Practical cardiovascular risk calculator for asymptomatic patients with type 2 diabetes mellitus: PRECISE‐DM risk score

Abstract Background Obstructive coronary artery disease (OCAD) is a significant predictor of adverse clinical events in asymptomatic patients with type 2 diabetes mellitus (T2DM). Hypothesis We sought to develop an easy‐to‐use risk scoring system to predict OCAD and long‐term clinical outcome in asymptomatic patients with T2DM (PRECISE‐DM). Methods A total of 2799 asymptomatic patients with T2DM and no prior coronary disease were consecutively enrolled. OCAD was defined as ≥50% coronary artery stenosis on coronary computed tomography angiography (CCTA). A new risk scoring system was developed in 933 patients undergoing CCTA (derivation cohort) and its performance to predict OCAD and major adverse cardiac and cerebrovascular event (MACCE) was compared with other risk estimates. The scoring system was externally validated in 1899 patients not undergoing CCTA (validation cohort). Results The PRECISE‐DM scoring system was created using seven variables that were associated with increased risk of OCAD, with scores ranging from 0 to 9. The scoring system predicted presence of OCAD with a C‐statistic of 0.680 and risk of MACCE with a C‐statistic of 0.708. The UKPDS risk engine and the Framingham risk score showed unreliable performance in prediction of OCAD (C‐statistics 0.531 and 0.577, respectively). Calcium score was highly predictive for OCAD (C‐statistic 0.825) but showed only modest accuracy in predicting MACCE (C‐statistic 0.675). In the external validation cohort, the PRECISE‐DM score showed acceptable discrimination for prediction of MACCE (C‐statistic 0.707). Conclusions The PRECISE‐DM scoring system accurately predicted presence of OCAD and risk of MACCE in asymptomatic patients with T2DM.


| INTRODUCTION
Despite advancement in medical treatment options including antiplatelet agents and statins, coronary artery disease (CAD) is still a significant threat to patients with diabetes in terms of morbidity and mortality. 1 Accurate prediction and early detection of obstructive CAD in asymptomatic patients are particularly important because CAD often progresses without symptoms in diabetic patients. The prevalence of silent significant CAD in diabetic patients was 22% to 33% in studies using myocardial perfusion imaging, 2,3 and was up to 50% in an autopsy study. 4 Multi-slice coronary computed tomography angiography (CCTA) provides accurate non-invasive imaging of the extent and severity of CAD. Among asymptomatic patients with type 2 diabetes mellitus (T2DM), CCTA detected obstructive CAD in 40% of the subjects. 5 The presence of obstructive CAD on CCTA also showed a significant correlation with future cardiovascular events. 5,6 However, the FACTOR-64 randomized trial found that routine indiscriminate screening by CCTA in asymptomatic patients with T2DM failed to improve clinical outcome. 7 Considering the low prevalence of severe coronary stenosis (overall 10.6%) and cardiac death rate (1.5%) reported in the FACTOR-64 study, intensive diagnostic or therapeutic approach would be only beneficial in selective patients at high risk of obstructive CAD in an asymptomatic diabetic population.
In real world practice, a reliable risk prediction model using only clinical variables may confer higher cost-benefit in identifying individuals who need early intervention for cardiovascular disease. However, the well-known cardiovascular risk prediction models in the general population, such as the Framingham risk estimate 8 or DECODE, 9 gave an unreliable performance in asymptomatic diabetic patients with a greater than 30% underestimation of CAD risk. 10 The UKPDS risk engine was developed as a more diabetes-specific risk prediction model for CAD. 11 However, it showed only modest accuracy in predicting coronary heart disease events in external validation studies and has a significant disadvantage as a practical usage tool because of the complex computation process. 12,13 This study aimed to develop a new scoring system for PREdicting

| External validation cohort
Within 3 months of the primary enrollment, two age-and sexmatched patients per enrolled patient in the derivation cohort were enrolled as an external validation cohort. The inclusion and exclusion criteria were same as the derivation cohort but CCTA was not performed in patients enrolled in the validation cohort. This study was approved by the institutional review board of the Seoul St. Mary's Hospital and performed in accordance with Strengthening the Reporting of Observational Studies in Epidemiology guidelines. 16 The written informed consent from the patients was waived by the institutional review board as only anonymized data were accessed and analyzed. All scans were analyzed by two experienced radiologists who were blinded to patient clinical information. In accordance with the guidelines of the Society of Cardiovascular Computed Tomography, coronary segments were visually scored for the presence of coronary plaques using a 16-segment coronary artery model in an intent-todiagnose manner. 17 Segments were included in the analysis if the diameter was >1.5 mm. The severity of luminal diameter stenosis was scored as none (0% luminal stenosis), nonobstructive (plaques with a lumen narrowing <50%), or obstructive (plaques with maximum stenosis ≥50%). Obstructive CAD in the diagonal branches, obtuse marginal branches, and posterolateral branches was regarded as part of the corresponding major epicardial coronary artery system. The number of diseased vessels was categorized as one, two, three, or left main (LM) coronary artery vessels. The severity of coronary artery calcification was scored using the method developed by Agatston. 18

| Data collection and outcome analysis
Included patients underwent a structured interview for past medical history, laboratory testing and 12-lead ECG before the CCTA examination. The diagnosis of T2DM was based on the 2010 criteria of the American Diabetes Association, and was defined as fasting glucose ≥126 mg/dL, HbA1c ≥6.5%, and/or postchallenge glucose ≥200 mg/ dL. 19 Patients with a self-reported or documented history of T2DM under treatment with oral hypoglycemic agents or insulin were also considered to have diabetes. Chronic kidney disease was defined as estimated glomerular filtration rate <60 mL/min/1.73 m 2 . Abnormal

| Statistical analyses
Continuous variables are presented as mean ± SD and compared using Student's t-tests. Categorical variables are presented as counts with percentages (%) and compared by the Chi-square test or Fisher's exact test. We used multiple imputations to replace missing values using fully conditional specification approaches based on all candidate predictors and conducted 20 multiple imputations with 50 resampling replications, creating 1000 full datasets. Multivariate logistic regression analysis was used to adjust the risk of obstructive CAD and to identify independent predictors among baseline variables. All significant variables in univariate analysis were considered candidate predictors for the final multivariate logistic regression model. Continuous variables including age, diabetes duration, and HbA1c were categorized by the cutoff with the best discrimination value in the receiver operating characteristic curve.
Eight variables (age, sex, prior stroke, hypertension, diabetes duration, HbA1c, use of clopidogrel, and abnormal ECG) were retained in the multivariate model but use of clopidogrel was excluded to avoid multicollinearity. Performance of the final prediction model was evaluated using area under the curve (AUC) analysis and the Hosmer-Lemeshow goodness-of-fit test. The risk score was calculated by dividing each regression coefficient (β) by the smallest regression coefficient from the final model and then rounding that number to the nearest integer. The total risk score was calculated for each patient by summation of the score points. The internal validity of the scoring system was assessed by the simulation study, which was formed with 1000 iterations of random partitioning of the data into training and validation sets (50:50 train/test split). The risk score obtained from the training data was applied to the samples in the validation set and the corresponding risk strata were predicted for each sample. This process was iterated 1000 times and the average prediction rate was calculated. 22 Survival analysis using Cox regression was used to assess the risk of clinical endpoints. Discrimination values of the prediction model for MACCE and all-cause death were estimated using the Harrell's overall C-index. 23 All analyses were two-tailed, and P-values<.05 were considered to indicate statistical significance. Statistical analyses were performed using the SAS software, version 9.4 (SAS Institute, Cary, North Carolina).

| Baseline characteristics
A total of 933 patients were enrolled in the derivation cohort and underwent CCTA. Baseline characteristics of the derivation cohort are summarized in Table S1. Mean age was 63.4 (±9.6) and 556 (59.6%) were male.
Mean duration of diabetes was 11.7 (±9.2) years. Obstructive CAD was detected by CCTA in 374 (40.1%) patients. Among the baseline variables, older age, male sex, longer duration of diabetes, higher HbA1c level, history of hypertension, and prior stroke were significantly associated with presence of obstructive CAD. There was no difference in BMI, smoking ratio, prevalence of dyslipidemia, or serum cholesterol level between patients with and without obstructive CAD. ECG abnormality was more frequently observed in patients with obstructive CAD (25.4% vs 13.9% in the group with obstructive CAD and without, respectively, P < .001).
More patients with obstructive CAD were receiving insulin therapy (29.7% vs 17.7%, P < .001) and clopidogrel (8.8% vs 1.9%, P < .001). The prescription rates of aspirin, beta-blockers, angiotensin-converting enzyme inhibitor/angiotensin receptor blockers, and statins did not differ between patients with and without obstructive CAD.
The external validation cohort consisted of 1866 patients who did not undergo CCTA at enrollment. The patients had a shorter duration of diabetes, lower HbA1c level, a lower prevalence of dyslipidemia and a higher prevalence of chronic kidney disease compared to the patients in the derivation cohort (Table S2). The prescription rate of aspirin and statin was lower, and the rate of insulin therapy was higher in the validation cohort.

| Development of the prediction model
Among all significant predictors for presence of obstructive CAD in univariate logistic regression analysis, we found 7 factors (age ≥ 70, male gender, hypertension, prior stroke, diabetes duration ≥10 years, HbA1c ≥7.0, and abnormal ECG) that were associated with increased risk of obstructive CAD in a multivariate analysis (Table 1)

| Validation of the PRECISE-DM risk score in the internal derivation cohort
The C-statistic of the PRECISE-DM risk score was 0.680   Table 3 and Figure 2).

T A B L E 1
Univariate and multivariate analyses of the predictors for obstructive CAD

| Comparison between PRECISE-DM score and other risk predictors
In the internal derivation cohort, the risk of coronary heart disease event calculated by the UKPDS risk estimates was 10.1 (±8.6) %. The C-statistic of the UKPDS risk estimates was 0.531 for prediction of obstructive CAD and 0.618 for prediction of MACCE in the derivation cohort (Table S3). In the validation cohort, the UKPDS risk estimate showed also lower performance than PRECISE-DM score in predicting MACCE (C-statistic 0.653).

| DISCUSSION
This new PRECISE-DM risk score was developed and validated using risk factors associated with obstructive CAD on CCTA and applied to prediction of long-term clinical outcome in asymptomatic patients with T2DM. This score was intentionally designed to select patients at high-risk of obstructive CAD among an asymptomatic diabetic population, that may benefit from early diagnostic or therapeutic intervention. The PRECISE-DM score is an easy-to-perform, user-friendly

| Limitations
First, the sample size of our study is modest compared to the previous studies for risk model development, mainly because of exclusive inclusion of asymptomatic diabetic patients undergoing CCTA. Second, patients with moderate to severe chronic kidney disease in whom the probability of obstructive CAD is high were excluded due to requirement of contrast agent administration for CCTA.

| CONCLUSION
We developed and validated the PRECISE-DM risk score as a straightforward and practical clinical scoring system for predicting both a

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of this article.