Large Language Model-informed ECG Dual Attention Network for Heart Failure Risk Prediction.

Importance:Despite the availability of disease-modifying therapies, scalable strategies for heart failure (HF) risk stratification remain elusive. Portable devices capable of recording single-lead electrocardiograms (ECGs) can enable large-scale community-based risk assessment.Objective:To evaluate an artificial intelligence (AI) algorithm to predict HF risk from noisy single-lead ECGs.Design:Multicohort study.Setting:Retrospective cohort of individuals with outpatient ECGs in the integrated Yale New Haven Health System (YNHHS) and prospective population-based cohorts of UK Biobank (UKB) and Brazilian Longitudinal Study of Adult Health (ELSA-Brasil).Participants:Individuals without HF at baseline.Exposures:AI-ECG-defined risk of left ventricular systolic dysfunction (LVSD).Main Outcomes and Measures:Among individuals with ECGs, we isolated lead I ECGs and deployed a noise-adapted AI-ECG model trained to identify LVSD. We evaluated the association of the model probability with new-onset HF, defined as the first HF hospitalization. We compared the discrimination of AI-ECG against two risk scores for new-onset HF (PCP-HF and PREVENT equations) using Harrel’s C-statistic, integrated discrimination improvement (IDI), and net reclassification improvement (NRI).Results:There were 192,667 YNHHS patients (age 56 years [IQR, 41–69], 112,082 women [58%]), 42,141 UKB participants (65 years [59–71], 21,795 women [52%]), and 13,454 ELSA-Brasil participants (56 years [41–69], 7,348 women [55%]) with baseline ECGs. A total of 3,697 developed HF in YNHHS over 4.6 years (2.8–6.6), 46 in UKB over 3.1 years (2.1–4.5), and 31 in ELSA-Brasil over 4.2 years (3.7–4.5). A positive AI-ECG screen was associated with a 3- to 7-fold higher risk for HF, and each 0.1 increment in the model probability portended a 27–65% higher hazard across cohorts, independent of age, sex, comorbidities, and competing risk of death. AI-ECG’s discrimination for new-onset HF was 0.725 in YNHHS, 0.792 in UKB, and 0.833 in ELSA-Brasil. Across cohorts, incorporating AI-ECG predictions in addition to PCP-HF and PREVENT equations resulted in improved Harrel’s C-statistic (ΔPCP-HF=0.112–0.114; ΔPREVENT=0.080–0.101). AI-ECG had IDI of 0.094–0.238 and 0.090–0.192, and NRI of 15.8%−48.8% and 12.8%−36.3%, vs. PCP-HF and PREVENT, respectively.Conclusions and Relevance:Across multinational cohorts, a noise-adapted AI model defined HF risk using lead I ECGs, suggesting a potential portable and wearable device-based HF risk-stratification strategy.

Despite the availability of disease-modifying therapies, scalable strategies for heart failure (HF) risk stratification remain elusive. Portable devices capable of recording single-lead electrocardiograms (ECGs) may enable large-scale community-based risk assessment. To evaluate whether an artificial intelligence (AI) algorithm can predict HF risk from noisy single-lead ECGs. A retrospective cohort study of individuals without HF at baseline was conducted among individuals with conventionally obtained outpatient ECGs in the integrated Yale New Haven Health System (YNHHS) and prospective population-based cohorts of the UK Biobank (UKB) and the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Data analysis was performed from September 2023 to February 2025. AI-ECG-defined risk of left ventricular systolic dysfunction (LVSD). Among individuals with ECGs, lead I ECGs were isolated and a noise-adapted AI-ECG model (to simulate ECG signals from wearable devices) trained to identify LVSD was deployed. The association of the model probability with new-onset HF, defined as the first HF hospitalization, was evaluated. The discrimination of AI-ECG was compared against 2 risk scores for new-onset HF (Pooled Cohort Equations to Prevent Heart Failure [PCP-HF] and Predicting Risk of Cardiovascular Disease Events [PREVENT] equations) using the Harrel C statistic, integrated discrimination improvement, and net reclassification improvement. There were 192 667 YNHHS patients (median [IQR] age, 56 [41-69] years; 111 181 women [57.7%]), 42 141 UKB participants (median [IQR] age, 65 [59-71] years; 21 795 women [51.7%]), and 13 454 ELSA-Brasil participants (median [IQR] age, 51 [45-58] years; 7348 women [54.6%]) with baseline ECGs. A total of 3697 (1.9%) developed HF in YNHHS over a median (IQR) of 4.6 (2.8-6.6) years, 46 (0.1%) in UKB over a median (IQR) of 3.1 (2.1-4.5) years, and 31 (0.2%) in ELSA-Brasil over a median (IQR) of 4.2 (3.7-4.5) years. A positive AI-ECG screening result for LVSD was associated with a 3- to 7-fold higher risk for HF, and each 0.1 increment in the model probability was associated with a 27% to 65% higher hazard across cohorts, independent of age, sex, comorbidities, and competing risk of death. AI-ECG's discrimination for new-onset HF was 0.723 (95% CI, 0.694-0.752) in YNHHS, 0.736 (95% CI, 0.606-0.867) in UKB, and 0.828 (95% CI, 0.692-0.964) in ELSA-Brasil. Across cohorts, incorporating AI-ECG predictions alongside PCP-HF and PREVENT equations was associated with a higher Harrel C statistic (difference in addition to PCP-HF, 0.080-0.107; difference in addition to PREVENT, 0.069-0.094). AI-ECG had an integrated discrimination improvement of 0.091 to 0.205 vs PCP-HF and 0.068 to 0.192 vs PREVENT; it had a net reclassification improvement of 18.2% to 47.2% vs PCP-HF and 11.8% to 47.5% vs PREVENT. Across multinational cohorts, a noise-adapted AI-ECG model estimated HF risk using lead I ECGs, suggesting a potential HF risk-stratification strategy requiring prospective study using wearable and portable ECG devices.

Heart failure (HF) is a global pandemic with a growing prevalence and is a growing burden on the healthcare system. Machine learning (ML) has the potential to revolutionize medicine and can be applied in many different forms to aid in the prevention of symptomatic HF (stage C). HF prevention currently has several challenges, specifically in the detection of pre-HF (stage B). HF events are missed in contemporary models, limited therapeutic options are proven to prevent HF, and the prevention of HF with preserved ejection is particularly lacking. ML has the potential to overcome these challenges through existing and future models. ML has limitations, but the many benefits of ML outweigh these limitations and risks in most scenarios. ML can be applied in HF prevention through various strategies such as refinement of incident HF risk prediction models, capturing diagnostic signs from available tests such as electrocardiograms, chest x-rays, or echocardiograms to identify structural/functional cardiac abnormalities suggestive of pre-HF (stage B HF), and interpretation of biomarkers and epigenetic data. Altogether, ML is able to expand the screening of individuals at risk for HF (stage A HF), identify populations with pre-HF (stage B HF), predict the risk of incident stage C HF events, and offer the ability to intervene early to prevent progression to or decline in stage C HF. In this narrative review, we discuss the methods by which ML is utilized in HF prevention, the benefits and pitfalls of ML in HF risk prediction, and the future directions.

Stroke prevention in heart failure and sinus rhythm: where do we go from here?

Heart failure (HF) risk and the underlying risk factors vary by race. Traditional models for HF risk prediction treat race as a covariate in risk prediction and do not account for significant parameters such as cardiac biomarkers. Machine learning (ML) may offer advantages over traditional modeling techniques to develop race-specific HF risk prediction models and to elucidate important contributors of HF development across races. We performed a retrospective analysis of 4 large, community cohort studies (ARIC [Atherosclerosis Risk in Communities], DHS [Dallas Heart Study], JHS [Jackson Heart Study], and MESA [Multi-Ethnic Study of Atherosclerosis]) with adjudicated HF events. The study included participants who were >40 years of age and free of HF at baseline. Race-specific ML models for HF risk prediction were developed in the JHS cohort (for Black race-specific model) and White adults from ARIC (for White race-specific model). The models included 39 candidate variables across demographic, anthropometric, medical history, laboratory, and electrocardiographic domains. The ML models were externally validated and compared with prior established traditional and non-race-specific ML models in race-specific subgroups of the pooled MESA/DHS cohort and Black participants of ARIC. The Harrell C-index and Greenwood-Nam-D'Agostino χ2 tests were used to assess discrimination and calibration, respectively. The ML models had excellent discrimination in the derivation cohorts for Black (n=4141 in JHS, C-index=0.88) and White (n=7858 in ARIC, C-index=0.89) participants. In the external validation cohorts, the race-specific ML model demonstrated adequate calibration and superior discrimination (Black individuals, C-index=0.80-0.83; White individuals, C-index=0.82) compared with established HF risk models or with non-race-specific ML models derived with race included as a covariate. Among the risk factors, natriuretic peptide levels were the most important predictor of HF risk across both races, followed by troponin levels in Black and ECG-based Cornell voltage in White individuals. Other key predictors of HF risk among Black individuals were glycemic parameters and socioeconomic factors. In contrast, prevalent cardiovascular disease and traditional cardiovascular risk factors were stronger predictors of HF risk in White adults. Race-specific and ML-based HF risk models that integrate clinical, laboratory, and biomarker data demonstrated superior performance compared with traditional HF risk and non-race-specific ML models. This approach identifies distinct race-specific contributors of HF.

Genetic susceptibility to various chronic diseases has been shown to influence heart failure (HF) risk. However, the underlying biological pathways, particularly the role of leukocyte telomere length (LTL), are largely unknown. We investigated the impact of genetic susceptibility to chronic diseases and various traits on HF risk, and whether LTL mediates or modifies the pathways. We conducted prospective cohort analyses on 404 883 European participants from the UK Biobank, including 9989 incident HF cases. Multivariable Cox regression was used to estimate associations between HF risk and 24 polygenic risk scores (PRSs) for various diseases or traits previously generated using a Bayesian approach. We assessed multiplicative interactions between the PRSs and LTL previously measured in the UK Biobank using quantitative PCR. Causal mediation analyses were conducted to estimate the proportion of the total effect of PRSs acting indirectly through LTL, an integrative marker of biological aging. We identified 9 PRSs associated with HF risk, including those for various cardiovascular diseases or traits, rheumatoid arthritis (P = 1.3E-04), and asthma (P = 1.8E-08). Additionally, longer LTL was strongly associated with decreased HF risk (P-trend = 1.7E-08). Notably, LTL strengthened the asthma-HF relationship significantly (P-interaction = 2.8E-03). However, LTL mediated only 1.13% (P < 0.001) of the total effect of the asthma PRS on HF risk. Our findings shed light onto the shared genetic susceptibility between HF risk, asthma, rheumatoid arthritis, and other traits. Longer LTL strengthened the genetic effect of asthma in the pathway to HF. These results support consideration of LTL and PRSs in HF risk prediction.

Are physicians neglecting the risk of heart failure in diabetic patients who are receiving sulfonylureas? Lessons from the TOSCA.IT trial.

Is preventive medicine responsible for the increasing prevalence of heart failure?

To summarise existing heart failure (HF) risk prediction models and describe the risk factors for HF-related adverse outcomes in adult patients with congenital heart disease (CHD). We performed a systematic...

The implantable cardioverter-defibrillator (ICD) was devised to satisfy the unmet need for an effective, instantaneous therapy to prevent sudden cardiac death (SCD) due to ventricular fibrillation (VF) in at-risk, ambulatory patients. That therapy was a high-voltage electric shock delivered directly into the heart muscle. More than 3 decades later, shocks are still the defining operating characteristic of ICDs, and no other instantaneously effective therapy for VF exists. This elite status was clinched by large randomized clinical trials1,2 which demonstrated that ICDs improved mortality in patients with reduced left ventricular ejection fraction, regardless of pathogenesis or accompanying symptoms of heart failure (HF), by primary prevention of SCD due to ventricular tachyarrhythmia (VTA). Like bradycardia pacemakers for asystole, the ICD resides as a therapy genre of one, with no peer, and no competitor on the horizon. These sibling therapies for lethal heart rhythm disturbances will stand prominently among the greatest medical achievements of the 20th century. The ICD is a mature technology, and neither the technique nor the tools have changed much for several decades. Despite a certain evolutionary elegance of the operating system, the ICD is still a blunt instrument. Although it is true that some innovation has occurred, it is still a matter of a shock delivered by insulated metal conductors residing somewhere in direct proximity to the heart. No innovation beyond the fundamental of a timed shock for VF has proven to enhance mortality benefit. The basic design persists simply because no one can think of a suitable alternative and the self-satisfying aphorism that “shocks save lives.” Yet there is a growing intellectual dissatisfaction with the unintended consequences of this powerful, irreplaceable therapy. The stimulus for this self-inspection is an awareness of the very high morbidity risk overhead borne by the primary prevention patient, in particular, …

Introduction: Immune dysfunction is a mechanism involved in atherosclerosis; atherosclerotic cardiac events contribute to heart failure (HF) risk. We investigated whether immune cell subtypes were associated with HF and HF types in people with and without HIV. Hypothesis: Six CD4+ and CD8+ T-cell subsets previously associated with atherosclerosis will be associated with increased HF with reduced ejection fraction (HFrEF) risk. Other lymphocyte subsets will be associated with increased HF with preserved ejection fraction (HFpEF) risk. Methods: We measured lymphocyte subsets (T-, B-, and natural killer cells) in the Veterans Aging Cohort Study (VACS) Biomarker Cohort. Primary analyses used Cox proportional hazards models to estimate hazard ratios of HFpEF or HFrEF per standard deviation increment in each of the 6 pro-atherosclerotic T-cell subsets. Secondary analyses used survival random forest analysis to rank the importance of a broader under-studied set of lymphocyte subsets in HF risk prediction and Cox models to estimate HF risk. Results: Among 2174 (66% PWH) participants without prevalent cardiovascular disease, 269 had incident HF (22% HFrEF, 47% HFpEF) over 8.8 years median follow-up time. Participants were predominantly male (&gt;90%) and black (&gt;66%). After adjusting for age, race/ethnicity, cytomegalovirus status, and HIV status, we did not detect associations the 6 T-cell subsets with HFrEF. We detected positive associations of CD4+CD28-, CD4+CD45RA+CD28-CD57+, TH1, TH17 cells with HFpEF. Survival random forest analysis revealed the relative importance of lymphocyte subsets for prediction of HF beyond those previously associated with atherosclerosis (Fig 1). Conclusion: Contrary to hypotheses, pro-atherosclerotic subsets of CD4+ and CD8+ T-cells were associated with HFpEF but not HFrEF. Other lymphocyte subsets previously-understudied for incident human cardiac outcomes were associated with HF incidence. Future work will examine whether these associations differ by HIV status.

Heart failure (HF) is a global pandemic affecting over 64 million people worldwide. Its prevalence is on an upward trajectory, with associated increasing healthcare expenditure. Organizations including the American College of Cardiology (ACC) and the American Heart Association (AHA) have identified HF prevention as an important focus. Recently, the ACC/AHA/Heart Failure Society of America (HFSA) Guidelines on heart failure were updated with a new Class IIa, Level of Evidence B recommendation for biomarker-based screening in patients at risk of developing heart failure. In this review, we evaluate the studies that have assessed the various roles and contributions of biomarkers in the prediction and prevention of heart failure. We examined studies that have utilized biomarkers to detect cardiac dysfunction or abnormality for HF risk prediction and screening before patients develop clinical signs and symptoms of HF. We also included studies with biomarkers on prognostication and risk prediction over and above existing HF risk prediction models and studies that address the utility of changes in biomarkers over time for HF risk. We discuss studies of biomarkers to guide management and assess the efficacy of prevention strategies and multi-biomarker and multimodality approaches to improve risk prediction.

The relationship between heart failure (HF) and Mediterranean and DASH diets is not well delineated. This meta-analysis aimed to assess the effectiveness of high adherence to Mediterranean and DASH diets compared to low adherence in reducing the risk of incident HF (primary prevention of HF) and reducing all-cause mortality in patients with HF (secondary prevention of HF). The reporting stages of this meta-analysis closely adhered to the PRISMA guidelines. A comprehensive literature search was undertaken for published papers in PubMed, Embase, EBSCO, ICTRP, and the NIH clinical trials databases. A total of 16 reports from 14 studies were included in this paper. A significant inverse association was identified between high adherence to the Mediterranean diet model (compared to low adherence) and the risk of incident HF (OR = 0.77, 95% CI: 0.63-0.93, p = 0.007) among patients without previous diagnosis of HF. Similarly, there was a significant and inverse relationship between high adherence to the DASH diet (compared to low adherence) and the risk of incident HF (OR = 0.83, 95% CI: 0.70-0.98, p = 0.03) among patients without previous diagnosis of HF. High adherence to the Mediterranean diet model (compared to low adherence) was associated with lower all-cause mortality (OR = 0.88, 95% CI: 0.78-0.99, p = 0.03) among patients with HF. This paper demonstrated that high adherence to Mediterranean and DASH diets significantly reduced the risk of incident HF among individuals without a previous diagnosis of HF, whereas only high adherence to the Mediterranean diet was associated with lower all-cause mortality among patients with HF.

A new technique for the prediction of heart failure risk driven by hierarchical neighborhood component-based learning and adaptive multi-layer networks

Effective heart failure (HF) prevention strategies are needed, given concerning epidemiological trends1 and continued difficulty in identifying disease-modifying therapies for HF with preserved ejection fraction (HFpEF).2 However, focusing strategies on those with established risk factors for HF presents challenges because of the large numbers involved. Moreover, most of these patients are at relatively low risk, never developing HF. It is essential, therefore, to phenotype those at-risk patients who transition to symptomatic HF, identifying predictors of transition risk, in order to target preventative strategies appropriately. The St Vincent's screening TO Prevent-Heart Failure (STOP-HF) programme,3 a clinically proven and cost-effective approach to HF prevention,4 offers a valuable opportunity to obtain insight into risk stratification and new-onset HF over time, based on longitudinal follow-up of an at-risk cohort. This programme, an extension of the STOP-HF trial,3 utilises B-type natriuretic peptide (BNP)-based risk stratification and collaborative care between community physicians and specialist cardiology services to optimise risk factor control, prevent HF and monitor for the onset of symptomatic HF. Herein, we report on the baseline phenotype of at-risk patients who later developed HF within the STOP-HF cohort, and examine their change in natriuretic peptide over time. Moreover, we investigate whether ongoing surveillance for HF symptoms results in earlier HF diagnosis compared with patients diagnosed through routine referral pathways from the community. A total of 237 patients within the STOP-HF programme (no baseline HF, >40 years of age with one or more risk factors for HF) were followed for a median of 4.1 years. At initial assessment and approximately yearly review,3 patient history, clinical examination, HF risk factors, medications, Doppler echocardiogram and blood biochemistry, including BNP, were recorded. New-onset HF diagnosis was made by a staff cardiologist using established criteria5 and these patients were designated ‘transitioners’. Transitioners were compared firstly with those who did not develop HF (non-transitioners) and subsequently, following HF diagnosis, with community-based patients referred by family physicians to our rapid-access new diagnostic HF clinic. Propensity score matching was applied to match transitioners to non-transitioners by age, gender and follow-up, and change in BNP over time was examined. Summary statistics were mean or median (± standard deviation or interquartile range) and number (%). Differences between groups were tested using parametric/non-parametric tests as appropriate. Logistic regression analysis and net reclassification index were used to assess associates of transition to HF. Bootstrapping analyses used 10 000 iterations, and 2-sided P-values <0.05 were considered statistically significant. During a median follow-up of 4.1 years, 86 of the 2037 patients transitioned to HF (4.2%, 8.3 per 1000 patient-years). The majority (70.1%) of transitioners developed HFpEF. Univariate baseline characteristics of transitioners vs. non-transitioners are presented in Table 1. Transitioners were older, more likely to be male, and had higher baseline body mass index, BNP and creatinine than non-transitioners. Co-morbidities, including hypertension, were more prevalent in transitioners, however both groups had similar usage of renin–angiotensin–aldosterone-modifying therapies. Doppler echocardiography showed lower left ventricular ejection fraction and higher left ventricular mass index, left atrial volume index and E/E' ratio (all P < 0.001) in transitioners at baseline. Baseline BNP was a strong, independent associate of transition to HF in univariate and multivariate analyses (P < 0.001). Baseline BNP and increase in BNP over time were greater in transitioners than non-transitioners (P < 0.001) (Figure 1). Each unit increase in log10 BNP tripled the likelihood of transitioning and the magnitude of this estimate did not change in multivariable or stepwise models, suggesting a robust estimate. Following transition to symptomatic HF, the 86 transitioners were compared with 607 contemporaneous community-referred patients with newly diagnosed HF (Table 1). Although transitioners had higher prevalence of diabetes, hypertension and vascular disease, they were younger, and had lower heart rates, systolic blood pressure and BNP than community-referred patients at time of HF diagnosis. Atrial fibrillation and chronic obstructive pulmonary disease (COPD) were more prevalent in the community-referred cohort, reflecting its older, multi-morbid profile, with evidence of higher cardiovascular risk on presentation. This report provides original data on the incidence of transition to new-onset HF within the STOP-HF prevention programme, and establishes a baseline phenotype of patients transitioning to HF in this setting, with baseline BNP emerging as the strongest predictor of transition, confirming previous results.6, 7 A novel finding is the potential of change in BNP over time to further refine risk prediction. Similar observations have been made regarding BNP,8 but not in a heightened-risk cohort such as STOP-HF. Although further work is needed to understand the prognostic importance of temporal changes in BNP in these patients, it may allow HF risk to be tracked, prognosis to be refined, and the impact of therapeutic strategies to be assessed over time. Whilst elevated natriuretic peptides correlate with prognosis in both HF with reduced and preserved ejection fraction, natriuretic peptide levels are lower and can be normal in HFpEF, reflecting, among other issues, the complicating impact of co-morbidity in this phenotype, in particular the impact of obesity in lowering natriuretic peptides. Therefore, combining natriuretic peptides, left atrial volume index and, indeed, change in natriuretic peptides over time, may be of particular value in this setting. Furthermore, the data suggest that diagnosis of HF in the STOP-HF programme occurs at an earlier stage in the natural history of the syndrome, with less atrial fibrillation, less COPD and lower BNP than community-referred, newly diagnosed, HF patients. Delayed onward-referral from the community may be explained, in part, by the subtle, non-specific nature of presenting symptoms. Earlier HF diagnosis within a prevention programme provides opportunities to improve outcome using self-care advice and disease-modifying therapies where applicable. These findings underline the role of HF prevention programmes, such as STOP-HF, and may have particular relevance to primary care, where the vast majority of at-risk patients are managed. Traditional HF risk profiling in the community involves large numbers of patients, the majority of whom are at low transition risk. Characterization of the baseline phenotype of transitioners, and the observation that baseline BNP is the strongest independent predictor of transition risk and tracks risk over time, may enable community physicians to better define risk through serial BNP measurement in at-risk patients. Coupled with enhanced access to specialist cardiology care for patients at high transition risk, this may provide a platform for cost-effective and efficacious HF prevention efforts. Conflict of interest: none declared.