u003e\n \u003cp\u003eOncology\u003c/p\u003e\n \u003cp\u003eDermatology\u003c/p\u003e\n \u003cp\u003eOphthalmology\u003c/p\u003e\n \u003cp\u003ePublic health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e56\u003c/p\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e72.7\u003c/p\u003e\n \u003cp\u003e11.7\u003c/p\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003cp\u003e3.9\u003c/p\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePurpose\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eEvaluation\u003c/p\u003e\n \u003cp\u003eGovernance\u003c/p\u003e\n \u003cp\u003eEthical principles\u003c/p\u003e\n \u003cp\u003eImplementation\u003c/p\u003e\n \u003cp\u003eGuidelines or standards\u003c/p\u003e\n \u003cp\u003eTranslation and integration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e23.4\u003c/p\u003e\n \u003cp\u003e20.8\u003c/p\u003e\n \u003cp\u003e16.9\u003c/p\u003e\n \u003cp\u003e14.3\u003c/p\u003e\n \u003cp\u003e14.3\u003c/p\u003e\n \u003cp\u003e10.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMethod of derivation^\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eLiterature review\u003c/p\u003e\n \u003cp\u003eResearch and consultation\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003cp\u003ePrevious framework(s)\u003c/p\u003e\n \u003cp\u003eSelf-created\u003c/p\u003e\n \u003cp\u003eExpert consensus\u003c/p\u003e\n \u003cp\u003eCase studies\u003c/p\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e38\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e49.4\u003c/p\u003e\n \u003cp\u003e16.9\u003c/p\u003e\n \u003cp\u003e16.9\u003c/p\u003e\n \u003cp\u003e16.9\u003c/p\u003e\n \u003cp\u003e9.1\u003c/p\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003cp\u003e13.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStructure\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eTheme-based\u003c/p\u003e\n \u003cp\u003eStage-based\u003c/p\u003e\n \u003cp\u003eTheme x stage\u003c/p\u003e\n \u003cp\u003eOther or N/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e39\u003c/p\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e50.6\u003c/p\u003e\n \u003cp\u003e16.6\u003c/p\u003e\n \u003cp\u003e27.3\u003c/p\u003e\n \u003cp\u003e5.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of key components included\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eOne component\u003c/p\u003e\n \u003cp\u003eTwo components\u003c/p\u003e\n \u003cp\u003eThree components\u003c/p\u003e\n \u003cp\u003eFour components\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e24.7\u003c/p\u003e\n \u003cp\u003e40.3\u003c/p\u003e\n \u003cp\u003e22.1\u003c/p\u003e\n \u003cp\u003e13.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 340px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInclusion of key components\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eComponent 1: Guiding principles\u003c/p\u003e\n \u003cp\u003eComponent 2: Assessment method\u003c/p\u003e\n \u003cp\u003eComponent 3: AI lifecycle stages\u003c/p\u003e\n \u003cp\u003eComponent 4: Oversight mechanism\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e72\u003c/p\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003cp\u003e35\u003c/p\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e93.5\u003c/p\u003e\n \u003cp\u003e64.9\u003c/p\u003e\n \u003cp\u003e45.5\u003c/p\u003e\n \u003cp\u003e19.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e*Perioperative medicine, operating rooms, otolaryngology-head and neck surgery\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e^Some frameworks used multiple methods/sources\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e\u0026dagger;\u003c/sup\u003eIncludes expert consultation, interviews, focus groups, and workshops\u003c/p\u003e\n\u003cp\u003eKey framework components\u003c/p\u003e\n\u003cp\u003eMost frameworks only included one or two out of the four key components (Table 1, Supplement S5). Of the 10 with all four components, seven were operational or had been tested (Table 2).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEach key component is presented in the subsequent sections.\u003c/p\u003e\n\u003cp\u003eComponent 1: Guiding principles\u003c/p\u003e\n\u003cp\u003eThe first key component was a set of guiding principles for safe and responsible practices, included in nearly all the frameworks reviewed (n=72, 93.5%). We identified a total of 25 key principles across two broad categories \u0026ndash; ethics principles and governance principles (Table 3, Supplement S3, S4). The most common ethics principle was transparency (n=61, 79.2%), followed by fairness (n=50, 64.9%) and privacy and security (n=48, 62.3%).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGovernance principles were distinct from ethics concerns, relating to the organisational, regulatory, and technical processes and practices that ensure safe and responsible use of AI tools. The most common governance principles were data selection and management (n=57, 74.0%), accuracy and model performance (n=55, 71.4%), and ongoing monitoring and maintenance (n=54, 70.1%) (Table 3, Supplement S3, S4).\u003c/p\u003e\n\u003cp\u003eTable 2. Summary of frameworks that include all four key components of AI governance (n=10).\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStudy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eYear\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eImplementation status\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGuiding principles\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAI lifecycle stages\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOversight mechanism\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eApfelbacher et al.\u003csup\u003e27\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eStatus not provided\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003eDetails/checklist not provided - describes principles in text. Principles include safety, reliability, integration, education and training, autonomy, and accountability.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003col\u003e\n \u003cli\u003eBefore implementation\u003c/li\u003e\n \u003cli\u003eDuring implementation\u003c/li\u003e\n \u003cli\u003eAfter implementation\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eSpecific AI governance group \u0026ndash; Suggests establishing an AI committee prior to implementation.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eCallahan et al.\u003csup\u003e28\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eOperational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStage 1:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eProblem, need and use case definition\u003c/li\u003e\n \u003cli\u003eUsefulness estimates by workflow simulation\u003c/li\u003e\n \u003cli\u003eFinancial projections\u003c/li\u003e\n \u003cli\u003eEthical considerations (Responsibility, Equity, Traceability, Reliability, Governance, Non-maleficence, Autonomy)\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003e\u003cstrong\u003eStage 2:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eModel formulation\u003c/li\u003e\n \u003cli\u003eModel training and testing\u003c/li\u003e\n \u003cli\u003eDeployment on SHC infrastructure\u003c/li\u003e\n \u003cli\u003eOrganisational integration\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003e\u003cstrong\u003eStage 3:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eMonitoring\u003c/li\u003e\n \u003cli\u003eProspective evaluation\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003col\u003e\n \u003cli\u003eWhat \u0026amp; Why\u003c/li\u003e\n \u003cli\u003eHow\u003c/li\u003e\n \u003cli\u003eImpact\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eExisting committee, new remit - An existing data science team at Stanford Health Care.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eCoalition for Health AI\u003csup\u003e29\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eStatus not provided\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e5 core principles for trustworthy health AI:\u003c/p\u003e\n \u003col\u003e\n \u003cli\u003eUsefulness, usability, and efficacy\u003c/li\u003e\n \u003cli\u003eFairness and equity\u003c/li\u003e\n \u003cli\u003eSafety and reliability\u003c/li\u003e\n \u003cli\u003eTransparency, intelligibility, and accountability\u003c/li\u003e\n \u003cli\u003eSecurity and privacy\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eStage 1: Define the problem\u003cbr\u003eStage 2: Design the AI system\u003cbr\u003eStage 3: Engineer the AI solution\u003cbr\u003eStage 4: Assess\u003cbr\u003eStage 5: Pilot\u003cbr\u003eStage 6: Deploy \u0026amp; monitor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eIndependent review - Suggests a combination of a local leadership structure and regular independent third-party review.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eEconomou-Zavlanos et al.\u003csup\u003e30\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eOperational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cul\u003e\n \u003cli\u003eClinical Value and Safety\u003c/li\u003e\n \u003cli\u003eFairness \u0026amp; Equity\u003c/li\u003e\n \u003cli\u003eUsability \u0026amp; Adoption\u003c/li\u003e\n \u003cli\u003eRegulatory Compliance\u003c/li\u003e\n \u003cli\u003eTransparency and Accountability\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003col\u003e\n \u003cli\u003eModel development\u003c/li\u003e\n \u003cli\u003eSilent evaluation\u003c/li\u003e\n \u003cli\u003eEffectiveness evaluation\u003c/li\u003e\n \u003cli\u003eGeneral deployment\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eSpecific AI governance group - ABCDS Oversight Committee and Review Committee.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eLiao et al.\u003csup\u003e31\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eOperational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003eSet of guiding principles endorsed by the AI Committee:\u003c/p\u003e\n \u003col\u003e\n \u003cli\u003ePredictive model\u003c/li\u003e\n \u003cli\u003eModel evaluation includes statistical measures and relevant operational health metrics\u003c/li\u003e\n \u003cli\u003eModel output follows the five rights of CDSs and is associated with interventions\u003c/li\u003e\n \u003cli\u003eModel monitoring\u003c/li\u003e\n \u003cli\u003eHealthcare ethics incorporated in all stages of model evaluation and validation\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eValue stream from initial presentation to the Committee to periodic review and update or decommissioning.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eSpecific AI governance group - Clinical-AI-Predictive Analytics (CAIPA) Committee and case-based sub-committees.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eMorley et al.\u003csup\u003e32\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eNot implemented\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003eKey questions at each stage:\u003c/p\u003e\n \u003col\u003e\n \u003cli\u003eIs this AI system the right solution to the problem?\u003c/li\u003e\n \u003cli\u003eHas the AI system been designed in the right way?\u003c/li\u003e\n \u003cli\u003eIs the AI system working in the right way?\u003c/li\u003e\n \u003cli\u003eIs the AI system having the right kind of impact?\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003col\u003e\n \u003cli\u003ePreclinical stage (Theoretical Stage)\u003c/li\u003e\n \u003cli\u003eExploratory stage (Validation Stage)\u003c/li\u003e\n \u003cli\u003eDefinitive Stage (Real-world Evaluation Stage)\u003c/li\u003e\n \u003cli\u003ePost-market surveillance\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eInclude AI governance in existing processes \u0026ndash; Suggests identifying and mapping which aspects of evaluation are already covered by existing bodies and governing processes.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eSaenz et al.\u003csup\u003e33\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eTested\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003eI. Model Assessment\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eData provenance for training and testing\u003c/li\u003e\n \u003cli\u003eRegulatory Compliance\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eII. Pre-Deployment (Shadow Deployment)\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eTesting phase\u003c/li\u003e\n \u003cli\u003eDemographic Evaluation\u003c/li\u003e\n \u003cli\u003eSystem Customization\u003c/li\u003e\n \u003cli\u003eModel Output\u003c/li\u003e\n \u003cli\u003eSecurity and Privacy\u003c/li\u003e\n \u003cli\u003eAutonomy and Human Oversight\u003c/li\u003e\n \u003cli\u003eTransparency\u003c/li\u003e\n \u003cli\u003eEducation\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eIII. Deployment and Continuous Monitoring\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eDocumentation/Communication\u003c/li\u003e\n \u003cli\u003ePerformance Metrics\u003c/li\u003e\n \u003cli\u003eRisk Management\u003c/li\u003e\n \u003cli\u003eEquity and Fairness\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eI. Model Assessment\u003cbr\u003eII. Pre-Deployment (Shadow Deployment)\u003cbr\u003eIII. Deployment and Continuous Monitoring\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eSpecific AI governance group \u0026ndash; recommends presence of an AI monitoring committee.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eSingapore MOH, HSA, and IHiS\u003csup\u003e34\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eTested\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGuiding principles\u003c/strong\u003e: Fairness, Responsibility, Transparency, Explainability, Patient-Centricity\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eGuideline sections\u003c/strong\u003e:\u003cbr\u003eDevelopment\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eDesign: clinical inputs, end-user inputs, understanding the current clinical practice, data, cybersecurity, explainability\u003c/li\u003e\n \u003cli\u003eBuild: development standards, self-validation\u003c/li\u003e\n \u003cli\u003eTest: Evaluation and monitoring of AI-MD, intended use and workflow\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eImplementation\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eUse: Clinical governance, operational workflows and processes, end-user communication\u003c/li\u003e\n \u003cli\u003eMonitor \u0026amp; respond: Post-deployment monitoring\u003c/li\u003e\n \u003cli\u003eReview: Review and tracking\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003col\u003e\n \u003cli\u003eDevelopment - design, build, test\u003c/li\u003e\n \u003cli\u003eImplementation - use, monitor \u0026amp; respond, review\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eNon-specific - suggests review by \u0026ldquo;Organisational Leadership\u0026rdquo; who then makes a decision to implement.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003evan der Vegt et al.\u003csup\u003e35\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eTested\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003col\u003e\n \u003cli\u003eSpecification\u003c/li\u003e\n \u003cli\u003eComponent development\u003c/li\u003e\n \u003cli\u003eCombination of components into systems\u003c/li\u003e\n \u003cli\u003eIntegration of system into environment\u003c/li\u003e\n \u003cli\u003eRoutine use\u003c/li\u003e\n \u003cli\u003eOrganisation engagement\u003c/li\u003e\n \u003cli\u003ePolicy domains\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eStage I - Definition\u003cbr\u003eStage II - Retrospective study\u003cbr\u003eStage III - Silent trial\u003cbr\u003eStage IV - Pilot trial\u003cbr\u003eStage V - Large trial/roll-out\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eNon-specific \u0026ndash; includes \u0026ldquo;Governance\u0026rdquo; as a key principle under \u0026ldquo;F. Organisation engagement\u0026rdquo; but does not provide specific structure.\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003eWhittaker et al.\u003csup\u003e36\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003cp\u003eOperational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cul\u003e\n \u003cli\u003eAppropriateness\u003c/li\u003e\n \u003cli\u003eConsumers/population perspectives\u003c/li\u003e\n \u003cli\u003eMāori perspectives\u003c/li\u003e\n \u003cli\u003eEquity and fairness\u003c/li\u003e\n \u003cli\u003eEthical principles\u003c/li\u003e\n \u003cli\u003eClinical perspectives\u003c/li\u003e\n \u003cli\u003eData availability, quality, appropriateness, and completeness\u003c/li\u003e\n \u003cli\u003eTechnical processes\u003c/li\u003e\n \u003cli\u003eContractual and legal issues\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20px;\"\u003e\n \u003cp\u003eChecklist for each domain across 3 stages:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eConcept development\u003c/li\u003e\n \u003cli\u003eAccess to data for pre-processing, labelling, or model development\u003c/li\u003e\n \u003cli\u003eValidation or implementation of an existing AI model\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eSpecific AI governance group - National AI Expert Advisory Group (NAIAEAG) and AI Lab.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 3. Summary of guiding ethics and governance principles for AI.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003ePrinciple\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003en (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003eReferences\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003eEthics principles\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eAccountability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Those responsible for the different phases of the AI system lifecycle should be identifiable and accountable for the outcomes of the AI systems, and human oversight of AI systems should be enabled.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e41 (53.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15-17,20,22,25,27-30,33-36,38-63\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eAutonomy \u0026amp; human-centred values\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Throughout their lifecycle, AI systems should respect human rights, diversity, and the autonomy of individuals.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e Includes the principles of consent (n=23, 29.9%).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e31 (40.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15,17,20,22,25,27,28,31,33,36,38,40,43-45,48,49,53,54,56-58,61,63-69\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eBeneficence/ Human, societal \u0026amp; environmental wellbeing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Throughout their lifecycle, AI systems should benefit individuals, society and the environment.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e40 (51.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,7,17,20,22,25,27-31,36,38-41,43-45,48-52,56,59-66,68-73\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eContestability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;When an AI system significantly impacts a person, community, group or environment, there should be a timely process to allow people to challenge the use or outcomes of the Ai system.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e10 (13.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,27,29,34,42,44,56,61,63,66\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eFairness \u0026amp; equity\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Throughout their lifecycle, AI systems should be inclusive and accessible and should not involve or result in unfair discrimination against individuals, communities or groups.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e Involves ensuring AI systems do not create or perpetuate bias.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e50 (64.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,7,15-17,20,22,25,28-31,33-36,38,40-51,53,54,56-58,60,61,63-69,71-77\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eNon-maleficence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;The injunction to \u0026ldquo;Do No Harm,\u0026rdquo; that is, that every reasonable effort shall be made to avoid, prevent, and minimize harm or damage to any stakeholder.\u0026rdquo;\u003csup\u003e56\u003c/sup\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e36 (46.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,20-22,28-31,34,36,38,40-42,44,49-52,56,57,59,61-65,68,69,72,74,75,77-79\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003ePrivacy \u0026amp; security\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Throughout their lifecycle, AI systems should respect and uphold privacy rights and data protection and ensure the security of data\u0026rdquo;.\u003csup\u003e37\u003c/sup\u003e Includes the principle of confidentiality (n=19, 24.7%).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e48 (62.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15-17,20-22,25,27,29,33-36,38,40,42-45,48-53,55,57-61,63-70,74,76,79-84\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eReliability \u0026amp; safety\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Throughout their lifecycle, AI systems should reliably operate in accordance with their intended purpose.\u0026rdquo;\u003csup\u003e37\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e43 (55.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15-17,20,21,25,27-32,34-36,38,40,42-50,52,53,55,56,58-61,63,66,68,70,71,79,81,85\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eSustainability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eAI should be responsibly maintained and continue to produce benefit over time with minimal harm to the environment and health system. \u003csup\u003e2,20,61\u003c/sup\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e16 (20.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,20,40,43-45,48,50,61,63,66,68,71,74,81,86\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eTransparency\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;There should be transparency and responsible disclosure so people can understand when they are being significantly impacted by AI, and can find out when an AI system is engaging with them\u0026rdquo;.\u003csup\u003e37\u003c/sup\u003e Includes explainability, interpretability, and traceability/auditability.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e61 (79.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,15-17,20,22,25,27-30,32-36,38,40-45,47-58,60-72,74-79,84,86-90\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eTrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;References to trust include calls for trustworthy AI research and technology, trustworthy AI developers and organizations, trustworthy \u0026lsquo;design principles\u0026rsquo;, or underline the importance of customers\u0026rsquo; trust\u0026rdquo; \u003csup\u003e19\u003c/sup\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e23 (29.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,17,20-22,29,32,36,40,46,48,49,54-56,60,63,74,82,86-88,91\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003eGovernance principles\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eAccuracy \u0026amp; model performance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Accuracy describes evaluation based on how many right or wrong decisions a system makes. Example measurements are algorithmic accuracy, area under the curve (AUC) values, F1 scores, recall (sensitivity), precision, and specificity\u0026rdquo;\u003csup\u003e80\u003c/sup\u003e AI models should demonstrate accuracy and perform in line with its intended purpose.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e55 (71.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,7,15,16,20,25,27-36,38,41,43-48,51,52,55,57,59,60,62,63,65,67,69,71-75,77-88,90,92,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eAppropriateness \u0026amp; relevance\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eAI should be the best solution to the identified problem, demonstrating fitness-for-purpose, superiority to non-AI solutions, and usefulness.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e47 (61.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e4,7,16,20,21,25,28-32,34,36,38,39,42,44-49,51,52,60-63,65-67,69,71-74,78-82,84,85,89,90,93,94\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eAwareness, education, \u0026amp; training\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eSystem users should receive appropriate training and education to interact with the system as intended.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e38 (49.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,16,17,20,27-30,33-36,38,43,44,46,48-52,54,56,62-66,71,73,74,76,77,81,88,91,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eData selection \u0026amp; management\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eData needs to be selected, handled, and governed safely and responsibly. Appropriate practices and protocols should be established and enforced.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e57 (74.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15,16,20,27,29-36,38,41-48,51,52,54,55,57-63,65-69,71-81,83-90\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eFinancial considerations\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eConsideration of requirements for the AI system\u0026rsquo;s funding, cost-effectiveness, and commercialisation.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e30 (39.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e16,17,21,25,27-31,35,42,43,45,48,51,52,60,62,65,66,68,71,72,74,81,84-86,92,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eInteroperability\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;Data interoperability refers to the ability to accurately interpret data that is exchanged between different systems or organizations. It involves ensuring that the data has clear and unambiguous meaning, is correctly mapped, and is formatted in the required form\u0026rdquo; \u003csup\u003e95\u003c/sup\u003e.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e16 (20.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e16,21,25,29,35,43,48,49,52,57,59,60,66,68,70,83\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eOngoing monitoring \u0026amp; maintenance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eAI systems should be monitored and maintained continuously across the life cycle. Includes risk management, appropriate reporting, and plans for decommissioning.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e54 (70.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,7,15,16,25,27-36,38,41-48,50-52,55-57,59-65,68,71-74,76-78,81-83,85,86,88-90\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eOrganisational integration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003e\u0026ldquo;The organisational aspects of deploying the AI model and workflow.\u0026rdquo;\u003csup\u003e28\u003c/sup\u003e Includes leadership, policy and strategy, and resources.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e31 (39.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e4,15,16,20,27-31,33-36,38,42,45-48,50,51,54,55,64,66,71,74,81,82,86,89,91,92\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eRegulatory compliance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eAI development and use should comply with legal, regulatory, contractual, or human rights laws and guidelines, e.g. privacy and confidentiality law, medical device regulations, clinical guidelines.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e41 (53.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,4,15,16,25,27,29-31,33-36,38,43-45,48,51-53,57,59-61,63,64,66,68-70,72,74,75,80-83,85,90,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eStakeholder engagement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eThe active involvement of stakeholders in decision making processes, e.g. inclusion of multidisciplinary stakeholder perspectives in the design, implementation, or evaluation of AI solutions.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e35 (45.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,15,16,20,29,32,34-36,38,42,47,48,50,51,53,55,59-61,63,64,66,72-75,77,80,81,83,84,90,92,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eTechnical integration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eIntegration of the AI tool into existing technical infrastructure (e.g. Electronic Health Records (EHR)).\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e29 (37.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e27-30,35,36,43,46,48,52,54,57,59,60,62,65,66,72-75,78,79,85,88,89,92-94\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eTraining \u0026amp; validation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eAI models should be trained and validated on appropriate and representative data, following responsible processes and practices.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e45 (58.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e2,16,17,25,28-36,38,41,45-49,51,52,57,60,62,63,65,67,68,71-73,77-80,83-87,89,90,92,93\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eUsability \u0026amp; adoption\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eWhen integrating an AI solution, it should be designed to ensure it is intuitive, accessible, learnable, efficient, and user-centric. Usability has direct impacts on user experience, satisfaction, and uptake.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e35 (45.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e16,17,21,25,29,30,34-36,42,43,47,48,50,52,57,59,62,65,68,70,71,74,75,77,79-81,84,85,88,89,92-94\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eWorkflow integration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 50px;\"\u003e\n \u003cp\u003eIntegration of the AI tool into existing clinical workflows with minimal disruption. May also include workflow redesign to align with new needs of the AI tool and its users.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 7px;\"\u003e\n \u003cp\u003e39 (50.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e\u003csup\u003e4,7,21,27-32,34-36,38,39,42,43,46-48,52,54,55,57,62,65-67,71,73-75,77,78,81,82,85,88,89,94\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eComponent 2: Assessment method\u003c/p\u003e\n\u003cp\u003eThe second key component was a method to obtain information about AI tools and assess their alignment to predefined standards. This was included in 50 (64.9%) frameworks and typically comprised of a set of conceptual questions, or for more practical frameworks, a checklist or form to operationalise guiding principles into actionable practices or assessable criteria. For example, to assess fairness, the ABCDS framework required submission of a bias analysis and management plan with results of AI tool performance across subgroups.\u003csup\u003e30\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eFrameworks like the ABCDS framework,\u003csup\u003e30\u003c/sup\u003e FURM assessment,\u003csup\u003e28\u003c/sup\u003e and Clinical-AI-Predictive Analytics model review form\u003csup\u003e31\u003c/sup\u003e used formats similar to an ethics submission form, where developers or implementers must provide details about proposed AI tools. This method was often used in frameworks which were operational in organisations with an oversight committee who would use the form to review AI tools.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOther frameworks\u003csup\u003e22,60,68,91,93\u003c/sup\u003e proposed a checklist structure, where the presence or absence of criteria are indicated in a checkbox or binary yes/no style. Some frameworks\u003csup\u003e66,79,84\u003c/sup\u003e included a scoring method where the AI tool\u0026rsquo;s consideration of each principle is given a numerical value (e.g. between 0-5, with 0 indicating no evidence of consideration and 5 indicating evidence of extensive consideration of a principle). A total score for the AI tool overall and/or multiple scores for different domains were then calculated. Checklists and scoring methods were often used for self-assessment by relevant stakeholders (e.g. researchers, developers, or implementers).\u003c/p\u003e\n\u003cp\u003eComponent 3: AI life cycle stages\u003c/p\u003e\n\u003cp\u003eThe third key component was a review timeline or consideration of the AI life cycle stages which was included in 35 (45.5.%) frameworks. Effective governance acknowledges the changing requirements and challenges across different stages of the AI life cycle. Frameworks assigned different recommendations to different stages of AI development (e.g. FUTURE-AI framework\u003csup\u003e48\u003c/sup\u003e) or adopted a multi-stage review process requiring submission of new materials at different stages as AI tools were further developed and evaluated (e.g. ABCDS framework and FURM assessments\u003csup\u003e28,30\u003c/sup\u003e). Five key life cycle stages are summarised in the following sections.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStage 1: Problem identification\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis involves identifying and defining the issue and planning an AI-driven solution. Problems may be identified by clinicians who raise pain points to organisational leaders or by leaders who identify opportunities for growth.\u003csup\u003e46\u003c/sup\u003e This typically involved:\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eIdentifying and engaging with stakeholders to define problems\u003csup\u003e29,34,42,46,48,55,56,60,82,83,90\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eAssessing whether AI is an appropriate solution\u003csup\u003e28,32,34,36,56,57,60,61,82\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eDefining use cases and potential impacts of solutions\u003csup\u003e28,29,48,55-57,59,72,82,85\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eEvaluating resource requirements and feasibility\u003csup\u003e28,29,48,55,60,82\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003ePreliminary ethics and/or risk assessments\u003csup\u003e28,34,45,48,59-61,63\u003c/sup\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eOnce the problem has been defined, and an AI tool has been deemed an appropriate solution, a decision needs to be made to either procure an existing AI tool or develop a new one (in-house or in partnership with external developers).\u003csup\u003e29,46,60\u003c/sup\u003e Clearly defining and understanding problems can ensure AI tools that are procured or developed are purpose-driven and not merely guided by technological trends or availability.\u003csup\u003e29,42,82\u003c/sup\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStage 2: Design and development\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis involves development or procurement of suitable AI tools. Though each method of obtaining an AI tool may involve case-specific activities, this stage typically involves the consideration of technical and workflow requirements, the selection and preparation of data, and ensuring tool design is tailored to its intended use.\u003csup\u003e29,35,45,60,83\u003c/sup\u003e Ethics principles such as privacy, security, autonomy, and fairness should be embedded during design.\u003csup\u003e29,63\u003c/sup\u003e Risk management, monitoring, and contestability plans should also be developed.\u003csup\u003e29\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStage 3: Training, validation, and evaluation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis involves training, validation, and evaluation of AI tools, and ensures that ethics and governance principles are effectively applied prior to clinical or administrative use. For traditional machine learning, AI models need to be trained and validated in an iterative process using appropriate and representative data to ensure the model is refined according to its purpose, performs as intended, and does not create or perpetuate any existing bias.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEvaluation often involves a series of studies from silent or small-scale evaluation to real-world or clinical studies.\u003csup\u003e29,30,62,67,85\u003c/sup\u003e Outcome measures may include statistical outcomes like accuracy and reliability, clinical outcomes, usability and adoption, cost-analyses, or ethics analyses (e.g. fairness evaluation).\u003csup\u003e21,29,30,32,41,48,52,57,67,72,85\u003c/sup\u003e Performance metrics should be carefully selected so they are relevant to the given task and models can be reliably evaluated.\u003csup\u003e2,32,52,72,79\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eAI tool performance and evaluation should be reported in line with established reporting standards such as MINIMAR,\u003csup\u003e87\u003c/sup\u003e DECIDE-AI,\u003csup\u003e77\u003c/sup\u003e CONSORT-AI,\u003csup\u003e78\u003c/sup\u003e TRIPOD-AI, STARD-AI, SPIRIT-AI.\u003csup\u003e62\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStage 4: Implementation and integration\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis is the stage at which AI tools are deployed, and often involves:\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eTechnical integration: integration of AI tools into existing technical infrastructure and workflows (e.g. with EHRs) and changes to technical infrastructure.\u003csup\u003e28,65,85,89\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eClinical workflow integration: integration of AI tools into existing workflows, which can include workflow redesign and introduction or removal of key roles.\u003csup\u003e42,57\u003c/sup\u003e Effective integration often involves extensive stakeholder engagement and regular feedback.\u003c/li\u003e\n \u003cli\u003eOrganisational integration and governance: ensures organisations have the appropriate routines, competencies, and resources\u003csup\u003e4\u003c/sup\u003e to successfully implement AI tools, and that implementation is in line with organisational structures, systems, and values.\u003csup\u003e29,55,81,86,91\u003c/sup\u003e This includes the establishment of leadership and management structures, such as an oversight committee and local champions. This may also involve new policies and procedures that must be communicated to all stakeholders.\u003csup\u003e29,42,46,81\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eTraining and education ensures that all stakeholders build literacy for safe and appropriate use.\u003csup\u003e2,16,48,50,61\u003c/sup\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cem\u003eStage 5: Monitoring and maintenance\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis stage involves ongoing oversight of AI tool performance post-deployment, including processes for risk or error identification and management, and mechanisms for updating or decommissioning tools when appropriate.\u003csup\u003e46,47,56\u003c/sup\u003e An effective monitoring strategy should encompass a range of indicators, such as the monitoring of technical components, clinical outcomes, unintended consequences and harms, privacy and transparency concerns, and feedback on usability and acceptance.\u003csup\u003e35,46,48\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eComponent 4: Oversight mechanism\u003c/p\u003e\n\u003cp\u003eMechanisms for implementing and managing AI governance were only included by 15 (19.5%) frameworks. While others provided guidance on their use by suggesting target audience/s or assigning stakeholders to different tasks, they did not provide guidance on management and organisational oversight. When included, the most common approach was a specific AI governance group which tested or used the framework. Three frameworks\u003csup\u003e30,31,36\u003c/sup\u003e reported establishing a specific AI governance group and six \u003csup\u003e27,33,42,50,54,89\u003c/sup\u003e recommended establishing one. All emphasised the need for diverse, multidisciplinary membership by ensuring the governance group was representative of key stakeholders (e.g. clinical, ethics, legal, IT, operations, research, consumers). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOne framework\u003csup\u003e32\u003c/sup\u003e suggested mapping AI governance requirements onto existing organisational governance bodies and processes, instead of creating a new group. Another\u003csup\u003e28\u003c/sup\u003e was being used by an existing team in the organisation who expanded their scope of responsibility and introduced new governance processes for AI. One framework suggested a combination of local oversight and an independent review process, where AI tools are reviewed by an external third party at regular intervals.\u003csup\u003e29\u003c/sup\u003e Three frameworks\u003csup\u003e34,35,51\u003c/sup\u003e included organisational-level leadership and oversight as key components, but did not specify how this was facilitated.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eDespite rapid development and extensive research, guidance for healthcare organisations to navigate ethical and organisational challenges, and ensure safe and responsible AI adoption remains limited. To address this gap and integrate AI approaches with existing healthcare governance, we conducted a scoping review and identified ethical and governance principles, along with four key components for AI governance frameworks:\u0026nbsp;guiding principles; an assessment method; consideration of the AI lifecycle stages; and an oversight mechanism.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThere was extensive literature on AI governance principles, however, depending on the purpose and focus of the frameworks, principles ranged from ethics principles to principles focused on safe and responsible organisational processes.\u0026nbsp;We proposed a distinction between ethics and governance principles, where governance principles are a separate category related to the organisational, regulatory, and technical factors that impact AI at the organisational level, such as data selection and management, regulatory compliance, integration, and stakeholder engagement.\u0026nbsp;Existing frameworks often had a stronger focus on one category while lacking depth in the other, such as ethics only frameworks, or frameworks including broad ethics principles without further elaboration. (e.g. \u003csup\u003e35,86\u003c/sup\u003e)\u0026nbsp;Our distinction provides clarity on governance needs at the organisational level and narrows the gap between conceptual ethics frameworks and practical AI governance requirements.\u003c/p\u003e\n\u003cp\u003eThe most common ethics principles such as transparency, fairness and equity, privacy and security, accountability, and beneficence aligned closely with established AI ethics frameworks.\u003csup\u003e19,96\u003c/sup\u003e However, subthemes specifically relevant to healthcare, such as consent, confidentiality, and medicolegal liability were not commonly included. Noticeably, some critical ethics principles such as contestability, sustainability, and cultural safety were absent from more than three-quarters of the frameworks.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe principle of contestability, which is critical in ensuring public trust in AI and upholding the principles of justice and autonomy,\u003csup\u003e97\u003c/sup\u003e was only included in 10 frameworks (13.0%).\u003csup\u003e2,27,29,34,42,44,56,61,63,66\u003c/sup\u003e All stakeholders, including clinicians, patients, and caregivers require assurance that efficient and accessible processes exist to lodge complaints and receive appropriate and adequate redress. The ability for consumers to contest decisions or the use of AI in care delivery acknowledges the potential fallibility of AI tools, the importance of autonomy in healthcare, and the necessity for regular human oversight or, where necessary, intervention.\u003csup\u003e97\u003c/sup\u003e\u0026nbsp; If the process for contesting decisions and seeking redress is too difficult, justice is compromised and there is little opportunity for similar harms to be avoided in the future.\u003csup\u003e63\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eSustainability was a key principle in only 16 (20.8%) frameworks. References to sustainability included health system sustainability, environmental sustainability, economic sustainability, and social sustainability more broadly.\u003csup\u003e2,20,40,61,71\u003c/sup\u003e Sustainability is a crucial concern in healthcare, and related issues like climate change pose an existential\u0026nbsp;threat to humans, health systems, and health infrastructure.\u003csup\u003e98,99\u003c/sup\u003e For healthcare AI to be sustainable, organisations should be adequately resourced with the ability to fully integrate, maintain, and update tools over time.\u003csup\u003e61\u003c/sup\u003e Without consideration of sustainability in AI design and monitoring, potential benefits and harms of AI tools may be missed, including harms to human and environmental wellbeing, impacts on intergenerational equity, and workplace disruptions like job changes or losses.\u003csup\u003e19,20,61\u003c/sup\u003e Patient trust in the organisation may also be compromised, especially if sustainability continues to become a public priority. Measures to improve sustainability include the use of energy efficient AI, renewable energy, and collaboration to reduce redundancy.\u003csup\u003e100,101\u003c/sup\u003e Various tools and frameworks could also be used to address sustainability.\u003csup\u003e102-104\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eAlthough most frameworks included themes or discussion surrounding fairness, equity, and data bias, only three frameworks, two from New Zealand\u003csup\u003e36,51\u003c/sup\u003e and one from the USA\u003csup\u003e33\u003c/sup\u003e incorporated specific criteria regarding culturally safe and appropriate practices. The New Zealand frameworks were the only ones to specifically address governance concerns for Indigenous populations. Culturally appropriate AI is a relevant concern for many healthcare organisations globally, and Indigenous and culturally and linguistically diverse groups have expressed the need for better inclusion and consideration of cultural safety in AI governance, beyond issues with data or algorithmic bias.\u003csup\u003e105,106\u003c/sup\u003e For example, in the Australian context, Aboriginal and or Torres Strait Islander peoples still experience disparities in health outcomes and access due to the ongoing effects of colonisation and continued discrimination and exclusion. There are specific guidelines on data governance and Data Sovereignty\u003csup\u003e107,108\u003c/sup\u003e and culturally safe healthcare\u003csup\u003e109\u003c/sup\u003e for Aboriginal and or Torres Strait Islander peoples that require consideration when designing and implementing safe and inclusive AI. Incorporating Indigenous principles and leadership into AI governance in healthcare organisations has the potential to address national priorities for digital inclusion and health equity (e.g. National Agreement on Closing the Gap, Priority Reform 4 and Targets 1, 2, 14, 17\u003csup\u003e110\u003c/sup\u003e) and promote further benefit from AI for both Indigenous and non-Indigenous populations.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe approach adopted by Whittaker et al.\u003csup\u003e36\u003c/sup\u003e to develop a framework specific to the Aotearoa New Zealand context represents a thoughtful approach that could inform future work, particularly as it highlights the insufficiencies of existing international frameworks in addressing Māori data governance and guardianship (kaitiakitangi) rights. Their resulting framework and governance group embeds Māori data sovereignty principles, ethics principles (tikanga), and philosophies (matauranga) to honour Te Tiriti (the Treaty of Waitangi) and prioritise consumer needs. Whittaker et al. demonstrate the value of going beyond inclusion and consultation to fully integrate and embed Indigenous principles in their framework to ensure value and safety for all stakeholders.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe also identified governance considerations such as commercialisation that were not commonly included in frameworks, despite their importance.\u003csup\u003e20,70,72\u003c/sup\u003e Only seven frameworks (10.6%)\u003csup\u003e20,30,36,70,72,82,86\u003c/sup\u003e included discussion on the commercialisation of AI. This may have been due to the overall lack of consideration of practical AI governance needs in the literature. When commercialisation is an organisational goal, it is important that agreements recognise institutional contributions (e.g. the provision of patient data), benefits are fairly distributed, intellectual property rights are defined, and lines of accountability are clear.\u003csup\u003e72\u003c/sup\u003e This includes considerations around ethics principles such as privacy and fairness, which may be sidelined during commercialisation.\u003csup\u003e20\u003c/sup\u003e Commercialisation may also impact consent. For example, patients may be comfortable with the use of their data for internal quality improvement but may desire a more formal informed consent process if their data were to be provided to an external organisation. Institutional needs and pathways for commercialisation should be clarified in the planning stages.\u003csup\u003e20,72\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eSome frameworks (e.g.\u003csup\u003e7,28,47,66\u003c/sup\u003e) identified the potential limitations in their application to diverse use cases and technologies including generative AI. Five (6.5%) frameworks\u003csup\u003e22,53,63,89,93\u003c/sup\u003e focused specifically on generative AI governance, and one\u003csup\u003e21\u003c/sup\u003e focused on conversational agents incorporating generative AI, but had limited applicability to AI technologies more broadly. Generative AI presents new challenges and risks with its ability to generate original material in a variety of mediums (e.g. text, images, audio) as opposed to traditional uses of AI for tasks like classification, analysis, or prediction.\u003csup\u003e111\u003c/sup\u003e These risks include a greater potential for misuse, errors, and inaccuracies (e.g. hallucinations), less transparency around its processes, and more complexity surrounding liability.\u003csup\u003e112,113\u003c/sup\u003e As AI continues to evolve, organisations and regulatory bodies are finding it increasingly difficult to envision the extent of emerging technological capability and associated ethical and regulatory requirements.\u003csup\u003e114\u003c/sup\u003e A responsive and sustainable framework would anticipate the unique risks and challenges of generative AI and other emerging technologies.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAn earlier review on clinical AI implementation frameworks by van der Vegt et al.\u003csup\u003e35\u003c/sup\u003e found that most existing frameworks only included one of either theme-based (determinant) or stage-based (process) elements, which presented a critical gap in the availability of comprehensive and actionable AI implementation frameworks. Many of the 77 frameworks we examined addressed this gap by including both theme-based and stage-based elements. However, we found that most were still missing components that could enable their real-world implementation, and identified the additional components of an assessment method and oversight mechanism. Only 10 out of 77 (13.0%)\u003csup\u003e27-36\u003c/sup\u003e included all four of the key components we identified, with the least common being an oversight mechanism.\u003c/p\u003e\n\u003cp\u003eOrganisational oversight dictates how human oversight will be maintained, who has the decision-making power to approve or deny the development and/or deployment of new AI tools in the organisation, and where stakeholders can obtain expert advice. Having a clear organisational apparatus for AI governance and embedding standardised processes and policies allows organisations and relevant stakeholders to be confident that deployed AI tools have met an established standard that aligns with the organisation\u0026rsquo;s values and purpose.\u003csup\u003e30\u003c/sup\u003e This transparency in processes may also encourage ethical and responsible practices from the early stages and can help promote trust in AI. Furthermore, an oversight structure that includes membership from diverse fields (e.g. clinical, operational, legal/ethical, technical, research, consumers) facilitates consideration of a variety of relevant concerns, builds AI literacy and expertise, and ensures alignment across the organisation.\u003csup\u003e31,54\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eThe most common practice or recommendation for an oversight mechanism was for an AI-specific governance group to oversee all AI projects and tools within the organisation. The primary role of these committees is to identify, review, and monitor AI tools.\u003csup\u003e115\u003c/sup\u003e Liao et al.\u003csup\u003e31\u003c/sup\u003e and Economou et al.\u003csup\u003e30\u003c/sup\u003e each describe versions of this approach, where there is a broader AI-specific governance committee that oversees the review process, and a secondary group/s that assist with the ongoing development of AI tools. University of Wisconsin Health\u003csup\u003e31\u003c/sup\u003e has an institution-level Clinical AI and Predictive Analytics Committee from which separate subcommittees are created to assist with specific AI projects on a case-by case basis. Similarly, at Duke Health,\u003csup\u003e30\u003c/sup\u003e the broader ABCDS Oversight Committee has three subcommittees (ABCDS Regulatory Subcommittee, ABCDS Evaluation Subcommittee, and ABCDS Implementation and Monitoring Subcommittee) overseeing specific checkpoints at each AI lifecycle stage. The dual approach allows these organisations to provide specificity and appropriate expertise for the wide variety of technologies (e.g. traditional versus generative AI) and use cases (e.g. operational versus clinical, or different clinical areas) that a multidisciplinary setting may need to consider, while maintaining a centralised process to ensure appropriate oversight and consistency across the organisation. This approach also allows for more effective risk stratification and resource allocation so benefits can be maximised.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe did not find any frameworks that were systematically evaluated for their feasibility and effectiveness in facilitating safe and responsible AI. Though preliminary learnings and measures of success were recorded for some (e.g. number of successful deployments, retirements, and non-deployments;\u003csup\u003e31\u003c/sup\u003e number of AI models under the organisation\u0026rsquo;s governance portfolio;\u003csup\u003e30,116\u003c/sup\u003e use cases advancing to implementation \u003csup\u003e28\u003c/sup\u003e), systematic evaluation of AI governance frameworks is needed to identify the barriers, enablers, gaps, and unintended impacts of implementing different frameworks on organisational, health, and patient experience outcomes. Future research involving systematic evaluation, including engagement with relevant stakeholders such as clinicians and consumers will aid in consolidating our understanding of best practice for AI governance.\u003c/p\u003e\n\u003cp\u003eWe aimed to conduct a comprehensive review of AI governance frameworks for healthcare organisations, however, there were several limitations. First, we excluded primary and home care which are integral to care delivery. Their focus on accessibility and community-centred care may have provided additional insights and limits the applicability of our findings to these contexts, despite the need for governance in areas such as home care, which is being revolutionised by AI. Second, we were limited to frameworks available in English. The key components and principles represent a mostly Western perspective, with most frameworks originating from North America, the UK, and Australia. Future research inclusive of frameworks in other languages may provide a broader or more culturally responsive view. Lastly, this review will likely miss relevant published work that could influence findings, especially as AI evolves rapidly and publications grow exponentially. It is also likely that we have missed internally developed frameworks that are not publicly or freely available. We sought to mitigate these limitations by updating the search close to the time of submission, and including multiple bibliographic databases and search methods.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis review highlights both ethics and governance principles that healthcare organisations should consider to support safe and responsible AI implementation. We identified four key components of AI governance frameworks including guiding principles, an assessment method, AI lifecycle stages, and an oversight mechanism. The findings provide a sound foundation for healthcare organisations to develop locally tailored AI governance frameworks that are both comprehensive and practical.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cp\u003eWe used the framework proposed by Arksey and O'Malley\u003csup\u003e117\u003c/sup\u003e and advanced by Levac et al.\u003csup\u003e118\u003c/sup\u003e which recommends five-stages for scoping reviews: identifying the research question; identifying relevant studies; study selection; data charting; and collating, summarising, and reporting results. We developed a protocol prior to commencement, and used it to guide eligibility criteria, search strategies, data extraction, and overall objectives. Reporting was guided by the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist.\u003csup\u003e119\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eEligibility criteria\u003c/p\u003e\n\u003cp\u003eThe eligibility criteria were adapted from van der Vegt et al.\u003csup\u003e35\u003c/sup\u003e and adjusted to widen the scope from implementation to the broader concept of governance (Table 4).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 4. Eligibility criteria.\u003c/p\u003e\n\u003ctable width=\"100%\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\"\u003e\n\u003cp\u003e\u003cstrong\u003eComponent\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e\u003cstrong\u003eInclusions\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e\u003cstrong\u003eExclusions\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\"\u003e\n\u003cp\u003eConcept\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Studies reporting on either proposed or enforced AI governance.\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Where a framework is not proposed.\u003c/p\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Frameworks that are not specific to AI or do not include a specific AI component.\u003c/p\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Frameworks that are not about governance or an aspect of governance.\u003c/p\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Specific technology/IT \u0026lsquo;frameworks\u0026rsquo; rather than generic.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\"\u003e\n\u003cp\u003eContext\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Proposed AI governance frameworks must be aimed to be implemented in acute care settings.\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003eNon-acute settings:\u003c/p\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Primary care.\u003c/p\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Home care.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\"\u003e\n\u003cp\u003eParticipants\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Targeting any health condition or patient group.\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\"\u003e\n\u003cp\u003eType of evidence\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Published peer-reviewed or grey literature.\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"39%\"\u003e\n\u003cp\u003e-\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Not available in English.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInformation sources and search\u003c/p\u003e\n\u003cp\u003eBibliographic databases including MEDLINE, Embase, and Scopus were searched in April 2024 and updated in March 2025. Databases were selected in consultation with a clinical librarian. Appropriate search strings were developed by linking concept clusters relating to each component of the question (artificial intelligence, governance, and a healthcare setting) with Boolean operators \u0026ldquo;AND\u0026rdquo; and \u0026ldquo;OR\u0026rdquo; (Supplement S1). Searches were also conducted on Google Scholar, where the first 200 results were included for review based on guidance for grey literature searching.\u003csup\u003e120\u003c/sup\u003e Reference lists of key papers were searched for additional peer-reviewed and grey literature.\u003c/p\u003e\n\u003cp\u003eFigure 1. PRISMA flowchart of the stages of the review.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSelection of sources of evidence\u003c/p\u003e\n\u003cp\u003eSearch results were imported into Covidence, where a two-stage review process was conducted by a team of independent reviewers (SF, TJ, AW). 6,391 studies were retrieved from the searches (Figure 1). After removal of duplicates, 5,194 studies remained. An initial title and abstract screening was conducted by two reviewers to exclude non-relevant studies. The remaining studies then underwent a second screening of the full text by two reviewers. An additional 23 studies from citation searching and grey literature were included at full text screening, resulting in 315 studies to review. After exclusions, 77 frameworks were analysed. Conflicts were resolved by two-way consensus. Further exclusions were based on researcher consensus during data charting and extraction.\u003c/p\u003e\n\u003cp\u003eData extraction and synthesis\u003c/p\u003e\n\u003cp\u003eIncluded studies were exported from Covidence to Excel for data charting and extraction by AW, with verification by SF. A data charting table was developed to ensure extraction of relevant information (Supplement S2). For each included study descriptive information about the framework including name, setting, purpose, and method of derivation were extracted. The following data were also examined:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eStructure\u003c/em\u003e: Framework structure was examined using a previously published classification\u003csup\u003e35\u003c/sup\u003e based on two of Nielsen\u0026rsquo;s five implementation framework categories\u003csup\u003e121\u003c/sup\u003e which we adapted to the context of governance:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eTheme-based (determinant), where frameworks are structured around determinants that influence governance outcomes (e.g. \u003csup\u003e16,53,61\u003c/sup\u003e)\u003c/li\u003e\n\u003cli\u003eStage-based (process), where frameworks are structured by steps or stages (e.g.\u003csup\u003e45,83,90\u003c/sup\u003e)\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eWe added a third classification (theme x stage) for frameworks with both stage-based and theme-based elements (e.g.\u003csup\u003e30,35,79\u003c/sup\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eKey components\u003c/em\u003e: Upon initial reading and synthesis, we identified four components that form a practical AI governance framework:\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eSet of guiding principles.\u003c/li\u003e\n\u003cli\u003eMethod to review or assess AI tools (e.g. questions, checklist items, supporting materials).\u003c/li\u003e\n\u003cli\u003eReview timeline or consideration of the AI life cycle stages.\u003c/li\u003e\n\u003cli\u003eOrganisational oversight mechanism (e.g. governance committee).\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cem\u003eKey \u003c/em\u003e\u003cem\u003eprinciples\u003c/em\u003e: A content analysis was conducted to determine the key AI ethics and governance principles in each framework. An inductive approach was applied, where principles were added to the matrix throughout analysis (Supplement S3, S4). Principles were marked as present only when included as a discrete framework component.\u003c/p\u003e\n\u003cp\u003eA narrative synthesis then integrated findings into descriptive summaries for each of the four key components. \u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCOMPETING INTERESTS\u003c/h2\u003e\u003cp\u003eThe authors have no competing interests to declare.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAW and SF conceived this review, designed, and conducted the analysis with advice and input from FM. AW and SF selected the studies. AW drafted the manuscript with input from all authors. All authors provided revisions for intellectual content. All authors have approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe wish to acknowledge the contributions of Tamasha Jayawardena (TJ) who assisted with the study screening.\u003c/p\u003e\u003ch2\u003eDATA AVAILABILITY\u003c/h2\u003e\u003cp\u003eAll data relevant to the analysis are included in the article and online supplementary material. There are no new data associated with this article.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBaig, M. A., Almuhaizea, M. A., Alshehri, J., Bazarbashi, M. S. \u0026amp; Al-Shagathrh, F. Urgent Need for Developing a Framework for the Governance of AI in Healthcare. \u003cem\u003eStud Health Technol Inform.\u003c/em\u003e \u003cstrong\u003e272\u003c/strong\u003e, 253-256 (2020). https://doi.org/10.3233/shti200542\u003c/li\u003e\n\u003cli\u003eSolanki, P., Grundy, J. \u0026amp; Hussain, W. Operationalising ethics in artificial intelligence for healthcare: A framework for AI developers. \u003cem\u003eAI and Ethics.\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 223-240 (2023). https://doi.org/10.1007/s43681-022-00195-z\u003c/li\u003e\n\u003cli\u003eCarter, S. M.\u003cem\u003e et al.\u003c/em\u003e How should artificial intelligence be used in Australian health care? Recommendations from a citizens\u0026rsquo; jury. \u003cem\u003eMed J Aust.\u003c/em\u003e \u003cstrong\u003e220\u003c/strong\u003e, 409-416 (2024). https://doi.org/10.5694/mja2.52283\u003c/li\u003e\n\u003cli\u003eNovak, L. L.\u003cem\u003e et al.\u003c/em\u003e Clinical use of artificial intelligence requires AI-capable organizations. \u003cem\u003eJAMIA Open.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, ooad028 (2023). https://doi.org/10.1093/jamiaopen/ooad028\u003c/li\u003e\n\u003cli\u003eLyell, D., Wang, Y., Coiera, E. \u0026amp; Magrabi, F. More than algorithms: an analysis of safety events involving ML-enabled medical devices reported to the FDA. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 1227-1236 (2023). https://doi.org/10.1093/jamia/ocad065\u003c/li\u003e\n\u003cli\u003eLabkoff, S.\u003cem\u003e et al.\u003c/em\u003e Toward a responsible future: recommendations for AI-enabled clinical decision support. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 2730-2739 (2024). https://doi.org/10.1093/jamia/ocae209\u003c/li\u003e\n\u003cli\u003eMcCradden, M. D., Joshi, S., Anderson, J. A. \u0026amp; London, A. J. A normative framework for artificial intelligence as a sociotechnical system in healthcare. \u003cem\u003ePatterns.\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 100864 (2023). https://doi.org/10.1016/j.patter.2023.100864\u003c/li\u003e\n\u003cli\u003eDavenport, T. H. \u0026amp; Glaser, J. P. Factors governing the adoption of artificial intelligence in healthcare providers. \u003cem\u003eDiscov Health Syst.\u003c/em\u003e \u003cstrong\u003e1\u003c/strong\u003e, 4 (2022). https://doi.org/10.1007/s44250-022-00004-8\u003c/li\u003e\n\u003cli\u003eMaleki Varnosfaderani, S. \u0026amp; Forouzanfar, M. The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century. \u003cem\u003eBioengineering.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e (2024). https://doi.org/10.3390/bioengineering11040337\u003c/li\u003e\n\u003cli\u003eMagrabi, F., Lyell, D. \u0026amp; Coiera, E. Automation in Contemporary Clinical Information Systems: a Survey of AI in Healthcare Settings. \u003cem\u003eYearb Med Inform.\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 115-126 (2023). https://doi.org/10.1055/s-0043-1768733\u003c/li\u003e\n\u003cli\u003eBatool, A., Zowghi, D. \u0026amp; Bano, M. AI governance: a systematic literature review. \u003cem\u003eAI and Ethics.\u003c/em\u003e (2025). https://doi.org/10.1007/s43681-024-00653-w\u003c/li\u003e\n\u003cli\u003ePapagiannidis, E., Mikalef, P. \u0026amp; Conboy, K. Responsible artificial intelligence governance: A review and research framework. \u003cem\u003eThe Journal of Strategic Information Systems.\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, 101885 (2025). https://doi.org/10.1016/j.jsis.2024.101885\u003c/li\u003e\n\u003cli\u003eMorley, J., Floridi, L., Kinsey, L. \u0026amp; Elhalal, A. From What to How: An Initial Review of Publicly Available AI Ethics Tools, Methods and Research to Translate Principles into Practices. \u003cem\u003eSci Eng Ethics.\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 2141-2168 (2020). https://doi.org/10.1007/s11948-019-00165-5\u003c/li\u003e\n\u003cli\u003eBirkstedt, T., Minkkinen, M., Tandon, A. \u0026amp; M\u0026auml;ntym\u0026auml;ki, M. AI governance: themes, knowledge gaps and future agendas. \u003cem\u003eInternet Research.\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 133-167 (2023). https://doi.org/10.1108/INTR-01-2022-0042\u003c/li\u003e\n\u003cli\u003eKhanna, S., Khanna, I., Srivastava, S. \u0026amp; Pandey, V. AI Governance Framework for Oncology: Ethical, Legal, and Practical Considerations. \u003cem\u003eQuarterly Journal of Computational Technologies for Healthcare.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 1-26 (2021). \u003c/li\u003e\n\u003cli\u003eStogiannos, N.\u003cem\u003e et al.\u003c/em\u003e Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK. \u003cem\u003eBr J Radiol.\u003c/em\u003e \u003cstrong\u003e96\u003c/strong\u003e, 20221157 (2023). https://doi.org/10.1259/bjr.20221157\u003c/li\u003e\n\u003cli\u003eČartolovni, A., Tomičić, A. \u0026amp; Lazić Mosler, E. Ethical, legal, and social considerations of AI-based medical decision-support tools: A scoping review. \u003cem\u003eInt J Med Inform.\u003c/em\u003e \u003cstrong\u003e161\u003c/strong\u003e, 104738 (2022). https://doi.org/10.1016/j.ijmedinf.2022.104738\u003c/li\u003e\n\u003cli\u003eGoirand, M., Austin, E. \u0026amp; Clay-Williams, R. Implementing Ethics in Healthcare AI-Based Applications: A Scoping Review. \u003cem\u003eSci Eng Ethics.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 61 (2021). https://doi.org/10.1007/s11948-021-00336-3\u003c/li\u003e\n\u003cli\u003eJobin, A., Ienca, M. \u0026amp; Vayena, E. The global landscape of AI ethics guidelines. \u003cem\u003eNature Machine Intelligence.\u003c/em\u003e \u003cstrong\u003e1\u003c/strong\u003e, 389-399 (2019). https://doi.org/10.1038/s42256-019-0088-2\u003c/li\u003e\n\u003cli\u003eSiala, H. \u0026amp; Wang, Y. SHIFTing artificial intelligence to be responsible in healthcare: A systematic review. \u003cem\u003eSocial Science \u0026amp; Medicine.\u003c/em\u003e \u003cstrong\u003e296\u003c/strong\u003e, 114782 (2022). https://doi.org/10.1016/j.socscimed.2022.114782\u003c/li\u003e\n\u003cli\u003eDing, H., Simmich, J., Vaezipour, A., Andrews, N. \u0026amp; Russell, T. Evaluation framework for conversational agents with artificial intelligence in health interventions: a systematic scoping review. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 746-761 (2023). https://doi.org/10.1093/jamia/ocad222\u003c/li\u003e\n\u003cli\u003eNing, Y.\u003cem\u003e et al.\u003c/em\u003e Generative artificial intelligence and ethical considerations in health care: a scoping review and ethics checklist. \u003cem\u003eLancet Digit Health.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, e848-e856 (2024). https://doi.org/10.1016/S2589-7500(24)00143-2\u003c/li\u003e\n\u003cli\u003eFernando, M.\u003cem\u003e et al.\u003c/em\u003e Using Theories, Models, and Frameworks to Inform Implementation Cycles of Computerized Clinical Decision Support Systems in Tertiary Health Care Settings: Scoping Review. \u003cem\u003eJ Med Internet Res.\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, e45163 (2023). https://doi.org/10.2196/45163\u003c/li\u003e\n\u003cli\u003eLehoux, P., Rivard, L., de Oliveira, R. R., M\u0026ouml;rch, C. M. \u0026amp; Alami, H. Tools to foster responsibility in digital solutions that operate with or without artificial intelligence: A scoping review for health and innovation policymakers. \u003cem\u003eInt J Med Inform.\u003c/em\u003e \u003cstrong\u003e170\u003c/strong\u003e, 104933 (2023). https://doi.org/10.1016/j.ijmedinf.2022.104933\u003c/li\u003e\n\u003cli\u003eSegur-Ferrer, J., Molto-Puigmarti, C., Pastells-Peiro, R. \u0026amp; Vivanco-Hidalgo, R. M. Methodological Frameworks and Dimensions to Be Considered in Digital Health Technology Assessment: Scoping Review and Thematic Analysis. \u003cem\u003eJ Med Internet Res.\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, e48694 (2024). https://dx.doi.org/10.2196/48694\u003c/li\u003e\n\u003cli\u003eMunn, Z.\u003cem\u003e et al.\u003c/em\u003e Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. \u003cem\u003eBMC Med Res Methodol.\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 143 (2018). https://doi.org/10.1186/s12874-018-0611-x\u003c/li\u003e\n\u003cli\u003eApfelbacher, T., Kocman, S. E., Prokosch, H. U. \u0026amp; Christoph, J. A Governance Framework for the Implementation and Operation of AI Applications in a University Hospital. \u003cem\u003eStud Health Technol Inform.\u003c/em\u003e \u003cstrong\u003e316\u003c/strong\u003e, 776-780 (2024). https://dx.doi.org/10.3233/SHTI240527\u003c/li\u003e\n\u003cli\u003eCallahan, A.\u003cem\u003e et al.\u003c/em\u003e Standing on FURM ground: A framework for evaluating Fair, Useful, and Reliable AI Models in healthcare systems. \u003cem\u003eNEJM Catal Innov Care Deliv.\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e (2024). https://doi.org/10.1056/CAT.24.0131\u003c/li\u003e\n\u003cli\u003eCoalition for Health AI (CHAI). \u003cem\u003eAssurance Standards Guide and Reporting Checklist\u003c/em\u003e, https://chai.org/assurance-standards-guide/ (2024).\u003c/li\u003e\n\u003cli\u003eEconomou-Zavlanos, N. J.\u003cem\u003e et al.\u003c/em\u003e Translating ethical and quality principles for the effective, safe and fair development, deployment and use of artificial intelligence technologies in healthcare. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 705-713 (2024). https://dx.doi.org/10.1093/jamia/ocad221\u003c/li\u003e\n\u003cli\u003eLiao, F., Adelaine, S., Afshar, M. \u0026amp; Patterson, B. W. Governance of Clinical AI applications to facilitate safe and equitable deployment in a large health system: Key elements and early successes. \u003cem\u003eFront Digit Health.\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 931439 (2022). https://doi.org/10.3389/fdgth.2022.931439\u003c/li\u003e\n\u003cli\u003eMorley, J., Morton, C., Karpathakis, K., Taddeo, M. \u0026amp; Floridi, L. Towards a framework for evaluating the safety, acceptability and efficacy of AI systems for health: an initial synthesis. \u003cem\u003eArXiv.\u003c/em\u003e (2021). https://doi.org/10.48550/arXiv.2104.06910\u003c/li\u003e\n\u003cli\u003eSaenz, A. D.\u003cem\u003e et al.\u003c/em\u003e Establishing responsible use of AI guidelines: a comprehensive case study for healthcare institutions. \u003cem\u003eNPJ Digit Med.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 348 (2024). https://doi.org/10.1038/s41746-024-01300-8\u003c/li\u003e\n\u003cli\u003eSingapore Ministry of Health. Artificial Intelligence in Healthcare Guidelines. 2021. https://www.moh.gov.sg/licensing-and-regulation/artificial-intelligence-in-healthcare.\u003c/li\u003e\n\u003cli\u003evan der Vegt, A. H.\u003cem\u003e et al.\u003c/em\u003e Implementation frameworks for end-to-end clinical AI: derivation of the SALIENT framework. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 1503-1515 (2023). https://dx.doi.org/10.1093/jamia/ocad088\u003c/li\u003e\n\u003cli\u003eWhittaker, R.\u003cem\u003e et al.\u003c/em\u003e An example of governance for AI in health services from Aotearoa New Zealand. \u003cem\u003eNPJ Digit Med.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 164 (2023). https://doi.org/10.1038/s41746-023-00882-z\u003c/li\u003e\n\u003cli\u003eAustralian Government Department of Industry \u0026amp; Science and Resources. Australia\u0026rsquo;s AI Ethics Principles. (2022). https://www.industry.gov.au/publications/australias-artificial-intelligence-ethics-framework/australias-ai-ethics-principles.\u003c/li\u003e\n\u003cli\u003eAustralian Medical Association. Position Statement - Artificial Intelligence in Healthcare. 2023. https://www.ama.com.au/sites/default/files/2023-08/Artificial%20Intelligence%20in%20Healthcare%20-%20AMA.pdf.\u003c/li\u003e\n\u003cli\u003eBellini, V., Cascella, M., Montomoli, J. \u0026amp; Bignami, E. From Big Data\u0026apos;s 5Vs to clinical practice\u0026apos;s 5Ws: enhancing data-driven decision making in healthcare. \u003cem\u003eJ Clin Monit Comput.\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 1423-1425 (2023). https://dx.doi.org/10.1007/s10877-023-01007-3\u003c/li\u003e\n\u003cli\u003eCawthorne, D. \u0026amp; Robbins-van Wynsberghe, A. An Ethical Framework for the Design, Development, Implementation, and Assessment of Drones Used in Public Healthcare. \u003cem\u003eSci Eng Ethics.\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 2867-2891 (2020). https://dx.doi.org/10.1007/s11948-020-00233-1\u003c/li\u003e\n\u003cli\u003eChar, D. S., Abramoff, M. D. \u0026amp; Feudtner, C. Identifying Ethical Considerations for Machine Learning Healthcare Applications. \u003cem\u003eAm J Bioeth.\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 7-17 (2020). https://dx.doi.org/10.1080/15265161.2020.1819469\u003c/li\u003e\n\u003cli\u003eDixit, A., Quaglietta, J. \u0026amp; Gaulton, C. Preparing for the future: How organizations can prepare boards, leaders, and risk managers for artificial intelligence. \u003cem\u003eHealthc Manage Forum.\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, 346-352 (2021). https://dx.doi.org/10.1177/08404704211037995\u003c/li\u003e\n\u003cli\u003eJacob, C.\u003cem\u003e et al.\u003c/em\u003e AI for IMPACTS Framework for Evaluating the Long-Term Real-World Impacts of AI-Powered Clinician Tools: Systematic Review and Narrative Synthesis. \u003cem\u003eJ Med Internet Res.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, e67485 (2025). https://dx.doi.org/10.2196/67485\u003c/li\u003e\n\u003cli\u003eKatirai, A. The ethics of advancing artificial intelligence in healthcare: analyzing ethical considerations for Japan\u0026apos;s innovative AI hospital system. \u003cem\u003eFront Public Health.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 1142062 (2023). https://dx.doi.org/10.3389/fpubh.2023.1142062\u003c/li\u003e\n\u003cli\u003eKim, J.\u003cem\u003e et al.\u003c/em\u003e Developing a Framework for Self-regulatory Governance in Healthcare AI Research: Insights from South Korea. \u003cem\u003eAsian Bioeth Rev.\u003c/em\u003e 1-16 (2024). https://doi.org/10.1007/s41649-024-00281-w\u003c/li\u003e\n\u003cli\u003eKim, J. Y.\u003cem\u003e et al.\u003c/em\u003e in \u003cem\u003eProceedings of the 2023 ACM conference on fairness, accountability, and transparency (FAccT \u0026apos;23).\u003c/em\u003e 1396-1417.\u003c/li\u003e\n\u003cli\u003eKim, J. Y.\u003cem\u003e et al.\u003c/em\u003e Development and preliminary testing of Health Equity Across the AI Lifecycle (HEAAL): A framework for healthcare delivery organizations to mitigate the risk of AI solutions worsening health inequities. \u003cem\u003ePLOS Digit Health.\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, e0000390 (2024). https://doi.org/10.1371/journal.pdig.0000390\u003c/li\u003e\n\u003cli\u003eLekadir, K.\u003cem\u003e et al.\u003c/em\u003e FUTURE-AI: International consensus guideline for trustworthy and deployable artificial intelligence in healthcare. \u003cem\u003eArXiv.\u003c/em\u003e (2023). https://doi.org/10.48550/arXiv.2309.12325\u003c/li\u003e\n\u003cli\u003eMirbabaie, M., Hofeditz, L., Frick, N. R. J. \u0026amp; Stieglitz, S. Artificial intelligence in hospitals: providing a status quo of ethical considerations in academia to guide future research. \u003cem\u003eAI Soc.\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 1361-1382 (2022). https://doi.org/10.1007/s00146-021-01239-4\u003c/li\u003e\n\u003cli\u003eNasir, S., Khan, R. A. \u0026amp; Bai, S. Ethical Framework for Harnessing the Power of AI in Healthcare and Beyond. \u003cem\u003eIEEE Access.\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 31014-31035 (2024). https://doi.org/10.1109/ACCESS.2024.3369912\u003c/li\u003e\n\u003cli\u003eNew Zealand Ministry of Health. Emerging Health Technology: Introductory Guidance. (2019). https://www.tewhatuora.govt.nz/assets/Uploads/introductory_guidance_-_algorithms_v0.4_-_web.pdf.\u003c/li\u003e\n\u003cli\u003eOmoumi, P.\u003cem\u003e et al.\u003c/em\u003e To buy or not to buy\u0026mdash;evaluating commercial AI solutions in radiology (the ECLAIR guidelines). \u003cem\u003eEur Radiol.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 3786-3796 (2021). https://doi.org/10.1007/s00330-020-07684-x\u003c/li\u003e\n\u003cli\u003eOniani, D.\u003cem\u003e et al.\u003c/em\u003e Adopting and expanding ethical principles for generative artificial intelligence from military to healthcare. \u003cem\u003eNPJ Digit Med.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 225 (2023). https://dx.doi.org/10.1038/s41746-023-00965-x\u003c/li\u003e\n\u003cli\u003eReddy, S., Allan, S., Coghlan, S. \u0026amp; Cooper, P. A governance model for the application of AI in health care. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 491-497 (2020). https://doi.org/10.1093/jamia/ocz192\u003c/li\u003e\n\u003cli\u003eRoski, J., Maier, E. J., Vigilante, K., Kane, E. A. \u0026amp; Matheny, M. E. Enhancing trust in AI through industry self-governance. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 1582-1590 (2021). https://dx.doi.org/10.1093/jamia/ocab065\u003c/li\u003e\n\u003cli\u003eSolomonides, A. E.\u003cem\u003e et al.\u003c/em\u003e Defining AMIA\u0026apos;s artificial intelligence principles. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 585-591 (2022). https://dx.doi.org/10.1093/jamia/ocac006\u003c/li\u003e\n\u003cli\u003eTaribagil, P., Hogg, H. D. J., Balaskas, K. \u0026amp; Keane, P. A. Integrating artificial intelligence into an ophthalmologist\u0026rsquo;s workflow: obstacles and opportunities. \u003cem\u003eExpert Rev Ophthalmol.\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 45-56 (2023). https://doi.org/10.1080/17469899.2023.2175672\u003c/li\u003e\n\u003cli\u003eThe Royal Australian and New Zealand College of Radiologists. Ethical Principles for AI in Medicine. 2023. https://www.ranzcr.com/search/ethical-principles-for-ai-in-medicine.\u003c/li\u003e\n\u003cli\u003eUK National Health Services (NHS). Digital Technology Assessment Criteria (DTAC). 2021. https://transform.england.nhs.uk/key-tools-and-info/digital-technology-assessment-criteria-dtac/.\u003c/li\u003e\n\u003cli\u003eUS Department of Health and Human Services (DHHS). Trustworthy AI Playbook. 2021. https://www.hhs.gov/sites/default/files/hhs-trustworthy-ai-playbook.pdf.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance., (World Health Organization, Geneva, 2021). https://www.who.int/publications/i/item/9789240029200.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Generating evidence for artificial intelligence-based medical devices: a framework for training, validation and evaluation., (World Health Organization, Geneva, 2021). https://www.who.int/publications/i/item/9789240029200.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. (World Health Organization, Geneva, 2024). https://www.who.int/publications/i/item/9789240084759.\u003c/li\u003e\n\u003cli\u003eAbujaber, A. A. \u0026amp; Nashwan, A. J. Ethical framework for artificial intelligence in healthcare research: A path to integrity. \u003cem\u003eWorld J Methodol.\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 94071 (2024). https://doi.org/10.5662/wjm.v14.i3.94071\u003c/li\u003e\n\u003cli\u003eBrady, A. P.\u003cem\u003e et al.\u003c/em\u003e Developing, purchasing, implementing and monitoring AI tools in radiology: Practical considerations. A multi-society statement from the ACR, CAR, ESR, RANZCR \u0026amp; RSNA. \u003cem\u003eJ Med Imaging Radiat Oncol.\u003c/em\u003e \u003cstrong\u003e68\u003c/strong\u003e, 7-26 (2024). https://doi.org/10.1111/1754-9485.13612\u003c/li\u003e\n\u003cli\u003eLehoux, P.\u003cem\u003e et al.\u003c/em\u003e A Comprehensive, Valid, and Reliable Tool to Assess the Degree of Responsibility of Digital Health Solutions That Operate With or Without Artificial Intelligence: 3-Phase Mixed Methods Study. \u003cem\u003eJ Med Internet Res.\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, e48496 (2023). https://doi.org/10.2196/48496\u003c/li\u003e\n\u003cli\u003eMcCradden, M. D.\u003cem\u003e et al.\u003c/em\u003e A Research Ethics Framework for the Clinical Translation of Healthcare Machine Learning. \u003cem\u003eAm J Bioeth.\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 8-22 (2022). https://doi.org/10.1080/15265161.2021.2013977\u003c/li\u003e\n\u003cli\u003eNebeker, C., Torous, J. \u0026amp; Bartlett Ellis, R. J. Building the case for actionable ethics in digital health research supported by artificial intelligence. \u003cem\u003eBMC Med.\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 137 (2019). https://doi.org/10.1186/s12916-019-1377-7\u003c/li\u003e\n\u003cli\u003eThe Australasian College of Dermatologists. Position Statement - Use of Artificial Intelligence in Dermatology in Australia. 2022. https://www.dermcoll.edu.au/wp-content/uploads/2022/11/ACD-Position-Statement-Use-of-Artifical-Intelligence-in-Dermatology-in-Australia-Nov-2022.pdf.\u003c/li\u003e\n\u003cli\u003eUK Government Department of Health and Social Care. A guide to good practice for digital and data-driven health technologies. 2021. https://www.gov.uk/government/publications/code-of-conduct-for-data-driven-health-and-care-technology/initial-code-of-conduct-for-data-driven-health-and-care-technology.\u003c/li\u003e\n\u003cli\u003eUK Government National Institute for Health and Care Excellence (NICE). Evidence standards framework (ESF) for digital health technologies. 2022. https://www.nice.org.uk/corporate/ecd7/resources/evidence-standards-framework-for-digital-health-technologies-pdf-1124017457605.\u003c/li\u003e\n\u003cli\u003eVollmer, S.\u003cem\u003e et al.\u003c/em\u003e Machine learning and artificial intelligence research for patient benefit: 20 critical questions on transparency, replicability, ethics, and effectiveness. \u003cem\u003eBMJ.\u003c/em\u003e \u003cstrong\u003e368\u003c/strong\u003e, l6927 (2020). https://doi.org/10.1136/bmj.l6927\u003c/li\u003e\n\u003cli\u003eYan, A. P.\u003cem\u003e et al.\u003c/em\u003e A roadmap to implementing machine learning in healthcare: from concept to practice. \u003cem\u003eFront Digit Health.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e (2025). https://doi.org/10.3389/fdgth.2025.1462751\u003c/li\u003e\n\u003cli\u003eHassan, M., Borycki, E. M. \u0026amp; Kushniruk, A. W. Artificial intelligence governance framework for healthcare. \u003cem\u003eHealthc Manage Forum.\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 125-130 (2024). https://dx.doi.org/10.1177/08404704241291226\u003c/li\u003e\n\u003cli\u003eKawamoto, K., Finkelstein, J. \u0026amp; Del Fiol, G. Implementing Machine Learning in the Electronic Health Record: Checklist of Essential Considerations. \u003cem\u003eMayo Clin Proc.\u003c/em\u003e \u003cstrong\u003e98\u003c/strong\u003e, 366-369 (2023). https://dx.doi.org/10.1016/j.mayocp.2023.01.013\u003c/li\u003e\n\u003cli\u003eNwosu, O. I., Crowson, M. G. \u0026amp; Rameau, A. Artificial Intelligence Governance and Otolaryngology-Head and Neck Surgery. \u003cem\u003eLaryngoscope.\u003c/em\u003e \u003cstrong\u003e133\u003c/strong\u003e, 2868-2870 (2023). https://doi.org/10.1002/lary.31013\u003c/li\u003e\n\u003cli\u003eVasey, B.\u003cem\u003e et al.\u003c/em\u003e Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. \u003cem\u003eBMJ.\u003c/em\u003e \u003cstrong\u003e377\u003c/strong\u003e, e070904 (2022). https://doi.org/10.1136/bmj-2022-070904\u003c/li\u003e\n\u003cli\u003eLiu, X.\u003cem\u003e et al.\u003c/em\u003e Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. \u003cem\u003eLancet Digit Health.\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, e537-e548 (2020). https://doi.org/10.1016/S2589-7500(20)30218-1\u003c/li\u003e\n\u003cli\u003eReddy, S.\u003cem\u003e et al.\u003c/em\u003e Evaluation framework to guide implementation of AI systems into healthcare settings. \u003cem\u003eBMJ Health Care Inform.\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e (2021). https://dx.doi.org/10.1136/bmjhci-2021-100444\u003c/li\u003e\n\u003cli\u003eBertl, M., Ross, P. \u0026amp; Draheim, D. Systematic AI support for decision-making in the healthcare sector: Obstacles and success factors. \u003cem\u003eHealth Policy and Technology.\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 100748 (2023). https://doi.org/10.1016/j.hlpt.2023.100748\u003c/li\u003e\n\u003cli\u003eCresswell, K., Williams, R. \u0026amp; Sheikh, A. Developing and applying a formative evaluation framework for health information technology implementations: Qualitative investigation. \u003cem\u003eJ Med Internet Res.\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 15068 (2020). https://dx.doi.org/10.2196/15068\u003c/li\u003e\n\u003cli\u003eUK National Health Services (NHS). A Buyer\u0026apos;s Guide to AI in Health and Care. 2020. https://transform.england.nhs.uk/ai-lab/explore-all-resources/adopt-ai/a-buyers-guide-to-ai-in-health-and-care/.\u003c/li\u003e\n\u003cli\u003eVan de Sande, D.\u003cem\u003e et al.\u003c/em\u003e Developing, implementing and governing artificial intelligence in medicine: a step-by-step approach to prevent an artificial intelligence winter. \u003cem\u003eBMJ Health Care Inform.\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e (2022). https://doi.org/10.1136/bmjhci-2021-100495\u003c/li\u003e\n\u003cli\u003eMarconi, L., Pirovano, E. \u0026amp; Cabitza, F. CLARITY AI: A Comprehensive Checklist Integrating Established Frameworks for Enhanced Research Quality in Medical AI Studies. Bolzano, Italy, 2025. \u003c/li\u003e\n\u003cli\u003eSendak, M. P.\u003cem\u003e et al.\u003c/em\u003e A Path for Translation of Machine Learning Products into Healthcare Delivery. \u003cem\u003eEuro Med J Innov.\u003c/em\u003e (2020). https://doi.org/10.33590/emjinnov/19-00172\u003c/li\u003e\n\u003cli\u003eTruong, T., Gilbank, P., Johnson-Cover, K. \u0026amp; Ieraci, A. A Framework for Applied AI in Healthcare. \u003cem\u003eStud Health Technol Inform.\u003c/em\u003e \u003cstrong\u003e264\u003c/strong\u003e, 1993-1994 (2019). https://dx.doi.org/10.3233/SHTI190751\u003c/li\u003e\n\u003cli\u003eHernandez-Boussard, T., Bozkurt, S., Ioannidis, J. P. A. \u0026amp; Shah, N. H. MINIMAR (MINimum Information for Medical AI Reporting): Developing reporting standards for artificial intelligence in health care. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 2011-2015 (2020). https://dx.doi.org/10.1093/jamia/ocaa088\u003c/li\u003e\n\u003cli\u003eKwong, J. C. C., Nickel, G. C., Wang, S. C. Y. \u0026amp; Kvedar, J. C. Integrating artificial intelligence into healthcare systems: more than just the algorithm. \u003cem\u003eNPJ Digit Med.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e (2024). https://doi.org/10.1038/s41746-024-01066-z\u003c/li\u003e\n\u003cli\u003eReddy, S. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. \u003cem\u003eImplement Sci.\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 27 (2024). https://dx.doi.org/10.1186/s13012-024-01357-9\u003c/li\u003e\n\u003cli\u003eWiens, J.\u003cem\u003e et al.\u003c/em\u003e Do no harm: a roadmap for responsible machine learning for health care. \u003cem\u003eNat Med.\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 1337-1340 (2019). https://doi.org/10.1038/s41591-019-0548-6\u003c/li\u003e\n\u003cli\u003eSmith, T. G.\u003cem\u003e et al.\u003c/em\u003e Creating a Practical Transformational Change Management Model for Novel Artificial Intelligence-Enabled Technology Implementation in the Operating Room. \u003cem\u003eMayo Clin Proc Innov Qual Outcomes.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 584-596 (2022). https://dx.doi.org/10.1016/j.mayocpiqo.2022.09.004\u003c/li\u003e\n\u003cli\u003eChae, A.\u003cem\u003e et al.\u003c/em\u003e Strategies for Implementing Machine Learning Algorithms in the Clinical Practice of Radiology. \u003cem\u003eRadiology.\u003c/em\u003e \u003cstrong\u003e310\u003c/strong\u003e (2024). https://doi.org/10.1148/radiol.223170\u003c/li\u003e\n\u003cli\u003eRaza, M. M., Venkatesh, K. P. \u0026amp; Kvedar, J. C. Generative AI and large language models in health care: pathways to implementation. \u003cem\u003eNPJ Digit Med.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 62 (2024). https://doi.org/10.1038/s41746-023-00988-4\u003c/li\u003e\n\u003cli\u003eSalwei, M. E.\u003cem\u003e et al.\u003c/em\u003e Workflow integration analysis of a human factors-based clinical decision support in the emergency department. \u003cem\u003eApplied Ergonomics.\u003c/em\u003e \u003cstrong\u003e97\u003c/strong\u003e, 103498 (2021). https://dx.doi.org/10.1016/j.apergo.2021.103498\u003c/li\u003e\n\u003cli\u003eBelete, G. F., Voinov, A. \u0026amp; Laniak, G. F. An overview of the model integration process: From pre-integration assessment to testing. \u003cem\u003eEnvironmental Modelling \u0026amp; Software.\u003c/em\u003e \u003cstrong\u003e87\u003c/strong\u003e, 49-63 (2017). https://doi.org/10.1016/j.envsoft.2016.10.013\u003c/li\u003e\n\u003cli\u003eBeauchamp, T. L. \u0026amp; Childress, J. F. \u003cem\u003ePrinciples of biomedical ethics\u003c/em\u003e. Eighth edn, (Oxford University Press, 2019).\u003c/li\u003e\n\u003cli\u003eAlfrink, K., Keller, I., Kortuem, G. \u0026amp; Doorn, N. Contestable AI by Design: Towards a Framework. \u003cem\u003eMinds and Machines.\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 613-639 (2023). https://doi.org/10.1007/s11023-022-09611-z\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. \u003cem\u003eClimate change\u003c/em\u003e, https://www.who.int/news-room/fact-sheets/detail/climate-change-and-health (2023).\u003c/li\u003e\n\u003cli\u003eCoiera, E. \u0026amp; Magrabi, F. What did you do to avoid the climate disaster? A call to arms for health informatics. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 1997-1999 (2022). https://doi.org/10.1093/jamia/ocac185\u003c/li\u003e\n\u003cli\u003eDoo, F. X.\u003cem\u003e et al.\u003c/em\u003e Environmental Sustainability and AI in Radiology: A Double-Edged Sword. \u003cem\u003eRadiology.\u003c/em\u003e \u003cstrong\u003e310\u003c/strong\u003e, e232030 (2024). https://doi.org/10.1148/radiol.232030\u003c/li\u003e\n\u003cli\u003eBraithwaite, J.\u003cem\u003e et al.\u003c/em\u003e Strategies and tactics to reduce the impact of healthcare on climate change: systematic review. \u003cem\u003eBMJ.\u003c/em\u003e \u003cstrong\u003e387\u003c/strong\u003e, e081284 (2024). https://doi.org/10.1136/bmj-2024-081284\u003c/li\u003e\n\u003cli\u003eAnthony, L. F. W., Kanding, B. \u0026amp; Selvan, R. Carbontracker: Tracking and predicting the carbon footprint of training deep learning models. \u003cem\u003eArXiv.\u003c/em\u003e (2020). https://doi.org/10.48550/arXiv.2007.03051\u003c/li\u003e\n\u003cli\u003eThelisson, E., Mika, G., Schneiter, Q., Padh, K. \u0026amp; Verma, H. Toward Responsible AI Use: Considerations for Sustainability Impact Assessment. \u003cem\u003eArXiv.\u003c/em\u003e (2023). https://doi.org/10.48550/arXiv.2312.11996\u003c/li\u003e\n\u003cli\u003eRohde, F.\u003cem\u003e et al.\u003c/em\u003e Broadening the perspective for sustainable artificial intelligence: sustainability criteria and indicators for Artificial Intelligence systems. \u003cem\u003eCurrent Opinion in Environmental Sustainability.\u003c/em\u003e \u003cstrong\u003e66\u003c/strong\u003e, 101411 (2024). https://doi.org/10.1016/j.cosust.2023.101411\u003c/li\u003e\n\u003cli\u003eBarrowcliffe, R.\u003cem\u003e et al.\u003c/em\u003e Envisioning Aboriginal and Torres Strait Islander AI Futures Communique: March 2025.\u003cem\u003e Journal of Global Indigeneity.\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e (2025). https://doi.org/10.54760/001c.133656\u003c/li\u003e\n\u003cli\u003eLewis, J. E. Indigenous Protocol and Artificial Intelligence Position Paper. (The Initiative for Indigenous Futures and the Canadian Institute for Advanced Research (CIFAR). Honolulu, Hawai\u0026apos;i, 2020). \u003c/li\u003e\n\u003cli\u003eCommonwealth of Australia. Framework for Governance of Indigenous Data. NIAA, (2024). https://www.niaa.gov.au/resource-centre/framework-governance-indigenous-data.\u003c/li\u003e\n\u003cli\u003eMaiam nayri Wingara Indigenous Data Sovereignty Collective. Taking Control of Our Data: A Discussion Paper on Indigenous Data Governance for Aboriginal and Torres Strait Islander People and Communities. (Lowitja Institute, Melbourne, 2024). https://www.lowitja.org.au/resource/taking-control-of-our-data-a-discussion-paper-on-indigenous-data-governance-for-aboriginal-and-torres-strait-islander-people-and-communities/.\u003c/li\u003e\n\u003cli\u003eDe Zilva, S., Walker, T., Palermo, C. \u0026amp; Brimblecombe, J. Culturally safe health care practice for Indigenous Peoples in Australia: A systematic meta-ethnographic review. \u003cem\u003eJ Health Serv Res Policy.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 74-84 (2022). https://doi.org/10.1177/13558196211041835\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eNational Agreement on Closing the Gap\u003c/em\u003e, https://www.closingthegap.gov.au/national-agreement.\u003c/li\u003e\n\u003cli\u003eCoiera, E. \u0026amp; Fraile-Navarro, D. AI as an Ecosystem \u0026mdash; Ensuring Generative AI Is Safe and Effective. \u003cem\u003eNEJM AI.\u003c/em\u003e \u003cstrong\u003e1\u003c/strong\u003e, AIp2400611 (2024). https://doi.org/10.1056/AIp2400611\u003c/li\u003e\n\u003cli\u003eDuffourc, M. \u0026amp; Gerke, S. Generative AI in Health Care and Liability Risks for Physicians and Safety Concerns for Patients. \u003cem\u003eJAMA.\u003c/em\u003e \u003cstrong\u003e330\u003c/strong\u003e, 313-314 (2023). https://doi.org/10.1001/jama.2023.9630\u003c/li\u003e\n\u003cli\u003eMoulaei, K.\u003cem\u003e et al.\u003c/em\u003e Generative artificial intelligence in healthcare: A scoping review on benefits, challenges and applications. \u003cem\u003eInt J Med Inform.\u003c/em\u003e \u003cstrong\u003e188\u003c/strong\u003e, 105474 (2024). https://doi.org/10.1016/j.ijmedinf.2024.105474\u003c/li\u003e\n\u003cli\u003eThe Lancet Regional Health \u0026ndash;, E. Embracing generative AI in health care. \u003cem\u003eThe Lancet Regional Health \u0026ndash; Europe.\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e (2023). https://doi.org/10.1016/j.lanepe.2023.100677\u003c/li\u003e\n\u003cli\u003eParker, V. J., Economou, N. J. \u0026amp; Silcox, C. AI Governance in Health Systems: Aligning Innovation, Accountability, and Trust. Duke-Margolis Institute for Health Policy, 2024. https://healthpolicy.duke.edu/publications/ai-governance-health-systems-aligning-innovation-accountability-and-trust.\u003c/li\u003e\n\u003cli\u003eBedoya, A. D.\u003cem\u003e et al.\u003c/em\u003e A framework for the oversight and local deployment of safe and high-quality prediction models. \u003cem\u003eJ Am Med Inform Assoc.\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 1631-1636 (2022). https://doi.org/10.1093/jamia/ocac078\u003c/li\u003e\n\u003cli\u003eArksey, H. \u0026amp; and O\u0026apos;Malley, L. Scoping studies: towards a methodological framework. \u003cem\u003eInt J Soc Res Methodol.\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 19-32 (2005). https://doi.org/10.1080/1364557032000119616\u003c/li\u003e\n\u003cli\u003eLevac, D., Colquhoun, H. \u0026amp; O\u0026apos;Brien, K. K. Scoping studies: advancing the methodology. \u003cem\u003eImplement Sci.\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 69 (2010). https://doi.org/10.1186/1748-5908-5-69\u003c/li\u003e\n\u003cli\u003eTricco, A. C.\u003cem\u003e et al.\u003c/em\u003e PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. \u003cem\u003eAnn Intern Med.\u003c/em\u003e \u003cstrong\u003e169\u003c/strong\u003e, 467-473 (2018). https://doi.org/10.7326/m18-0850\u003c/li\u003e\n\u003cli\u003eHaddaway, N. R., Collins, A. M., Coughlin, D. \u0026amp; Kirk, S. The Role of Google Scholar in Evidence Reviews and Its Applicability to Grey Literature Searching. \u003cem\u003ePLoS One.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, e0138237 (2015). https://doi.org/10.1371/journal.pone.0138237\u003c/li\u003e\n\u003cli\u003eNilsen, P. Making sense of implementation theories, models and frameworks. \u003cem\u003eImplement Sci.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 53 (2015). https://doi.org/10.1186/s13012-015-0242-0\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7209096/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7209096/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis scoping review synthesises existing knowledge about AI governance in healthcare organisations by identifying key ethics and governance principles, and key components of AI governance frameworks. Following PRISMA-ScR guidelines, we searched MEDLINE, Embase, and Scopus (April 2024, updated March 2025) for AI governance frameworks in acute care. Seventy-seven frameworks were identified and examined for: 1) Guiding principles (ethics or governance-related); 2) Assessment methods; 3) AI life cycle stages; and 4) Oversight mechanisms. Most lacked real-world applicability and missed key principles or components such as an oversight mechanism. Only 10 frameworks (13.0%) included all four framework components, with oversight mechanisms (e.g. AI-specific governance committee) being the least common (n\u0026thinsp;=\u0026thinsp;15, 19.5%). No framework had been evaluated for effectiveness in enabling safe and responsible AI. There is a need to move beyond principles to implementing AI governance frameworks in healthcare organisations and assessing their real-world impact.\u003c/p\u003e","manuscriptTitle":"Governance for safe and responsible AI in healthcare organisations: A scoping review of frameworks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-19 14:59:15","doi":"10.21203/rs.3.rs-7209096/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-10T19:47:41+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-07T19:25:47+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-03T14:35:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"291988225573478146944943046337998740060","date":"2025-08-26T11:37:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"280087920925987918893410094316518644870","date":"2025-08-26T09:19:03+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-11T13:15:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-01T16:02:00+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-01T08:41:44+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2025-07-25T00:03:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ca16bf51-37df-4b49-b1e4-7981c9b5e823","owner":[],"postedDate":"August 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":53151093,"name":"Health sciences/Health care"},{"id":53151094,"name":"Health sciences/Medical research"},{"id":53151095,"name":"Scientific community and society/Scientific community"}],"tags":[],"updatedAt":"2026-05-04T15:59:26+00:00","versionOfRecord":{"articleIdentity":"rs-7209096","link":"https://doi.org/10.1038/s41746-026-02679-2","journal":{"identity":"npj-digital-medicine","isVorOnly":false,"title":"npj Digital Medicine"},"publishedOn":"2026-05-01 15:57:27","publishedOnDateReadable":"May 1st, 2026"},"versionCreatedAt":"2025-08-19 14:59:15","video":"","vorDoi":"10.1038/s41746-026-02679-2","vorDoiUrl":"https://doi.org/10.1038/s41746-026-02679-2","workflowStages":[]},"version":"v1","identity":"rs-7209096","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7209096","identity":"rs-7209096","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}