Back to Annual Symposium Program

Introduction: The goal of our study was to construct and validate a robust risk prediction model for patients undergoing carotid endartrectomy (CEA).
Methods: Vascular Group of New England (VSGNE) and Vascular Quality Initiative (VQI) databases were queried for patients who underwent CEA. Pre-operative variables including age, gender, pre-admission living, diabetes, dialysis, history of congestive heart failure, coronary artery disease, peripheral arterial disease, use of statins, beta-blockers, urgency of the operation and symptomatic neurological status were entered into a logistic regression model as predictors of the composite adverse outcomes. Backward elimination (alpha level of 0.5) was then used to select a more parsimonious model. Calibration was performed to measure how closely predicted outcomes agree with observed outcomes. To assess calibration we used Hosmer-Lemeshow for dichotomous outcomes, R2 for continuous outcomes and a modified version of Hosmer-Lemeshow for time to event data. The predictive value of the model was assessed via C-statistic. The external validation was then performed using VQI sample after excluding those in VSGNE sample (VQI-VSGNE) following similar method. Chi-square test was used to compare the two groups.
Results: A significantly higher rate of adverse outcomes was noted for the VQI sample (5.21%, n=12,075) compared with VSGNE sample (4.49%,n=8,661) (p<0.017). Table below summarizes the risk models in the VSGNE cohort. The discriminating ability of the model on the VSGNE is substantial (C=0.712) and the model fit is good (Hosmer-Lemeshow lack of fit test p=0.494). The discriminative ability of the VSGNE model remained substantial in the external data (VQI-VSGNE) (C=0.703) (Figure).
CONCLUSION: The internally validated VSGNE CEA risk model was externally validated by testing it against the remainder of VQI on patients who underwent CEA by a diverse array of physicians. This tool provides a simple and reliable method to risk stratify CEA patients using their pre-operative conditions. A risk score based on this model can reliably stratify patients according to their risk of adverse outcomes after CEA.
Table -VSGNE model: Parsimonious 0.5 level (C=0.712, Hosmer-Lemeshow p=0.494)

	Adj. Means Ratio	Lower Confidence Limit	Upper Confidence Limit	p value
Age: 75+ vs. <75	1.6947	1.3720	2.0933	<0.0001
Male vs. Female	0.9011	0.7284	1.1148	0.3374
Pre admission: Nursing Home vs. Home	7.7144	4.7317	12.5773	<0.0001
Diabetes	1.3992	1.1256	1.7393	0.0025
Congestive heart failure	1.5379	1.1220	2.1079	0.0075
Peripheral arterial disease	1.2545	0.9892	1.5909	0.0614
[[Unsupported Character - Symbol Font β]]-blockers	1.2277	0.9367	1.6092	0.1372
Symptomatic	2.0953	1.6734	2.6235	<0.0001
Urgent vs. Elective	2.8054	2.1790	3.6118	<0.0001