Decomposing the onset prediction task reveals that the model often misclassifies patients who were diagnosed later as positive, suggesting the patient journey embeddings encode disease state more reliably than care timing.
Abstract
Hypertrophic and dilated cardiomyopathy (HCM and DCM) carry substantial morbidity and mortality, yet diagnosis may be delayed, particularly when presentation is nonspecific. Existing machine-learning approaches to cardiomyopathy phenotyping, genotype prediction, and risk stratification commonly rely on disease-specific, hand-engineered features drawn from echocardiography, cardiac MRI, ECG, or curated clinical variables. We evaluated whether a general-purpose clinical foundation model, CLMBR-T-base, pre-trained via next-clinical-event prediction with no cardiomyopathy-specific supervision, could produce linearly separable embeddings for all three case/control cohorts. Using EHR data from the Penn Medicine BioBank, we constructed cohorts for (1) prediction of a first recorded qualifying HCM/DCM diagnosis at 1-, 3-, and 6-month horizons, decomposed into eventual-versus-never-case and imminent-versus-eventual comparisons; (2) genetic carrier status prediction among diagnosed patients with completed gene panels; and (3) prediction of heart-failure hospitalization, and all-cause mortality as both binary and time-to-event outcomes. Linear probes fitted to frozen embeddings achieved AUROCs of 0.75-0.82 for onset prediction, 0.74-0.75 for genotype status, and Harrell's concordance of 0.65-0.80 for time-to-event outcomes. Decomposing the onset prediction task reveals that the model often misclassifies patients who were diagnosed later as positive, suggesting the patient journey embeddings encode disease state more reliably than care timing. These results suggest that a single, generically pretrained EHR embedding can support multiple clinically motivated prediction problems in CM without disease-specific feature engineering.
It is emphasized that successful integration of AI into cardiovascular care requires rigorous prospective validation, transparent algorithmic governance, equitable data representation, and human-AI collaborative frameworks, provided its meaningful clinical implication is demonstrated through improved patient outcomes.
Xu Xia, Wasim Ullah Khan, Q. Khan et al.· Trends in cardiovascular med...· 0 citations
Cardiomyopathies are a leading cause of heart failure, arrhythmia and sudden cardiac death, yet patients sharing a diagnosis, and even a causal variant, follow markedly different clinical courses. This thesis argues that disease expression depends on the biological context in which genetic susceptibility acts, and that studying this requires experimental models containing the relevant context. Each chapter adds a layer.
The starting point is genetic. Among nearly 470,000 UK Biobank participants, carriers of TNNT2 p.Arg288Cys showed no excess hypertrophic cardiomyopathy after correction for multiple testing and cumulative penetrance below 1% by age 80, while displaying subtle functional differences on cardiac MRI. Dutch probands presented with more severe disease, consistent with ascertainment bias and enrichment for modifiers. The variant is therefore intermediate in effect, insufficient alone but capable of contributing in a permissive context. The question shifts from whether a variant is pathogenic to under which conditions it becomes so.
Cellular identity is the first layer. Endothelial and smooth muscle cells from nine canine macrovessels retained location-associated transcriptional programmes through at least three passages, with embryonic positional transcription factors providing the strongest discrimination. An epigenetic mechanism is plausible. Cell source is thus a design decision rather than a practical detail.
Local tissue context was added in a fibrin-based co-culture, where endothelial cells self-assembled into lumenised networks alongside hiPSC-derived cardiomyocytes and remained stable for at least three weeks in a commercially available microfluidic device. The positional identity findings simultaneously define its main limitation, since umbilical vein endothelial cells were used.
Environmental context was introduced as lipid overload in an isogenic MYH7 p.Arg403Leu pair. A model providing context is of limited value if the readout averages that context away. Across 44,403 single-cell measurements of calcium handling and contraction, within-well variability accounted for 76 to 96% of total variance, and clustering identified a lipid-enriched subpopulation invisible in well averages. Model complexity and measurement resolution are two sides of the same problem.
Some layers cannot be engineered. Lifelong exposure, genetic diversity and spontaneous progression are inherent to naturally occurring disease in companion animals. Canine dilated cardiomyopathy returns genetic susceptibility to the foreground with full context attached, with breed-level uniformity supporting bidirectional discovery. In 37 cats with hypertrophic cardiomyopathy, digital pathology combined with Nanopore sequencing revealed structure only after outcome-based reclassification, distinguishing arterial thromboembolism from congestive heart failure.
The same morphometric approach across 46 vertebrate species established a physiological reference frame, since a tenfold increase in body mass corresponded to roughly 8% greater cardiomyocyte diameter and no species approached dimensions reported in hypertrophic cardiomyopathy.
The contribution is both methodological, favouring continuous measurement analysed with models matched to nested data structures, and conceptual, in showing that genetic findings acquire meaning only within their spatial, metabolic and species-specific context. No single model suffices, but complementary models each supply a different layer.
Overall, AI-empowered echocardiography holds substantial promise for advancing precision diagnosis, risk stratification, and personalized management of HCM, facilitating a transition toward more intelligent and individualized cardiovascular care.
Miao Zhang, Shan-Shan Yuan, Hong-Yan Dai et al.· Frontiers in Cardiovascular...· 0 citations
BACKGROUND
Periodic reinterpretation of variants associated with hypertrophic cardiomyopathy (HCM) is recommended in the light of evolving knowledge, but it requires considerable resources, and its clinical impact is unresolved.
OBJECTIVES
We here report the results of a systematic variant reclassification of HCM-associated variants identified at a national referral center, and the impact on clinical profiling.
METHODS
A total of 805 consecutive probands with a definite HCM diagnosis genotyped in 1998-2023 (overall 276 variants: 162 pathogenic/likely pathogenic -P/LP-, 109 variants of uncertain significance -VUS-, and 5 benign/likely benign - B/LB) underwent variant reclassification. All were analyzed for all-cause death, ventricular arrhythmia composite (sudden death, cardiac arrest, appropriate implantable cardioverter-defibrillator therapy), atrial fibrillation or cerebrovascular events, heart failure composite (NYHA class III/IV, left ventricular ejection fraction <35%, cardiac transplantation).
RESULTS
After a median follow-up of 8.8 [6.1-11.8] years, among the 276 variants, 61 (22.1%) were reclassified: 29 variants from P/LP to VUS or B/LB, 21 from VUS to P/LP, and 11 from VUS to B/LB, affecting 69 patients overall. The overall yield of genetic testing (% with P/LP) changed from 58.7 to 55.4%. At survival analysis, HRs associated with P/LP status improved after reclassification for both all-cause death and ventricular arrhythmia. No changes were observed for the other outcomes.
CONCLUSION
Systematic reclassification of genetic variants led to a refinement of variant classification accuracy due to downgrading of 5.5% of P/LP variants, although 18.3%VUS/B/LB were upgraded to P/LP. Reclassification more accurately identified risks associated with P/LP status, compared to the initial adjudication.
A. Del Franco, Valeria Setti, Federica Colio et al.· International Journal of Car...· 0 citations
A fully automated ML model identifies area-derived LACI at end-diastole at end-diastole as a robust feature associated with disease progression, providing improved risk stratification for pre-symptomatic HCM.
Antoine Olivier, Auriane Riou, T. D'humières et al.· Frontiers in Cardiovascular...· 0 citations
Aims To characterize cardiac magnetic resonance (CMR)-derived phenotypes in a population-based cohort and to evaluate their incremental prognostic value for major adverse cardiovascular events (MACE) beyond traditional risk factors. Methods Participants from the UK Biobank imaging cohort without prior cardiovascular diseases were included. Unsupervised k-means clustering and Kaplan–Meier analyses identified distinct CMR-derived phenotypes. Sequential XGBoost models were developed to evaluate the incremental predictive value of CMR features over conventional risk factors, with SHAP used for interpretation. Cross-modality transportability was assessed in an independent Asian cohort using echocardiographic data as a surrogate for CMR-derived variables. Results Among 27,254 participants, 785 (2.88%) experienced MACE over a median follow-up of approximately 5 years. Two phenotypes were identified, with the higher-risk phenotype characterized by increased cardiac volumes, impaired cardiac function, a tendency toward myocardial hypertrophy, and reduced myocardial strain. The CMR-enhanced model improved MACE prediction compared with traditional risk factors alone (AUC: 0.76, 95% CI: 0.73–0.79; vs. AUC: 0.61, 95% CI: 0.58–0.64, P < 0.001), with strong performance for heart failure (AUC: 0.90, 95% CI: 0.86–0.94). In the cross-modality validation cohort, the simplified model yielded an AUC of 0.69 (95% CI: 0.56–0.83). Higher age, left ventricular mass index, global wall thickness, and lower levels of high-density lipoprotein cholesterol, left atrial ejection fraction, right atrial stroke volume were the key contributors to increased MACE risk. Conclusion CMR-derived phenotypes provide incremental prognostic value beyond traditional risk factors and may improve early identification of individuals at elevated cardiovascular risk.
Pei Liu, Chang Liu, Yao Ma et al.· Frontiers in Medicine· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
