CAD-RADS may underestimate coronary plaque progression as detected by serial CT angiography

Abstract Aims We wished to assess whether different clinical definitions of coronary artery disease (CAD) [segment stenosis and involvement score (SSS, SIS), Coronary Artery Disease—Reporting and Data System (CAD-RADS)] affect which patients are considered to progress and which risk factors affect progression. Methods and results We enrolled 115 subsequent patients (60.1 ± 9.6 years, 27% female) who underwent serial coronary computed tomography angiography (CTA) imaging with >1year between the two examinations. CAD was described using SSS, SIS, and CAD-RADS. Linear mixed models were used to investigate the effects of risk factors on the overall amount of CAD and the effect on annual progression rate of different definitions. Coronary plaque burdens were SSS 4.63 ± 4.06 vs. 5.67 ± 5.10, P < 0.001; SIS 3.43 ± 2.53 vs. 3.89 ± 2.65, P < 0.001; CAD-RADS 0:8.7% vs. 0.0% 1:44.3% vs. 40.9%, 2:34.8% vs. 40.9%, 3:7.0% vs. 9.6% 4:3.5% vs. 6.1% 5:1.7% vs. 2.6%, P < 0.001, at baseline and follow-up, respectively. Overall, 53.0%, 29.6%, and 28.7% of patients progressed over time based on SSS, SIS, and CAD-RADS, respectively. Of the patients who progressed based on SSS, only 54% showed changes in CAD-RADS. Smoking and diabetes increased the annual progression rate of SSS by 0.37/year and 0.38/year, respectively (both P < 0.05). Furthermore, each year increase in age raised SSS by 0.12 [confidence interval (CI) 0.05–0.20, P = 0.001] and SIS 0.10 (CI 0.06–0.15, P < 0.001), while female sex was associated with 2.86 lower SSS (CI −4.52 to −1.20, P < 0.001) and 1.68 SIS values (CI −2.65 to −0.77, P = 0.001). Conclusion CAD-RADS could not capture the progression of CAD in almost half of patients with serial CTA. Differences in CAD definitions may lead to significant differences in patients who are considered to progress, and which risk factors are considered to influence progression.


Introduction
Coronary artery disease (CAD) is a chronic disease with dynamic temporal changes in plaque size and composition. 1,2 Coronary computed tomography angiography (CTA) is a robust, non-invasive diagnostic test to characterize CAD. 3,4 Coronary CTA allows the accurate assessment of the presence, extent and severity of coronary atherosclerosis, and has received class I recommendation (level of evidence B) for the evaluation of CAD among patients with stable chest pain in the latest guidelines. 5 Recent studies have proposed CAD progression as a prognosticator of adverse events, irrespectively from stenosis severity using CTA. 1 However, many clinical definitions and metrics of coronary plaque progression (PP) have been proposed. 6,7 Nevertheless, there are still limited data on the predictors of coronary PP with respect to different metrics of CAD quantity.
Therefore, we aimed to evaluate the predictors of CAD progression using different clinical definitions of plaque burden on serial computed tomography (CT) imaging. In addition, we sought to compare the identification of PP by three different semiquantitative definitions of CAD including the stenosis classification of the Coronary Artery Disease-Reporting and Data System (CAD-RADS).

Patient population
From 7233 patients in our structured reporting registry of coronary CTA, 316 had at least two coronary CTAs with at least 1 year between the two examinations using the same 256-slice scanner (Philips Brilliance iCT, Best, The Netherlands) between 1 January 2015 and 6 January 2020. Indications for serial testing were the planning of radiofrequency ablation in patients with paroxysmal atrial fibrillation (23.4%) or recurrent angina after initial CTA (76.6%). Exclusion criteria for our analysis were myocardial infarction, percutaneous coronary intervention, heart transplantation, or coronary bypass graft surgery prior to any CTA imaging and nondiagnostic CT image quality. Overall, we found 115 symptomatic and stable patients who met the inclusion and exclusion criteria ( Figure 1). Anamnestic, anthropometric data and CTA findings were recorded in a structured reporting platform (Axis, Neumann Medical Ltd. Budapest, Hungary).
The study was approved by the institutional review board and informed consent was waived due to the retrospective nature of the study. All procedures used in this study were in accordance with local and federal regulations, and the Declaration of Helsinki.

Demographics and comorbidities
Medical chart review was performed to gather data on patient demographics and comorbidities. Both at baseline and follow-up, patients underwent detailed interview on risk factors, medical history, and medication. Hyperlipidaemia was defined as having elevated plasma cholesterol levels (total cholesterol >200 mg/dL) or the use of lipid-lowering therapy. Hypertension was defined as systolic blood pressure >140 mmHg and/or diastolic blood pressure >90 mmHg or antihypertensive medication use verified by medical records. Smoking was defined as prior tobacco use within 1 year prior CTA (both baseline and follow-up time point evaluated). Diagnosis of diabetes mellitus (DM) was established based on elevated plasma glucose levels (fasting plasma glucose > _ 126 mg/dL; HbA1c > _ 6.5%) or the use of antidiabetic medication or insulin therapy. Statin use was recorded at baseline and follow-up at the time of the CT scanning and was defined as the use of statin at the time of the scans.

CT Image acquisition
We performed prospectively triggered CTA of the heart according to the guidelines of the SCCT. 8 Per os metoprolol was administered in case of heart rate exceeded 65 beats per minute 1 h before the coronary CTA examination and if not contraindicated. All patients received 0.8 mg of sublingual nitroglycerine prior to the CTA and intravenous betablocker was also administered if heart rate was still above 60 beats/min. Image acquisition was performed at diastole (75-81% of the R-R interval) or at systole (37-43% of the R-R interval) if heart rate was above 70 bpm despite premedication. The following scanner settings were used: 270 ms rotation time, 128 Â 0.625 mm collimation, tube voltage of 100-120 kVp, and tube current of 200-300 mAs based on patient's body mass index (BMI). Axial images were reconstructed with 0.4 mm slice thickness using iterative reconstruction.

Coronary CTA analysis
A single reader with 7 years of experience in cardiovascular CT imaging (B.S.) assessed the location, morphology, and severity of coronary lesions using an 18-segment coronary tree model.
Coronary plaque was defined on the CTA based on former publications by Mahabadi et al. and Achenbach et al. 9,10 Calcified and non-calcified plaques were defined as any discernible structure with a density of > _130 and <130 Hounsfield units, respectively, which were assigned to the vessel wall in at least two independent image planes. Coronary segments with a diameter of >1.5 mm were analysed. The reader evaluated baseline and follow-up images simultaneously to detect changes in plaque composition (on visual evaluation) or stenosis severity and was blinded to patient characteristics, CTA date, comorbidities, and medical treatment.

Statistical analysis
Continuous variables are presented as mean and standard deviation, whereas categorical parameters are presented as frequency with percentages. Paired sample t-test was used to compare parameters describing coronary plaque burden of the two CTA examinations. First, we used linear regression analysis to identify predictors of annualized PP. Furthermore, we applied a more complex approach using linear mixed models to analyse repeated observations at non-standardized intervals. [13][14][15] By analysing intra-individual changes over time we can simultaneously estimate the effect of a predictor on (i) overall amount of the outcome (SSS, SIS, and CAD-RADS) and (ii) effect on annual progression rate. 15,16 Using this information, the model is able to provide an estimate on the effect of a covariate on overall amount of the outcome, irrespective of time (e.g. statin users tend to have more plaque, as statin is initiated in patients with increased plaque burdens), but it also provides estimates on how the covariate affects temporal changes (e.g. statin users will have slower progression as compared to the average). This method also already accounts for the total plaque burden at baseline, therefore no additional adjustment is needed in the models. We calculated univariate linear mixed models to assess the effect of each predictor on the outcome (CAD definitions) and annual progression of the outcome. If either effect had a P-value <0. 10, we included that predictor in a multivariate model. Clinical predictors of coronary PP were included in the models as predictors with SIS, SSS, and CAD-RADS as outcomes. Inter-observer reproducibility was assessed in 25 patients by two observers using weighted kappa. We selected five plaques per stenosis category (minimal 1-24%, mild 25-49%, moderate 50-69%, severe 70-99%, and occluded 100%), including one plaque per patient. The k values were interpreted as  All analyses were conducted in the R environment v3. 6.1 and STATA v13.0. A two-sided P-value smaller than 0.05 was considered statistically significant.

Results
The baseline characteristics of the 115 enrolled patients (mean age 60.1 ± 9.6 years, 27% female) are summarized in Table 1. On average, 2.6 ± 1.1 years has passed between the CTA examinations. A total of 1763 coronary artery segments were evaluated at both time points. Mean effective radiation dose was 5.07 and 5.09 mSv (P = 0.822) for baseline and follow-up.

Characteristics of coronary PP
A total of 105 (91.3%) patients had any CAD at baseline. The remaining 10 patients (8.7%) had no plaque on baseline and all developed minimal stenosis on follow-up (CAD-RADS 1). We detected no progression of disease in 54 patients with CAD (46.7%), furthermore, plaque regression was not observed in this patient population.
Representative case of CAD progression is seen on Figure 2, changes in SSS and CAD-RADS during follow-up are summarized on a Sankey diagram ( Figure 3). SSS-based progression is depicted among those with no changes in CAD-RADS severity on a Sankey diagram ( Figure 4).
We found excellent agreement across stenosis categories evaluating 25 plaques (5 from each stenosis category) by two observers (weighted kappa = 0.903).

Predictors of coronary PP
Linear regression models were carried out to identify predictors of PP described as annualized progression of SSS or SIS. Age was associated with PP based on SSS [b = 0.013, confidence interval (CI) 0.01-0.024, P = 0.034] but showed no association with SIS progression (P = 0.784). Importantly, none of the other predictors showed any association with the annualized progression rate of SSS, SIS, or CAD-RADS (all P > 0.05). However, this approach does not account for the temporal changes of covariates over time and does not account for the one's baseline value, therefore we also analysed our data using linear mixed models.

Linear mixed models for the analysis of coronary PP and progression rate
Univariate linear mixed models showed that age and gender affect the value of SSS and SIS in our population, whereas diabetes increased the annual progression rate of SSS by 0.34/year (CI 0.01-0.66; P = 0.036) (see Table 3).
On multivariate analysis, patients who smoked had significantly increased annual progression rate of SSS by 0.37/year (CI 0.07-0.67, P = 0.017) and higher total extent of CAD as described by SIS as   Table 4.

Discussion
In our longitudinal observation cohort study, we found that age and gender influenced the severity and the extent of CAD after correcting for the effects of other conventional cardiovascular risk factors.
Smoking and DM were significant factors increasing annual progression rates of SSS, reflecting both the severity and the extent of the disease. Importantly, among those with increased SSS and SIS on follow-up CTA, in 46% and 41% of patients the CAD-RADS severity classification did not change. On the other hand, none of the risk factors seem to influence CAD-RADS, the currently recommended clinical classification framework for reporting CAD. Coronary atherosclerosis is a dynamic and progressive disease that may lead to the obstruction of the coronary lumen and induce ischaemia. 17 CTA is a uniquely suited imaging modality to monitor changes in the extent and severity of CAD and underlying plaque composition. 18 Quantifying coronary plaque burden improves risk assessment using both semi-quantitative (SSS, SIS) or quantitative (volumetric) plaque metrics. 19,20 Recent studies incorporated the degree of stenosis, plaque morphology, and SSS for the detection of PP.
Although quantitative plaque analysis (volumetric change in atheroma burden) is not used routinely in clinical setting, this biomarker has been increasingly utilized for monitoring anti-atherosclerotic drug therapy. 21 Despite increasing number of studies utilizing CT imaging for the detection of the plaque development, there are limited data on the predictors of atherosclerosis progression. Moreover, there is a huge variety of definitions in use to quantify CAD progression and to characterize coronary atherosclerosis on CTA per se. Motoyama et al. 1 defined disease progression as either an increase in stenosis by at least 1 grade or an increase in the remodelling index ratio of >1.1, and found that increase in plaque burden might be the strongest predictor of adverse events. In other studies, progression was defined based on newly diagnosed cases with 50% or more coronary stenosis by the person-years of follow-up, 22 whereas Gu et al. 23 used coronary calcium score, SSS, and SIS for describing progression. More recent investigations focused on quantitative plaque analysis and defined PP based on volumetric changes (> _10%) compared to baseline volume 24 or simply evaluated the change in plaque volume. 25 Notably, most of these changes in plaque volume are relatively small and would not change clinical scoring systems such as SSS or CAD-RADS classification and clinical decision-making. Other limitations of using volumetric plaque analysis include insufficient image quality, lack of standardized protocols regarding acquisition protocols and iterative reconstruction. 26,27 Furthermore, many concerns have been raised regarding the inter-vendor, scan-rescan, inter-software, and inter-reader variability of quantitative plaque analysis which limits its widespread clinical use. [28][29][30] Importantly, the methodology of the aforementioned studies differed as compared to our analysis, since we are the first to use linear mixed models to define the effect of risk factors on CAD metrics and the progression rate. Most of the prior studies analysed two timepoints defining progression as the change in mean values of plaque parameters of a given population, but this approach might not account for the one's baseline value (i.e. individuals with more plaque may progress faster unless normalizing for the baseline plaque volume) or the possible changes in covariates over time (i.e. someone may develop DM over the follow-up period). Moreover, some studies enrolled patients with non-standardized follow-up times for serial scanning, which-also based on our observation of PP rate-can significantly influence results. Linear mixed models provide flexible modelling of intra-subject changes and enable to assess overall and individual patterns in time. This method enables to assess the impact of predictors in both timepoints on the outcomes. 15 To incorporate the main findings of coronary CTA in a unified and standardized framework, experts of the field have proposed the use of the CAD-RADS classification in clinical practice. Maroules et al. 31 found high inter-observer agreement for both experienced and early career readers when assessing CAD-RADS categories, except for vulnerability. Importantly, the currently used CAD-RADS . classification scheme only describes the plaque with the largest degree of stenosis and does not account for the extent of disease. For example, patients with mild or moderate stenosis (CAD-RADS category 2 or 3)-who are classified as non-obstructive patients-could also demonstrate extensive CAD affecting more than four coronary segments. Bittencourt et al. 32 demonstrated that extensive, but nonobstructive CAD patients have comparable risk profile as those with obstructive CAD. CAD burden scores (SSS and SIS) used in current study also incorporate the number of segments affected with plaques. Figure 2 shows a representative case of our study highlighting the importance of using different scores when describing PP. In this case, the patient had de novo plaques on two additional segments whereas the CAD-RADS category remained the same and thus might create the false impression that CAD did not progress throughout the years. Coronary PP is a strong prognosticator of events and therefore should also be implemented in CAD-RADS recommendation.
Proper risk assessment is crucial to better define a subset of patients who require a more aggressive secondary prevention therapy or downstream testing. We detected substantial changes in SSS and SIS during follow-up imaging; however, the CAD-RADS classification was not able to identify changes in the extent and severity of CAD and thus might not recognize increased risk for future events. CAD-RADS classification has been proposed to guide further testing and secondary prevention therapy; however, based on our data, it would miss almost every second patient with increased plaque severity or extent which have been previously shown to increase adverse events. The use of CAD-RADS underestimates coronary PP as it heavily focuses on the worst plaque of a given patient and only includes three-vessel obstructive > _70% disease in grade 4B when assessing plaque burden. Coronary CTA provides, however, rich information on the distribution, extent, and vulnerability of coronary atherosclerosis. The novel version of the classification scheme could improve risk assessment by combining the stenosis severity and plaque burden, moreover, experts of the field also suggested adding flow-limiting lesions and coronary artery calcium score information as well. Further outcome studies are warranted to evaluate the prognostic value of the current CAD-RADS classification and possible future versions in terms of plaque severity vs. the extent of CAD. Also, PP should be included in the classification and its predictive value should be tested in large contemporary cohort of outpatients with stable angina.
Prior investigations report conflicting results regarding the predictors of PP, which might originate from the differences in the definition of progression and the use of CAD scoring systems. Increased PP might be promoted by male gender, obesity, high low-density lipoprotein cholesterol level, DM, and smoking. 33,34 Among conventional cardiovascular risk factors, DM, BMI, and smoking induced PP in other studies. 35,36 Smoking can promote atherogenesis involving pathways of inflammation, endothelial dysfunction, platelet function, cholesterol metabolism, and thrombotic factors. [37][38][39] We also found that male gender was associated with larger SSS and SIS, moreover, patients who smoked had larger extent of CAD. Among cardiovascular risk factors smoking and DM were related to the progression rate of SSS using linear mixed models.
We acknowledge the limitations of our study. First, we retrospectively analysed patients who underwent serial CTA at a single tertiary centre for Cardiology with different time interval between the two ..
coronary CTA examinations. We therefore used linear mixed models that account for differences in follow-up times and provide detailed information on the factors of PP. Another limitation is the incidental statin use without routine cholesterol measurements and strict control of adherence to the drug. Also, prospective, multicentre trials are warranted to analyse whether the observed discrepancies between different CAD definitions could influence patient outcomes. Furthermore, smoking was defined as prior tobacco use within 1 year prior CTA reflecting current smoking and not life-long accumulated hazards of smoking.

Conclusion
CAD-RADS did not capture the progression of CAD in almost half of stable angina patients with serial CT imaging. Age and gender influenced segment stenosis and plaque extent as described by segment stenosis and involvement scores. Smoking and diabetes affected the progression rate of PP based on SSS. Differences in CAD definitions may lead to significant differences in patients who are considered to progress. Therefore, unified plaque metrics are needed that are capable to properly describe the extent and severity of CAD for CTbased risk prediction and clinical management.

Data availability
The data underlying this article will be shared on reasonable request to the corresponding author.