Coronary atherosclerosis scoring with semiquantitative CCTA risk scores for prediction of major adverse cardiac events: propensity score -based analysis of diabetic and non -diabetic patients

Aims


Introduction
Diabetes mellitus (DM) is a well-established risk factor for coronary artery disease (CAD) as described by global guidelines. 1,2Patients with DM exhibit higher burden of coronary plaque and have higher adverse event-rates as compared to patients without DM. 3,4Risk scores derived from general chest-pain populations are often challenging to use in DM patients, because of many confounders that are associated with CAD. 1 For example, the universal percentiles of coronary artery calcium from the Multi-Ethnic Study of Atherosclerosis (MESA) do not directly apply to diabetic patients. 57][8][9] Coronary plaque characteristics by CCTA (e.g.1][12][13][14] Recently, it has been demonstrated that the Leiden CCTA risk score, which grades all the aforementioned features of coronary atherosclerosis, improves prediction and reclassification of adverse events as compared to the stenosis severity component of the coronary artery disease-reporting and data system (CAD-RADS). 15However, it remains uncertain if comprehensive atherosclerotic scores derived from general chest-pain populations apply well to a specific cohort of DM patients.Therefore, the aim of the current study is to compare a subset of semiquantitative CCTA risk scores and their prognostic value between patients with and without DM.

Study design and population
Out of a 14,895-patient cohort from the Leiden University Medical Center (LUMC) and the COronary CT Angiography EvaluatioN For Clinical Outcomes: an InteRnational Multicenter (CONFIRM) registry at 17 sites in 9 countries with 5-year follow-up data, we performed a secondary analysis in 2,900 DM patients with suspected CAD who were clinically referred for CCTA between 2002 and 2015. 15,16Diagnosis of DM was defined as fasting glucose ≥126 mg/dL and/or treatment with insulin or hypoglycemic medication. 1 Demographic and clinical data were prospectively collected from departmental electronic information systems and retrospectively analyzed, which was approved by institutional review boards or ethics committees at each participating site.All patients provided written informed consent, except for the patients at the LUMC where this need was waived by the institutional review board.For the present study, the following additional exclusion criteria were applied: 1) an uninterpretable CCTA examination, 2) prior percutaneous coronary intervention or coronary artery bypass grafting, 3) CCTA in the setting of suspected acute coronary syndrome, 4) missing coronary plaque data for score calculation, and 5) missing follow-up data (n = 2,168).Thus, 732 DM patients were included and 1:1 propensity-matched with non-DM patients from the original cohort.

CCTA acquisition and image analysis
Patients were scanned with ≥64-slice CT scanners, and protocols with regard to the acquisition and post-processing of scans were previously published. 6,15,16At the LUMC, scans were analyzed according to a 17-segment modified American Heart Association (AHA) model of the coronary artery tree by consensus of experienced physicians. 10Qualitative analysis of all diseased segments was performed. 13Coronary plaque composition was defined as calcified for plaques with high density, non-calcified for plaques with lower density than the contrastenhanced lumen and mixed for plaques with both characteristics.Stenosis severity was categorized as normal, <30%, 30-50%, 50-70%, 70-99% and 100%.System dominance was determined upon the origin of the posterior descending artery as part of either the right coronary artery or left circumflex artery.For the CONFIRM registry, image analysis was systematically performed according to the Society of Cardiovascular Computed Tomography (SCCT) guidelines at the time. 17,18

Semiquantitative CCTA risk scores
For all patients, a subset of 7 semiquantitative CCTA risk scores was calculated: 1) any stenosis ≥50%, 2) any stenosis ≥70%, 3) stenosis severity component of the CAD-RADS, 4) segment involvement score (SIS), 5) segment stenosis score (SSS), 6) CT-adapted Leaman score (CT-LeSc), and 7) Leiden CCTA risk score.Any stenosis ≥50% or ≥70% was scored in a binary fashion.The stenosis severity component of the CAD-RADS was stratified into 3 groups according to previously published methods for reasons of uniformity and sample size 15 : no to minimal CAD (i.e.CAD-RADS 0-1), moderate CAD (i.e.CAD-RADS 2-3) and severe CAD (i.e.CAD-RADS 4-5).No high-risk plaque features were incorporated into this classification as these were not consistently evaluated in this study population.The SIS corresponded to the total number of diseased segments, irrespective of stenosis severity (range 0-17). 19The SSS graded stenosis severity from 0 to 3 in each individual segment and summed this into a continuous score (range 0-48). 19The CT-LeSc graded composition, stenosis and location in each individual segment and merged this into a continuous score (range 0-33). 14,20The Leiden CCTA risk score graded in each individual segment in consecutive order: presence and composition (i.e.plaque weight factor, range 0-1.3), stenosis (i.e.stenosis weight factor, range 1.0-1.4)and location according to system dominancy, major epicardial artery and distance from ostium (i.e.location weight factor, range 0-6) (Appendix Supplement 1, online calculator available at http://18.224.14.19/ calcApp/). 15The 3 weight factors were multiplied to compute individual segment scores, and summation of these scores resulted in a continuous score (range 0-42).Further, this continuous score was stratified into 3 groups that were proven to discriminate adverse events best: 0-5, 6-20 and > 20. 15 Moreover, plaque weights, stenosis weights and location weights were summed to create per-patient weight scores.Per-patient weight scores were divided by the number of segments with coronary plaque to create per-segment weight scores (only when plaque was observed).

Study endpoints
The primary endpoint was a composite of all-cause death and nonfatal myocardial infarction (MI).Non-fatal MI was defined according to standard definitions and/or current guidelines. 21,22Methodology on how mortality and follow-up data were documented was previously reported. 15

Statistical analysis 2.4.1. Propensity-matching
Propensity-matching of DM and non-DM patients was performed in a 1:1 ratio in order to detect the pure effect of diabetes on the CCTA risk scores.A propensity score was calculated to predict DM from the probabilities of a multivariable logistic regression model including age, sex, cardiovascular risk factors and medication.In case of missing variables, relaxed models were used to create as many matches as possible.A total of 732 DM patients was matched to 732 non-DM patients with this propensity score by the Matchit nearest-neighbor matching algorithm. 23,24In all matched patients the balancing property was satisfied.

General
Continuous data are reported as means ± standard deviations (SD), independent upon distribution for reasons of uniformity.Categorical data are reported as counts with percentages.Continuous data were compared with the paired T test or paired Wilcoxon Signed Rank test, where appropriate.Categorical data were compared with the McNemar's test.Uni and multivariable Cox-regression analysis was performed to assess the association between the 7 semiquantitative CCTA risk scores and the primary endpoint.To avoid overfitting of the multivariable model, backward selection with the Akaïke information criterion was used for selection of clinical variables.In DM-patients, also area under the receiver-operating characteristics curves (AUC) between the scores were compared with the DeLong's test to evaluate discriminatory ability.With regard to the Leiden CCTA risk score, also survival analysis with the Kaplan-Meier method was performed.Eventfree survival curves were compared with the log-rank test.All statistical tests were 2-sided and a p-value of <0.05 indicated statistical significance.All analyses were performed with R (version 3.3.2,R Development Core Team, Vienna, Austria) and SPSS software (version 25, SPSS IBM Corp., Armonk, New York).

Study population
A total of 1,464 DM and non-DM patients (mean age 58 ± 12 years, 40% women) underwent CCTA and had a median follow-up of 5.1 years (interquartile range 2.2-6.2years).The primary endpoint was documented in 155 (11%) patients, of which 95 (7%) and 60 (4%) in patients with and without DM, respectively.DM patients were largely comparable to non-DM patients with regard to age, sex, cardiovascular risk factors and medication, except for the prevalence of hypercholesterolemia and statin therapy (Table 1).However, on CCTA, patients with DM demonstrated more obstructive CAD than patients without DM, whilst no or non-obstructive CAD was less frequently observed (p < 0.001) (Table 2).

Discussion
The present study is a propensity score-based analysis of 1,464 DM and non-DM patients with suspected CAD, who were clinically referred for CCTA and followed for all-cause death and non-fatal MI during a median of 5.1 years.We compared 7 semiquantitative CCTA risk scores -which score presence, extent, composition, stenosis and/or location of coronary atherosclerosis -and their prognostic value between both groups.All scores were independently associated with the primary endpoint in both patients with and without DM, with non-significant interaction between the scores and diabetes.Particularly, the discriminatory ability of semiquantitative CCTA risk scores that weighted stenosis and incorporated the full extent of CAD, such as the CAD-RADS, SIS, SSS, CT-LeSc and Leiden CCTA risk score, was superior to the binary evaluation of obstructive stenosis in patients with DM.These findings might be clinically useful for risk stratification of this specific patient population.

Semiquantitative CCTA risk scores in DM
Prior studies evaluated semiquantitative CCTA risk scores in diabetic patients, mainly the presence of obstructive stenosis, the SIS and the SSS. 4,25Hadamitzky et al. studied 1,922 patients without known CAD who underwent a clinically indicated CCTA at a single center: 140 patients with DM and 1,782 patients without DM. 25By analyzing both groups, more obstructive CAD (48% vs. 26%, p < 0.001) and a higher SIS (5.2 ± 3.7 vs. 2.9 ± 3.2, p < 0.001) were observed in DM patients as compared to non-DM patients.After a mean follow-up of 33 months, the SIS remained independently predictive of the primary endpoint of all-cause death, non-fatal MI and unstable angina requiring hospitalization in all patients.In addition to this, in a previous analysis of the CONFIRM registry, Rana et al. selected 11,110 patients without known CAD who underwent CCTA. 4 A total of 3,370 patients with DM were 1:2 propensity-matched with 6,740 patients without DM by age, sex and cardiovascular risk factors.DM patients demonstrated less no or nonobstructive CAD (63% vs. 73%, p = 0.041), more obstructive CAD (37% vs. 27%, p < 0.001), a higher SIS (2 vs. 1, p < 0.001) and a higher SSS (3 vs. 2, p < 0.001) as compared to non-DM patients.Our results, which 1) included additional patients from the LUMC to the CONFIRM registry, and 2) had greater restrictions with regard to system dominancy, composition, stenosis and location for score calculation, were overall very consistent with the aforementioned findings.
Only a few studies investigated the CT-LeSc next to the presence of obstructive stenosis, the SIS and the SSS in diabetic patients.For instance, Gonçalves et al. evaluated 581 patients without known CAD who underwent CCTA at a single center: 85 patients with DM and 496 patients without DM. 26Comparable to the abovementioned findings and our analysis, DM patients demonstrated less no or non-obstructive CAD (68% vs. 90%, p < 0.001), more obstructive CAD (32% vs. 10%, p < 0.001) and a higher prevalence of SIS>5 (37% vs. 13%, p < 0.001), SSS>5 (25% vs. 5%, p < 0.001) and CT-LeSc>8.3(41% vs. 16%, p < 0.001) as compared to non-DM patients.Whether these CCTA risk scores were predictive of adverse events was not tested in this study.However, the long-term prognostic value of the CT-LeSc with regard to hard endpoints (e.g.non-fatal MI, all-cause death, cardiac death) has been established in other patient populations, such as a general chest-pain population and patients with non-obstructive CAD. 14,20o prior studies evaluated the Leiden CCTA risk score in diabetic patients.Recently, van Rosendael et al. established the prognostic importance of the Leiden CCTA risk score for adverse events (i.e.all-cause death, non-fatal MI) in a large observational study of 2,134 patients with suspected but without known CAD. 15 When the Leiden CCTA risk score was added to a selection of classical cardiovascular risk factors, both the discrimination of adverse events (AUC 0.768 vs. 0.742, p = 0.001) and reclassification of patients (net reclassification improvement 12.4%, p < 0.001) increased compared to the stenosis severity component of the CAD-RADS plus the same risk factors.Also, this discriminatory ability was reproduced in an external validation cohort.To this end, we hypothesized that the Leiden CCTA risk score might not be applicable to DM patients, because of various confounders that are associated with CAD. 1,5Our analysis proved that the Leiden CCTA risk score was independently predictive of adverse events, and importantly, that this predictive value did not differ in the presence or absence of DM.Also, the Leiden CCTA risk score and per-segment and per-patient weight scores were evaluated in DM and non-DM patients, in order to determine the contribution of its components to the total score.By doing so, it was demonstrated that the Leiden CCTA risk score and all the per-patient weight scores were significantly higher in patients with DM, whilst the per-segment location weight score was lower.8][29][30] For example, per-patient weight scores will increase in case of more diseased segments, as they are by definition the sum of all plaque weights, stenosis weights and location weights within a patient.When normalized for the total extent of CAD, only the per-segment location weight score remained significantly lower in diabetic patients.Though, this difference was numerically very modest.

Clinical implications
2][33][34] Accordingly, current global guidelines underlined that patients with DM should be considered at high-risk for cardiovascular disease, and at very high-risk with ≥1 other cardiovascular risk factors or end organ damage. 1,35Hence, multiple studies declared DM as an equivalent of CAD. 2,3,36,37These statements were evaluated through a large systematic review and meta-analysis by Bulugahapitiya et al. 38 In this analysis, 13 cohort-and observational studies involving 45,108 patients were included: 21,675 DM patients and 23,433 non-DM patients.This study, with mean follow-up of 13.4 years, demonstrated that for DM patients without prior MI the risk for fatal or non-fatal MI was 43% lower than in non-DM patients with prior MI.Thus, they did not support the hypothesis of DM as a CAD-equivalent.
Although our results showed that patients with DM exhibited higher overall burden of coronary atherosclerosis, all semiquantitative CCTA risk scores were still able to predict the primary endpoint of all-cause death and non-fatal MI accurately.Especially scores incorporating the total extent of CAD performed particularly well.Additional results with regard to the extent atherosclerotic disease and the survival of diabetic and non-diabetic patients are available in Appendix Supplement 3.

Limitations
First, this was a nested case-control study with all the intrinsic limitations of an observational cohort study like unmeasured confounding factors and selection bias.Second, the event-rate of the primary endpoint was relatively low, and therefore the present study was underpowered to enter a multitude of variables into the multivariable model.Though, by employing the backward selection method overfitting of this model was avoided.Third, we only performed qualitative analysis or visual categorization of all diseased segments within patients.Quantitative analysis of coronary plaque might capture the full extent of atherosclerotic disease more precisely or will provide additional information. 39Fourth, recent studies addressed the value of serial CCTA to detect not only coronary plaque growth but also the progression of high-risk or vulnerable plaques in order to evaluate the natural history of the atherosclerotic process in patients with DM. 40 Our study only ascertained scans at a single timepoint.

Table 1
Baseline characteristics of study population.
Abbreviations: ACE, angiotensin converting enzyme-inhibitor; BMI, body mass index; CAD, coronary artery disease; DM, diabetes mellitus.Definitions: * Blood pressure ≥140/90 mmHg and/or treatment with antihypertensive medication; † Total cholesterol ≥230 mg/dL or triglycerides ≥200 mg/dL and/or treatment with lipid-lowering medication; ‡ Presence of coronary artery disease in first-degree family members at age <55 years in males and <65 years in females.Abbreviations: CAD, coronary artery disease; DM, diabetes mellitus; LM, left main artery.

Table 3
Semiquantitative CCTA risk scores stratified by DM.

Table 4
Leiden CCTA risk score and weight scores stratified by DM.

Table 5
Cox-regression analysis stratified by DM.