This retrospective analysis included 1,955 consecutive subjects who underwent both CAC scanning and CCTA at the Lundquist Institute at Harbor-UCLA Medical Center. Patients were referred for evaluation of coronary atherosclerosis or risk assessment. Exclusion criteria included prior coronary revascularization (percutaneous coronary intervention or bypass surgery), incomplete imaging data, or technically inadequate studies. To focus on individuals with measurable calcification, subjects with CAC = 0 were excluded.

Clinical variables, including age, sex, diabetes mellitus (DM), hypertension (HTN), and statin use, were extracted from medical records. The study was approved by the institutional review board, and informed consent was waived due to the retrospective design.

All scans were performed using 256 GE Revolution CT scanners with prospective ECG-gating. CAC scoring was acquired with non-contrast imaging (120 kVp, 2.5 mm slice thickness) and quantified using the Agatston method. CCTA was performed using iodinated contrast (80–100 mL) with dose-modulated protocols optimized for heart rate control (target < 65 bpm) using oral or intravenous beta-blockers. Sublingual nitroglycerin was administered before image acquisition.

AI-based quantitative analysis was conducted using Cleerly (Cleerly Inc., New York, NY), which automatically segments coronary arteries, identifies plaque components, and quantifies volumetric plaque measures (mm³). TPB was defined as the sum of all calcified, fibrous, and lipid-rich plaque volumes across major coronary vessels. The software has been previously validated against expert manual assessment and intravascular reference standards.

CAC was categorized as 1–99, 100–299, and ≥ 300 Agatston units. A secondary analysis was performed with adjusted cutoffs of 1–99, 100–399, and ≥ 400. TPB was categorized as 0, 1–250, 251–750, or > 750 mm³.

Continuous variables were expressed as mean ± standard deviation, and categorical variables as counts and percentages. Agreement between CAC and TPB categories was evaluated using Cohen’s weighted Kappa (κ) with 95% confidence intervals. The strength of agreement was defined as poor (< 0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (> 0.80). The continuous relationship between CAC and TPB was assessed using Spearman’s rank correlation coefficient (ρ). Subgroup analyses were performed stratified by age, sex, diabetes, hypertension, and statin use. Statistical significance was set at a two-tailed p < 0.05.

Continuous variables were expressed as mean ± standard deviation (SD), and categorical variables as frequencies and percentages. Agreement between CAC and TPB categories was quantified using Cohen’s weighted Kappa (κ) with 95% confidence intervals. Kappa values were interpreted as follows: poor (< 0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), and almost perfect (> 0.80). The correlation between continuous CAC and TPB values was assessed using Spearman’s rank correlation coefficient (ρ). Subgroup analyses were conducted by age (≤ 63.9 vs > 63.9 years), sex, presence of diabetes or hypertension, and statin use. All analyses were performed using SPSS version 29 (IBM Corp., Armonk, NY). A two-tailed p-value < 0.05 was considered statistically significant.

Among 1,955 subjects, the mean age was 66.5 ± 10.4 years, and 69% were male. Diabetes mellitus was present in 16%, hypertension in 54%, and 68% were on statin therapy. Demographic and clinical characteristics are summarized in Table 1.

The categorical distributions of Coronary Artery Calcium (CAC) and Total Plaque Burden (TPB) are summarized in Table 2. Among the study population, 39% had CAC scores between 1–99, 20% fell within the 100–299 range, and 41% had scores ≥ 300. In comparison, TPB was distributed as follows: 67% of individuals had a burden between 1–250 mm³, 27% fell within the 251–750 mm³ range, and 12% had total plaque volumes exceeding 750 mm³.

Overall, weighted Kappa was 0.495 (95% CI 0.471–0.519), representing moderate agreement between CAC and TPB categories. When CAC thresholds were adjusted to 1–99, 100–399, and ≥ 400, Kappa increased modestly to 0.535 (95% CI 0.511–0.559).

Spearman correlation between continuous CAC and TPB was very strong (ρ = 0.92, p < 0.001).

We assessed the categorical agreement between Coronary Artery Calcium (CAC) categories and AI-derived Total Plaque Burden (TPB) across clinically relevant subgroups using Cohen’s weighted Kappa statistic. The overall agreement was moderate (κ ≈ 0.50), with notable variability based on age, sex, and cardiovascular risk factors.

Agreement was slightly higher in older individuals (age > 63.9 years) compared to those ≤ 63.9 years, suggesting better concordance between calcification and total plaque burden in more advanced disease. Patients on statin therapy also demonstrated higher Kappa values than those not receiving statins, consistent with statin-associated promotion of plaque calcification. Conversely, individuals without statin use, diabetes, or hypertension exhibited wider confidence intervals and lower Kappa values, indicating greater discordance between CAC and total plaque burden.

Sex-based differences were particularly prominent. Female participants showed the lowest categorical agreement (κ < 0.45), reflecting the higher prevalence of non-calcified plaque in women and reinforcing the limitations of relying solely on CAC for risk assessment in this population.

These subgroup differences are visually summarized in Fig. 1, which displays Kappa estimates with 95% confidence intervals for each clinical category.

In this large, single-center study of nearly 2,000 patients undergoing both CAC scoring and AI-enabled quantitative CCTA, we observed a moderate categorical agreement (κ ≈ 0.50) but very strong continuous correlation (ρ = 0.92) between CAC scores and total plaque burden (TPB). This dichotomy highlights an important clinical insight: although CAC and TPB both aim to quantify the extent of coronary atherosclerosis, they do so through fundamentally different lenses. The moderate categorical overlap implies that applying discrete risk thresholds (e.g., CAC > 100, TPB > 250 mm³) may not reliably classify the same individuals into concordant risk categories, underscoring the potential for clinically meaningful discordance in plaque assessment (15, 16).

The discrepancy in categorical agreement stems from the inherent limitation of CAC: it detects only calcified plaque, neglecting the fibrous and lipid-rich components that characterize earlier and potentially more vulnerable stages of atherosclerosis (17, 18). Consequently, two patients with identical CAC scores (e.g., CAC = 100) may have markedly different atherosclerotic burdens if one harbors predominantly non-calcified plaque. This limitation is especially pertinent in younger individuals and women, where non-calcified plaque tends to predominate (19–21). The strong continuous correlation (ρ = 0.92) suggests that overall plaque accumulation, both calcified and non-calcified, tends to progress in tandem, but the observed categorical misalignment indicates that a single metric may miss important nuances in disease characterization (22, 23).

Importantly, these findings reaffirm that CAC and TPB are complementary rather than interchangeable. CAC provides a surrogate of disease chronicity and stability, while AI-based TPB offers a more nuanced, volumetric, and compositional view of the full atherosclerotic landscape.

Sex-specific analysis revealed lower agreement in women, consistent with known differences in plaque biology. Prior research has demonstrated that women tend to exhibit less calcified and more non-calcified plaque than men at equivalent levels of overall atherosclerosis. This contributes to the underestimation of risk when relying solely on CAC scores in female populations (24–27).

Similarly, statin therapy may alter the CAC–TPB relationship. Statins promote plaque stabilization and calcification, a phenomenon referred to as the “statin paradox,” where CAC increases due to plaque healing, despite overall risk reduction. This transformation of non-calcified plaque into denser, more calcified forms can lead to an apparent “discrepancy” between high CAC and low-risk morphology. Conversely, patients on long-term statin therapy may show elevated CAC despite relatively modest TPB. Recognizing these modifying effects is critical when interpreting imaging results in both clinical and research settings (28–31).

Our results corroborate prior observations from the Multi-Ethnic Study of Atherosclerosis (MESA), which revealed that up to 15% of individuals with CAC = 0 may still harbor non-calcified plaque detectable by CCTA. Similarly, Budoff et al. and Williams et al. demonstrated that total plaque burden assessed by CCTA, especially lipid-rich or low-attenuation plaque, is independently predictive of major adverse cardiovascular events (MACE), even in the absence of CAC (32, 33).

By explicitly quantifying categorical agreement using weighted Kappa statistics, our study extends this literature, offering a formal metric for the degree of concordance between these two commonly used imaging biomarkers. This benchmark may help frame future studies assessing integrated risk models that combine CAC and AI-derived plaque metrics.

Our findings have several important clinical implications. First, they validate the incremental value of AI-enabled CCTA in detecting subclinical atherosclerosis that may be invisible to calcium scoring alone. In an era of precision prevention, where cardiovascular risk stratification increasingly guides therapeutic intensity, failing to identify patients with high non-calcified plaque burden could delay timely intervention.

Second, in patients with discordant findings, for instance, low or zero CAC but elevated TPB, physicians may consider intensifying lifestyle or pharmacologic therapy. Conversely, high CAC with low TPB may prompt further evaluation of plaque morphology or stability, rather than immediate escalation of therapy. This nuanced interpretation allows for a more tailored approach to preventive cardiology, avoiding both over- and under-treatment.

Finally, with growing availability and regulatory approval of AI-based tools, integration of TPB analysis into routine clinical workflows may become feasible and cost-effective, especially when bundled with routine CCTA performed for other indications (e.g., evaluation of chest pain). Combining CAC and TPB may enable multi-dimensional risk profiling, which better captures both plaque quantity and quality (34–36).

Several limitations merit consideration. This was a retrospective, single-center study without follow-up outcomes; thus, prognostic implications could not be assessed. CAC = 0 subjects were excluded, limiting generalizability to early subclinical disease. Imaging was performed on different CT systems, although standardized acquisition protocols were used. AI-derived plaque quantification depends on image quality and reconstruction parameters, which could introduce variability. Future multicenter studies should validate these findings across diverse populations and correlate CAC–TPB discordance with cardiovascular events.

Defining plaque volume thresholds corresponding to “low-risk” or “trivial” disease, analogous to CAC = 0, could enhance the interpretation of AI-derived metrics. Longitudinal analyses integrating CAC and TPB progression may further elucidate how plaque transformation, from non-calcified to calcified, relates to clinical outcomes. Combining AI-based plaque quantification with traditional calcium scoring has the potential to refine risk prediction models and optimize preventive therapy allocation.