Artificial intelligence (AI)-enhanced ECG (AI-ECG) models are often designed to detect specific anatomical and functional cardiac abnormalities. Understanding the selectivity of their phenotypic associations is essential to inform their clinical use. Here, we sought to assess whether AI-ECG models function as condition-specific classifiers or broader cardiovascular risk markers. We included 4 distinct study populations drawn from both electronic health records and prospective cohort studies. We deployed 6 image-based AI-ECG models: 5 validated models for the detection of left ventricular systolic dysfunction, aortic stenosis, mitral regurgitation, left ventricular hypertrophy, and a composite model for structural heart disease; and 1 negative control AI-ECG model for biological sex. Additionally, we developed 6 experimental models designed to identify noncardiovascular conditions. Diagnosis codes from electronic health records and cohorts were transformed into interpretable phenotypes using a phenome-wide association study framework. We assessed associations of AI-ECG probabilities with cross-sectional phenotypes using logistic regression and with new-onset cardiovascular diseases using Cox regression. Pearson correlation coefficients were calculated to compare phenotypic signatures. The study included one random ECG from 233 689 individuals (mean age 59±18 years, 130 084 [56%] women) across sites. Each of the 5 AI-ECG models for structural and functional cardiac disorders was more likely to be associated with cardiovascular phenotypes compared with other phenotype groups (odds ratios ranging from 2.16 to 4.41, P<10⁻⁶), whereas the sex model did not show a similar pattern. All AI-ECG models were significantly associated with their respective target phenotype but also showed similar or stronger associations with a broad range of other cardiovascular phenotypes. Phenotypic associations were similar across AI-ECG models trained for different conditions, which was not observed in models for noncardiovascular conditions. Correlation of phenotype association patterns between models was high (0.67-0.96). This pattern was consistent across all models and external data sets and in both cross-sectional and prospective analyses. Despite being developed to detect specific cardiovascular conditions, AI-ECG models detect the presence and predict the future development of a broad range of cardiovascular diseases with similar propensity. This challenges their role as binary diagnostic tools and instead supports their use as broader cardiovascular biomarkers.

Use of cytotoxic chemotherapy at end-of-life (EOL) is associated with adverse quality of life, increased health care utilization, and lower hospice rates. Although EOL cytotoxic chemotherapy use has declined in recent years, EOL novel (immunotherapy and targeted therapy) use has increased. The association between use of novel therapies at EOL and health care utilization has not been widely studied. We identified patients within SEER-Medicare who had part D coverage (excluding those with Medicare Advantage) age 66 years and older, and breast, colorectal, lung, prostate, bladder, cervical, kidney, liver, ovarian, pancreatic, melanoma, or uterine cancer. Patients were diagnosed between 2005 and 2019 and died between 2015 and 2020. We analyzed associations between EOL systemic anticancer therapy (SACT) use (overall and by subtype), and health care utilization in the last 30 days of life (emergency department [ED], hospitalization, intensive care unit [ICU], and inpatient death), and hospice with multivariable regression, controlling for sociodemographic and cancer covariates. Of 315,089 beneficiaries, 23,970 (7.6%) received SACT within 30 days of death. The breakdown by type was cytotoxic therapy 50.6%, immunotherapy 20.8%, targeted therapy 18%, and combination therapies 10.6%. After adjusting for covariates, any SACT use at EOL was associated with higher ED use (odds ratio [OR], 3.05 [95% CI, 2.95 to 3.15]), hospital admissions (OR, 2.64 [95% CI, 2.56 to 2.72]), ICU admission (OR, 1.78 [95% CI, 1.72 to 1.83]), hospital death (OR, 2.02 [95% CI, 1.96 to 2.08]), and lower hospice use (OR, 0.51 [95% CI, 0.50 to 0.53]) compared with no SACT. All subtypes of SACT were individually associated with higher health care utilization and lower hospice use (P < .001). All subtypes of SACT use were associated with markers of worse-quality EOL care. These data can inform decisions for current care guidelines and efforts to reduce overutilization.

Coronary artery disease (CAD) is the leading cause of heart failure with reduced ejection fraction (HFrEF). Coronary artery bypass grafting (CABG) improves long-term mortality in HFrEF. Percutaneous coronary intervention (PCI) is often performed as an alternative to CABG in patients at high surgical risk. However, in patients with HFrEF and limited myocardial reserve, PCI may result in haemodynamic instability, increasing risk and precluding optimal revascularisation. Mechanical circulatory support (MCS) during high-risk PCI may enhance haemodynamic stability during the procedure and enable complete revascularisation. We thus performed the PROTECT IV trial to determine whether PCI with routine use of the Impella CP microaxial flow pump improves early and late outcomes in patients with HFrEF and complex CAD compared with PCI with or without use of an intra-aortic balloon pump (IABP). PROTECT IV is a prospective, multicentre, randomised, parallel-controlled, open-label, superiority trial with an adaptive design. Patients with complex CAD and left ventricular ejection fraction ≤40% (n=1,252) deemed at excessive surgical risk for bypass grafting by the Heart Team will be randomised in a 1:1 ratio to PCI with Impella CP versus PCI with or without an IABP. The primary endpoint is the composite of all-cause death, stroke, myocardial infarction, unplanned clinically driven revascularisation, durable left ventricular assist device implant or heart transplant, or other hospitalisation for cardiovascular causes at 3-year follow-up, with at least 1-year follow-up in all patients. Prespecified substudies will evaluate the impact of MCS on renal function, the procedural role of right heart catheterisation, and the utility of myocardial viability assessment. The PROTECT IV trial will determine whether routine MCS with Impella CP during high-risk PCI improves the prognosis of patients with complex CAD and HFrEF.

Background: Artificial intelligence (AI) applied to routine electrocardiograms (ECGs) offers promise for screening of structural heart disease (SHD), yet broad clinical integration remains limited by high false positive rates and the lack of tailored deployment strategies.Methods: We developed TARGET-AI, a multimodal AI-enabled pipeline that integrates longitudinal electronic health record (EHR) data with ECG images to identify optimal intersections of healthcare encounters and patient phenotypes for targeted AI-ECG screening of SHD. The approach is built on (1) a pretrained EHR foundation model (CLMBR-T) applied to 118 million coded events from 159,322 individuals to generate temporal patient embeddings and identify high-risk screening candidates, followed by (2) a novel contrastive vision-language model trained on 754,533 ECG image-echocardiogram report pairs to detect SHD subtypes with tunable performance characteristics. We evaluated this sequential, gated strategy in 5,198 individuals referred for their first transthoracic echocardiogram (TTE) within 90 days of an ECG (temporal validation), as well as in geographically distinct cohorts, including 33,518 UK Biobank participants undergoing protocolized ECG and cardiac magnetic resonance imaging, and a geographically distinct inpatient EHR cohort of 3,628 patients with ECG-TTE pairs (MIMIC-IV). Significance was determined by comparing metric differences between targeted and untargeted strategies, with bootstrap-derived 95% confidence intervals excluding zero considered significant.Results: Our pre-trained AI-ECG image foundation model discriminated 26 SHD subtypes, including left ventricular systolic dysfunction (AUROC of 0.90), severe aortic stenosis (AUROC of 0.85) and elevated right ventricular systolic pressure (AUROC of 0.82). Compared with untargeted AI-ECG screening, targeted screening in the temporal validation set (n=5,198) was associated with a significant increase in F1 scores (median of 0.25 [range: 0.09 to 0.75]) and decrease in false positives (median of -303 [range: -715 to -77]) across 26 SHD labels. Similar increases in F1 scores and reductions in false positives were seen in the UK Biobank (n=33,518; median change in false positives of -819 [range: -3,521 to -459] across 7 SHD labels) and MIMIC-IV (n=3,628; median false positive change of -255 [range: -716 to -86] across 5 SHD labels).Conclusions: TARGET-AI may guide the targeted deployment of AI-ECG for SHD screening by integrating longitudinal EHR phenotypes with multimodal ECG-echocardiogram representations in an interoperable framework, enabling adaptive, data-driven screening strategies across health systems.

Delays in stroke diagnosis contribute to long-term disability. Many patients still face barriers to effective risk factor management, timely detection, and access to poststroke rehabilitation. The emergence of artificial intelligence-enabled, consumer-facing health technologies offers a transformative opportunity to address these gaps across the stroke care continuum. This review examines the evolving role of artificial intelligence-powered devices, including smartwatches, smartphones, wearable sensors, and ambient home-based technologies, in enabling precision stroke care. In stroke prevention, these tools facilitate scalable monitoring of cardiometabolic and stroke-specific risk factors. For early detection, artificial intelligence algorithms applied to multimodal sensor data can identify subtle neurological impairments and support real-time triage. In recovery, artificial intelligence-enhanced remote monitoring and virtual supervision offer scalable models for delivering personalized rehabilitation outside of specialized centers. Although most of these innovations remain in early development, they signal a paradigm shift toward accessible, individualized, and data-driven stroke prevention and management.

Pulmonary arteriovenous malformation (pAVM) is an anatomical intrapulmonary left-to-right vascular shunt, and it may be a cause of paradoxical stroke. However, the risk of stroke associated with pAVMs is unknown. In this study, we seek to quantify the risk of stroke recurrence among cryptogenic stroke (CS) patients with pAVM compared with those with patent foramen ovale (PFO). This was a retrospective analysis of the 2016-2022 Nationwide Readmissions Database. Patients with acute ischemic stroke were identified, and those with embolic, thrombotic, lacunar, or other specified stroke etiologies were excluded to identify patients with CS. Recurrent stroke risk was compared between CS patients with PFO and pAVM using Cox regression models with multivariable adjustments for demographics, stroke severity, and stroke risk factors. A total of 65,611 CS patients with PFO and 471 with pAVM were included. By 300 days, patients with pAVM had a 5.3% risk of recurrent stroke, which was not significantly different from 2.9% among patients with PFO after multivariable adjustments (HR 1.46 [95% CI 0.45-4.67], p = 0.53). Among patients with CS, the presence of a pAVM may be reasonably considered equivalent to a PFO when estimating the risk of stroke recurrence.