Daily Cardiology Research Analysis
Three impactful cardiology studies stood out today: a deep learning model accurately detected past and near-future bradyarrhythmias from 24-hour ambulatory ECG segments; a PROSPERO-registered meta-analysis showed SGLT2 inhibitors produce cardiac reverse remodeling on CMR; and a five-cohort analysis confirmed that adding depression, anxiety, and insomnia to the SMART2 score does not improve recurrent ASCVD risk prediction.
Summary
Three impactful cardiology studies stood out today: a deep learning model accurately detected past and near-future bradyarrhythmias from 24-hour ambulatory ECG segments; a PROSPERO-registered meta-analysis showed SGLT2 inhibitors produce cardiac reverse remodeling on CMR; and a five-cohort analysis confirmed that adding depression, anxiety, and insomnia to the SMART2 score does not improve recurrent ASCVD risk prediction.
Research Themes
- AI-enabled arrhythmia detection from ambulatory ECG
- Heart failure therapeutics and cardiac reverse remodeling on CMR
- Risk prediction in secondary prevention and the role of psychological factors
Selected Articles
1. Deep Learning Can Unmask Conduction Tissue Disease From an Ambulatory ECG.
Using 320,959 ambulatory ECG recordings, a deep learning model detected prior bradyarrhythmias from the last 24 hours of clean ECG with AUCs up to 0.93 and NPVs up to 99.9%, and also predicted events in the next 13 days (AUC 0.88). This approach could triage patients with intermittent conduction disease more efficiently than standard monitoring.
Impact: This is a large, externally validated AI study that repurposes routine ambulatory ECG to reveal otherwise hidden bradyarrhythmias and forecast near-term risk.
Clinical Implications: Could reduce time to diagnosis and guide earlier pacing or monitoring decisions in patients with syncope or intermittent palpitations using existing single-lead ECG workflows.
Key Findings
- External validation AUCs: 0.89 (daytime sinus pause ≥3 s), 0.87 (anytime sinus pause ≥6 s), 0.93 (complete heart block), 0.89 (composite).
- Negative predictive values ranged from 97.9% to 99.9% across bradyarrhythmia endpoints.
- Predictive modeling using the first 24 hours forecasted bradyarrhythmias over the next 13 days with AUC 0.88.
Methodological Strengths
- Very large, unselected real-world dataset with internal and external validation.
- Clear pre-specified endpoints and robust performance on both detection and short-term prediction.
Limitations
- Retrospective design without prospective clinical outcomes testing.
- Single-lead ECG limits morphological information; generalizability to other devices/settings requires validation.
Future Directions: Prospective clinical trials to assess impact on diagnostic yield, time-to-pacing, and outcomes; integration with clinical workflows and cost-effectiveness analyses.
BACKGROUND: Bradyarrhythmia is a common and potentially serious cause of syncope, often difficult to detect due to its intermittent nature. Traditional ECG monitoring methods either provide low diagnostic accuracy or delay diagnosis, increasing the risk of recurrence. We hypothesized that a deep learning-enabled, 24-hour, single-lead ECG could detect past episodes of bradyarrhythmia. METHODS: Using unselected 14-day single-lead ambulatory ECG recordings, we developed a deep learning model to identify patients with prior asystole from sinus arrest or complete heart block. The model was trained using the last 24 hours of each recording, free of bradyarrhythmias, to identify daytime sinus pause of ≥3 s, anytime sinus pause of ≥6 s, complete heart block, or a composite of these bradyarrhythmias from the previous 13 days. RESULTS: A total of 320 959 unselected 14-day ambulatory ECG recordings (mean age, 60.5±17.8 years; 60% female) were split into training (n=189 414), tuning (n=45 982), internal validation (n=43 390), and external validation (n=42 173) sets. External validation of prior daytime sinus pause ≥3 s, anytime sinus pause ≥6 s, complete heart block, and a composite end point demonstrated an area under the receiver operating characteristic curve of 0.89, 0.87, 0.93, and 0.89, respectively, with negative predictive values between 97.9 and 99.9%. In addition to this approach of uncovering past events, our model was also tested for its ability to predict bradyarrhythmias within the following 13 days using the first 24 hours of ECG data, achieving an AUC of 0.88 for the composite end point. CONCLUSIONS: A deep learning-enabled ambulatory ECG is capable of unmasking underlying conduction tissue system disease. This tool may help identify patients with significant intermittent bradyarrhythmia, potentially improving timely diagnosis and management.
2. Effect of SGLT2 inhibitors on cardiac structure and function assessed by cardiac magnetic resonance: a systematic review and meta-analysis.
Across 23 studies (n=1008), SGLT2 inhibitors reduced LV end-diastolic volume by 7 mL, LV mass by 4 g, and epicardial adipose tissue by ~5 mL on CMR; stroke volume improved in reduced EF subgroups. Findings support a structural basis for clinical benefits across age, sex, and diabetes status.
Impact: Provides quantitative, imaging-based evidence of reverse remodeling with SGLT2 inhibitors, strengthening mechanistic plausibility for their benefits in heart failure.
Clinical Implications: Supports SGLT2i use across HF phenotypes with objective CMR markers of benefit; imaging endpoints may guide patient selection and monitoring strategies.
Key Findings
- LV end-diastolic volume decreased by −7.10 mL (95% CI −13.01 to −1.19).
- LV mass decreased by −4.24 g (95% CI −7.88 to −0.60).
- Epicardial adipose tissue decreased by −4.94 mL (95% CI −9.06 to −0.82).
- Subgroup with reduced LVEF showed improved LV stroke volume; age, sex, and diabetes prevalence did not modify effects.
Methodological Strengths
- PROSPERO-registered systematic review with random-effects pooling and meta-regression.
- Broad CMR endpoints including volumes, mass, and tissue characterization across 23 studies.
Limitations
- Heterogeneity in study designs and populations; many included studies are nonrandomized or small.
- Imaging surrogates rather than hard clinical outcomes; follow-up durations vary.
Future Directions: Prospective CMR-guided trials to link remodeling magnitude to clinical endpoints and to identify responders across HF phenotypes.
BACKGROUND AND AIM: Sodium-glucose cotransporter-2 inhibitors (SGLT2i) improve outcomes in patients with heart failure (HF) but underlying mechanisms remain incompletely understood. Cardiac magnetic resonance (CMR) is key in evaluating cardiac structure and function, enabling accurate assessment of reverse remodeling. Aim of this systematic review and meta-analysis was to assess the effects of SGLT2i on cardiac remodeling evaluated by CMR changes. METHODS: We conducted a systematic review and meta-analysis of studies assessing changes in CMR parameters in patients treated with SGLT2i (PROSPERO registration: CRD42024574302). Databases were searched through April 30, 2025. Random-effects models were used to pool mean changes in left and right ventricular volumes, mass, function, stroke volume, global longitudinal strain, left atrial volume, and tissue characterization indices. Meta-regression and sensitivity analyses were performed to evaluate potential sources of heterogeneity. RESULTS: Twenty-three studies and 1008 patients were included. Treatment with SGLT2i was associated with significant reductions in left ventricular (LV) end-diastolic volume (- 7.10 mL; 95% CI: -13.01 to - 1.19, p = 0.023), left ventricular mass (- 4.24 g; 95% CI: -7.88 to - 0.60, p = 0.027) and epicardial adipose tissue (-4.94 ml; 95% CI: -9.06, -0.82, p = 0.019). A subgroup analysis in patients with reduced LV ejection fraction showed improvement in LV stroke volume. Meta-regression revealed no significant effect of age, male sex or diabetes prevalence on pooled estimates. CONCLUSIONS: SGLT2i are associated with reductions in LV volumes and mass in line with an overall favorable reverse remodeling effects as assessed by CMR.
3. Psychological factors beyond the SMART2 model for predicting recurrent events in atherosclerotic cardiovascular disease patients.
In 20,050 patients across five European cohorts, adding depression, anxiety, or insomnia (symptoms, diagnoses, or medications) to SMART2 did not improve recurrent event prediction (ΔC-stat −0.0003 to 0.0011) or calibration. SMART2 retains adequate performance in patients with psychological conditions.
Impact: Clarifies that routine inclusion of psychological factors does not enhance established secondary prevention risk models, reducing model complexity and preventing overfitting.
Clinical Implications: SMART2 can be used without adjustment for depression, anxiety, or insomnia in secondary prevention; resource allocation should prioritize modifiable cardiometabolic risk factors.
Key Findings
- Among 20,050 patients, 2,987 recurrent CV events occurred; psychological factors were common by self-report but uncommon by diagnosis codes.
- No significant incremental predictive value beyond SMART2; ΔC-statistic ranged from −0.0003 (depression symptoms) to 0.0011 (anxiolytic use).
- Model calibration and discrimination remained adequate (C-statistics 0.61–0.70) among patients with psychological factors.
Methodological Strengths
- Large, multicohort analysis using Fine–Gray competing risk models with SMART2 coefficients as offset.
- Comprehensive assessment across symptoms, diagnoses, and treatments with calibration checks.
Limitations
- Psychological factors were self-reported in part; residual confounding and misclassification are possible.
- European cohorts may limit generalizability to other populations and care systems.
Future Directions: Explore dynamic risk updating, causal mediation, and targeted interventions for patients with psychological comorbidity outside risk-score augmentation.
BACKGROUND: Guidelines recommend using cardiovascular (CV) risk prediction models to support treatment decisions. Psychological factors such as depression, anxiety, and insomnia may affect CV risk. AIM: Evaluating the added value of psychological factors on top of the SMART2 model for estimating 10-year risk of recurrent CV events in patients with atherosclerotic CV disease. METHODS: Patients aged 40-80 with atherosclerotic CV disease were included from the UCC-SMART, HUNT3, HUNT4, NORCOR, and Nor-COAST cohorts. Potential added value of (self-reported) symptoms, diagnoses, and treatment for depression, anxiety, and insomnia on top of the SMART2 was investigated using Fine and Gray models, with SMART2 coefficients as an offset. Added value was assessed by change (Δ) in C-statistic and calibration. RESULTS: 20,050 patients were eligible, with 2,987 experiencing recurrent CV events. Psychological factors were present in 3-9%, 3-45% and 10-14% of patients for diagnosis, self-reported symptoms and prescribed treatment, respectively. No psychological factors were significantly associated with recurrent CV events on top of SMART2 and performance in patients with psychological factors was adequate (C-statistics 0.61 to 0.70). Added value of psychological factors was minimal, with the ΔC-statistic ranging from -0.0003 [95% confidence interval: -0.0005-0.0001] for symptoms of depression, to 0.0011 [0.0011-0.0011] for prescribed treatment with anxiolytics. There was no relevant difference in calibration. CONCLUSIONS: SMART2 model reliably estimates recurrent CV event risk in patients with psychological factors. Psychological factors have no added value to CV risk prediction when integrated on top of the SMART2 risk score. Models for prediction of cardiovascular disease risk inform treatment decisions but are thought to underestimate risk in patients with depression, anxiety, and insomnia, possibly contributing to undertreatment in these vulnerable patients. In this study involving 20,050 patients from five large European cohorts, the SMART2 model for estimation of risk of cardiovascular disease in patients with established cardiovascular disease was found to accurately estimate risk even in those with depression, anxiety, and insomnia. Extending the SMART2 model with information about diagnosis or symptoms of or prescribed treatment for depression, anxiety, or insomnia did not lead to meaningful improvements of the predictions made by SMART2.These results indicate that the SMART2 model can be used reliably in patients regardless of presence of psychological factors, providing reliable risk assessment for patients with psychological conditions.