This observational study retrospectively recruited consecutive patients who had consultation between January 2019 and December 2021 at the Department of Cardiology, First Affiliated Hospital of Zhengzhou University. Data were extracted from the hospital’s files and were anonymized. The study population consisted of patients diagnosed with PVCs and presented to our hospital for routine follow-up. The inclusion criteria were as follows: 1) patients who underwent non-contrast CT, 2) patients symptomatic with frequent PVCs (>10,000/24 h), 3) patients >18 years of age, and 4) absence of severe valvular heart disease and hypertrophic heart disease. The exclusion criteria were follows: 1) patients with a history of prior radiofrequency ablation, 2) patients with poor/insufficient CT images, 3) patients with atrial fibrillation (AF), and 4) patients with coronary heart disease, including angina pectoris, myocardial infarction, percutaneous coronary intervention, and coronary artery bypass surgery. The control group (n = 402) matched with the PVC patients during the same time period, which had no detected PVCs using the 24-h Holter monitoring. Patients with a diagnosis of PVCs were identified using the International Classification of Disease (ICD) code in our hospital’s electronic health record systems. Frequent PVC was defined as a total PVC count of >10,000 beats during a 24-h Holter monitoring. Non-contrast CT was performed for patients who wanted examination for various reasons including smoking histories, shortness of breath, or screening for lung disease in our institution. The study complied with the Declaration of Helsinki. The study protocol was authorized by the local institution’s ethics committee and waived the need for written informed consent (2022-KY-0288).

The following data were collected from all patients: demographic parameters, comorbidities, echocardiography parameters [left ventricular ejection fraction (LVEF), left atrial diameter (LAD), E/A ratios], CT parameter (EAT volume), and burden levels of PVCs on admission. PVC burden = total ventricular premature beats/total beats × 100%. Body mass index (BMI) ≥24 kg/m² was defined as overweight or obese. Smokers and drinkers were either former or active smokers and drinkers.

All individual examinations use a dual-source CT system (Somatom Force, Siemens Healthineers, Forchheim, Germany). Non-contrast CT was performed with 120 kV, and the tube current depended on body habitus and heart rate. The images were reconstructed with a slice thickness of 0.5 mm, a reconstruction increment of 0.5 mm with a medium soft-tissue convolution kernel (B26F), and a reconstructed matrix size of 512 × 512.

EAT volume was quantified using a dedicated semiautomatic software (syngo.via Frontier Cardiac Risk Assessment, version 1.2.3, Siemens Healthineers, Germany), as shown in Figure 1 . Previous studies used the sophisticated software to quantify EAT volume (8–10). EAT is defined as non-contrast CT density ranging from −195 Hounsfield units (HU) to −45 HU (8, 10, 11). The software automatically delineates and identifies EAT and manually adjusts the contour of EAT volume if necessary. EAT volume measurement was performed by two independent radiologists who were unaware of the patient’s clinical data.

The values of the continuous variables were described as the mean ± standard deviation or median (Q1, Q3 quartiles) and compared among the groups using Student’s t-test or Mann–Whitney U test, depending on whether the data were normally distributed. Categorical variables are presented as numbers (percentage) and compared between two groups by the Pearson chi-squared test or Fisher exact test. The propensity score matching (PSM) 1:1 was applied in the observational case–control study to reduce bias in selecting the case controls. Matching of the two groups was performed for basic characteristics (age, sex), and variables subjected to univariate analysis showed p <0.10 (BMI ≥ 24 kg/m², hypertension, diabetes mellitus, fasting blood glucose, total cholesterol, E/A ratio <1, LVEF, and LAD). We calculated the standardized mean difference to assess the balance of baseline characteristics after PSM. Receiver operating characteristic (ROC) curve analysis was performed to analyze discriminative power, specificity, and sensitivity to predict the frequent PVC risk based on the EAT volume.

The associations between EAT volume and continuous variables were determined using Pearson’s correlation analysis. The cutoff values of the PVC burden levels were determined by the median index. Variables subjected to univariable analysis showed p-value <0.10, and important clinical risk factors were included in the multivariable analysis. Multivariable logistic analysis was performed to investigate factors associated with PVC risk and burden levels after adjusting other confounders. We also used restricted cubic splines (RCS) with four knots at the 5th, 25th, 75th, and 95th centiles to flexibly model the relationship between EAT volume and PVC burden levels. Furthermore, we compared EAT volume by sextile of the PVC burden levels, with the first sextile serving as the reference. Statistical analysis was performed using R language version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria). A two-sided p-value of <0.05 was considered statistically significant.

Four hundred ninety-nine frequent PVC patients were screened for eligibility. Four hundred and two PVC patients were included in the final analysis, as shown in Supplementary Figure 1 . The gap time between the non-contrast CT and the 24-h Holter monitoring scan performed was 2 (1–3) days in frequent PVC patients. The patients with frequent PVCs had significantly higher proportions of hypertension, diabetes mellitus, E/A ratio <1, and BMI ≥24 kg/m², compared with control patients. In the unadjusted cohort, PVC patients also had significantly larger age, BMI, fasting blood glucose, total cholesterol, LVEF, and LAD, compared with the control patients.

PSM was used to reduce bias in the selection of potential clinical confounders. Following the 1:1 matching protocol, differences in clinical characteristics between the two groups were eliminated. Compared with the control patients, patients’ age, hypertension, diabetes mellitus, E/A ratio <1, BMI ≥24 kg/m², BMI, fasting blood glucose, total cholesterol, LVEF, and LAD showed no significant differences between the two groups in the PSM cohort, as shown in Table 1 . Based on the propensity score, PSM achieved an optimal balance between the two groups. The standardized mean difference values were all under 0.1, which indicated that confounding factor bias was attenuated ( Supplementary Figure 2 ).

The distribution of EAT volume is shown in Figure 2 . PVC patients had significantly larger EAT volume compared with control patients in the unadjusted cohort and PSM cohort. The cutoff points of EAT volume for the prediction of frequent PVCs were calculated as 102.81 ml, and the AUC was 0.660 (95% CI: 0.622–0.697). The specificity and sensitivity were 60.2% and 66.9%, respectively, as depicted in Supplementary Figure 3 . EAT volume was significantly correlated with age (r = 0.396, p = 0.001), BMI (r = 0.388, p < 0.001), and LAD (r = 0.322, p < 0.001) in the 402 PVC patients ( Supplementary Table 1 ). EAT volume was significantly larger in male PVC patients with BMI ≥24 kg/m², diabetes mellitus, and E/A ratio <1, as described in Figure 3 . Notably, the multivariable logistic analysis found that EAT volume (per 10 ml increase) was an independently associated factor for PVCs after adjusting different confounders, as indicated in Table 2 .

The 402 PVC patients were divided into two groups—those with high burden levels of PVCs and those with low burden levels of PVCs—according to the median burden level of 19.32%. Compared with patients with low burden levels of PVCs, patients with high burden levels had significantly larger age, LAD, and EAT volume and higher proportions of being a smoker, having diabetes mellitus, and having an E/A ratio <1 ( Supplementary Table 2 ). Variables included in the univariable analysis showed p-value <0.10, and clinical risk factors (fasting blood glucose, total cholesterol) were included in the multivariable analysis, which included age, female gender, smoker, fasting blood glucose, total cholesterol, diabetes mellitus, LAD, E/A ratio <1, and EAT volume (per 10 ml increase). EAT volume (per 10 ml increase; HR: 1.06; 95% CI: 1.01–1.10; p = 0.008) remained independently associated with high burden levels of PVCs, as described in Table 3 . The risk of high PVC burden levels increased with the rising of EAT volume (p for non-linearity = 0.084) by RCS, using the median of EAT volume as the reference (123.38 ml), as illustrated in Figure 4 . Moreover, EAT volume with high PVC burden levels significantly increased in PVC patients with burden levels of sextiles 3–6, with sextile 1 serving as the reference ( Figure 5 ).

Our study presented some limitations. First, this single-center study was a cross-sectional, observational study, which may lead to selection bias. It is difficult to determine any causal relationship between PVCs and EAT volume. Second, patients with coronary artery disease and AF were excluded from the study, and we considered the close association of these diseases with EAT, which leads to the inaccuracy of the research results. Third, we selected patients with high burden levels of PVCs with more symptoms and impaired cardiac function, limiting the generalizability of the results. However, it is more meaningful to study patients with frequent PVCs, which need more treatment and follow-up. Moreover, our study did not include any reference standard, such as histopathology, to confirm absolute EAT volume. However, we applied a sophisticated CT software for automatic quantification of EAT volume. Finally, we were not able to assess the potential association between PVCs and the ventricular region of EAT. To address this issue, further research may be needed using CMR to quantify the EAT volume in particular areas and investigate the impacts of different EAT areas on PVC patients. Our study could not include all confounding factors, which may alter the power of the statistical analysis and the study’s conclusion. Further prospective studies involving matched groups are needed to confirm the findings.