Prognostic value of CAD-RADS classification by coronary CTA in patients with suspected CAD

This study population consisted of 11,356 consecutive suspected CAD patients who underwent coronary CTA for evaluation from January 2012 to December 2019 (Fig. 1). Patients included in our analysis meet the following inclusion criteria: (1) adults who were 18 years old or older; (2) refer for coronary CTA using a ≥ 64-detector row scanner; (3) good quality images acquired that could be diagnosed with standardized reporting of segmental coronary stenosis [10]. The following exclusion criteria were used: (1) repeated coronary CTA examination in the database (n = 876) and patients younger than 18 years (n = 7); (2) patients had congenital heart disease (n = 44); (3) patients had known CAD (prior MI, angiographically confirmed CAD, PCI or CABG, n = 561); (4) Coronary CTA datasets were not available (n = 44) or non-diagnostic image quality (including insufficient image quality, motion of one or more vessels, bad contrast and wrong bolus timing, n = 144) and (5) loss of follow-up (n = 86). At last, a total of 9625 patients were included in the final analysis.

Coronary CTA was performed using one of the four CT systems: Aquilion 64, Toshiba Medical Systems, Otawara, Japan; Philips Brilliance iCT, Philips Medical Systems, Best, the Netherlands; uCT 760, United Imaging, China; SOMATOM Definition AS, Siemens Healthineers, Germany which has been reported previously [11]. Heart rate control (HR ≥ 65 beats/min) was performed with beta-blockers before the scan. Scanning parameters were as following: tube voltage 120 kV, tube current 280–300 mAs. For contrast enhancement, 60–80 mL of iopromide (370 mgI/mL, Bayer Schering Pharma, Germany) followed by 30–40 mL of pure saline with a flow rate of 4–5 mL/s. The iodine contrast agent was automatically triggered into descending aorta of 100 HU threshold units. Then the scanning was performed during an inspiratory breath hold of 8–14 s after delay of 2 s. The reconstruction images were automatic send to one of the four workstations VITREA 2 (version 6.1, Vital Images, Inc, Minnesota, America), Intelligence space portal (Version 6.0.4, Philips Medical Systems, Best, the Netherlands), UIH Advanced Workstation (uWS-CT, R004, United Imaging Healthcare, Shanghai, China), SyngoVIA (Siemens Medical Solutions, Forchheim, Gemany).

First, according to the CAD-RADS consensus document [7], the standardized CAD-RADS classification was based on the highest degree of coronary stenosis and defined as follows: CAD-RADS 0: 0% stenosis, CAD-RADS 1: 1–24% stenosis, CAD-RADS 2: 25–49% stenosis, CAD-RADS 3: 50–69% stenosis, CAD-RADS 4A: 70–99% stenosis in 1 or 2 vessels, CAD-RADS 4B: Left main > 50% stenosis or 3-vessel disease, 70–99% stenosis, CAD-RADS 5: 100% stenosis or total occlusion. In order to restrict our analysis to patients without previously known CAD or revascularization, we did not include CAD-RADS modifiers to describe patients with stents (modifier S), or grafts (modifier G) in our analysis. Second, three traditional stenosis classifications and Duke Prognostic CAD Index were described as following [8, 9, 12]: Traditional CAD Classification 1, no CAD: 0% stenosis, mild CAD: 1–49% stenosis, moderate CAD: 50–69% stenosis in any major vessels/branch, severe CAD: ≥ 50% stenosis in LM or ≥ 70% stenosis in any major vessels/branch. Traditional CAD Classification 2, normal: 0% stenosis, mildly abnormal: 1–69% stenosis in any major vessels/branch, or 1–49% stenosis in LM, moderately abnormal: ≥ 70% stenosis in any major vessels/branch, severely abnormal: ≥ 70% stenosis in 2 or more vessel, or ≥ 50% stenosis in LM, or ≥ 70% stenosis in proximal left anterior descending (pLAD). Traditional CAD Classification 3, no CAD: 0% stenosis, nonobstructive CAD: 1–49% stenosis in any major vessels/branch, 1 vessel obstructive CAD: ≥ 50% stenosis in 1 major vessels/branch, 2 vessel obstructive CAD: ≥ 50% stenosis in 2 major vessels/branch, 3 vessel/LM obstructive CAD: ≥ 50% stenosis in 3 major vessels/branch, or ≥ 50% stenosis in LM. Duke Prognostic CAD Index, Duke CAD 0: 0% stenosis in all vessels, Duke CAD 1: 1–24% stenosis, or at most 1 with 25–49% stenosis, Duke CAD 2: ≥ 2 stenosis 25–49%, Duke CAD 3: 1 vessel with 50–69% stenosis, Duke CAD 4: 2 stenosis 50–69%, or 1 vessel with ≥ 70% stenosis, Duke CAD 5: 3 stenosis 50–69%, or 2 vessels with ≥ 70% stenosis, or pLAD stenosis ≥ 70%, Duke CAD 6: 3 vessels ≥ 70% stenosis, 2 vessels with ≥ 70% stenosis with pLAD, Duke CAD 7: LM stenosis ≥ 50%.

The primary endpoint was all-cause mortality. Follow-up procedures were approved by our hospital’s institutional review board (the Institutional Review Board of the Clinical Research Institute at The Central Hospital of Wuhan). Death status was ascertained by querying the local Community Health Service Centers. For death outside of the city, event was determined through telephone call, or review of medical records. The deadline date of follow-up was December 31, 2020.

Continuous variables were presented as mean ± SD. Categorical variables are presented as frequencies and percentages. We used Student’s t test for continuous variables between groups. Categorical variables were compared using a chi-square test or Fisher’s exact test as appropriate. Cox proportional hazards regression models were used to calculated time to death of all cause and hazard ratios (HR) with 95% confidence intervals (95% CI). Kaplan–Meier method was used to estimate cumulative event-free survival. Time dependent Receiver-operating characteristic (ROC) curves were used to evaluate the discriminatory value of traditional CAD classifications, Duke Prognostic CAD Index and CAD-RADS grading for the outcome. The calibration of models compares the agreement between the predicted and observed event rates. The Hosmer–Lemeshow goodness-of-fit test was employed to evaluate calibration. This test divided patients into deciles according to the risk scores of models and compared the predicted versus the observed rates of mortality. A significant P value indicates a lack-of-fit and suboptimal calibration. A two-tailed P < 0.05 was considered statistically significant. All statistical analyses were performed using Stata version 16 (StataCorp LP, College Station, Texas) and R statistical package (version 3.6.3, R foundation for Statistical Computing, Vienna, Austria).

Overall, 9625 patients were included in the final analysis. Figure 1 showed the reason for exclusion. Of the 9625 patients, the average age was 59.8 ± 10.7 years, and 44.3% (4262 of 9619) were male, the baseline clinical characteristics and coronary CTA characteristics of the study population in each group based on three traditional CAD classifications, Duke Prognostic CAD Index or CAD-RADS classification were shown in Table 1.

A total of 540 patients died from all causes with a median follow-up of 4.3 ± 2.1 years. Kaplan–Meier survival curves showed the cumulative events increased significantly associated with CAD-RADS, three traditional CAD classifications and Duke Prognostic CAD Index (log-rank test, all P < 0.001, Fig. 2). The univariate Cox regression analysis showed that compared to CAD-RADS 0 group, CAD-RADS 1, 2, 3, 4A, 4B, 5 were significantly associated with all-cause mortality (all P < 0.001).

In multivariate Cox regressions that were adjusted for gender and age, the risk for the all-cause death increased from HR 0.861 (95% CI 0.420–1.764) for CAD-RADS 1 to HR 2.761 (95% CI 1.961–3.887) for CAD-RADS 4B&5, using CAD-RADS 0 as the reference group. The relative HRs for all-cause death increased proportionally with the grades of the three traditional CAD classifications and Duke Prognostic CAD Index (Table 2).

The prognostic performance of CAD-RADS compared to the three traditional CAD classification of characterizing CAD extent/severity and Duke Prognostic CAD Index were shown in Fig. 3. The area under the time dependent ROC curve for prediction of all-cause death was 0.7917, 0.7805, 0.7991 for CAD-RADS in 1 year, 3 year, 5 year, respectively, which was similar compared to the traditional CAD classification 1 (0.7928 in 1 year, 0.7819 in 3 year, 0.7987 in 5 year; all P > 0.05), traditional CAD classification 2 (0.7945 in 1 year, 0.78 in 3 year, 0.7965 in 5 year; all P > 0.05), traditional CAD classification 3 (0.7926 in 1 year, 0.7812 in 3 year, 0.7996 in 5 year; all P > 0.05) and Duke Prognostic CAD Index (0.7930 in 1 year, 0.7811 in 3 year, 0.799 in 5 year; all P > 0.05). The five Classifications showed good calibration in the overall cohort (Hosmer–Lemeshow P value were 0.732, 0.71, 0.39, 0.505 and 0.91 for traditional CAD classification 1 traditional CAD classification 2, traditional CAD classification 3, Duke Prognostic CAD Index and CAD-RADS, respectively) (Additional file 1: Figure S1).