Abstract Electronic health records (EHRs) are central to patient-centered care. However, many current EHR systems do not represent, exchange, or reuse genomic findings in structured, interoperable, and clinically actionable forms. Instead, genomic results are often retained in static reports. This limits their use in routine clinical care and clinical decision support. It also restricts their future reinterpretation as scientific knowledge evolves, as well as their reuse for research within appropriate governance frameworks. To support precision and personalized medicine, we advocate that clinically relevant genomic findings should be represented within EHRs as structured, phenotype-linked, interoperable, and versioned data. Raw and intermediate genomic files, however, should remain in appropriately governed laboratory or genomics repositories. This viewpoint is intended for health policymakers, health informatics leaders, EHR developers, research and clinical genomics laboratories, and clinicians involved in implementing genomics into routine care. It is informed by selected published literature and relevant technical standards. It is also informed by operational experience at the Cyprus Institute of Neurology and Genetics and multidisciplinary discussions from the Gen2EHR Implementation Guidelines Workshop conducted within the Extended EHR@EU Data Space and European Health Data Space for Secondary Use @CY initiatives. We examine technical, organizational, clinical, ethical, legal, and governance challenges, including inconsistent reporting formats and unclear clinical responsibility. We propose a staged, workshop-informed implementation pathway for evaluation within the European Health Data Space. The pathway combines Health Level 7 Fast Health Interoperability Resources Genomics Reporting with complementary standards and terminologies. It also includes transitional dual PDF and structured Fast Health Interoperability Resources reporting, governance for access, version control, reinterpretation, and clinical responsibility, as well as multidisciplinary pilot validation in real diagnostic workflows. The clinical effectiveness and transferability of the proposed pathway across European health systems remain to be established.
Abstract Background Cross-domain chest radiograph report generation may produce formally valid reports that are nevertheless repetitive, weakly case-specific, or missing clinically relevant findings. These failures can be missed when evaluation relies on a single language-overlap or clinical-label score. Objective This study aimed to evaluate cross-domain report generation with a bidirectional layered audit, characterize direction-specific and end point–specific failure profiles under a composite training-side control and 3 decoding policies, and explore whether selected audit dimensions remained informative for a contemporary medical vision language model. Methods A fixed R2Gen-style system was evaluated bidirectionally on 575 IU X-ray and 401 MIMIC-CXR (Medical Information Mart for Intensive Care Chest X-Ray) target records. The audit included surface validity, complete-report recurrence and concentration, a predefined prefix end point, lexical diversity, official Stanford CheXpert/NegBio labels, paired cluster-bootstrap inference, and blinded review by 2 medical trainees. The primary human audit included 210 reports from 70 randomly sampled cases, with blinded adjudication of unresolved primary end point disagreements. A 60-report targeted panel was used for failure-mode analysis. Separately, a fixed MedGemma 1.5 4B configuration generated reports for all 976 records. Results In MIMIC-CXR–to–IU X-ray transfer, out of 575 reports, ON (protocol-defined training-side control enabled)+B4 (beam search with a beam size of 4) reduced the predefined recurrent prefix from 575 (100%) to 8 (1.4%; Holm-adjusted P =.002). However, 569 (99.0%) ON+B4 outputs still belonged to exact duplicate groups. In IU X-ray–to–MIMIC-CXR transfer, the 14-category CheXpert/NegBio micro- F 1 was higher for ON+B4 than for OFF (protocol-defined training-side control disabled)+B4 (0.1687 vs 0.0744; Holm-adjusted P =.01). This apparent advantage was not broad-based: after cardiomegaly was excluded in a post hoc descriptive sensitivity analysis, the difference changed from +0.0944 to −0.0610. Text-level tracing showed that the original contrast was mainly driven by a labeler response to recurrent, semantically normal heart-size wording. Across the primary human-audit cells, major omissions occurred in 12 to 15 out of 35 reports, and major unsupported findings occurred in 10 to 16 of 35 reports, with no consistent configuration ordering observed. The exploratory MedGemma micro- F 1 was 0.4722 on IU X-ray and 0.4340 on MIMIC-CXR, but 381 out of 575 (66.3%) IU outputs belonged to exact duplicate groups. Conclusions No configuration performed consistently well across end points, and no single metric captured all forms of cross-domain degradation. The MedGemma stress test suggested that recurrence and label-traceability checks remain useful across architectures, although it was not a matched performance comparison. Cross-domain report generators should undergo bidirectional evaluation, final-text verification, and image-grounded human review before external validation or clinical deployment.
Abstract Background Health AI investment is accelerating across the Gulf Cooperation Council (GCC), yet the region’s health AI research output has never been systematically quantified. National research capacity is a structural precondition for the safe local development and governance of clinical AI, but no scientometric baseline exists for the 6 GCC states. Objective We characterized the volume, growth, collaboration structure, and thematic composition of GCC health AI research and examined whether national output aligns with an existing measure of informatics readiness. Methods We conducted an ecological scientometric analysis of all health AI publications affiliated with the 6 GCC states (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab Emirates) for 2011 to 2024 (the last complete indexing year at extraction) using the open bibliographic database OpenAlex. A publication was eligible if it included both an AI family and a health or medicine family concept above a 0.30 relevance threshold. Bibliometric indicators (output, compound annual growth, per capita intensity, citation impact, international and intraregional collaboration, and institutional concentration) were computed; coauthorship and concept co-occurrence networks were constructed and clustered by modularity. National health AI intensity was correlated (Spearman) with the Health Informatics Readiness Index (HIRI) and Global Digital Health Monitor (GDHM) phase. Results Of 112,480 GCC-affiliated AI publications, 3512 met the health AI definition. Regional output rose from 18 (2011) to 904 (2024) publications per year (compound annual growth=35.2%), with 86.4% (3034/3512) published from 2020 onward and a sharp inflection from 2019 to 2021. Saudi Arabia contributed 59.6% (2094/3512) of the output by itself and 81.5% (2862/3512) together with the United Arab Emirates; Qatar led on a per capita basis (approximately 136 publications per million population). International collaboration (2551/3512, 72.6% of the papers) predominated over intra-GCC collaboration (164/3327, 4.9%), with the latter being inversely related to national output. The principal external partners were India, the United States, Pakistan, the United Kingdom, and Egypt. Output spanned 3186 institutions; concentration ranged from 89.4% (328/367) of a country’s output among the top 3 institutions for Qatar to 34.4% (721/2094) for Saudi Arabia. Concept co-occurrence was classified into 4 themes—classic machine learning clinical prediction, medical imaging and oncology, COVID-19 response, and diabetic eye disease. Health AI intensity (ρ=+0.60) and, more strongly, absolute output (ρ=+0.89) correlated with the HIRI; absolute output was also correlated with GDHM phase (ρ=+0.83). Conclusions GCC health AI research has grown rapidly but unevenly, is heavily dependent on international collaboration, is concentrated in medical imaging, and—outside Saudi Arabia—remains anchored in single national institutions. Informatics readiness was associated with both the intensity and the volume of national research output. The findings establish a reproducible baseline for monitoring regional health AI research capacity as GCC health systems move toward clinical AI deployment.
