The validity and applicability of CAD-RADS in the management of patients with coronary artery disease

Coronary artery disease (CAD) is the most common cause of death worldwide, and thus, early recognition is essential to avoid its related complications and improve prognosis [1]. Invasive coronary angiography (ICA) is considered the gold standard for the anatomical evaluation of CAD, but an invasive and costly procedure, with periprocedural morbidity and mortality [2]. Recently, coronary computed tomography angiography (CCTA) has emerged as a powerful modality to exclude obstructive coronary artery stenosis in the diagnostic workup of patients with suspected CAD [3]. Using CCTA has increased because of its non-invasive nature, the high negative predictive value in ruling out significant CAD, and recent technological advances [4].

Because of the discrepancy and high variability in CCTA reporting, a structured CAD evaluation template has long been required [5]. The structured reporting system for CCTA was introduced in 2016 by Curye et al. and is called the “coronary artery disease reporting and data system” (CAD-RADS) [6]. It is the first attempt to provide a simple, concise, and accurate classification of CAD. This new classification is expected to improve the communication and facilitate understanding of the CCTA report with the added inputs of further investigations and management recommendations. Although many institutions across the world have been using CAD-RADS over the last 2 years, it is still not widely accepted [7]. Moreover, the clinical impact on patient prognosis has not been extensively studied [8].

Several recent studies [9‐11] have examined the diagnostic validity of CCTA in the evaluation of CAD. However, no prior research has evaluated such validity in reference to the CAD-RADS classification system. Consequently, we performed the current prospective study to estimate the diagnostic validity, applicability, and reproducibility of CAD-RADS for predicting patients at high risk for significant CAD, and assess the value of such classification for decision-making in clinical settings.

The current study enrolled 287 patients with suspected CAD. We successfully performed all CCTA and ICA examinations without any side effects (the mean time from CCTA to ICA was 37 ± 12.3 days). Based on ICA results, we have 156 patients with non-significant CAD (CAD-RADS 0, 1, 2, and 3), and 131 patients with significant CAD (CAD-RADS 4 and 5). The prevalence rate for significant CAD was 45.6%.

In an attempt to enhance the clinical impact and quality of medical imaging, structured reporting systems have become increasingly common, providing physicians with a direct relationship between objective information in the radiology report and evidence-based management recommendations [16]. The CAD-RADS classification is one of these reporting systems that have been emerged recently. A few numbers of recent studies have been published explaining the CAD-RADS in details and clarifying the pitfalls and limitations of this classification system [17‐19]. However, since the pioneering report from Curye et al. in 2016, no study has been performed for external validation of the CAD-RADS. The current study is an attempt to validate the CAD-RADS and compare the results with ICA. We also assessed the application of such a reporting system for decision-making in clinical practice. The overall results are encouraging and show that this reporting system has excellent performance for categorization of CAD and predicting patients with significant CAD. However, more importantly in our opinion, we found that this classification system was highly appreciated by treating consultants as they could understand better the report of the CCTA examination and make more clear to them the findings observed as well as they felt more confident to take clinical decisions with their patients.

In terms of diagnostic validity, the CAD-RADS performed well, with a very high sensitivity, specificity, and accuracy (100%, 96.8 to 98.7%, and 98.3 to 99.3%, respectively, depending on the reviewer). This result is not surprising taking into consideration that it is based on CCTA findings, which have been extensively verified in many recent studies [20‐22], and proven to be a robust modality for CAD diagnosis and determining the severity of coronary artery stenosis. However, the relatively high prevalence of significant CAD (45.6%) in our study might be a potential selection bias which could affect the calculation of sensitivity and specificity. Moreover, this high prevalence of significant CAD could explain the substantially higher positive predictive values in our study (96.3–98.5%) than those reported in most previous CT studies. The higher prevalence of significant CAD in our patient group may be explained by the collection of our patients from a large central institution which received complicated patients. The higher sensitivities and higher negative predictive values in our study were attributed to many factors: First, we used a dedicated Philips workstation with specific software able to eliminate the calcium. Second, we excluded patients with high calcium score. Third, we excluded patients with a bad image quality (patients with arrhythmia or irregular heart rate, and patients unable to sustain a breath-hold). Fourth, we excluded patients with non-diagnostic examination either due to motion/respiratory artifacts or calcific plaques hindered full plaque assessment. Fifth, we reviewed all images by highly experienced reviewers.

The CAD-RADS is easy to understand and apply. However, one of the most critical limitations of the present study was that CAD-RADS is still uncommon and unfamiliar by many clinicians. Consequently, the rationale and goal of the CAD-RADS have to be clarified to referring consultants in several scientific meetings before the start of the study. We found more encouraging results about how referring consultants evaluated this reporting system and its utility for clinical decision-making. Our simple survey among referring consultants involved in patient management showed that 87.5% of referring consultants considered CAD-RADS a helpful classification system for clinical decision-making and referral. Moreover, the three consultants who did not encourage the use of CAD-RADS in their management protocol attributed that to the novelty of this reporting system and time required for learning and training.

Without better evidence for reproducibility of CAD-RADS, the results of the study become unusable for clinical practice, and we remain uncertain whether this is a reporting system we should be applied or not. So, we performed inter-reviewer reproducibility in our study. The overall results were considered highly satisfactory. Our reporting of CAD-RADS categories and modifiers showed excellent IRA (ICC = 0.9862 and 0.8064, respectively). These results are very similar to that of the previous two studies [23, 24], which assessed the inter-observer agreement of CAD-RADS. However, as regards vulnerable plaque features (modifier “V”), the IRA was moderate (ICC = 0.5453). This finding is in line with the recent literature [25], which reports moderate reproducibility and high variability among readers regarding vulnerable plaque features. As the proper assessment of vulnerable plaque features on CCTA is still challenging, a further modification of CAD-RADS classification regarding the “V” modifier is mandatory.

Although ICA is the gold standard for the diagnosis of CAD, it is not risk-free, and the costs are considerable. Most of the related complications are mild, but even in the absence of severe CAD, critical complications may also occur. So, the non-invasive investigation may be the most convenient method for the detection of intermediate lesions [26]. Integrating CAD-RADS-guided recommendations into clinical practice may help to reduce much-referral for ICA and encourage further appropriate follow-up care for patients undergoing a diagnostic workup of CAD. This is very important in stable patients with intermediate lesions (CAD-RADS 3) as we identified high rates of early ICA in these patients soon after CCTA and likely before adequate trials of medical therapy.

Using the ROC curve, all the reviewers in this study strongly agreed that the optimal cutoff value of CAD-RADS for predicting patients with significant CAD was < CAD-RADS 3. This cutoff value was associated with 100% sensitivity. However, we were not capable of comparing our cutoff value as no available previous studies provided cutoff value for the CAD-RADS. Thus, we recommend further similar studies with a larger population to confirm or refute our cutoff value.

Finally, based on our findings, and in keeping with reports of previous authors, the CAD-RADS classification is considered a categorized reporting of CCTA findings with several advantages. Therefore, we strongly encourage the incorporation of CAD-RADS into CCTA reports. Nevertheless, the CAD-RADS still has drawbacks as some essential data are not listed in the assignment of CAD-RADS categories (e.g., myocardial CT perfusion and fractional flow reserve, non-coronary cardiac and extra-cardiac CT findings and interaction with other established reporting tools, Agatston calcium score, pretest probability, and coronary artery anomalies). Therefore, the CAD-RADS needs further modification to become accurate, useful, and comprehensive of all relevant descriptors and definitions. Future large longitudinal studies on the long-term clinical outcomes are still required to show the added value of these data to CAD-RADS classification and the concepts of referring clinicians on this reporting system.

There were some limitations to the study. First, this study was performed in a single center. Therefore, confirmation by larger multi-center studies is required. Second, the higher prevalence of significant CAD in our study might be a potential selection bias which could affect the calculation of sensitivity and specificity. Third, all CCTA examinations were analyzed by experienced reviewers; this is potentially affecting diagnostic performance. Thus, further studies about the performance of this reporting system when applied by non-experienced reviewers are needed. However, we observed that the reproducibility of this reporting system was high. Fourth, a CAD-RADS management protocol was provided to referring consultants before initiating the study. This could have biased their decision on how patients should be managed. Fifth, the exclusion of patients who were classified as CAD-RADS N from the study may represent a source of bias. Sixth, our study has a high number of CAD-RADS modifiers. The interpretation of CCTA in patients with previously known CAD needs high experience, especially in the intermediate lesions and in-stent restenosis that may be overestimated. However, all reviewers in our study had high experience. Finally, the CAD-RADS is still uncommon and unfamiliar by many clinicians and is still under modification.

The CAD-RADS is valuable for improving CCTA structural reports with high diagnostic accuracy and high reproducibility, and very helpful for referring consultants for clinical decision-making.