European heart journal. Digital health最新文献

Aims: Artificial Intelligence (AI) models applied to standard 12-lead ECGs enable estimation of biological age (AI-ECG age), which has shown prognostic value in general populations. However, its clinical utility in high-risk patients with cardiovascular disease (CVD) or acute medical conditions remains insufficiently explored.

Methods and results: We analysed the first ECG of 48 950 consecutive patients presenting to a tertiary care centre with CVD or acute illness between 2000 and 2021. AI-ECG age was derived using a validated deep learning model. Δ-age, defined as the difference between AI-ECG and chronological age, was analysed categorically (±8 years) and continuously using multivariable Cox models adjusted for clinical and ECG variables. Primary endpoint was long-term total mortality (up to 10 years). Saliency map analysis was performed to identify input regions that the model was most sensitive to. AI-ECG age correlated strongly with chronological age (r = 0.72, P < 0.001), though this correlation weakened in patients with multiple comorbidities. Patients with a positive Δ-age (≥+8 years) had significantly higher 10 year mortality risk (HR: 1.45, P < 0.001), while those with a negative Δ-age (≤-8 years) had lower risk (HR: 0.88, P < 0.001). These associations were consistent across care settings and remained robust when Δ-age was analysed continuously. Saliency maps indicated that the AI model was most sensitive to the P-wave.

Conclusion: AI-ECG age is a strong and independent predictor of long-term mortality in cardiovascular and acute care patients.

Aims: Cardiovascular disease is the leading global cause of mortality. Traditional face-to-face cardiovascular care, while effective, poses challenges such as travel burdens and accessibility issues. Video consultations offer a modern solution, improving access and efficiency while reducing patient strain. This study investigates patient acceptance of video consultations in cardiovascular care using a survey-based approach, assessing key factors influencing their integration into routine practice.

Methods and results: A cross-sectional study including patients attending a cardiological university hospital was conducted. Acceptance of video consultations and its associated factors were assessed using a modified assessment instrument based on the unified theory of acceptance and use of technology. The study comprised 337 participants (M = 61.13 years, SD = 14.54), 54.6% male. Acceptance was moderate (M = 2.88, SD = 1.37), with 30.27% of the participants reporting high acceptance, 28.19% reporting moderate acceptance, and 41.54% low acceptance. Only 3% had used video consultations before. eHealth literacy was high, while digital confidence was moderate. Analysis showed that higher education, digital confidence, and eHealth literacy predicted greater acceptance of video consultations. Effort expectancy, performance expectancy (PE), and social influence (SI) accounted for most of the variance in acceptance (R ² = 0.724).

Conclusion: We identified moderate acceptance of video consultations in cardiology, with education, digital confidence, eHealth literacy, and PE as key factors associated with acceptance. Despite low prior use, perceived ease of use and SI were most strongly associated with acceptance. Addressing technical concerns and promoting digital literacy may enhance adoption, improving access to remote cardiac care.

Aims: Timely and accurate detection of pericardial effusion and assessment of cardiac tamponade remain challenging and highly operator dependent.

Objectives: Artificial intelligence has advanced many echocardiographic assessments, and we aimed to develop and validate a deep learning model to automate the assessment of pericardial effusion severity and cardiac tamponade from echocardiogram videos.

Methods and results: We developed a deep learning model (EchoNet-Pericardium) using temporal-spatial convolutional neural networks to automate pericardial effusion severity grading and tamponade detection from echocardiography videos. The model was trained using a retrospective dataset of 1 427 660 videos from 85 380 echocardiograms at Cedars-Sinai Medical Centre (CSMC) to predict PE severity and cardiac tamponade across individual echocardiographic views and an ensemble approach combining predictions from five standard views. External validation was performed on 33 310 videos from 1806 echocardiograms from Stanford Healthcare (SHC). In the held-out CSMC test set, EchoNet-Pericardium achieved an AUC of 0.900 (95% CI: 0.884-0.916) for detecting moderate or larger pericardial effusion, 0.942 (95% CI: 0.917-0.964) for large pericardial effusion, and 0.955 (95% CI: 0.939-0.968) for cardiac tamponade. In the SHC external validation cohort, the model achieved AUCs of 0.869 (95% CI: 0.794-0.933) for moderate or larger pericardial effusion, 0.959 (95% CI: 0.945-0.972) for large pericardial effusion, and 0.966 (95% CI: 0.906-0.995) for cardiac tamponade. Subgroup analysis demonstrated consistent performance across ages, sexes, left ventricular ejection fraction, and atrial fibrillation statuses.

Conclusion: Our deep learning-based framework accurately grades pericardial effusion severity and detects cardiac tamponade from echocardiograms, demonstrating consistent performance and generalizability across different cohorts. This automated tool has the potential to enhance clinical decision-making by reducing operator dependence and expediting diagnosis.

Aims: The aim of the current study was to assess the utility of a state-of-the-art large language model (LLM) based on curated, defined clinical practice recommendations to support clinicians in obtaining point-of-care guidelines for individual patient treatment while maintaining transparency.

Methods and results: We combined cloud-based and locally run LLMs with versatile open-source tools to form a multi-query, multimodal, retrieval-augmented generation chain that closely reflects European cardiology guidelines in its responses. We compared the performance of this generation chain to other LLMs including GPT-3.5 and GPT-4 on a 306-question multiple-choice exam with questions consisting of short patient vignettes from various cardiology subspecialties, originally written to prepare candidates for the European Exam in Core Cardiology. On the multiple-choice test, our system demonstrated overall accuracy of 73.53%, while GPT-3.5 and GPT-4 had overall accuracies of 44.03 and 62.26%, respectively. Our system outperformed GPT-3.5 and GPT-4 for the following categories of questions: coronary artery disease, arrhythmia, other, valvular heart disease, cardiomyopathies, endocarditis, adult congenital heart disease, pericardial disease, cardio-oncology, pulmonary hypertension, and non-cardiac surgery. For maximum transparency, the system also provided reference quotes for its recommendations.

Conclusion: Our system demonstrated superior performance in question-answering tasks on a set of core cardiology questions as compared with contemporary publicly available chat models. The current study illustrates that LLMs can be tailored to provide documented and accountable guideline recommendations towards actual clinical needs while ensuring recommendations are derived from up-to-date, trustable, and traceable documents.

Background: Echocardiographic image data accumulating in echo labs are a highly valuable but underutilized resource for cardiac imaging research. Despite the availability of large image databases, quantitative measurements required for clinical analysis and research remain limited. Retrospective manual measurements are highly time-consuming and susceptible to operator-related variability. Moreover, data curation and quality control metrics are needed to prepare real-world data for analysis.

Methods: Deep learning-based image analysis can provide fully automated, rapid, and consistent extraction of measurements, given that the data have been properly curated. In this work, we develop an automated pipeline for data curation of a large echo database of 14 326 exams from 9678 patients and evaluate automated measurements of left ventricular ejection fraction (LVEF) and left atrial volume index (LAVI) as a use case.

Results: In validation subsample of 1763 subjects with varying image quality and cardiac diseases and 1488 healthy subjects, the pipeline output was compared with manual measurements. Bland-Altman analysis revealed a bias [standard deviation (SD)] of -1.8% (7.6%) for LVEF and 3.3 mL/m² (8.1 mL/m²) for LAVI and demonstrated robust performance for varying image quality and pathological conditions. Additionally, in the large part of the database of 9678 exams without clinical measurements, the automated data curation and measurement quality control resulted in 79% measured data with high confidence.

Conclusion: This work highlights the potential of deep learning-based automated measurements in echocardiography for data mining in large real-world databases, paving the way for advancements in cardiac imaging research and diagnostics.

Aims: Heart failure with preserved ejection fraction (HFpEF) accounts for approximately half of all heart failure cases, with high levels of morbidity and mortality. However, many patients who meet diagnostic criteria for HFpEF do not have a documented diagnosis, particularly in non-White populations where conventional risk scores may underestimate risk. Our aim was to develop and validate a diagnostic prediction model to detect HFpEF based on ESC criteria, AIM-HFpEF.

Methods and results: We applied natural language processing (NLP) and machine learning methods to routinely collected electronic health record (EHR) data from a tertiary centre hospital trust in London, UK, to derive the AIM-HFpEF model. We then externally validated the model and performed benchmarking against existing HFpEF prediction models (H2FPEF and HFpEF-ABA) for diagnostic power on the entire external cohort and in patients of non-White ethnicity and patients from areas of increased socioeconomic deprivation. An XGBoost model combining demographic, clinical, and echocardiogram data showed strong diagnostic performance in the derivation dataset [n = 3173, AUC = 0.88, (95% CI, 0.85-0.91)] and validation cohort [n = 5383, AUC: 0.88 (95% CI, 0.86-0.90)]. Diagnostic performance was maintained in patients of non-White ethnicity [AUC = 0.89 (95% CI, 0.85-0.93)] and patients from areas of high socioeconomic deprivation [AUC = 0.90 (95% CI, 0.85-0.95)]. In contrast, AIM-HFpEF demonstrated favourable performance relative to the H2FPEF and HFpEF-ABA models. AIM-HFpEF model probabilities were associated with an increased risk of death, hospitalization, and stroke in the external validation cohort (P < 0.001, P = 0.01, P < 0.001, respectively, for highest vs. middle tertile).

Conclusion: AIM-HFpEF represents a validated equitable diagnostic model for HFpEF, which can be embedded within an EHR to allow for fully automated HFpEF detection.

Aims: Patients with primary left ventricular hypertrophy (LVH) often experience a diagnostic delay of several years, largely related to fragmented knowledge among different specialties and the rarity of the conditions. We developed and validated a digital support tool to guide the physician in the differential diagnostic process of patients presenting with primary LVH.

Methods and results: A total of 818 patients with definitive diagnosis of sarcomeric hypertrophic cardiomyopathy (HCM) or one of its phenocopies [479 (62%) males, 48 ± 24 years] were included. Pre-specified disease-specific red flags (RFs) were categorized into five domains: family history, signs/symptoms, electrocardiography, echocardiographic, and laboratory. Each patient's characteristics were inserted by two independent and blind investigators into the app. The diagnostic outcome, based on the presence/absence of RF, was categorized as follows: (i) most likely diagnosis, (ii) possible diagnosis, and (iii) less likely diagnosis. A total of 2979 RFs were identified and non-sarcomeric phenocopies exhibited a higher RF burden than sarcomeric HCM (3.9 vs. 2.7 RFs per patient, P = 0.007), with systemic features and extracardiac findings being strong predictors of non-sarcomeric disease. Thick-Heart App correctly classified 93% of cases into the most likely diagnosis category (sensitivity of 88-100%, specificity 97%). The positive predictive value (PPV) for TTR amyloidosis reached 92%, while Friedrich's ataxia was correctly identified in all cases (PPV = 100%).

Conclusion: The Thick-Heart App correctly classified 93% of cases into the most-likely diagnosis category (sensitivity 88-100%, specificity 97%). Our study underscores the potential clinical value of digital decision support tools to enable timelier identification of specific cardiomyopathies, by promoting awareness in non-reference settings.

Introduction: Integrating machine learning (ML) models into wearable or connected devices to deliver early warning alerts prior to atrial fibrillation (AF) onset may represent an effective preventive strategy. Machine learning algorithms applied to two-lead Holter electrocardiogram (ECG) recordings can support the development of predictive models capable of detecting imminent paroxysmal AF episodes within short-term windows. This approach could facilitate a more targeted 'pill-in-the-pocket' (PITP)-like intervention strategy, potentially enhancing timely therapeutic actions and improving patient outcomes.

Aim: This study aimed to identify patients currently in sinus rhythm who will experience an AF episode within the subsequent hours by analysing 24-h Holter ECG recordings with ML.

Methods: We established a novel database comprising 95 871 manually analysed Holter ECG recordings, identifying 1319 episodes of paroxysmal AF from 872 patients. Among these, 835 AF episodes from 506 recordings had more than 60 min of normal sinus rhythm prior to AF onset and more than 10 min of sustained AF following onset. Patients were stratified into five age groups: all patients combined, under 60 years, 60-70 years, 70-80 years, and over 80 years. Additionally, 365 recordings from 347 patients without rhythm abnormalities were identified and classified, from which two ECG segments were selected. Two deep learning (DL) models were trained on raw ECG data to predict AF onset. To compare DL models with traditional ML approaches using heart rate variability (HRV) parameters, we employed a random forest classifier and a gradient-boosted decision tree model (XGBoost, XGB).

Results: The decision trees models trained on HRV parameters delivered the best predictive performance. The most significant results were observed for episodes lasting more than 5 min of AF, achieving an area under the receiver operating characteristic curve of 0.919 (95% CI: 0.879-0.958) and an area under the precision-recall curve of 0.919 (95% CI: 0.879-0.958) for XGB. At a decision threshold of 0.5, accuracy was 84.5% (81.2-87.8), sensitivity was 83.0% (79.5-86.4), specificity was 86.6% (79.3-93.9), positive predictive value was 90.2% (85.5-94.9), negative predictive value was 78.4% (74.7-82.1), and the F1 score was 86.2% (83.5-89.0).

Conclusion: These findings indicate that HRV parameters contain crucial information for the short-term prediction of AF onset, supporting preventive strategies. Integration of such predictive models into wearable mHealth technologies could facilitate a PITP-like preventive approach, potentially reducing AF-related morbidity. Prospective studies are warranted to validate these promising results further.

Aims: Despite the excellent clinical outcomes of the HeartMate 3 (HM3) left ventricular assist device, the current pump monitoring limits in-depth pump data analysis. This study investigated HM3 pump parameters collected non-invasively with HM3 Snoopy during orthostatic transitions (OTs).

Methods and results: In this single-centre cohort study, a standardized OT protocol was developed, involving postural changes between supine, seated, and standing. Data were recorded using the HM3 Snoopy and a Holter electrocardiogram. Pump flows (Q_MIN, Q_MEAN, Q_MAX), pulsatility index (PI), pump speed, MagLev parameters, and heart rate were synchronized per second. The primary outcome was the identification of distinct orthostatic pump flow response phenotypes. Further, a binary classifier using MagLev parameters, to differentiate between supine and upright patient positions, was developed and assessed. In 25 HM3 patients (age: 61.2 ± 9.6 years, female: 12%, body mass index: 26.8 ± 4.7 kg/m²), greater flow alterations were observed during transitions from supine to standing vs. seated to standing, with most significant changes in Q_MIN [3 (-13; 10)%]. Phenotypes were identified across 75 OTs as no flow response (60%), undesired unloading with a loss in Q_MIN ≥ 50% (20%), and loss of pulsatility ≥ 50% (16%). The MagLev patient position classifier achieved a median sensitivity of 88% and specificity of 86% across the entire cohort.

Conclusion: Three HM3 pump flow response phenotypes were identified in response to OTs, challenging the utilization of PI events to detect undesired unloading events. A MagLev-based position classifier revealed potential for differentiation of HM3 patient position.