Deep Learning-Derived Pericardial Adipose Tissue by Electrocardiogram-Gated Cardiac Computed Tomography Predicts Cardiovascular Events Beyond Coronary Calcium Score

Q: What are the key findings of this study?

Deep learning-derived pericardial adipose tissue volume in the highest tertile (>285 cm3) independently increased the risk of total cardiovascular events by 30% (HR 1.30, p=0.006) beyond CAC score.

Q: What is the clinical evidence from this study?

Study design: Cohort. Population: Primary cardiovascular disease prevention (n=11897). Intervention: Highest tertile of deep learning-derived pericardial adipose tissue (PAT) volume (>285 cm3) vs. Lowest tertile of PAT volume (<182 cm3). Primary outcome: Composite of total cardiovascular disease events (myocardial infarction, ischemic stroke, heart failure, and cardiovascular death) (HR 1.30, 95% CI 1.08-1.58, p=0.006).

Q: What does this research mean for the field?

Deep learning-derived pericardial adipose tissue volume in the highest tertile independently increases the risk of total cardiovascular events by 30% beyond coronary artery calcium scoring and clinical risk equations. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Zahra Esmaeili; Jose R. Medina-Inojosa; Elham Mahmoudi; L M Ortega-Aviles; Betsy Medina-Inojosa; Bradley J. Erickson; Francisco Lopez-Jimenez

doi:10.1016/j.ajpc.2026.101549

This research is Trending!

March 1, 2026American Journal of Preventive CardiologyOpen Access

Deep Learning-Derived Pericardial Adipose Tissue by Electrocardiogram-Gated Cardiac Computed Tomography Predicts Cardiovascular Events Beyond Coronary Calcium Score

View Full Paper

Ask AI

Bookmark

Share

Key result

High pericardial adipose tissue volume is linked to ~30% higher cardiovascular risk beyond CAC score.

HR 1.30
95% CI 1.08-1.58
P=0.006
n=11,897

Figures

Why the study?

To evaluate whether pericardial adipose tissue volume improves prediction of cardiovascular events beyond Coronary Artery Calcium score and the AHA PREVENT equation.

Does automated deep learning-derived pericardial adipose tissue volume from non-contrast ECG-gated cardiac CT improve the prediction of total cardiovascular disease events beyond coronary artery calcium scoring and the AHA PREVENT equation in adults without prior cardiovascular disease?

Key Points

To evaluate whether pericardial adipose tissue (PAT) volume measured via deep learning improves cardiovascular disease event prediction beyond coronary artery calcium (CAC) score and the PREVENT equation.
Analyzed 11,897 Olmsted County, MN, residents without known heart failure or atherosclerotic cardiovascular disease who underwent non-contrast, ECG-gated CAC scans.
Quantified PAT volume and density using an automated deep learning segmentation model (Dice score 0.94).
Evaluated incident total cardiovascular disease (myocardial infarction, ischemic stroke, heart failure, and cardiovascular death) over a median follow-up of 16.4 years using Cox proportional hazards models.
Total cardiovascular events occurred in 9.6% of participants (1142 subjects; 0.66 per 100 person-years), with progressively lower event-free survival across higher PAT volume tertiles (log-rank p < 0.001).
Higher PAT volume conferred incremental risk even among patients with low CAC scores (HR = 1.24, 95% CI: 1.05–1.48) and improved risk stratification within borderline and intermediate PREVENT categories.
Participants in the highest PAT volume tertile (> 285 cm³) exhibited a 30% higher risk of cardiovascular events compared to the lowest tertile (< 182 cm³) after adjusting for risk factors and CAC score (HR = 1.30, 95% CI: 1.08–1.58).

Study Design

Type

Cohort (n=11,897)

Multicenter

Yes

Structured PICO

Does automated deep learning-derived pericardial adipose tissue volume from non-contrast ECG-gated cardiac CT improve the prediction of total cardiovascular disease events beyond coronary artery calcium scoring and the AHA PREVENT equation in adults without prior cardiovascular disease?

P

Population

11,897 adult Olmsted County residents (mean age 54.6 years, 34.4% female) free of known heart failure or atherosclerotic cardiovascular disease and not on statin therapy who underwent non-contrast ECG-gated CAC CT scans, followed for a median of 16.4 years.

E

Exposure

Automated deep learning (DL)-derived measurement of pericardial adipose tissue (PAT) volume (cm³) and density (Hounsfield units, range -190 to -30 HU) from non-contrast, ECG-gated cardiac CT scans, stratified by tertiles (<182, 182–285, >285 cm³) or top quartile (>319 cm³).

C

Comparator

Coronary Artery Calcium (CAC) score (cut-off 100 Agatston units or categories 0, 1–99, 100–299, >300 AU) and the 2023 AHA Predicting Risk of Cardiovascular Disease Events (PREVENT™) 10-year risk equation categories (low <5%, borderline 5–7.4%, intermediate 7.5–19.9%, high ≥20%).

O

Outcome

Composite of total cardiovascular disease (CVD) events—including myocardial infarction, ischemic stroke, heart failure, and cardiovascular death—over a median follow-up of 16.4 years.composite

Main Result

Hazard Ratio: 1.3 (95% CI 1.08–1.58)

Absolute Event Rate: 13.2% vs 6.7%

p-value: p=0.006

Automated deep learning quantification of pericardial adipose tissue volume from routine non-contrast CAC CT scans independently predicts long-term cardiovascular events and improves risk stratification beyond CAC scoring and the AHA PREVENT equation, particularly in borderline- to intermediate-risk and low-CAC patients.

Limitations

Deep learning algorithms are in early stages of development and implementation, limiting generalizability and real-world clinical adoption.
Predominance of White individuals (97%) in Olmsted County, Minnesota limits extrapolation to other racial and ethnic groups.
Comorbidities were gathered using ICD-9 and ICD-10 codes, introducing a possible source of misclassification bias.
The segmentation model captures total pericardial adipose tissue without distinguishing between epicardial and paracardial fat components.
Potential temporal confounding arising from a more than 20-year enrollment window (1997 to 2020), including changes in CT scanner technology.

Key expert perspectives

Captured external expert commentary on this paper, strongest first. Original sources are linked where available.

ZE Zahra EsmaeiliCardiovascular medicine research fellow, Mayo Clinic

“Pericardial fat has been recognized as a marker of cardiovascular risk, but this study shows how we can now measure it automatically and use it to meaningfully improve risk prediction, especially in patients at borderline or intermediate risk where clinical decisions are often less clear.”

Mayo ClinicNews Coverage

FL Francisco Lopez-JimenezPreventive cardiologist, co-director of AI in Cardiovascular Medicine, Mayo Clinic

“Because this measurement comes from imaging that many patients are already receiving, it represents a practical and scalable way to enhance cardiovascular risk assessment. It could help clinicians intervene earlier and more effectively.”

Mayo ClinicNews Coverage

ZE Zahra EsmaeiliCardiovascular medicine research fellow, Mayo Clinic

“When we added PAT volume and density to a model that already included PREVENT risk factors and CAC score, the C-statistic went from 0.72 to 0.76. That's a significant improvement.”

Mayo ClinicNews Coverage

Implication

Higher PAT volume may refine stratification beyond CAC in low-score patients; leaves open prospective validation before routine adoption.

Abstract

Pericardial adipose tissue (PAT) is an emerging imaging biomarker of cardiovascular disease (CVD) risk. We evaluated whether PAT volume improves prediction of cardiovascular events beyond Coronary Artery Calcium (CAC) score and the AHA Predicting Risk of Cardiovascular Disease Events (PREVENT TM ) equation. We analyzed data of 11,897 Olmsted County, MN, residents without known history of heart failure and atherosclerotic cardiovascular disease (ASCVD) events who underwent non-contrast, ECG-gated CAC scans. PAT volume (cm 3 ) and density (Hounsfield units) were measured using a validated automated deep learning (DL) segmentation model (Dice score 0.94). We stratified patients into groups based on PAT volume (cut-off: top quartile 319 cm 3 ) and CAC score (cut-off: 100 Agatston units). Cox proportional hazard models evaluated associations with the composite outcome of total CVD (myocardial infarction, ischemic stroke, heart failure, and cardiovascular death), while adjusting for cardiovascular risk factors, and CAC score across PREVENT TM risk categories. Over a median of 16.4 years, 9.6 % experienced total CVD events. Kaplan-Meier curves demonstrated progressively lower CVD-free survival across PAT and CAC groups (p < 0.001). PAT volume stratified risk across CAC categories, with higher volume conferring incremental risk even at low CAC (HR = 1.24, 95% CI:1.05–1.48), and enhanced risk stratification within borderline and intermediate PREVENT TM categories. PAT volume provides prognostic value for total CVD events beyond PREVENT TM and CAC scoring.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Esmaeili et al. (2026) conducted a cohort in Primary cardiovascular disease prevention (n=11,897). Highest tertile of deep learning-derived pericardial adipose tissue (PAT) volume (>285 cm3) vs. Lowest tertile of PAT volume (<182 cm3) was evaluated on Composite of total cardiovascular disease events (myocardial infarction, ischemic stroke, heart failure, and cardiovascular death) (HR 1.30, 95% CI 1.08-1.58, p=0.006). Deep learning-derived pericardial adipose tissue volume in the highest tertile (>285 cm3) independently increased the risk of total cardiovascular events by 30% (HR 1.30, p=0.006) beyond CAC score.

synapsesocial.com/papers/69c4cc85fdc3bde448917d85 https://doi.org/10.1016/j.ajpc.2026.101549

View Full Paper

Ask AI

Bookmark

Share

Deep Learning-Derived Pericardial Adipose Tissue by Electrocardiogram-Gated Cardiac Computed Tomography Predicts Cardiovascular Events Beyond Coronary Calcium Score

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Key expert perspectives

Implication

Abstract

Discussion

Cite This Study

Deep Learning-Derived Pericardial Adipose Tissue by Electrocardiogram-Gated Cardiac Computed Tomography Predicts Cardiovascular Events Beyond Coronary Calcium Score

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Key expert perspectives

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading