The analysis included participants who had undergone CCTA at least twice, with intervals of 2 years or more, between January 2006 and December 2019 at Seoul National University Hospital (SNUH) and SNUH Gangnam Center. During this period, a total of 4,431 participants were extracted from the SNUH Patients Research Environment, which serves as a clinical data warehouse. Of the total participants, 814 individuals with analyzable images were selected for further analysis, while 52 were excluded for the following reasons: inadequate clinical information (n = 21), insufficient image quality (n = 18), history of percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG) (n = 11), and indistinct EAT delineation due to severe motion artifacts or pericardial effects (n = 2). Ultimately, a total of 762 participants were included in the study (Fig. 1).

Baseline and follow-up CCTA images were retrieved from the Picture Archiving and Communication System of SNUH, stored in the Digital Imaging and Communications in Medicine format. These images were then downloaded and transferred to the Department of Radiology at Seoul National University Bundang Hospital, which served as the core lab for EAT volume measurements. Additionally, CCTA images were separately analyzed by two experienced radiology researchers using a dedicated workstation (SyngoVia; Siemens Healthcare, Erlangen, Germany). If discrepancies exceeding 10% of EAT volume were detected, a third measurement was performed. The analysis was conducted independently, without interference from the other researchers involved. All CCTA images were measured at the same cardiac cycle to ensure consistency, as the volume of EAT could vary between systolic and diastolic periods. Experienced researchers manually marked an epicardial line on CCTA images, setting thresholds at -190 Hounsfield unit (HU; lower) and -30 HU (upper). This marking was conducted on axial scans of 0.9 mm thickness, spanning carina to heart base. Moreover, EAT measurements, highlighted in purple, were semi-automated. Researchers checked and adjusted the outlined area for accuracy. Cases with unclear EAT lines caused by severe motion artifacts or pronounced pericardial effects (pericardial effusion exceeding 10 mm) were excluded. Additionally, images with insufficient scan range -starting below the carina level or not reaching the heart base-were also excluded, as they could impact EAT volume measurements. The EAT volume was computed by summing EAT measurements from each scanned image, as determined through the aforementioned methodology. To counter the influence of height and weight variations, EAT was normalized by dividing it by the body surface area (BSA) for further analysis. The annual EAT change was computed as (follow-up EAT - baseline EAT) / baseline EAT × 100 / year. For participants who underwent CCTA imaging more than three times within the study period, the initial and final CCTA images were chosen for analysis. A written signed informed consent was obtained from each patient and the research was carried out with approval from Institutional Review Board of SNUH (IRB no H-2022-174-1107).

Clinical data collection for these participants was manually secured through Electronic Medical Records (EMR). Utilizing EMR at SNUH, various clinical details were gathered, including information from CCTA and laboratory data such as complete blood count (CBC) and medication history. This comprehensive data compilation was aimed at assessing each participant’s cardiovascular risk. Additional medical records were collected, encompassing sex, age, body mass index (BMI), BSA, systolic blood pressure, diastolic blood pressure, pulse rate, and lifestyle information such as smoking and drinking histories. Medical conditions, including diagnosed hypertension and diabetes, were also included. Moreover, participants’ cardiovascular family histories, comprising coronary artery disease, angina, and stroke, were gathered. Cases of hypertension and diabetes were restricted to those diagnosed or under medication while also excluding individuals with a family history of hypertension and diabetes. Smoking history was categorized as never smoked, ex-smoker, or current smoker while drinking history was classified as none, social drinker, or heavy drinker. Measurements for CBC, glycated hemoglobin (HbA1c), total cholesterol, triglyceride (TG), high-density lipoprotein (HDL), and low-density lipoprotein (LDL) were obtained. Data points obtained within a month prior to the timing of the two CCTA scans were chosen. In instances where this data was not available, the previous data was employed using the principle of last observation carried forward. In cases where no data was present, those instances were treated as missing values.

Lastly, a review was conducted to determine whether coronary artery disease had manifested during the follow-up period. CVD events and new cases of coronary artery disease were identified based on the presence of 50% stenosis in the CCTA findings at the follow-up assessment or if coronary artery lesions exhibited progression during the follow-up period and necessitated either PCI or CABG.

Data are presented as mean ± standard deviation, or a number with percentages, or as an odds ratio (OR) with a 95% confidence interval (CI). The Agatston score for calcium was categorized as follows: 0 HU, 0.1–100 HU, 100.1–400 HU, and > 400 HU [26]. The atherosclerotic cardiovascular disease (ASCVD) 10-year risk score percentages were classified into the following categories: 0–5%, 5–7.5%, 7.5–20%, and > 20% [27].

To analyze the relationship between EAT (divided into quartiles) and risk factors, t-tests and Mann–Whitney U-tests were employed. The association between changes in EAT and risk factors over time was examined using simple regression models, receiver operating characteristic curve (ROC) curve analysis, and Pearson’s chi-square test. Statistical significance was determined as p < 0.05. All analyses were performed using the SPSS software package (version 26.0; IBM Corp., Armonk, NY, USA).

A total of 762 participants (age 58.7 ± 6.6 yr, 91.5% males) were included (Table 1). The interval duration between the two computed tomography (CT) scans was 2,237 ± 984 days. Participants were grouped into quartiles according to the baseline EAT volume, < 84.00 mL, 84.00–107.77 mL, 107.77–135.39 mL, and > 135.39 mL, respectively. Approximately, 42.7% of the participants were obese with a BMI of 24.5 ± 2.6 kg/m², and the prevalence of hypertension and diabetes mellitus was 42.3% (p < 0.001) and 17.8% (p = 0.033), respectively. A significant difference was observed between Q1 and Q4 for hypertension (28.8% vs. 51.8%, respectively) and diabetes (11.5% vs. 19.9%, respectively). Height, weight, and BMI exhibited a positive association with EAT volume at baseline. The association between smoking and EAT displayed no significance (p = 0.302), while for alcohol consumption (p = 0.012), heavy drinkers were prevalent in Q4.

Classical risk indicators, such as coronary calcium score (p = 0.013) and the ASCVD 10-year risk score, exhibited a significant correlation with EAT. From laboratory results, HbA1c, glucose, TGs, HDL, and TG/HDL were associated with EAT (all p < 0.001), while total cholesterol (p = 0.294) and LDL (p = 0.569) were not. Among these, except for HDL, which demonstrated a negative association, HbA1c, glucose, and TGs, TG/HDL exhibited a positive association with EAT volume.

At baseline, the average EAT was 113.1 ± 39.2 mL, with EAT/BSA at 62.1 ± 19.7 mL/m² (Table 2). During follow-up, both EAT and EAT/BSA increased to 124.6 ± 44.9 mL and 68.6 ± 22.5 mL/m², respectively. Weight and BMI displayed no significant change between the baseline and follow-up measurement (weight: from 70.1 ± 9.3 kg to 69.8 ± 9.5 kg; BMI: from 24.5 ± 2.6 kg/m² to 24.5 ± 2.5 kg/m²). Unlike weight, BMI, and BSA remained relatively unchanged (BSA: from 1.8 ± 0.2 m² to 1.8 ± 0.1 m²). Furthermore, all groups exhibited an increased mean EAT and EAT/BSA at follow-up. Significant differences in BSA were observed among each quartile at both baseline and follow-up, except for between Q2 and Q3 in both instances.

Groups with initially low EAT levels exhibited a pronounced yearly increase in EAT. Specifically, Q1 demonstrated a change of 2.7 ± 4.5% per year, while Q4 displayed a change of 1.3 ± 2.9% per year. Upon comparison among the groups, statistically significant differences were observed solely in EAT changes between Q1 and Q3/4. The p-value for the comparison between Q1 and Q3 was 0.012, and between Q1 and Q4, the value was 0.003.

The participant population was divided according to the EAT change per year; +5%/yr and -5%/yr were selected arbitrarily as criteria for clinical and statistical importance. Clinical characteristics of participants according to EAT change per year were compared (Table 3). EAT increased in parallel with BMI, TG/HDL, and glucose over time (all p < 0.001). Other risk factors did not exhibit a significant difference. Moreover, BMI, TG/HDL, and glucose were considered significant risk factors, and LDL, calcium score, and ASCVD 10-year risk score were selected as comparative risk factors for simple regression analysis. The association and correlation coefficients (R2) between EAT change per year and BMI, TG/HDL, glucose, LDL, calcium score, and ASCVD 10-year risk score are illustrated by scatter plots (Fig. 2). Although BMI, TG/HDL, and glucose demonstrated significant association with EAT change, R2 of the single parameter was very low.

We analyzed the association between baseline EAT, follow-up EAT, EAT/BSA, and CVD events stratified by statin use. However, no statistically significant relationship was observed between changes in EAT and the occurrence of CVD events based on statin use (Table 4). To compare the risk-predicting power of EAT, the association between CVD event and calcium score, ASCVD 10-year risk score, baseline EAT, baseline EAT/BSA, and change of EAT per year was plotted as ROC curve (Fig. 3). Only the calcium score and ASCVD 10-year risk score demonstrated significant predictability (area under the curve: 0.753 vs. 0.574, p < 0.001 for calcium score and p = 0.007 for ASCVD 10-year risk score, respectively). Neither baseline EAT nor EAT change over time displayed any significant predictability of CVD events. According to the baseline characteristics, an increase in EAT quartile was associated with an increase in calcium score. Subsequently, we conducted multiple regression analyses based on the change in calcium score, confirming that HbA1c, weight, baseline TG, and age were independent predictors of calcium score change (β > 0, all p < 0.05) (Supplementary Table 1).

Lastly, to determine if adding EAT to an existing risk prediction engine has additional effects, the CVD event rate was re-analyzed by dividing the change of EAT into three groups according to calcium score and ASCVD 10-year risk score (Fig. 4). The re-analyzed group with the lowest calcium score displayed a p value of 0.508, followed by > 0.999 and 0.673 for the intermediate and highest groups, respectively. Similarly, for the ASCVD 10-year risk score, the lowest, intermediate, and highest re-analyzed groups had p values of 0.585, > 0.999, and > 0.999, respectively. In summary, no additive effect of EAT change per year on top of the conventional risk prediction markers was observed in any group.