Citations

Abstract Patients with rare genetic disorders often wait years for a correct diagnosis, highlighting the urgent need for early and efficient triage for those at risk. Electronic health records (EHRs) are rich in useful information, yet the extent to which this data is sufficient to generate prediagnostic signals is uncertain. We trained machine learning models on billing codes and phenotypes extracted from clinical notes and applied these across eight conditions in a longitudinal cohort of roughly 3 million patient records from the Mayo Clinic, seeking early identification of conditions. The number of patients identified early varied by condition, ranging from 10% to 89% at 99% specificity. Median lead times exceeded one year prior to the first diagnostic billing code for most conditions, including Fabry disease (1,408 days), hereditary angioedema (2,755 days), neurofibromatosis type 1 (1,067 days), and hereditary hemorrhagic telangiectasia (2,320 days). We show that ML models trained on EHR data can prioritize patients for clinician review and confirmatory genetic testing and that the integration of diverse phenotypic data sources provides superior predictive value.

Background. Rare diseases collectively affect an estimated 300 million people worldwide, yet patients wait an average of 4-8 years for a correct diagnosis after visiting multiple physicians, accumulating unexplained findings, and suffering preventable disease progression before a unifying diagnosis is reached. Electronic health records (EHRs) capture this pre-diagnostic trajectory in rich detail, but conventional methods typically collapse years of longitudinal phenotypic signals into a single static snapshot; our approach instead models this temporal structure directly. Methods. We conducted a retrospective cohort study using longitudinal EHR data from approximately 3 million patients at Mayo Clinic Platform_Accelerate. Cases were patients with confirmed diagnoses of 10 rare diseases; controls (~5,200 per disease) had no record of any rare disease. HPO phenotype terms were extracted from clinical notes and laboratory results using a medspaCy NLP pipeline; ICD codes were grouped at three-character level. We developed and trained two temporal models, GRU-Attn and Conformer, that encode patient histories as quarterly time-binned sequences spanning up to 30 years, and compared these against three static baselines (CatBoost, XGBoost, logistic regression). Models were evaluated systematically at prediction horizons h = 1-10 years before diagnosis. Results. In the full held-out cohort benchmark, temporal models outperformed static baselines across the large majority of diseases and prediction horizons, with the advantage widening substantially as the horizon lengthened. At 1-2 years prior to diagnosis, mean AUROC was 0.928 (GRU-Attn) versus 0.871 (CatBoost); at 6-10 years prior to diagnosis, 0.726 versus 0.693, with static models degrading to near-chance for some diseases. Combining HPO and ICD features yielded the strongest performance, with a mean gain of +0.063 AUROC over the best single source at 1-2 years prior to diagnosis. Separately, in a manually curated cohort of 143 confirmed-undiagnosed cases with extended EHR histories, GRU-Attn identified 138 cases before first disease mention and all 143 cases before formal diagnosis at the prespecified 99th-percentile specificity threshold, with median lead times of 1.1-8.3 years and 1.9-21.7 years, respectively. Conclusions. By learning from the temporal trajectory of phenotypic signals rather than their static aggregate, temporal EHR models improved retrospective early detection performance across several rare diseases, most powerfully for conditions where pre-diagnostic findings accrue gradually and heterogeneously over years. These results show that longitudinal EHR trajectories carry rare-disease signal beyond static phenotypic burden and support prospective evaluation of temporal EHR modeling for rare-disease risk stratification.

Importance Elevated low-density lipoprotein cholesterol (LDL-C) is a modifiable risk factor for cardiovascular disease, the leading cause of premature death worldwide. Assessing the LDL-C–related burden is critical for guiding prevention and treatment strategies. Objectives To estimate the global, regional, and national burden of ischemic heart disease and ischemic stroke attributable to elevated LDL-C (relative to 35-54 mg/dL) from 1990 to 2023 and to quantify the contributions of population growth, aging, risk-deleted burden, and exposure changes to burden trends. Design, Setting, and Population This comparative risk assessment, part of the Global Burden of Disease Study 2023, estimated population-level LDL-C exposure and associated health loss in 204 countries and territories. Mean LDL-C levels were estimated using spatiotemporal gaussian process regression based on 806 studies across 161 countries. Relative risks were derived from meta-analyses of 38 randomized clinical trials. Population-attributable fractions for deaths and disability-adjusted life-years (DALYs) were estimated by age and sex for adults aged 25 years or older from 1990 to 2023, with 95% uncertainty intervals. Exposure Population-level LDL-C concentrations. Main Outcomes and Measures Population-attributable fractions, counts, and rates (all ages and age standardized per 100 000) of LDL-C–attributable deaths and DALYs from ischemic heart disease and ischemic stroke, with uncertainty intervals. Results In 2023, elevated LDL-C accounted for 3.6 million deaths (95% uncertainty interval, 2.2-5.4 million; 6.0% of global mortality) and 90.7 million DALYs (95% uncertainty interval, 58.9-123.3 million; 3.2% of DALYs). Although global all-ages rates remained stable, age-standardized death and DALY rates decreased by 45.6% and 39.5%, respectively, since 1990. In 2023, age-standardized LDL-C–attributable DALY rates were highest in Eastern Europe and lowest in high-income Asia-Pacific. One-third of the global LDL-C burden occurred in India and China. Population growth and aging drove the increasing burden, with notable regional disparities in LDL-C exposure and risk-deleted DALY rates shifting toward middle-sociodemographic settings. Conclusions and Relevance Despite declining age-standardized rates, the absolute LDL-C burden has increased since 1990 due to demographic changes and has shifted toward middle-sociodemographic countries. Measurement and surveillance gaps persist. Strengthened prevention, diagnosis, and treatment access strategies are essential to mitigate the health burden of LDL-C.

Importance Multidomain lifestyle interventions slow cognitive decline in at-risk older adults, but effects on cerebrovascular injury and whether any effect differs by age or across magnetic resonance imaging (MRI) markers remain unclear. Objectives To test whether a structured multidomain lifestyle intervention, vs a self-guided program, alters trajectories of cerebrovascular MRI markers and whether baseline diffusion MRI estimates subsequent vascular injury. Design, Setting, and Participants POINTER Imaging was a prespecified secondary imaging analysis of the US POINTER (US Study to Protect Brain Health Through Lifestyle Intervention to Reduce Risk) randomized clinical trial conducted at multiple US sites. Starting August 1, 2020, and continuing until March 1, 2023, all newly enrolled US POINTER trial participants who met criteria, including being 60 to 79 years of age and at increased risk for cognitive decline, were given the opportunity to enroll in POINTER Imaging. Annual MRIs were performed at baseline and up to 2 follow-up visits at 12 and 24 months. The imaging study ended May 14, 2025. Interventions Structured multidomain lifestyle arm (supervised exercise, dietary counseling, cognitive and social engagement, and health coaching) vs self-guided educational arm. Main Outcomes and Measures Primary outcomes were cerebrovascular MRI markers of global white matter free water (FW), fractional anisotropy (FA), peak width of skeletonized mean diffusivity (PSMD), analysis along the perivascular space (ALPS) index, white matter hyperintensity (WMH) volume, and incident cerebral microbleeds (CMBs). Age-group analyses (<70 vs ≥70 years) were prespecified. Linear mixed-effects models tested time, intervention arm, age group, and interactions for continuous outcomes; logistic regression evaluated incident CMBs. All analyses were adjusted for sex, hypertension, diabetes, and smoking and were conducted for the evaluable population. Data were analyzed September 5 to December 11, 2025. Results For 959 participants included in the analysis, the mean (SD) age was 68 (5) years, and 594 (61.9%) were female. A significant time × arm × age group interaction was observed for FW (β = 0.050; SE = 0.018; P = .006). In participants younger than 70 years, the structured intervention, compared with the self-guided intervention, attenuated the FW increase over time (β = −0.031; SE = 0.012; P = .009); no intervention arm difference was observed in participants aged 70 years or older (β = 0.019; SE = 0.014; P = .18). No intervention-related interactions were detected (eg, time × intervention arm interaction) for WMH (β = 0.002; SE = 0.016; P = .90), FA (β = 0.016; SE = 0.016; P = .30), PSMD (β = −0.011; SE = 0.021; P = .60), or ALPS (β = 0.006; SE = 0.021; P = .79). Overall, greater odds of incident CMBs were associated with higher baseline FW (odds ratio, 1.63 [95% CI, 1.24-2.14]; P < .001) and PSMD (odds ratio, 1.40 [95% CI, 1.08-1.81]; P < .001). Greater WMH progression was associated with higher baseline FW (β = 0.014; SE = 0.006; P = .02) and lower baseline FA (β = −0.018; SE = 0.006; P = .004). Conclusions and Relevance In this secondary imaging analysis of a randomized clinical trial, FW was the cerebrovascular MRI marker most responsive to lifestyle intervention in adults 60 to 70 years of age, showing attenuated small-vessel disease–related injury. Findings suggested greater microvascular benefit when intensive lifestyle intervention is implemented earlier in late life. Trial Registration ClinicalTrials.gov Identifier: NCT03688126

Abstract Rationale The impacts of COVID-19 vaccination and SAR-CoV-2 infection on brain tissues and structures remain controversial. This study examines whether COVID-19-related exposures are associated with measurable changes in structural magnetic resonance imaging (MRI) using data from the US general population-based Collaborative Cohort of Cohorts for COVID-19 Research (C4R) Study. Methods The C4R study is an observational meta-cohort study of 14 NIH-funded population-based cohorts, with directly measured data on COVID-19-related exposures, including vaccination and infection. For this analysis, we included participants with C4R data and paired brain MRI scans from both before and after the onset of the COVID-19 pandemic (defined as March 1, 2020) from two cohorts: the Atherosclerosis Risk in Communities study (ARIC) and Framingham Heart Study (FHS). MRI features examined as outcomes were white matter volume (WMV), gray matter volume (GMV), total brain volume (BV), hippocampal volume (HV), and white matter hyperintensity volume (WMH), a marker of microvascular disease. Exposures included SARS-CoV-2 infections identified via questionnaires, serosurveys, and medical record reviews from 2020-2025. Infections resulting in hospitalization were classified as severe. Time-to-recovery, symptoms of long COVID, and vaccination status were self-reported by questionnaire. Analysis of covariance (ANCOVA) estimated associations between COVID-19-related exposures and pandemic MRI-measured outcomes, with adjustment for: the corresponding pre-pandemic MRI measure, the time between the two MRI scans, age, sex, systolic blood pressure, diabetes, and cardiovascular disease. Results Of 270 participants with available data, mean (SD) age was 80.6 years (10.6); 43.0% were male; and the mean (SD) interval between paired MRI scans was 4.2 (2.0) years. Most (90%) were vaccinated (of whom 6.7% vaccinated before infection) and 7.9% reported SARS-CoV-2 infection (of which 13.6% were severe infections) prior to their pandemic MRI (conducted March 2020 - June 2022). In fully adjusted models, there was no statistically significant association for change in WMV, GMV, BV, HV, or WMH with SAR-CoV2 infection status, infection severity, vaccination status, or vaccination timing (Table, P > 0.05 for all comparisons). Self-reported symptoms of long COVID and vaccine manufacturer were also not associated with MRI outcomes. Conclusions Based on repeated (pre-pandemic and pandemic era) structural MRI scans, we found no evidence that SARS-CoV-2 infection or COVID-19 vaccination were associated with changes in brain structure or brain microvascular disease. These findings are consistent with prior data supporting COVID-19 vaccine safety and suggest resilience of brain structures following SARS-CoV-2 exposures. Nonetheless, further research is warranted using more sensitive measures and assessing longer term effects. This abstract is funded by: OT2HL156812

Abstract INTRODUCTION To examine the generalizability of Alzheimer's disease (AD) biomarker models in real‐world older adults, we examined AD biomarker relationships with cognition in two multicenter cohorts that differ with respect to recruitment approach and health risk factors but were matched on a variety of characteristics. METHODS We compared harmonized health and demographic data, AD and cerebrovascular biomarkers, and cognitive performance in the community‐recruited U.S. Study to Protect Brain Health Through Lifestyle Intervention to Reduce Risk (U.S. POINTER) Imaging substudy and a matched sample from the Alzheimer's Disease Neuroimaging Initiative (ADNI) which recruited primarily from academic specialty clinics. RESULTS Elevated β‐amyloid (Aβ) and tau were associated with cognitive performance in ADNI but not U.S. POINTER. Findings were consistent across different cohort matching schemes, and were not explained by discrepancies in vascular risk. DISCUSSION The role of Aβ and tau in cognitive performance may be reduced in real world samples compared to academic specialty clinics.

Objective Meaningful assessments of how large language models (LLMs) incorporate clinical guidelines require large-scale testing over many queries. Here, we evaluate the prevalence of clinical guideline omissions and hallucinations in a large sample of diagnostic LLM outputs. Methods We used simulated case vignettes and zero-shot prompting to generate diagnostic outputs and rationales from GPT-4.1 and DeepSeek-V3. English case vignettes were created for hypercholesterolaemia and type-2 diabetes mellitus. Each vignette contained identical medical information, while sociodemographic characteristics varied in terms of sex, ethnicity and location. We calculated the prevalence of existing and hallucinated clinical guidelines in LLM outputs across disease, LLM and sociodemographic characteristics. Results We analysed a total of 12 197 LLM outputs, which quantifies three hazard areas: omissions (up to 97% for DeepSeek-V3 and 46% for GPT-4.1), hallucinations (up to 9%) and inconsistencies (guideline citation rate ranging from 0% to 78.39% across sociodemographic vignettes). Omission and hallucination rates were generally similar across vignettes with different sex or ethnicity data, yet were particularly sensitive to patient location. Discussion This study highlights significant variability in clinical guideline prediction across two different diseases, three different sociodemographic variables and two LLMs, even when the LLMs were instructed by identical prompts, establishing clinical guideline prediction in LLM outputs as a stochastic event. Conclusion The stochastic nature of LLMs creates a unique challenge for evidence generation and clinical deployment. Being able to measure and capture this stochasticity within high-quality research designs will be a prerequisite to advancing the responsible deployment of LLMs in healthcare.

Importance Brain imaging biomarkers may (1) identify underlying mechanisms of intervention effects and (2) define brain characteristics of those most likely to benefit cognitively from the intervention. Objective To test whether (1) a lifestyle intervention is related to brain changes, (2) brain changes are correlated with intervention-specific cognition changes, and (3) brain imaging biomarkers provide information about who is likely to benefit from a lifestyle intervention. Design, Setting, and Participants In this randomized clinical trial, the POINTER Imaging study was an ancillary study of the 2-year, single-blind multicenter US POINTER clinical trial in which neuroimaging data were collected on eligible enrolled participants. The study was conducted from May 2019 to March 2023 (final follow-up, May 14, 2025). Participants were enrolled into the POINTER Imaging study following a standard procedure at 5 clinical sites in the US. Participant eligibility criteria included an age of 60 to 79 years, sedentary lifestyle, and suboptimal diet, plus at least 2 additional risk criteria for cognitive decline and no contraindications for neuroimaging. Interventions Participants were randomly assigned to receive the structured or self-guided intervention. Both interventions emphasized increased physical and cognitive activity, healthy nutrition, social engagement, and cardiovascular health monitoring, but they differed in intensity and accountability. Main Outcomes and Measures There were 4 primary imaging outcomes: global β-amyloid (Aβ) burden, tau burden in the entorhinal cortex (ERC), hippocampal (HC) volume, and white matter hyperintensity volume. The primary cognitive measure was a global cognitive composite. Results Among 1943 parent trial participants who were assessed for eligibility, 983 underwent at least 1 magnetic resonance imaging or positron emission tomography scanning session. The mean (SD) age was 68.4 (5.2) years; 605 participants (61.5%) were female and 378 male (38.5%). Five hundred sixteen participants received the structured intervention and 467 the self-guided intervention. There were no intervention group differences in longitudinal cognitive or imaging outcomes. Intervention group differences were not associated with or moderated by Aβ status or accumulation. There was a negative association between change in ERC tau and change in global cognition in the self-guided group that was attenuated in the structured group (difference in association: 0.289; 95% CI, 0.029 to 0.550; interaction P = .03). Lower baseline HC volume was associated with greater cognitive benefit for participants in the structured group vs the self-guided group (lower HC: 0.077 SD; 95% CI, 0.022 to 0.132; higher HC: 0.002 SD; 95% CI, −0.037 to 0.041; interaction P = .03). Conclusion and Relevance This study found that a high-intensity multidomain lifestyle intervention did not affect brain biomarker trajectories and was not associated with Aβ pathology, but older adults with specific at-risk brain characteristics, including lower baseline HC volume, showed a greater cognitive benefit of the structured intervention. Trial Registration ClinicalTrials.gov Identifier: NCT03688126

Abstract Background Women with severe aortic stenosis (AS) are diagnosed later and experience poorer outcomes than men, partly because clinical approaches rely on 2D, valve-centric thresholds derived from male-predominant cohorts that underutilize information from 3D left ventricular (LV) geometry. We hypothesize that a sex-specific computational framework integrating statistical shape analysis (SSA) of pre-TAVR CT with machine learning would improve prediction of 1-year LV mass regression (LVMR). Objective To develop a computational framework leveraging 3D LV geometry and evaluate whether it improves sex-specific prediction of 1-year LVMR after TAVR. Methods We studied 339 patients with severe AS who underwent TAVR from 2013 to 2020 and had pre-TAVR CT and 1-year post-TAVR echocardiography. LV geometries were segmented into digital twins, and shape modes predictive of LVMR were extracted using SSA and partial least squares. These modes were incorporated into support vector regression models and compared with conventional echocardiographic predictors, including pre-TAVR LVEF, LVMI, and E/A ratio. Performance was assessed using RMSE and R 2 . Results After one year, 65% of patients showed positive LVMR, with median regression of approximately 10%; regression was significant overall and within each sex ( p < 0.001) and similar between sexes ( p = 0.99). Predictive shape modes differed by sex ( p < 0.01), with women showing more localized variation and men broader geometric gradients. Sex-specific shape modes outperformed general modes and clinical metrics, particularly in women (R 2 = 0.80, RMSE = 0.09 vs. R 2 = 0.59, RMSE = 0.13; clinical-only baseline R 2 = 0.16, RMSE = 0.22). In men, sex-specific modes also performed strongly (R 2 = 0.89, RMSE = 0.08). Conclusion In severe AS, 3D LV geometry predicts post-TAVR reverse remodeling more accurately than conventional metrics and may improve risk stratification, particularly in women.