Daily Cardiology Research Analysis
Automation and AI are rapidly reshaping cardiology. A prospective, blinded study showed robotic process automation outperformed manual screening to identify patients needing perioperative myocardial injury surveillance while cutting costs. Deep learning on ECGs identified undiagnosed coronary artery disease and predicted adverse outcomes, and a large LAAO registry suggested single antiplatelet therapy may be a safe, bleeding-sparing alternative to dual therapy after Amulet implantation.
Summary
Automation and AI are rapidly reshaping cardiology. A prospective, blinded study showed robotic process automation outperformed manual screening to identify patients needing perioperative myocardial injury surveillance while cutting costs. Deep learning on ECGs identified undiagnosed coronary artery disease and predicted adverse outcomes, and a large LAAO registry suggested single antiplatelet therapy may be a safe, bleeding-sparing alternative to dual therapy after Amulet implantation.
Research Themes
- Automation/AI for cardiovascular risk identification
- Post-LAAO antithrombotic strategy optimization
- ECG-based deep learning for CAD detection and prognosis
Selected Articles
1. Robotic process automation to identify patients at high risk for perioperative myocardial infarction or injury: a prospective, blinded, paired reader-controlled single-centre study.
In a prospective, blinded, paired-reader study, robotic process automation outperformed manual screening to identify patients requiring perioperative myocardial infarction/injury surveillance, with higher sensitivity (0.97 vs 0.82) and markedly lower annual costs (−81%). Specificity remained high for both approaches, and the number needed to screen to gain one additional true positive with RPA was 6.
Impact: This pragmatic automation trial directly addresses a major implementation barrier to guideline-recommended perioperative myocardial injury surveillance by replacing manual screening with a scalable, higher-performing, lower-cost RPA workflow.
Clinical Implications: Hospitals can deploy RPA-driven screening to more reliably flag high-risk surgical patients for perioperative troponin monitoring and management pathways, improving compliance with recommendations while reducing personnel costs.
Key Findings
- Among 660 participants, 12% met criteria for surveillance; RPA identified 75/77 (97%) true positives vs 63/77 (82%) for manual screening.
- Relative true positive fraction favored RPA (1.19; 95% CI 1.08–1.32; P=0.004); number needed to screen to gain one additional true positive was 6.
- Sensitivity: RPA 0.97 vs manual 0.82; specificity high for both (0.98 vs 1.00).
- Estimated annual screening cost was 81% lower with RPA compared with manual screening.
Methodological Strengths
- Prospective, blinded, paired reader-controlled design with independent adjudication.
- Direct cost evaluation demonstrating substantial resource savings.
Limitations
- Single-centre study may limit generalizability.
- Did not measure downstream patient outcomes from enhanced surveillance.
Future Directions: Multicentre implementation studies assessing clinical outcomes (MI detection rates, complications, length of stay) and integration with EHRs; evaluation of model drift and governance for automated screening.
BACKGROUND: Although current guidelines recommend active surveillance for perioperative myocardial infarction, injury, or both in high-risk patients, implementation remains limited in most institutions worldwide because of a lack of resources. METHODS: We hypothesised that robotic process automation (RPA), a software technology that enables virtual bots to replicate human tasks within digital systems, could accurately replace experienced clinical staff. Manual screening by experienced clinical staff and RPA screening were carried out simultaneously and blinded to identify high-risk patients eligible for active surveillance for myocardial infarction/injury, according to predefined screening criteria. Discrepant identification was reviewed by an independent clinician blinded to the origin of the identification, generating a reference standard classification of paired reader-controlled patients to investigate the primary diagnostic endpoint: relative true positive fraction. RESULTS: In 660 participants (median age 60 yr, interquartile range 42-73 yr, 54.8% female), 77/660 (12%) were eligible for active surveillance for perioperative myocardial infarction or injury according to the reference standard classification. RPA screening achieved 75 (97%) true positive identifications, compared with 63 (82%) identified from manual screening (relative true positive fraction: 1.19, 95% confidence interval 1.08-1.32, P=0.004). The number needed to screen to identify one additional true positive using RPA screening was 6. RPA screening had a sensitivity of 0.97 (0.91-0.99), compared with 0.82 (0.72-0.89) for. Both approaches had high specificity (RPA screening: 0.98 [0.97-0.99], compared with manual screening: 1.0 [0.99-1.00]). The estimated annual cost of RPA screening was 81% lower compared with manual screening. CONCLUSIONS: RPA screening was superior to standard-of-care manual screening by experienced clinical staff in identifying patients at high risk for perioperative myocardial infarction or injury. CLINICAL TRIAL REGISTRATION: NCT02573532.
2. Electrocardiogram-Based Artificial Intelligence to Identify Coronary Artery Disease.
A deep learning model trained on 764,670 ECGs discriminated prevalent CAD across 3 cohorts (AUROC ~0.75–0.78) and provided incremental value beyond age/sex and Pooled Cohort Equations. In primary care, the highest risk quintile carried markedly higher hazards of MI, HF, and all-cause mortality, supporting ECG-based opportunistic screening for CAD and risk stratification.
Impact: This work operationalizes ECG-based AI for CAD detection with large-scale external validation and prognostic linkage, enabling practical, low-cost risk identification in primary care.
Clinical Implications: ECG2CAD could flag high-risk individuals for targeted diagnostic testing (e.g., coronary CTA) and aggressive preventive therapy, improving detection of silent CAD and guiding resource allocation.
Key Findings
- Discrimination of prevalent CAD across MGH, BWH, and UK Biobank: AUROC 0.782, 0.747, and 0.760; AUPRC 0.639, 0.588, and 0.155, respectively.
- Incremental performance beyond age/sex and Pooled Cohort Equations (P < 0.01) in MGH and BWH.
- Top ECG2CAD risk quintile associated with substantially higher hazards: MI HR 5.59, HF HR 10.49, all-cause mortality HR 2.68.
- Performance consistent across diverse primary care subgroups.
Methodological Strengths
- Very large training set with multi-cohort external validation.
- Prognostic linkage to incident MI, HF, and mortality, and benchmarking against guideline risk tools.
Limitations
- CAD labels derived from diagnostic codes may introduce misclassification.
- Retrospective design; prospective impact on clinical pathways not yet tested.
Future Directions: Prospective trials to assess clinical utility (changes in testing, therapy, and outcomes), calibration across devices/vendors, and integration with multimodal imaging biomarkers.
BACKGROUND: Coronary artery disease (CAD) results in substantial morbidity and mortality. OBJECTIVES: The purpose of this study was to develop a deep learning model to detect CAD defined using diagnostic codes ("ECG2CAD") and identify people at risk for adverse events using electrocardiograms (ECGs) in a primary care setting. METHODS: ECG2CAD was trained on 764,670 ECGs representing 137,199 individuals at Massachusetts General Hospital (MGH). Model performance for discrimination of prevalent CAD was measured using area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC), and compared against model of age and sex, and Pooled Cohort Equations, in 3 test sets: MGH, Brigham and Women's Hospital (BWH), and UK Biobank. Subgroups were assessed for incident CAD-related events in a BWH primary care cohort. RESULTS: ECG2CAD was evaluated in MGH (N = 18,706 [6,051 cases], age 57 ± 16 years), BWH (N = 88,270 [27,898 cases], age 57 ± 16 years), and UK Biobank (N = 42,147 [1,509 cases], age 65 ± 8 years). ECG2CAD consistently discriminated prevalent CAD (MGH AUROC: 0.782; AUPRC: 0.639; BWH: AUROC: 0.747; AUPRC: 0.588; UK Biobank AUROC: 0.760; AUPRC: 0.155) and incrementally vs models based on age and sex or Pooled Cohort Equations (P < 0.01) in MGH and BWH. In the BWH primary care subset, model performance was consistent across subgroups. Being in the highest quintile of ECG2CAD risk was associated with higher risk for adverse events compared with low-risk group (myocardial infarction HR: 5.59; 95% CI: 4.76-6.56, heart failure 10.49; 95% CI: 7.96-13.84, all-cause mortality 2.68; 95% CI: 2.32-3.10). CONCLUSIONS: Artificial intelligence-enabled analysis of the ECG may facilitate identification of individuals with possible undiagnosed CAD and inform downstream testing and preventive measures.
3. Outcomes for single antiplatelet, dual antiplatelet, or oral anticoagulation after Amulet: Insights from EMERGE LAA post-approval study.
In 11,445 Amulet LAAO patients, adjusted 6-month safety and effectiveness outcomes did not differ between DAPT, SAPT, and OAC. Despite higher baseline bleeding risk, SAPT showed numerically lower bleeding than DAPT, with similar device-related thrombus rates and >95% clinically relevant closure across groups.
Impact: These real-world data support SAPT as a reasonable alternative to DAPT after Amulet implantation, potentially reducing bleeding without compromising effectiveness, informing contemporary antithrombotic strategies in AF patients undergoing LAAO.
Clinical Implications: For high bleeding risk patients after Amulet LAAO, SAPT may be considered instead of routine DAPT, with careful individualized assessment until randomized data are available.
Key Findings
- Among 11,445 cases, discharge regimens were DAPT 81.7%, SAPT 5.3%, and OAC 13.0%.
-
95% achieved clinically relevant closure (≤3 mm peri-device leak) at 45 days across all groups.
- At 6 months, adjusted analyses showed no significant differences in safety or effectiveness endpoints among DAPT, SAPT, and OAC.
- SAPT had numerically lower bleeding than DAPT (3.9% vs 4.8%) despite higher baseline bleeding risk; device-related thrombus rates were identical (0.8%).
Methodological Strengths
- Large national registry with contemporary real-world practice and adjudicated endpoints over 6 months.
- Risk-adjusted analyses comparing three clinically relevant antithrombotic strategies.
Limitations
- Nonrandomized observational design with potential residual confounding and treatment selection bias.
- Follow-up limited to 6 months; longer-term thromboembolic and bleeding outcomes are unknown.
Future Directions: Randomized trials directly comparing SAPT vs DAPT post-Amulet, extended follow-up for late device-related thrombosis and bleeding, and subgroup analyses by bleeding/thrombotic risk.
BACKGROUND: Outcomes associated with different antithrombotic strategies after Amulet left atrial appendage occlusion (LAAO) are not well described. OBJECTIVE: This analysis compared outcomes from patients discharged on dual antiplatelet therapy (DAPT) versus single antiplatelet therapy (SAPT) or oral anticoagulation (OAC) following Amulet implant in the EMERGE LAA post-approval study. METHODS: Patients with a successful Amulet implant and discharged from the hospital between August 14, 2021 and December 15, 2023 and entered into the National Cardiovascular Data Registry (NCDR) LAAO Registry were included. A safety endpoint of all-cause death, stroke, major bleeding, or systemic embolism and effectiveness endpoint of ischemic stroke, systemic embolism, or cardiovascular death were assessed through 6 months as well as major adverse events. RESULTS: A total of 11,445 patients were included in this analysis with 9355 discharged on DAPT (81.7 %), 606 on SAPT (5.3 %), and 1484 on OAC (13.0 %). Patients in the SAPT group had more comorbid conditions and were at the greatest bleeding risk pre-Amulet implant. At 45-days, clinically relevant closure (≤3 mm peri-device leak) was achieved in >95 % of all patients. At 6 months, the safety endpoint rates were 8.8 %, 7.0 %, and 7.0 % in the DAPT, SAPT, and OAC groups, respectively (p = 0.045) and effectiveness endpoint rates were 2.1 %, 1.6 %, and 1.7 % in the DAPT, SAPT, and OAC groups, respectively (p = 0.511). Despite higher baseline bleeding risk, the SAPT group had numerically lower bleeding rates than the DAPT group through 6 months (DAPT 4.8 % vs. SAPT 3.9 %; HR 1.23 [0.78, 1.95]) with no difference in device-related thrombus rates (DAPT 0.8 % vs. 0.8 % SAPT; HR 0.91 [0.33, 2.50]). However, after adjusting for differences in baseline characteristics, no significant differences (p > 0.05) were noted for any clinical events between the three medication groups at 45 days or 6 months. CONCLUSION: In the present cohort of patients treated with Amulet LAAO there were high rates of LAA complete closure, and SAPT emerged as a viable alternative to the current DAPT regimen potentially reducing the risk of bleeding complications in patients at high risk without compromising effectiveness.