Citations

Abstract Rare Mendelian disorders affect 300-400 million people globally. Although genetic testing has become widely adopted, gene-specific evidence for tailored variant interpretation remains scattered across resources. We present Gene Portals, a framework for gene-centered multimodal knowledge bases that co-localize expert-harmonized clinical data, functional assays, population variation, structural annotations and gene-specific ACMG/AMP specifications within a single resource. A modular interface integrates this unified evidence with VCEP-refined ACMG specifications to enable automated gene-specific variant classification, infer molecular mechanisms, and support cross-gene analyses. We demonstrate the framework’s utility across five Gene portals spanning eleven neurodevelopmental disorder-associated genes, integrating data from 4,423 individuals with 2,838 unique variants, 36,149 ClinVar submissions, and 1,044 expert-curated molecular readouts. By organizing evidence that is otherwise dispersed across multiple sources into a unified, queryable framework, the SCN, GRIN, CACNA1A, SATB2 and SLC6A1 Gene Portals became widely used community resources and provide an extensible template for standardized rare-disease variant interpretation and mechanism-aware discovery.

Introduction To examine within-class sulfonylurea safety, we compared risks of major adverse cardiovascular events (MACE) and severe hypoglycemia among adults with type 2 diabetes (T2D) and moderate cardiovascular risk following sulfonylurea initiation. Research design and methods We conducted a target trial emulation including adults ≥21 years old with T2D and moderate cardiovascular risk who initiated glimepiride, glipizide or glyburide between 2014 and 2021, using claims data from Optum Labs Data Warehouse and the Medicare fee-for-service 100% sample. Study outcomes were MACE (primary), expanded MACE and its components and emergency department or hospital encounters for hypoglycemia, ascertained during follow-up through 2022. Inverse probability of treatment weighting (IPTW) was applied using propensity scores estimated using the super learner ensemble, and outcomes were examined using IPTW Cox proportional hazards models. Results The weighted study cohort comprised 314 699 patients (mean age 66.9 years, 52.0% men, 76.6% non-Hispanic white). At 1 year, MACE was experienced by 2.5%, 2.7% and 2.8% of patients starting glimepiride, glipizide and glyburide, respectively. Compared with glimepiride, glyburide and glipizide were associated with higher risk of MACE (HR 1.10, 95% CI 1.05 to 1.16 for glyburide; HR 1.05, 95% CI 1.03 to 1.07 for glipizide). At 1 year, severe hypoglycemia was experienced by 0.3%, 0.3% and 0.4% of patients starting glimepiride, glipizide and glyburide, respectively. Glyburide was associated with a greater risk of severe hypoglycemia compared with glipizide (HR 1.43, 95% CI 1.23 to 1.65), while glipizide was associated with a lower risk compared with glimepiride (HR 0.82, 95% CI 0.77 to 0.87). Conclusions Among adults with T2D and moderate cardiovascular risk, glimepiride was associated with lowest risk of MACE and glipizide with lowest risk of severe hypoglycemia. These results can help inform treatment selection if sulfonylureas are used for glucose-lowering.

Abstract Generative artificial intelligence (AI) is rapidly populating medical records with synthetic or partially AI-generated content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability of medical data at population scale. By analysing more than 800,000 synthetic data points across clinical text generation, vision–language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architectures. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence. Models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies that can be easily integrated into existing clinical workflows, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

Cancer health disparities represent profound inequalities in incidence, outcomes, and survivorship across populations. While traditionally examined through distinct lenses of either molecular biology or social epidemiology, these disparities arise from the complex interplay of genetic susceptibility, epigenetic dysregulation, and social determinants of health (SDoH). This review proposes that DNA damage and genomic instability serve as a critical mechanistic bridge, integrating exposures from the societal level to cellular dysfunction. We synthesize evidence demonstrating how SDoH—such as systemic inequities, environmental exposures, and chronic stress—converge with genetic and epigenetic factors to disproportionately increase DNA damage burden, impair repair mechanisms, and accelerate tumorigenesis in marginalized communities. Using the elevated gastrointestinal cancer rates among Native Hawaiians and Pacific Islanders (NH/PI) as a case study, we illustrate how historical, environmental, and socioeconomic factors interact with biological pathways to drive disparities. The review highlights key advances in DNA damage research—from somatic mutation theory to the modern understanding of chronic genomic stress—and explores how innovations in single-cell genomics, biomarker discovery, and computational modeling can unravel disparity etiologies. We argue that a translational framework linking social exposure data to molecular biomarkers of DNA damage is essential for moving beyond descriptive disparities to mechanistic understanding. Ultimately, addressing cancer equity requires interdisciplinary strategies that bridge molecular oncology, public health, and community-engaged research, targeting the root causes where social inequities become biologically embedded as genomic instability.

Abstract Generative artificial intelligence (AI) is rapidly populating medical records with synthetic or partially AI-generated content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability of medical data at population scale. By analysing more than 800,000 synthetic data points across clinical text generation, vision–language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architectures. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence. Models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies that can be easily integrated into existing clinical workflows, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

Abstract Background The mechanisms of left ventricular outflow tract obstruction (LVOTO) in hypertrophic cardiomyopathy (HCM) with moderate septal hypertrophy (≤ 18 mm) remain uncertain with therapeutic implications. This study investigated the role of three-dimensional echocardiography (3DE)-derived myocardial mechanics in LVOTO. Methods We retrospectively analyzed 216 HCM patients with moderate septal hypertrophy, stratified into nonobstructive (n = 38), provokable LVOTO (n = 63), and resting LVOTO (n = 115). Transthoracic echocardiography assessed LV geometry, LVOT diameter, anterior mitral leaflet length, and papillary muscle abnormalities (via a composite SubMV score). 3DE-derived strain parameters, including global longitudinal (GLS), circumferential, radial, area (GAS) strain, twist, and torsion, were quantified. Determinants of LVOTO were evaluated using multivariate regression, restricted cubic splines (RCS), and receiver operating characteristic (ROC) analysis. Results Both LVOTO subgroups showed significantly augmented strain mechanics versus nonobstructive patients, greatest in resting LVOTO (p < 0.05); twist and torsion were the strongest discriminators (p < 0.001). Multivariate regression showed that, beyond LVOT diameter and SubMV score, torsion (B = 9.47), twist (B = 1.92), and GAS (B = 1.29) independently predicted provoked LVOT gradients (all p < 0.05); RCS demonstrated nonlinear twist/torsion–LVOTO relationships, with obstruction risk rising sharply above twist = 15° and torsion = 3°/cm, especially under provocation. Torsion discriminated provokable LVOTO achieving 81% specificity (cutoff 2.4°/cm, AUC = 0.72), outperforming all structural parameters. Integrating 3DE-derived mechanics with structural metrics improved diagnostic accuracy, especially in provokable LVOTO (AUC 0.84 vs. 0.76, DeLong p = 0.003). Conclusions In moderate-hypertrophy HCM, 3DE-derived strain mechanics, especially twist and torsion, independently determine LVOTO beyond structural narrowing, particularly under provocation, strengthening mechanistic insight and clinical applicability.