The ethics-driven data governance framework for health-based AI (EDG-AI4H): an integrated approach for empowering explainable AI in health insurance risk prediction in Iran Authors: Rezvan Rahimi Published: 2026 ABSTRACT AbstractPurposeThe digital transformation of health data has enabled predictive models in insurance. In Iran, the absence of an indigenous data governance framework integrating ethical, legal, and technical considerations poses serious risks to responsible AI development. This study aimed to design an ethics-driven data governance framework to empower Explainable AI (XAI) in health insurance risk prediction in Iran.MethodsThis qualitative study (2024) used thematic analysis and framework development. Data came from: (1) a narrative review of international articles and Iranian policy documents; (2) semi-structured interviews with five experts in insurance IT, health data science, IT law, epidemiology, and insurance management. Data were analyzed using Braun and Clarke's six-step thematic analysis in MAXQDA 2022. The framework was validated through a focus group with the same five experts to ensure continuity and in-depth engagement with the framework design. The study adheres to COREQ guidelines, and a completed COREQ checklist is provided as Supplementary File 1.ResultsThe EDG-AI4H framework comprises four hierarchical and interactive layers: (1) Governance-Ethics (data ownership policies, dynamic consent management, Data Ethics Board, legal compliance); (2) Operational-Integration (metadata-driven catalog, standardized data exchange, data quality assurance, secure storage architecture); (3) Intelligence-Analytics (domain-driven feature engineering, mandatory XAI-by-design, active bias monitoring, model lifecycle governance); and (4) Value-Decision Making (multi-level decision support, preventive and personalized insurance, population health management, policyholder transparency and empowerment).ConclusionThe EDG-AI4H framework addresses a critical gap in Iran's health insurance sector by linking ethics, technology, and business. It offers a roadmap for responsible, transparent, and trustworthy AI in health risk prediction, potentially reducing ethical and legal risks while fostering public trust. The framework's operational feasibility requires empirical validation through future pilot implementation studies.Keywords: Ethics-driven data governance, Health insurance, Explainable Artificial Intelligence (XAI), Health risk prediction, Conceptual framework, Qualitative research Introduction Problem statement and its significance The global health insurance industry is undergoing a fundamental transformation driven by digital technologies and Artificial Intelligence (AI) [1]. The ability to analyze vast amounts of health data to predict future risks holds the promise of revolutionizing this industry's business models—from designing personalized products and preventive cost management to fraud detection [2]. However, this significant opportunity comes alongside unprecedented ethical and governance challenges. Health data is highly sensitive, and its use demands strict adherence to privacy, informed consent, and the prevention of discrimination [3]. In Iran, despite the existence of rich databases within insurance organizations and medical centers, the move towards utilizing AI for risk prediction faces numerous structural obstacles. The absence of a coherent data governance framework that can comprehensively address issues of data ownership, exchange standards, data quality, security, privacy, and especially AI ethics is considered the primary barrier [4]. Without such a framework, the development of AI models could not only lead to violations of policyholders' rights, unfair decision-making, and the reinforcement of existing biases but also face the risk of social non-acceptance and legal hurdles [5]. The necessity of addressing this issue becomes even more apparent when considering the gap in the domestic literature. A review of studies conducted in Iran shows that despite scattered research in health data mining, very few studies have focused on designing integrated, ethics-driven data governance frameworks to guide AI in the health insurance industry. This research aims to fill this important gap by designing an indigenous conceptual framework. Research objectives Primary objective: To design an integrated, indigenous conceptual framework for data governance in Iran's health insurance industry that enables the responsible development and deployment of Explainable AI (XAI) for predicting health risks. Secondary objectives: 1.Identifying key data governance and ethical challenges in using AI for risk prediction in Iranian health insurance from the perspective of experts. Identifying key data governance and ethical challenges in using AI for risk prediction in Iranian health insurance from the perspective of experts. 2.Extracting the essential requirements and components of an effective, ethics-driven data governance framework in this domain. Extracting the essential requirements and components of an effective, ethics-driven data governance framework in this domain. 3.Delineating the relationships between different layers of the framework (governance, operational, analytical, and value) and providing a roadmap for implementation. Delineating the relationships between different layers of the framework (governance, operational, analytical, and value) and providing a roadmap for implementation. Research innovation The innovation of this research lies in its deep interdisciplinary integration and the presentation of the first indigenous framework that links five key domains: 1) health insurance and risk management, 2) data science and explainable AI, 3) data governance and information architecture, 4) medical ethics and privacy, and 5) national laws and regulations. This framework is not merely theoretical but is designed considering the realities of Iran's health and insurance ecosystem. Theoretical foundations and background Data governance and AI ethics in the health domain Data governance refers to a set of processes, policies, standards, and roles that determine how data is managed, used, and protected within an organization [6]. In the health domain, this concept is intertwined with additional challenges, including the high sensitivity of health data, the necessity of privacy compliance (such as HIPAA and GDPR principles), and medical ethics requirements (principles of beneficence, non-maleficence, autonomy, and justice) [7],Simultaneously, the emergence of AI, particularly its subfields like machine learning, has enabled predictive analysis of health risks [8]. However, the “black box” nature of many complex AI models has raised serious concerns about transparency, accountability, algorithmic bias, and discrimination [9]. These challenges have positioned “Explainable AI” (XAI) as an essential paradigm for building trust, ensuring fairness, and facilitating social acceptance of these technologies in sensitive domains like health insurance [10]. Explainable AI (XAI): from concept to technology Explainable AI refers to a set of methods, techniques, and frameworks in the design and development of AI systems that make understanding, interpreting, and trusting the outputs and decisions of these systems possible for human stakeholders (from developers and regulators to end-users)[11]. In health risk prediction models, XAI answers questions like “Why was this individual classified as high-risk for disease X?” or “Which factors contributed most to this prediction?” Leading techniques in this area include intrinsic methods (like simpler, inherently interpretable models such as decision trees) and post-hoc methods (like LIME and SHAP) that generate explanations at the sample or model level after building more complex models [12], [13]. Successful XAI implementation requires its integration into the AI development lifecycle and data governance. Ethics-driven data governance: a bridge between technology and values An ethics-driven data governance framework is a mechanism for translating abstract ethical principles (such as privacy, fairness, transparency, and accountability) into policies, technical controls, and operational processes within an organization [14]. In the context of using XAI for risk prediction in health insurance, such a framework must address key questions: •Data Ownership and Control: Who owns and controls health data? How is informed consent obtained and managed? Data Ownership and Control: Who owns and controls health data? How is informed consent obtained and managed? •Data Quality and Bias: How is it ensured that training data is complete, accurate, and unbiased to prevent perpetuating or exacerbating health inequalities? Data Quality and Bias: How is it ensured that training data is complete, accurate, and unbiased to prevent perpetuating or exacerbating health inequalities? •Transparency and Explainability: What level of explanation is necessary for which stakeholder (policyholder, physician, regulator)? How should explanations be presented? Transparency and Explainability: What level of explanation is necessary for which stakeholder (policyholder, physician, regulator)? How should explanations be presented? •Accountability: In the event of a prediction error or discriminatory decision, who is responsible? What is the mechanism for complaint and redress? Accountability: In the event of a prediction error or discriminatory decision, who is responsible? What is the mechanism for complaint and redress? •Security and Privacy: How is data protected against unauthorized access and use throughout its lifecycle? Security and Privacy: How is data protected against unauthorized access and use throughout its lifecycle? International experiences: lessons for framework design Experiences from leading countries indicate that success in this path requires a holistic, multi-layered approach: •European Union: The GDPR regulation establishes the right to an “explanation” for individuals regarding automated decisions, and its proposed AI Act mandates high-risk systems (like health insurance) to comply with stringent requirements for privacy, transparency, and human oversight [15], [16]. European Union: The GDPR regulation establishes the right to an “explanation” for individuals regarding automated decisions, and its proposed AI Act mandates high-risk systems (like health insurance) to comply with stringent requirements for privacy, transparency, and human oversight [15], [16]. •Singapore: The Model AI Governance Framework [17] focuses on accountability, transparency, fairness, and human-centric ethics, offering an operational and testable approach for organizations [18]. Singapore: The Model AI Governance Framework [17] focuses on accountability, transparency, fairness, and human-centric ethics, offering an operational and testable approach for organizations [18]. •United Kingdom: The Information Commissioner's Office (ICO) and the Centre for Data Ethics and Innovation (CDEI) have published practical guidance on “explainability” in AI-based decisions, emphasizing the provision of “useful, understandable, and actionable” explanations for individuals [19], [20], [21]. United Kingdom: The Information Commissioner's Office (ICO) and the Centre for Data Ethics and Innovation (CDEI) have published practical guidance on “explainability” in AI-based decisions, emphasizing the provision of “useful, understandable, and actionable” explanations for individuals [19], [20], [21]. These experiences underscore the necessity of interdisciplinary collaboration (law, ethics, technology, medicine, and management), the development of technical standards for explainability, and the establishment of specialized regulatory bodies. Domestic background: identifying the research gap In Iran, despite the richness of health data in insurance organizations and medical centers, and some initial efforts in health IT, coherent research specifically focused on designing an ethics-driven data governance framework to empower XAI in health insurance is very limited. Existing studies often concentrate on technical aspects of data mining and machine learning in health [22] or general IT challenges in the health system [23]. A major gap in the domestic literature is the lack of an integrated, indigenous framework that can simultaneously and harmoniously address the technical requirements of XAI, ethical and religious principles, national laws (such as the Personal Data Protection Law), and the managerial and economic realities of Iran's health insurance industry. This research aims to take a step towards filling this gap by using qualitative methods and drawing on expert opinions, offering a roadmap for a responsible and trust-based movement towards utilizing AI in health insurance risk prediction. Research methodology Research design This research is a qualitative study with a conceptual framework development approach. The goal is to generate a deep understanding of the problem and provide a novel, indigenous theoretical structure for it. Participants and sampling Using purposive sampling with a maximum variation strategy, five prominent Iranian specialists were selected, representing the five key related domains: IT and data governance, AI and data science, IT law and privacy, epidemiology and health information systems, and strategic insurance management. Selection criteria included: 1) at least 10 years of specialized experience, 2) holding a policy-making position, senior management role, or recognized academic standing, and 3) having demonstrated understanding of data challenges in Iran's health or insurance domain. All five invited experts agreed to participate, and no dropouts occurred during the study. The experts' profiles are presented in Table 1. Table 1Profile of participating experts.Expert IDKey expertise domainExperience (years)Organizational roleInstitutional affiliationSectorSpecific expertiseH1IT and Data Governance in Insurance16Deputy of ITMajor Private Insurance CompanyPrivateHealth data architecture, data governanceH2AI and Data Science in Health13Head of Data Mining GroupTop-tier Medical Research CenterAcademicMachine learning, predictive modelingH3IT Law, Privacy, and Legislation11University Faculty Member; Legal ConsultantUniversity of TehranAcademicHealth data privacy, regulatory complianceH4Epidemiology and Health Information Systems19Professor; Senior AdvisorMinistry of HealthPublicPopulation health, health informaticsH5Strategic Management and Innovation in Insurance22Senior Advisor; Former Board MemberInsurance Industry AssociationPrivate/PublicInsurance business models, innovation strategy Profile of participating experts. The sample size of five experts was justified based on the exploratory and conceptual nature of this framework development study, which prioritizes depth of expertise over statistical generalizability. Given the highly specialized nature of the required expertise—spanning five distinct domains—the number of experts meeting all inclusion criteria in Iran is inherently limited. We assessed thematic saturation systematically after each interview [24]. By the fourth interview, no new codes or themes emerged, and the fifth interview confirmed saturation within the five disciplinary domains represented by the expert panel. The maximum variation strategy ensured that despite the small sample, perspectives from all relevant disciplinary and professional domains were captured. However, it is important to distinguish between thematic saturation within the domains sampled and the adequacy of disciplinary coverage. While the five experts provide breadth across five key domains, this does not necessarily establish saturation across the entire Iranian health-insurance ecosystem. The findings should be interpreted as exploratory and context-specific, providing a foundation for future empirical validation with broader stakeholder groups. Data collection Data were collected through two parallel streams: 1.Documentary/Narrative Review: A structured narrative review of international articles and relevant domestic policy documents was conducted to extract initial components and identify existing literature. We searched PubMed, Scopus, Web of Science, Google Scholar, and Persian databases (Magiran, SID) for publications between 2015 and 2024 using keywords including ‘health data governance,’ ‘AI ethics,’ ‘explainable AI,’ ‘health insurance,’ and ‘risk prediction.’ Relevant articles were screened and included for thematic extraction. Additionally, relevant Iranian policy documents, including the Personal Data Protection Law and health information exchange guidelines, were reviewed. Documentary/Narrative Review: A structured narrative review of international articles and relevant domestic policy documents was conducted to extract initial components and identify existing literature. We searched PubMed, Scopus, Web of Science, Google Scholar, and Persian databases (Magiran, SID) for publications between 2015 and 2024 using keywords including ‘health data governance,’ ‘AI ethics,’ ‘explainable AI,’ ‘health insurance,’ and ‘risk prediction.’ Relevant articles were screened and included for thematic extraction. Additionally, relevant Iranian policy documents, including the Personal Data Protection Law and health information exchange guidelines, were reviewed. 2.In-depth Semi-structured Interviews: For each expert, an individual interview session (in-person or virtual) lasting 60 to 90 min was conducted using a researcher-made interview guide. The interview guide was developed based on the literature review findings and was pilot-tested with two non-participant experts (one in health informatics and one in insurance management) to refine question wording, ensure clarity, and estimate appropriate interview duration. The final interview guide comprised 12 open-ended questions organized around four main themes: (a) Diagnosis and analysis of existing challenges in health insurance data governance; (b) Outlining requirements and principles of an ideal ethics-driven framework; (c) Visualizing components and their relationships across governance, operational, analytical, and value layers; and (d) Assessing implementation barriers and facilitators specific to Iran's context. All interviews were recorded with participants' full consent and subsequently transcribed verbatim. Potential participants were contacted via formal invitation letters and follow-up phone calls. Interviews were conducted at participants' workplace offices or via secure video conferencing platforms according to participant preference. No other individuals were present during the interviews. Field notes were taken during and immediately after each interview to capture contextual observations and reflections. In-depth Semi-structured Interviews: For each expert, an individual interview session (in-person or virtual) lasting 60 to 90 min was conducted using a researcher-made interview guide. The interview guide was developed based on the literature review findings and was pilot-tested with two non-participant experts (one in health informatics and one in insurance management) to refine question wording, ensure clarity, and estimate appropriate interview duration. The final interview guide comprised 12 open-ended questions organized around four main themes: (a) Diagnosis and analysis of existing challenges in health insurance data governance; (b) Outlining requirements and principles of an ideal ethics-driven framework; (c) Visualizing components and their relationships across governance, operational, analytical, and value layers; and (d) Assessing implementation barriers and facilitators specific to Iran's context. All interviews were recorded with participants' full consent and subsequently transcribed verbatim. Potential participants were contacted via formal invitation letters and follow-up phone calls. Interviews were conducted at participants' workplace offices or via secure video conferencing platforms according to participant preference. No other individuals were present during the interviews. Field notes were taken during and immediately after each interview to capture contextual observations and reflections. Data analysis The interview data were analyzed using Thematic Analysis based on Braun and Clarke's [25], [26] six-step model with the aid of MAXQDA software version 2022. The analysis process involved six stages: (1) deep familiarization with the data through repeated reading of the texts; (2) generating initial codes from prominent data features; (3) searching for and organizing themes by categorizing extracted codes; (4) reviewing themes at the level of codes and the entire dataset; (5) final definition and naming of themes; and (6) preparing an analytical report of the findings. Initial coding was performed independently by the primary researcher. To enhance reliability, a second researcher (Dr. Hanieh Zehtab Hashemi, PhD in Medical Informatics, Virtual University of Medical Sciences, Tehran, Iran), who was not involved in data collection, independently coded 20% of the interview transcripts (two randomly selected interviews). Inter-coder agreement was assessed through a qualitative consensus approach rather than a formal quantitative statistic. After independent coding, the two coders met to compare their coding decisions. All discrepancies and disagreements were discussed in joint sessions until full consensus was reached on all coding decisions. This consensus-based approach is widely recognized as an appropriate and rigorous method for establishing inter-coder agreement in qualitative thematic analysis [25], [27]. The final coding structure, comprising 84 initial codes, 16 sub-themes, and 4 main themes, was reviewed and refined through iterative team discussions. Ethical considerations and research quality assurance This research was conducted in accordance with the Declaration of Helsinki and after obtaining an ethics code from Research Ethics Committees of School of Public Health & Allied Medical Sciences- Tehran University of Medical Sciences (Code: IR.TUMS.SPH.REC.1402.171), and informed consent was obtained orally and in writing from all participants. Confidentiality and the right to withdraw at any stage were guaranteed. To enhance the validity and trustworthiness of the qualitative research, multiple strategies were employed, including member checking by sending extracted theme summaries to each expert for confirmation and revision, data triangulation by combining deep interviews and documentary study, providing thick descriptions of the research process and context, and holding a validation focus group session with the five experts. This 120-min session involved critiquing and refining the designed framework to increase its relevance and practicality. The reporting of this qualitative study adheres to the Consolidated Criteria for Reporting Qualitative Research (COREQ) guidelines. A completed COREQ checklist is provided as Supplementary File 1. The primary researcher (Female, PhD in Medical Informatics) has formal training in qualitative methods and thematic analysis. No prior relationship existed with any participant before the study. While the researcher's professional background in health data governance informed the inquiry, reflexive bracketing was applied throughout data collection and analysis to mitigate potential bias and ensure interpretive rigor. Findings The initial components of the EDG-AI4H framework were derived from the narrative review of international literature and Iranian policy documents. These preliminary components then served as a sensitizing framework for developing the interview guide. During the thematic analysis of interview data, these components were confirmed, refined, or rejected based on expert perspectives. The final framework presented in Table 2 represents a synthesis of both sources, with interview data playing a primary role in shaping the structure and sub-themes. Components that were derived primarily from the literature were confirmed and enriched by participant insights, while components that emerged inductively from the interviews were cross-referenced with the literature to ensure theoretical grounding. Table 2Themes, sub-themes, and representative quotes from expert interviews.Main theme (layer)Sub-themesRepresentative quoteInterpretation1. Governance, Ethics, and Compliance-Data Ownership, Access, and Sharing Policies-Dynamic Informed Consent Management-Data Ethics Board-Legal and Regulatory Compliance“Any technical move without resolving issues of ownership, consent, and a legal compliance framework is like building a palace on quicksand.” (H3)Experts emphasized that technical AI solutions must be grounded in a solid legal and ethical foundation; otherwise, they risk failure and loss of public trust.“The missing key to trust-building is dynamic consent—policyholders must have granular control over how their data is used.” (H4)Dynamic consent was seen as essential for empowering policyholders and building the trust necessary for data sharing and AI adoption.2. Data Operations and Integration-Metadata-Driven Data Catalog-Standardized Data Exchange-Data Quality Assurance-Secure Storage Architecture“Internal transparency about the ‘what’ and ‘how’ of data is the first step towards data health.” (H1)Data cataloging and metadata management were identified as foundational for understanding and governing organizational data assets.“A supra-organizational, industry-driven effort to develop standardized data exchange mechanisms aligned with national health information infrastructure is essential.” (H2)Experts called for industry-wide collaboration to establish interoperability standards, recognizing that no single organization can achieve this alone.3. Responsible Intelligence and Analytics-Domain-Driven Feature Engineering-XAI-by-Design Model Development-Active Bias Monitoring-Model Lifecycle Governance“Data scientists alone cannot build meaningful features. They need epidemiologists and clinicians to understand what actually matters for health risk.” (H5)Cross-disciplinary collaboration was emphasized as essential for developing clinically valid and fair predictive features.“Using treatment adherence index or pre-processing service utilization patterns is far more fair and clinically valuable than raw, potentially discriminatory variables.” (H4)Experts stressed the importance of constructing clinically meaningful features rather than relying on raw data that may perpetuate bias.4. Value Creation and Transparent Decision-Making-Multi-level DSS-Preventive & Personalized Insurance-Population Health Management-Policyholder Transparency“When a policyholder receives a clear, actionable health report based on their own data, the relationship transforms from transactional to partnership.” (H4)Policyholder empowerment through transparent health reporting was seen as the highest value of the framework.“This is the strongest factor in building trust and shifting from claims payment to health management.” (H1)Experts identified transparency as the key to transforming the traditional insurer-policyholder relationship. Themes, sub-themes, and representative quotes from expert interviews. -Data Ownership, Access, and Sharing Policies Data Ownership, Access, and Sharing Policies -Dynamic Informed Consent Management Dynamic Informed Consent Management -Data Ethics Board Data Ethics Board -Legal and Regulatory Compliance Legal and Regulatory Compliance -Metadata-Driven Data Catalog Metadata-Driven Data Catalog -Standardized Data Exchange Standardized Data Exchange -Data Quality Assurance Data Quality Assurance -Secure Storage Architecture Secure Storage Architecture -Domain-Driven Feature Engineering Domain-Driven Feature Engineering -XAI-by-Design Model Development XAI-by-Design Model Development -Active Bias Monitoring Active Bias Monitoring -Model Lifecycle Governance Model Lifecycle Governance -Multi-level DSS Multi-level DSS -Preventive & Personalized Insurance Preventive & Personalized Insurance -Population Health Management Population Health Management -Policyholder Transparency Policyholder Transparency The thematic analysis of interviews and documents led to the identification of 4 main themes (layers) and 16 sub-themes (key components) that form the cornerstone of the EDG-AI4H framework. Table 2 presents a comprehensive summary of these themes, sub-themes, and their supporting evidence, including representative participant quotes and interpretive analysis. This four-layer framework, hierarchical and interactive, charts the path from raw data to ethics-driven and transparent value. A diagram of the framework is presented in Fig. 1. EDG-AI4H framework structure: A four-layer ethical AI governance model for health insurance showing data flow (upward arrows), value feedback (downward arrows), and stakeholder interactions within a trust-transparency-accountability feedback loop. The Governance-Ethics layer forms the foundation, supporting Operational-Integration, Intelligence-Analytics, and Value-Decision Making layers, with bidirectional stakeholder engagement throughout. Layer one: governance, ethics, and compliance (the foundational layer) This layer establishes the legal, ethical, and policy foundation for the entire framework and is a prerequisite for any operational action. As H3 explicitly stated: ‘Any technical move without resolving issues of ownership, consent, and a legal compliance framework is like building a palace on quicksand.’ Similarly, H4 described dynamic consent as the ‘missing key to trust-building,’ emphasizing that policyholders must have granular control over how their data is used •Data Ownership, Access, and Sharing Policies: Clear definition of ownership and usage rights for data generated in the tripartite interaction among insurer, service provider, and policyholder. Developing governance models for secure data sharing between these entities while preserving stakeholders' rights. Data Ownership, Access, and Sharing Policies: Clear definition of ownership and usage rights for data generated in the tripartite interaction among insurer, service provider, and policyholder. Developing governance models for secure data sharing between these entities while preserving stakeholders' rights. •Dynamic Informed Consent Management: Designing electronic, understandable, and user-friendly mechanisms that allow policyholders to specify granularly which data, for what purpose, for how long, and with which entities is shared. This consent should be easily modifiable or revocable. Dynamic Informed Consent Management: Designing electronic, understandable, and user-friendly mechanisms that allow policyholders to specify granularly which data, for what purpose, for how long, and with which entities is shared. This consent should be easily modifiable or revocable. •Data Ethics Board (DEB): Establishing a permanent, independent, and interdisciplinary committee comprising representatives from insurance, medicine, ethics, law, technology, and civil society. This committee's duty is to review AI projects before implementation and provide continuous oversight, focusing on assessing bias, fairness, transparency, and socio-ethical implications. Data Ethics Board (DEB): Establishing a permanent, independent, and interdisciplinary committee comprising representatives from insurance, medicine, ethics, law, technology, and civil society. This committee's duty is to review AI projects before implementation and provide continuous oversight, focusing on assessing bias, fairness, transparency, and socio-ethical implications. •Legal and Regulatory Compliance Framework: Developing internal guidelines for aligning operations with domestic laws (such as the Free Access to Information Act) and international principles (like Privacy by Design and Security by Design). Legal and Regulatory Compliance Framework: Developing internal guidelines for aligning operations with domestic laws (such as the Free Access to Information Act) and international principles (like Privacy by Design and Security by Design). Layer two: data operations and integration This layer translates the policies of the governance layer into operational processes and technologies, creating an integrated technical infrastructure. H1 emphasized the importance of data cataloging: “Internal transparency about the ‘what’ and ‘how’ of data is the first step towards data health.” Experts also highlighted the need for industry-wide standards. H2 noted that a ‘supra-organizational, industry-driven effort to develop standardized data exchange mechanisms aligned with national health information exchange standards ‘is essential for meaningful interoperability •Metadata-Driven Data Catalog: Creating a central system that catalogs all organizational data assets along with rich metadata (source, quality, sensitivity, owner, access permissions, usage policy). Metadata-Driven Data Catalog: Creating a central system that catalogs all organizational data assets along with rich metadata (source, quality, sensitivity, owner, access permissions, usage policy). •Standardized and Secure Data Exchange Pipelines: Developing and implementing secure APIs and using global health data exchange standards like FHIR (Fast Healthcare Interoperability Resources) for seamless and standardized communication between insurer systems, hospitals, laboratories, and electronic health record systems. Standardized and Secure Data Exchange Pipelines: Developing and implementing secure APIs and using global health data exchange standards like FHIR (Fast Healthcare Interoperability Resources) for seamless and standardized communication between insurer systems, hospitals, laboratories, and electronic health record systems. •Data Quality & Interoperability Assurance: Defining, monitoring, and reporting key data quality indicators (accuracy, completeness, timeliness, consistency) and using standardized clinical vocabularies (like ICD-10 for diagnoses) to ensure consistent meaning of data across different sources. Data Quality & Interoperability Assurance: Defining, monitoring, and reporting key data quality indicators (accuracy, completeness, timeliness, consistency) and using standardized clinical vocabularies (like ICD-10 for diagnoses) to ensure consistent meaning of data across different sources. •Secure Storage and Processing Architecture: Designing a hybrid architecture including a Data Lake for storing raw and heterogeneous data while preserving details, and a Data Warehouse for cleansed, integrated, and modeled data for operational and analytical purposes, adhering to the highest security levels and access segregation. Secure Storage and Processing Architecture: Designing a hybrid architecture including a Data Lake for storing raw and heterogeneous data while preserving details, and a Data Warehouse for cleansed, integrated, and modeled data for operational and analytical purposes, adhering to the highest security levels and access segregation. Layer three: responsible intelligence and analytics This layer builds the analytical power of AI upon the previous layers—which ensure data quality and ethical foundations. H5 stressed the importance of domain collaboration: ‘Data scientists alone cannot build meaningful features. They need epidemiologists and clinicians to understand what actually matters for health risk.’ This aligns with H4's view that ‘using treatment adherence index or pre-processing service utilization patterns is far more fair and clinically valuable than raw, potentially discriminatory variables. •Domain-Driven, Collaborative Feature Engineering: Close collaboration between data science teams with domain experts from insurance (H5) and medicine (H4) to extract and construct meaningful, fair, and clinically valuable features. For example, using “treatment adherence index” or “pre-processing service utilization patterns” instead of raw, potentially discriminatory variables. Domain-Driven, Collaborative Feature Engineering: Close collaboration between data science teams with domain experts from insurance (H5) and medicine (H4) to extract and construct meaningful, fair, and clinically valuable features. For example, using “treatment adherence index” or “pre-processing service utilization patterns” instead of raw, potentially discriminatory variables. •Mandatory XAI-by-Design Model Development: Incorporating explainability techniques as integral and mandatory components of every model's development cycle, not optional add-ons. This includes:oPre-model XAI: Selecting inherently interpretable models (e.g., decision trees, logistic regression) where appropriate for clinical and insurance decision support.oIn-model XAI: Implementing model architectures that provide natural explanations (e.g., attention mechanisms in neural networks).oPost-model XAI: Applying post-hoc explanation techniques (SHAP, LIME, integrated gradients) to any “black box” models used. Mandatory XAI-by-Design Model Development: Incorporating explainability techniques as integral and mandatory components of every model's development cycle, not optional add-ons. This includes:oPre-model XAI: Selecting inherently interpretable models (e.g., decision trees, logistic regression) where appropriate for clinical and insurance decision support.oIn-model XAI: Implementing model architectures that provide natural explanations (e.g., attention mechanisms in neural networks).oPost-model XAI: Applying post-hoc explanation techniques (SHAP, LIME, integrated gradients) to any “black box” models used. oPre-model XAI: Selecting inherently interpretable models (e.g., decision trees, logistic regression) where appropriate for clinical and insurance decision support. Pre-model XAI: Selecting inherently interpretable models (e.g., decision trees, logistic regression) where appropriate for clinical and insurance decision support. oIn-model XAI: Implementing model architectures that provide natural explanations (e.g., attention mechanisms in neural networks). In-model XAI: Implementing model architectures that provide natural explanations (e.g., attention mechanisms in neural networks). oPost-model XAI: Applying post-hoc explanation techniques (SHAP, LIME, integrated gradients) to any “black box” models used. Post-model XAI: Applying post-hoc explanation techniques (SHAP, LIME, integrated gradients) to any “black box” models used. Important qualification: While SHAP and LIME are valuable tools for generating post-hoc explanations, they have known limitations, including sensitivity to input data distribution, computational complexity, and the fact that they provide approximations rather than ground-truth explanations. The appropriateness of any XAI method depends on the model type, the intended user, and the decision context. Therefore, we emphasize that SHAP/LIME availability is one component of a broader XAI strategy that should also include intrinsic interpretability and stakeholder-specific reporting oExplanation reporting: Every final model must produce a standardized “feature impact report” that explains prediction reasons in understandable language tailored to each stakeholder (policyholder, clinician, insurer). Explanation reporting: Every final model must produce a standardized “feature impact report” that explains prediction reasons in understandable language tailored to each stakeholder (policyholder, clinician, insurer). oAudit trails: Maintaining comprehensive logs of model inputs, outputs, explanations, and decisions for regulatory review and complaint handling. Audit trails: Maintaining comprehensive logs of model inputs, outputs, explanations, and decisions for regulatory review and complaint handling. •Active Model Bias Monitoring & Mitigation: Implementing continuous, automated tests to identify unwanted model biases (especially against specific gender, ethnic, geographic, or socioeconomic groups) and designing mechanisms to mitigate or eliminate these biases. Active Model Bias Monitoring & Mitigation: Implementing continuous, automated tests to identify unwanted model biases (especially against specific gender, ethnic, geographic, or socioeconomic groups) and designing mechanisms to mitigate or eliminate these biases. •Model Lifecycle Governance: Defining formal processes for model versioning, deployment, performance monitoring in operational environments, retirement, and audit. This ensures models remain updated, reliable, and compliant with regulations. Model Lifecycle Governance: Defining formal processes for model versioning, deployment, performance monitoring in operational environments, retirement, and audit. This ensures models remain updated, reliable, and compliant with regulations. Layer four: value creation and transparent decision-making This layer transforms the insights generated in the intelligence layer into tangible actions, innovative services, and value for all stakeholders. All experts considered the policyholder transparency component the pinnacle of the framework. H4 stated: ‘When a policyholder receives a clear, actionable health report based on their own data, the relationship transforms from transactional to partnership.’ H1 added: ‘This is the strongest factor in building trust and shifting from claims payment to health management.’ •Multi-level Decision Support Systems (DSS): Providing dashboards and analytical tools tailored to each group's needs: senior executives (macro view of risk and cost trends), insurance and case specialists (guiding targeted interventions for high-risk policyholders), and healthcare providers (complementary insights for clinical care). Multi-level Decision Support Systems (DSS): Providing dashboards and analytical tools tailored to each group's needs: senior executives (macro view of risk and cost trends), insurance and case specialists (guiding targeted interventions for high-risk policyholders), and healthcare providers (complementary insights for clinical care). •Preventive & Personalized Insurance: Using predictive insights to design incentive-based products (e.g., premium discounts for participation in digital health programs or achieving health goals) or targeted care plans for specific groups. Preventive & Personalized Insurance: Using predictive insights to design incentive-based products (e.g., premium discounts for participation in digital health programs or achieving health goals) or targeted care plans for specific groups. •Population Health Management (PHM): Identifying “risk hotspots” within the covered population and designing public health interventions and community-level prevention programs, which both improve public health and reduce the insurer's long-term costs. Population Health Management (PHM): Identifying “risk hotspots” within the covered population and designing public health interventions and community-level prevention programs, which both improve public health and reduce the insurer's long-term costs. •Policyholder Transparency & Empowerment: Providing each policyholder with a personal, understandable, and actionable health report. Based on the individual's own data, this report presents probable future health risks, influencing factors, and practical, personalized suggestions for reducing these risks. All experts considered this component the pinnacle of the framework and the strongest factor in building trust and transforming the traditional insurer-policyholder relationship into a health partnership. Policyholder Transparency & Empowerment: Providing each policyholder with a personal, understandable, and actionable health report. Based on the individual's own data, this report presents probable future health risks, influencing factors, and practical, personalized suggestions for reducing these risks. All experts considered this component the pinnacle of the framework and the strongest factor in building trust and transforming the traditional insurer-policyholder relationship into a health partnership. Discussion In the present research, the EDG-AI4H framework has been presented as a systematic and indigenous response to the identified gaps in both domestic and international literature. This framework encompasses several key conceptual and operational advancements that distinguish it from purely technical approaches. First, it emphasizes the precedence of ethics and governance over technology. In EDG-AI4H, the governance-ethics layer acts as a prerequisite for the technical layers, aligning with the global paradigm of “Responsible AI” [14]. The establishment of a “Data Ethics Board” and “Dynamic Consent Management” are objective mechanisms for institutionalizing this. This approach complements technical frameworks like SHINFDP, which focuses on fraud detection through the integration of blockchain, machine learning, 5G, and cloud computing [28]. While SHINFDP offers a six-layer architecture focusing on security and efficiency, EDG-AI4H adds the governance-ethics layer as a foundation, ensuring these technologies are employed responsibly with respect to privacy and transparency. The EDG-AI4H approach ensures this “intelligence” and “adaptability” are framed within an appropriate ethical and governance context. Second, it creates a common language and platform for interdisciplinary dialogue. By simultaneously encompassing legal, medical, technical, and business dimensions, this framework provides a shared conceptual map that enables constructive collaboration between sometimes disparate domains. This is particularly vital in the context of Health Insurance Claim (HIC) processing, which involves multiple stakeholders including Healthcare Service Providers (HSP), policyholders (HINS), and insurance companies (HIP) [28]. The integrated EDG-AI4H approach can reduce the coordination and transparency challenges present in the traditional HIC process. This perspective resonates with the interoperability between professional developers and citizen developers (Business Analysts and End Users). The proposed framework facilitates this collaboration in the complex domain of health insurance by establishing a common language. Third, it elevates explainability from a technical feature to a governance and ethical imperative. In the sensitive health domain, model accuracy alone is insufficient; explainability and auditability are necessary conditions for social acceptance [29]. The “XAI-by-Design” requirement in the third layer embeds this concept at the core of the development process. This is also addressed in the SHINFDP framework, where the Prediction Layer uses machine learning algorithms to classify legitimate and fraudulent claims [28]. Integrating XAI in EDG-AI4H ensures that automation decisions are not only accurate but also understandable and auditable for all stakeholders, including policyholders and regulators. EDG-AI4H ensures these analytical and decision-making capabilities are transparent and accountable. Fourth, it focuses on creating bilateral value and shifting the mindset. EDG-AI4H explicitly emphasizes policyholder empowerment and transparency, not just insurer benefit. The “Policyholder Transparency” component can transform the traditional relationship into an active partnership in health management. This aligns with the use of Wearable (WRD) technologies in case studies where policyholders receive incentives like premium reductions for sharing their health data [28]. The proposed framework, by building trust through transparency and privacy-centricity, enables broader adoption of such programs. EDG-AI4H operationalizes these goals in the health insurance domain, centered on value creation for all stakeholders. The “XAI-by-Design” requirement embedded in Layer Three operationalizes transparency as a governance imperative rather than a technical afterthought. This requirement translates into four concrete governance mechanisms: first, mandatory explanation documentation for every model deployment, which must be reviewed and approved by the Data Ethics Board before production release; second, stakeholder-specific explanation formats that tailor the presentation of model explanations to the needs of different audiences—simple, non-technical language for policyholders, detailed technical reports for data scientists and auditors, and clinically meaningful summaries for physicians; third, the establishment of explainability as a formal model acceptance criterion, meaning that any model failing to meet minimum explainability standards—defined as the ability to produce comprehensive feature impact reports (e.g., SHAP or LIME explanations) within acceptable computational time—is not approved for production deployment; and fourth, regular explanation audits conducted periodically. Importantly, the availability of SHAP or LIME explanations should not be treated as proof that a model is explainable, fair, or trustworthy. In high-stakes health insurance decisions, we recommend a combination of methods—including intrinsic interpretability, post-hoc explanation techniques, and human oversight—rather than reliance on any single explanation technique. Future research should explore how different stakeholders—including policyholders, clinicians, and insurers—perceive, interpret, and act upon these explanations in real-world decision-making contexts. Compared to established international frameworks, EDG-AI4H offers several distinctive contributions while complementing existing approaches. The OECD AI Principles [30] emphasize transparency, accountability, and human-centered values, which are embedded throughout our framework. The WHO Guidance on AI Ethics [31] provides high-level ethical principles for health AI, including protecting autonomy, promoting human well-being, and ensuring transparency. However, neither offers the operational specificity required for health insurance risk prediction in Iran's context. EDG-AI4H translates these high-level principles into concrete layers—from data ownership policies to XAI-by-design model development—that can be operationalized by insurers. To systematically compare the EDG-AI4H framework with these established international governance frameworks, we present a comparative analysis in Table 3. This table highlights the key focus, strengths, and gaps addressed by EDG-AI4H relative to the OECD AI Principles [30], WHO Guidance on AI Ethics [31], NIST AI Risk Management Framework [32], EU AI Act [16], and the SHINFDP framework [28]. Table 3Comparative analysis of EDG-AI4H with international AI governance frameworks.FrameworkKey focusStrengthsProposes operational and contextual specificity for Iran's health insurance contextOECD AI PrinciplesTransparency, accountability, human-centered valuesBroad principles endorsed by 40+ countriesLack operational specificity and health domain tailoringWHO AI Ethics GuidanceHealth-specific AI ethicsComprehensive health domain principlesLimited implementation mechanismsNIST AI RMFAI risk managementSector-agnostic, practical approachHealth insurance specificities not addressedEU AI ActHigh-risk AI regulationLegally enforceable requirementsImplementation mechanisms left to member statesSHINFDPHealth insurance fraud detectionTechnical architecture focusLacks explicit governance-ethics foundationEDG-AI4HIntegrated governance + XAI + health insuranceCulturally tailored, operational, multi-stakeholderFills the implementation and localization gap Comparative analysis of EDG-AI4H with international AI governance frameworks. Similarly, the NIST AI Risk Management Framework [32] provides comprehensive guidance for managing AI risks across sectors, but it does not address the unique challenges of health insurance data governance, such as dynamic consent management, FHIR-based data exchange, or policyholder empowerment through transparent health reporting. The EU AI Act [16] classifies health insurance AI as ‘high-risk’ and mandates transparency and human oversight, but leaves implementation mechanisms to member states. EDG-AI4H proposes greater contextual and operational specificity that could potentially address the implementation gap identified in international frameworks, although this remains to be empirically validated through future implementation studies. The SHINFDP framework [28] offers a six-layer architecture for smart health insurance with a focus on fraud detection through blockchain, machine learning, and 5G technologies. While SHINFDP emphasizes security and efficiency, EDG-AI4H adds an explicit governance-ethics layer as a foundational prerequisite, ensuring that even technically advanced solutions are deployed responsibly with respect to privacy, transparency, and fairness. This reflects a broader trend in responsible AI where governance frameworks must precede or accompany technical innovation [5], [14]. Implementation challenges and resource requirements While the EDG-AI4H framework provides a comprehensive governance roadmap, its successful implementation faces several significant challenges that require careful consideration: Financial and Resource Requirements: Implementing the full framework, including infrastructure development (data lake, warehouse, secure APIs), human capacity building, and ongoing governance (Data Ethics Board operations), requires substantial financial investment. For Iranian insurance organizations, many of which operate with limited IT budgets, prioritizing and sequencing implementation phases is essential. Future research should conduct cost-benefit analyses to estimate economic returns of robust data governance. Technical Infrastructure Readiness: As noted by Expert H2, Iran's health information systems currently rely on openEHR and ISO 13606 standards, with FHIR adoption still emerging. The framework's operational layer (metadata catalog, standardized data exchange) requires significant technical upgrades to existing systems and industry-wide coordination to develop national interoperability standards. Organizational and Cultural Barriers: Insurance organizations may face resistance from staff accustomed to traditional claims-based operations. The paradigm shifts from reactive “claims payment” to proactive “health management” requires cultural transformation, change management strategies, and leadership commitment. Additionally, data sharing between insurers, hospitals, and policyholders requires building trust through transparency and demonstrated value. Regulatory and Policy Gaps: While Iran's Personal Data Protection Law provides a foundation, specific regulations for AI governance, algorithm transparency, and accountability mechanisms are still under development. The framework's implementation will require coordination with regulatory bodies and potential legal reforms. Human Capacity Development: The interdisciplinary nature of the framework—combining data science, law, ethics, medicine, and insurance—requires professionals with cross-domain competencies. Training programs, workshops, and academic partnerships will be essential to build the necessary workforce. Implementation Sequencing: We recommend a phased approach: Phase 1—establish the Data Ethics Board and develop data governance policies; Phase 2—implement metadata catalog and data quality frameworks; Phase 3—develop XAI-compliant models with bias monitoring; Phase 4—roll out value-creation components with policyholder engagement. Evaluation metrics and key performance indicators To facilitate the operationalization and evaluation of the EDG-AI4H framework, we propose the following measurable indicators across the four layers: Governance compliance indicators: •Percentage of AI projects reviewed by the Data Ethics Board Percentage of AI projects reviewed by the Data Ethics Board •Policyholder consent form completion and modification rates Policyholder consent form completion and modification rates •Number of privacy incidents and resolution time Number of privacy incidents and resolution time •Regulatory compliance audit scores Regulatory compliance audit scores Explainability indicators: •Availability of stakeholder-specific explanation reports (policyholder, clinician, insurer versions) Availability of stakeholder-specific explanation reports (policyholder, clinician, insurer versions) •User satisfaction with explanation clarity and usefulness User satisfaction with explanation clarity and usefulness •Completeness of audit trails for model decisions Completeness of audit trails for model decisions Fairness indicators: •Bias detection test coverage across demographic groups Bias detection test coverage across demographic groups •Fairness metrics (e.g., equal opportunity difference, disparate impact ratio) Fairness metrics (e.g., equal opportunity difference, disparate impact ratio) •Number of bias mitigation interventions implemented Number of bias mitigation interventions implemented Performance indicators (for reference): •Model predictive accuracy, precision, recall, and F1-score Model predictive accuracy, precision, recall, and F1-score •Model drift detection and retraining frequency Model drift detection and retraining frequency Stakeholder outcome indicators: •Policyholder engagement with health reports (open rates, time spent) Policyholder engagement with health reports (open rates, time spent) •Participation rates in preventive health programs Participation rates in preventive health programs •Policyholder trust and satisfaction survey scores Policyholder trust and satisfaction survey scores •Cost savings from preventive interventions Cost savings from preventive interventions Note on Explainability Indicators: We acknowledge that the availability of SHAP or LIME explanations alone does not equate to full explainability. Rather, it is one of several approaches that contribute to explainability alongside intrinsic interpretability and stakeholder-specific reporting. The appropriateness of any explanation method depends on the model type, intended user, and decision context. Future empirical studies should validate and refine these indicators through implementation in real-world insurance settings, as the appropriateness of specific metrics may vary across organizational contexts. Conclusion This research demonstrated that the transition of Iran's health insurance industry towards the era of data and AI requires a solid theoretical and operational foundation in the domain of ethics-driven data governance. The EDG-AI4H conceptual framework, resulting from a synthesis of international knowledge, domestic documents, and the collective wisdom of five prominent and experienced experts, can serve as a national roadmap and practical guide for insurance organizations, regulators, and policymakers in this field. The gradual and intelligent implementation of this framework can lead to the following macro achievements: 1.Risk Reduction: Minimize ethical (discrimination, privacy violation), legal (non-compliance), and reputational (loss of public trust) risks associated with AI projects. Risk Reduction: Minimize ethical (discrimination, privacy violation), legal (non-compliance), and reputational (loss of public trust) risks associated with AI projects. 2.Increased Trust and Participation: By promoting transparency and empowering policyholders, actively engage them in managing their own health and increase the industry's social capital. Increased Trust and Participation: By promoting transparency and empowering policyholders, actively engage them in managing their own health and increase the industry's social capital. 3.Industry Paradigm Shift: Steer the industry from the passive and reactive model of “claims payment” towards an active and preventive model of “health and risk management.” Industry Paradigm Shift: Steer the industry from the passive and reactive model of “claims payment” towards an active and preventive model of “health and risk management.” 4.Empowering Responsible Innovation: By creating a secure and trustworthy framework, pave the way for the emergence and growth of responsible innovations in domestic Health FinTech and InsurTech. Empowering Responsible Innovation: By creating a secure and trustworthy framework, pave the way for the emergence and growth of responsible innovations in domestic Health FinTech and InsurTech. Research limitations This study has several limitations that should be considered when interpreting the findings: Sample Size and Selection: The framework is based on interviews with only five experts. While this sample was appropriate for the exploratory, conceptual nature of this study and was justified by the highly specialized expertise required, the findings may not capture the full diversity of perspectives across Iran's health insurance ecosystem. Additionally, while thematic saturation was achieved within the five disciplinary domains sampled, this does not establish saturation across the broader Iranian health-insurance ecosystem. The findings should be considered exploratory rather than exhaustive. Validation Limitation: A key methodological limitation is that the validation focus group comprised the same five experts who participated in the initial interviews. This approach may have introduced confirmation bias and limited the independence of the validation process. Ideally, validation should involve independent stakeholders. Lack of Empirical Validation: The framework remains conceptual and has not been empirically tested in real-world insurance settings. While this is appropriate for an exploratory framework development study, the practical feasibility, effectiveness, and operationalization of the framework's components require testing through field studies, action research, or pilot projects. Generalizability: The framework was developed specifically for Iran's health insurance context. While the underlying principles may be transferable to other developing countries with similar healthcare systems, the specific implementation mechanisms may require cultural and institutional adaptation. Implementation Constraints: The framework's implementation requires significant organizational, technical, financial, and human resources that may not be readily available in all insurance organizations, particularly in resource-constrained settings.” Suggestions for future research 1.Action Research: Implementing and evaluating one component of this framework (such as a pilot launch of a Data Ethics Board or designing a dynamic consent management system) in a volunteer insurance company as a case study. Action Research: Implementing and evaluating one component of this framework (such as a pilot launch of a Data Ethics Board or designing a dynamic consent management system) in a volunteer insurance company as a case study. 2.Pilot Implementation Study: Conducting a pilot implementation of the full framework in a selected insurance organization to assess feasibility, identify implementation barriers, and measure preliminary outcomes. Pilot Implementation Study: Conducting a pilot implementation of the full framework in a selected insurance organization to assess feasibility, identify implementation barriers, and measure preliminary outcomes. 3.Mixed-Methods Evaluation: Combining quantitative metrics (governance quality indicators, transparency scores, stakeholder trust measures) with qualitative stakeholder feedback to evaluate framework effectiveness. Mixed-Methods Evaluation: Combining quantitative metrics (governance quality indicators, transparency scores, stakeholder trust measures) with qualitative stakeholder feedback to evaluate framework effectiveness. 4.Developing Operational Guides and Metrics: Translating the present conceptual framework into a step-by-step manual, practical checklists, and measurable Key Performance Indicators (KPIs) for organizations. Developing Operational Guides and Metrics: Translating the present conceptual framework into a step-by-step manual, practical checklists, and measurable Key Performance Indicators (KPIs) for organizations. 5.Cross-Cultural and Organizational Studies: Conducting comparative studies on the barriers and facilitators of implementing such frameworks within different organizational cultures of state-owned, private, and cooperative insurance companies in Iran. Cross-Cultural and Organizational Studies: Conducting comparative studies on the barriers and facilitators of implementing such frameworks within different organizational cultures of state-owned, private, and cooperative insurance companies in Iran. 6.Independent Stakeholder Validation: Conducting validation workshops with diverse independent stakeholders (policymakers, insurers, clinicians, AI developers, patient advocates) to assess the framework's credibility, applicability, and acceptability from multiple perspectives. Independent Stakeholder Validation: Conducting validation workshops with diverse independent stakeholders (policymakers, insurers, clinicians, AI developers, patient advocates) to assess the framework's credibility, applicability, and acceptability from multiple perspectives. 7.Cost-Benefit Analysis: Conducting a quantitative study to estimate the long-term economic and social benefits of investing in robust data governance compared to short-term initial investment costs. Cost-Benefit Analysis: Conducting a quantitative study to estimate the long-term economic and social benefits of investing in robust data governance compared to short-term initial investment costs. In conclusion, this research is an initial but essential step to ensure that the data and AI revolution in Iran's health insurance becomes a powerful tool for enhancing access equity, service quality, and health for all members of society, rather than a factor deepening inequalities, reducing transparency, and eroding public trust. Generative AI statement During the preparation of this work, the authors used generative AI tools (Deep seek) solely for language editing and grammar refinement of the manuscript text. The AI was not used for data analysis, interpretation, framework development, or any substantive intellectual contribution. All conceptual work, thematic analysis, framework design, and conclusions are the original work of the authors. The authors take full responsibility for the content, accuracy, and integrity of this manuscript. CRediT authorship contribution statement Rezvan Rahimi: Writing – review & editing, Writing – original draft, Visualization, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Consent to publish The authors affirm that no individual person's data (including images) requiring consent for publication is presented in this manuscript. Ethics approval and consent to participate This study was performed in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committees of School of Public Health & Allied Medical Sciences- Tehran University of Medical Sciences (Code: IR.TUMS.SPH.REC.1402.171). Informed consent was obtained from all individual participants included in the study. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Declaration of competing interest The author declares no competing interests. Acknowledgements We sincerely thank the five expert participants for their invaluable contributions to the interviews and focus group validation. We also gratefully acknowledge Dr. Hanieh Zehtab Hashemi, PhD in Medical Informatics, Virtual University of Medical Sciences, for her independent coding of 20% of the transcripts, which enhanced the reliability of the thematic analysis. We sincerely thank all the esteemed experts who generously and meticulously shared their valuable knowledge and experiences for this research. Data availability The qualitative data generated and analyzed during this study, including anonymized interview transcripts and documentary analysis records, are not publicly available due to participant confidentiality and ethical restrictions approved by the relevant Research Ethics Committee. However, aggregated and de-identified data may be made available from the corresponding author upon reasonable request and subject to a data access agreement consistent with Iranian privacy regulations and the informed consent obtained. References Associated Data Data Availability Statement The qualitative data generated and analyzed during this study, including anonymized interview transcripts and documentary analysis records, are not publicly available due to participant confidentiality and ethical restrictions approved by the relevant Research Ethics Committee. However, aggregated and de-identified data may be made available from the corresponding author upon reasonable request and subject to a data access agreement consistent with Iranian privacy regulations and the informed consent obtained.