A Novel Risk Scoring Tool to Predict Saphenous Vein Graft Occlusion in patients with type 2 diabetes mellitus after Cardiac Artery Bypass Graft Surgery

Background: Coronary artery bypass grafting (CABG) success is reduced by graft occlusion. Patients with type 2 diabetes mellitus(T2DM) are more likely to develop graft occlusion. Understanding factors associated with graft occlusion may improve T2DM patient outcomes. The aim of this study was to develop a predictive risk score for saphenous vein graft (SVG) occlusion in T2DM patients after CABG. Methods: This retrospective cohort study enrolled 3716 CABG patients with T2DM from January 2012 to March 2013. The development cohort included 2477 patients and the validation cohort included 1239 patients. The baseline clinical data at index CABG was analyzed for their independent impact on graft occlusion in our study using Cox proportional hazards regression. The predictive risk scoring tool was weighted by beta coecients from the nal model. Concordance (c)-statistics and comparison of the predicted and observed probabilities of predicted risk were used for discrimination and calibration. Results: A total of 959 (25.8%) out of 3716 patients developed at least one SVG occlusion. Signicant risk factors for occlusion were male sex, estimated glomerular ltration rate<90, smoking (currently), hyperuricemia, dyslipidemia, prior percutaneous coronary intervention (PCI), a rising number of lesion vessels, and SVG. On-pump surgery, and the use of angiotensin-converting enzyme inhibitors (ACEI)/angiotensin receptor blockers (ARB) and calcium channel blockers (CCB) were protective factors. The risk scoring tool with 11 variables was developed from the derivation cohort, which delineated each patient into risk quartiles. The c-statistic for this model was 0.71 in the validation cohort. Conclusions: An easy-to-use risk scoring tool that used common clinical variables was developed and validated. The scoring tool accurately estimated the risk of late SVG occlusion in T2DM patients after CABG. The nal model was adjusted for age, obesity, hypertension, aspirin, P2Y12 inhibitor, complex coronary lesions. CI, Condence interval; SVG, saphenous vein graft; ACEI, angiotensin-converting enzyme inhibitors; ARB, angiotensin receptor blocker; CCB, calcium channel blocker; PCI, percutaneous coronary intervention, eGFR, estimated glomerular ltration rate. the large-scale cohort–based a predictive model occlusion in T2DM patients that could be used for risk stratication of CABG patients. The risk score system could inform clinical decision-making through calculation of individual risk for late SVG occlusion in T2DM patients. Assessment of the SVG risk score could improve surgical strategy and help in the development of personalized postoperative treatment plans. Proactive risk assessment and associated treatment may also be particularly cost-effective by reducing SVG occlusion and cardiovascular events in in T2DM patients after CABG.


Background
Coronary artery disease (CAD) is the leading cause of mortality in people with type 2 diabetes mellitus (T2DM) (1) .Coronary artery bypass grafting (CABG) surgery is a widely used treatment for complex CAD that improves patient outcomes and prognosis, which was associated with signi cantly lower long-term adverse clinical outcomes compared to PCI in patients with T2DM (2,3). However, some patients experience myocardial ischemia recurrence after CABG. A complex interaction is emerging between platelet function, antiplatelet drugs, coronary diseases and ischemia/reperfusion injury, especially in diabetic conditions (4). Studies have shown that the myocardial ischemic recurrence at 1 and 10 years after CABG are 17% and 63%, respectively (5,6). The main cause of myocardial ischemic recurrence is graft failure (7). Saphenous vein is the most widely used vascular conduit for CABG (8); however, the estimated rate of occlusion is as high as 42% at a mean follow-up of 7.5 years (9). Graft occlusion is associated with worse quality of life and reduced long-term survival. T2DM is one of the most important risk factors for graft occlusion. Although the patency of saphenous vein graft (SVG) has been assessed in several studies (10), the speci c risk factors for late SVG occlusion in patients with T2DM remain unclear. Previous studies have suggested that sex (11), chronic kidney disease (12,13), and off-pump surgery (14) affect graft patency. However, there is no precise scoring model for late SVG occlusion in patients with T2DM. The risk prediction model is an important tool for risk assessment. The risk scoring system is used to provide risk strati cation, identify high-risk patients, control risk factors, and inform strategies to reduce mortality and improve quality of medical care. The previous Saphenous Vein Graft Failure-An Outcomes Study in Coronary Artery Bypass Grafting (SAFINOUS-CABG) score is a simpli ed 12-variable risk scoring system that performs well in prediction of early SVG occlusion risk. However, the mechanisms of early and late occlusion are different. Late occlusion is due to development of atherosclerosis, which affects long-term clinical outcomes; the treatment is more di cult compared to early occlusion, and the risk factors are diverse. There is no standardized risk scoring system for late SVG occlusion. Therefore, it is of great clinical signi cance to establish a predictive risk scoring tool that accounts for speci city and accuracy. This study aimed to establish a risk scoring system that is suitable for CABG patients and evaluates the risk of late SVG occlusion.

Patients
Using the electronic medical system of Beijing Anzhen Hospital, we retrospectively identi ed 4021 patients with T2DM who underwent CABG surgery at our cardiac center between January 2012 and March 2013. Subsequently, we followed up with these patients between January 2017 and December 2017 to review and record their postoperative invasive angiography or coronary computed tomography angiography (CTA) results. Patients were excluded unless their medical records contained the following: (1) detailed preoperative angiographic results; (2) saphenous vein used for the graft; (3) results of postoperative invasive angiography or CTA; and (4) detailed information on patency and occlusion of SVG. A total of 3716 patients met the above criteria and were included in the nal analysis ( Fig. 1).2568 patients were symptomatic, 592 patients underwent invasive angiography or CTA because of acute coronary syndrome. 2526 patients underwent CTA, 1190 patients underwent invasive angiography. All data were retrieved from the electronic medical records system. Patient anonymity was ensured and this study was approved by the Institutional Review Board.

Endpoints
If a patient had at least one SVG occlusion on follow-up invasive angiography or CTA, we regarded the patient to have reached the primary endpoint. The invasive angiography or CTA result was reviewed by two or more experienced cardiologists and a radiologist independently.

Rationale for risk factor selection
Risk factors selection was based on previous studies (10,11,(15)(16)(17)(18) and clinical experience. Factors chosen were those easily measured and recorded. We recorded data on 38 relevant factors in this study (Table 1), and, after primary screening and Cox proportional hazards model analysis, 11 independent risk factors associated with late occlusion were selected. The construction of this predictive risk score was based on the method used in development of the Framingham risk score system (19). Development of the predictive risk score had three steps. First, to determine covariates that were independent risk factors for occlusion, we entered signi cant (P < 0.05) variables from the univariate analysis and/or those with clinical relevance into a multivariable Cox proportional hazards regression model using backward elimination with a critical P < 0.05. Second, to de ne the continuous risk factor, categories based on SVG number and number of vessels with lesions were used to determine a reference value (W ij ). Categorical risk factors were modelled using sets of indicator variables. The referent risk factor pro le (W iREF ) was considered "not at risk" of SVG occlusion, and de ned as category 1. We computed the distance from the category reference value to the referent value for each risk factor category by regression analysis using βi(W ij −W iREF ). The β coe cient of the SVG number (continuous variable) was used as a reference standard and assigned one point, with the constant B equaling 0.12. Finally, the points associated with each category of risk factor were calculated via Points ij = βi(W ij −W iREF )/B. The speci c risk of each score was then calculated according to the previously described formula (19).
Model validation had two steps: discrimination and calibration. Discrimination of the predictive risk model was assessed using the c-index, which is equivalent to the area under the receiver operating characteristic (ROC) curve for binary dependent variables (20,21), as an overall measure of model discrimination.

Missing data
Data for some variables were missing from our data set. Multiple imputation by chained equations (MICE) was used to impute missing values (23), which is superior to other methods (e.g. regression method, delete and mean method) and demonstrates stable performance.

Prediction modeling for late SVG occlusion
In the multivariate COX proportional hazard model, 11 variables were identi ed as independent predictors for SVG occlusion: sex, estimated glomerular ltration rate(eGFR) < 90, smoking, dyslipidemia, hyperuricemia, prior percutaneous coronary intervention (PCI), on-pump surgery, SVG number, lesion vessel number (including only vessels of the left anterior descending branch, circum ex artery, and right coronary artery), use of angiotensin-converting enzyme inhibitors (ACEI)/angiotensin receptor blockers (ARB), and use of calcium channel blockers (CCB; Table 2). Table 3 shows the β regression coe cient (βi), reference value (W ij ), referent risk factor pro le (W iREF ), nal point totals, and mean or proportion for each variable. Table 4 shows the cumulative risk score associated with risk of late occlusion, with the theoretical range of point values between − 9 to 24. Since there are few patients in the lower and upper ranges of the distribution, we shortened the risk table to avoid overstating the precision of the risk estimates. For example, a female patient (risk score = 4) with eGFR < 90 (2) and hyperuricemia (2), a history of percutaneous coronary intervention (PCI; 2), and was a triple-vessel lesion patient (8), would have a nal risk score of 18, with a predictive risk of 41% at 5 years of follow-up.

Summary measures of calibration and discrimination
Discrimination of the predictive risk model was assessed using the c-index. The nal predictive model had good performance for prediction of late SVG occlusion in the derivation cohort (c-index = 0.0.694; 95% CI, 0.674-0.714). The predictive model also had good performance in the validation cohort (c-index = 0.734; 95% CI, 0.7659-0.801).
Patients were classi ed into four groups representing the quartiles of risk. The rst to fourth quartiles contained 299, 342, 265, and 333 patients, respectively, and corresponded to a score of ≤ 5, 6-10, 11-13, and ≥ 14, respectively. Discrimination was good, as is shown in the plot of cumulative rate of late SVG occlusion for patients classi ed into each of the four risk groups (Fig. 2). The observed versus predicted rates of late SVG occlusion in the rst to fourth quartiles is shown in Fig. 3 There was an underprediction of SVG occlusion in the two lower risk groups, and a relatively precise prediction in the two higher risk groups. A modest underestimation in the lower probability range and a relatively precise estimation in the higher probability range of late SVG occlusion were also evident from calibration plots (Figs. 4 and 5). Figure 4 presents the observed late SVG occlusion rate versus model-predicted risk in groups based on the SVG occlusion risk score. In Fig. 5, the calibration plot presents the mean predicted risk of late SVG occlusion against the observed proportion of late SVG occlusion for 21 groups based on the calculated SVG occlusion risk score. Visually, it appeared to be a good calibration of both the predictive risk model and observed proportion across every risk score from 0-20.

Discussion
We developed a predictive risk score for saphenous vein graft SVG occlusion in patients with T2DM after CABG. We found that the model discrimination was good; in other words, the risk score was reliable in correctly classifying patients via risk strati cation. However, the calibration was not ideal, which caused a modest underestimation of SVG occlusion for low risk patients, while demonstrating a relatively precise predictive performance for high risk patients. There are a total of 11 independent predictive factors in the scoring system that leads to calculation of a CABG patient's personalized risk of developing late SVG occlusion. The risk score developed in this study could guide treatment strategy by focusing on the likelihood of late SVG occlusion in patients with T2DM after CABG with high-risk factors, such as hyperuricemia and the use of ACEI/ARB or CCB, which are not currently addressed in treatment recommendations. This risk score has signi cant implications for patients with T2DM after CABG, as those with higher risk scores should be managed with greater vigilance and intensive treatment to effectively mitigate cardiovascular risk factors.
CABG is one of the most effective revascularization strategies for CAD, especially for patients with T2DM and multivessel diseases, and has been shown to reduce mortality and improve quality of life (24). However, SVG occlusion has an adverse impact on the prognosis of patients and increases the economic burden of health care systems (9). Among all grafts, diabetes was associated with an increased risk of graft occlusion (25). SVG occlusion can be classi ed into two types: early and late. Early SVG occlusion is primarily attributed to a technical failure that results in graft thrombosis and hyperplasia as the SVG is arterialized. Late SVG occlusion is primarily due to generalized neointimal hyperplasia and atherosclerosis, which occurs over the injured endothelium (2), which was associated with diabetes progression. The risk factors associated with late SVG occlusion have been evaluated in some studies (10,12,14,18,26), but a widely accepted prediction model for late SVG occlusion in T2DM patients had not been previously developed. We have designed and validated a prediction model for late SVG occlusion in T2DM patients by using cohort data from a high-volume cardiac center. Previous studies have shown speci c risk factors from patient-related, graft-related, and surgery-related perspectives.
Female sex is an independent risk factor of SVG occlusion in early vein graft failure (11,15,17), possibly due to smaller target vessel diameter of female patients. Cardiovascular risk factors like smoking, dyslipidemia, and a history of PCI have also been identi ed as risk factors of late SVG occlusion (10,18). Uric acid level had never been considered relevant to SVG occlusion, however, clinical practice experience and prior research indicate that hyperuricemia may lead to kidney injury (27), and chronic kidney disease has been reported as a risk factor for vein graft disease (12,13). Off-pump surgery for CABG without cardioplegia has been associated with lower graft patency rates compared with on-pump surgery.
Additionally, the coagulopathy and platelet dysfunction induced by cardiopulmonary bypass can affect SVG patency (26,28). From the graft-related and surgery-related perspectives, any use of SVGs is independently associated with reduced survival after coronary artery bypass surgery (29), which is consistent with the risk factors we have derived. Most patients with diseases with multiple lesions have diffuse lesions, suggesting that the condition of the graft may be poor; these patients have a higher rate of late SVG occlusion. As for medications, the use of ACEI/ARB and CCB are both protective factors for SVG occlusion, which may be related to the dilation of blood vessels, antispasmodics, and increased graft diameter. Furthermore, the effects of antihypertensive medications may contribute to reduced risk of SVG hyperplasia, which has been demonstrated in a study on early SVG occlusion (30).
A variety of cardiac surgery risk prediction models have been established, including the Society of Thoracic Surgeons (STS) score (31), the American College of Cardiology/American Heart Association (ACC / AHA) score (32), the European EuroSCORE(European System for Cardiac Operative Risk Evaluation score) (33), and its modi ed version, the EuroSCORE II (34). These prediction models were primarily used for evaluating perioperative risk. For SVG occlusion, the SAFINOUS-CABG score (16)       Observed rate of occlusion with 95% con dence interval vs. model-predicted risk of occlusion in groups. Legend: SVG: saphenous vein graft.