2.1 Data
CEA was identified as a primary procedure using current procedure terminology code of 35301. Patients with additional procedures performed at the time of CEA were not excluded
a priori, but by considering CEA only as a principal procedure, cases where CEA was performed secondarily to a major procedure, such as coronary artery bypass graft surgery, were excluded. This process identified 34,493 CEAs performed between the years 2005 and 2010 from all participating institutions. These cases represent only a fraction of the total CEAs performed during these years since NSQIP only samples cases at participating hospitals and is not all inclusive. [
17- Fink A.S.
- Campbell Jr., D.A.
- Mentzer Jr., R.M.
- et al.
The National Surgical Quality Improvement Program in non-veterans administration hospitals: initial demonstration of feasibility.
,
18- Khuri S.F.
- Daley J.
- Henderson W.
- et al.
The Department of Veterans Affairs' NSQIP: the first national, validated, outcome-based, risk-adjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program.
]. Most of these procedures were performed by vascular surgeons (
N = 32,848) and general surgeons (
N = 1,645). Analyses were stratified by these two surgical specialties.
All patient characteristic and outcome data were taken from the NSQIP database collected using standard NSQIP methodology [
17- Fink A.S.
- Campbell Jr., D.A.
- Mentzer Jr., R.M.
- et al.
The National Surgical Quality Improvement Program in non-veterans administration hospitals: initial demonstration of feasibility.
,
18- Khuri S.F.
- Daley J.
- Henderson W.
- et al.
The Department of Veterans Affairs' NSQIP: the first national, validated, outcome-based, risk-adjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program.
]. Data were collected by a trained staff of surgical clinical nurse reviewers who worked in conjunction with the surgeon champion for accurate data collection. Uniformity was maintained through the use of an operation manual, which outlined data collection procedures and variable definitions, as well as routine conference calls, site visits, and annual meetings [
[19]- Boltz M.M.
- Hollenbeak C.S.
- Julian K.G.
- et al.
Hospital costs associated with surgical site infections in general and vascular surgery patients.
]. Surgeon specialty was assigned by the surgical clinical nurse reviewer using either the surgical service line most closely associated with the principal operative procedure or the surgeon's self-declared specialty [
[20]American College of Surgeons. (2013). ACS NSQIP Operations Manual: 26-27.
]. For CEA, if a surgeon was board certified in both vascular and general surgery, the surgeon was considered a vascular surgeon [
[20]American College of Surgeons. (2013). ACS NSQIP Operations Manual: 26-27.
].
Of the 60 patient characteristics collected for the NSQIP database, variables that had the greatest relevance to the CEA procedure were selected. Preoperative characteristics included age, sex, race/ethnicity, anesthesia type, American Society of Anesthesiologists (ASA) class, operation time, and comorbidities (
Table 1). Age was divided into quartiles including 20–64, 65–74, 75–79, and 80+ years. Similarly, operation time, recorded in minutes, was divided into quartiles including <84, 85–109, 110–139, and 140+. Race/ethnicity was stratified by white (non-Hispanic), black (non-Hispanic), Hispanic (including all Hispanic ethnicities), and other (including all other races recorded in the database: American Indian, Alaska Native, Asian, Pacific Islander, Native Hawaiian, or unknown race not of Hispanic origin). Anesthesia type was divided into four categories, the most common anesthesia type being general (
N = 29,077), followed by regional (
N = 3,709), then monitored (
N = 1,372), then all the other types (
N = 335), including spinal, epidural, other, and no anesthesia. Most patients were rated at an ASA class 3 (
N = 26,801) or class 4 (
N = 4,582), with very few rated at class 1 (
N = 54) or class 5 (
N = 12). For statistical analyses, ASA class was divided into two categories, “1, 2, and 3” and “4 and 5.” Comorbidities selected include diabetes, smoking, previous Percutaneous coronary intervention (PCI), previous previous cardiac surgery (PCS), hypertension requiring medication, and history of congestive heart failure (CHF), MI, angina, peripheral vascular disease (PVD), or CVA.
Table 1Summary statistics of patients undergoing carotid endarterectomy stratified by surgeon specialty.
Hx = History; HT = Hypertension.
In addition to LOS, 30-d postoperative mortality, and any outcome variables, we selected four of NSQIP's 17 perioperative outcomes that were relevant to CEA: SSI, MI, CVA, and blood transfusion requirement. We also created two composite outcome variables. The first measured the incidence of any of the 17 intra- or postoperative outcomes of interest. These 17 outcomes included cardiac arrest, CVA, blood transfusion requirement, intubation lasting >48 h, failure of graft/prosthesis, wound dehiscence, three types of SSI, MI, venous thromboembolism, urinary tract infection, renal insufficiency, sepsis, pneumonia, septic shock, and acute renal failure. The second composite outcome variable measured the incidence of 30-d mortality, MI, or CVA. The SSI outcome included superficial SSI, deep incision SSI, and organ space SSI that occurred within 30 d of the procedure. Superficial SSI included infections that involved only the skin or subcutaneous tissue of the incision. Deep incision SSI included infection of the deep soft tissue (muscle and fascia) of the incision, whereas organ space SSI included infections of any of the organs or spaces unconnected to the incision but which were manipulated during the procedure. Patients with any one of these types of SSI were regarded as having an SSI in our analyses. The outcome of MI was recorded in the incidence of any new acute MI that occurred during the procedure or within 30 d postoperatively. The CVA outcome was recorded in the incidence of the patient's development of symptoms lasting for >24 h within 30 d postoperatively. Mortality was recorded as any death occurring during the procedure or within 30 d postoperatively. This was an institutional review board exempt study.
2.2 Statistical analysis
Statistical analysis was performed primarily to determine whether surgical specialty was significantly associated with outcomes after controlling for patient and surgical characteristics. The first statistical analysis performed was univariate analysis to determine whether there were differences in patient characteristics across surgeon specialty. This was done using t-tests for continuous variables and chi-square tests for binary and categorical variables. Patient outcomes were also compared across surgical specialty using t-tests and chi-square tests, without controlling for any patient characteristics.
Logistic regression was then used to model the effect of surgical specialty on binary outcomes after controlling for patient and surgical characteristics. Areas under the receiver operating characteristic curves were calculated to assess model performance. Multivariate analysis of LOS was performed using a generalized linear regression model. This was done because LOS was highly skewed and clearly violated the normality assumption of classical linear regression. For the generalized linear model, we assumed a gamma family of distributions and a log link function. We report the marginal effects from the generalized linear models, which show the effect of a 1 unit change in the independent variable on the outcome. A deviance test was calculated to assess goodness-of-fit.
If a significant imbalance in patient covariates existed between general and vascular surgeons, then a regression model may not adequately control for covariates. Therefore, a propensity score matching analysis that dealt with potential covariate imbalance was performed. The propensity score model was fit using a logistic regression model with general surgical specialty as the dependent variable and controlled for covariates as previously described. Predicted probability of treatment by a general surgeon (i.e., the propensity score) was then computed from the fitted regression model. Patients of general surgeons were matched 1:5 to patients of vascular surgeons. Patients were matched based on a k-nearest neighbor match with a max-min common support restriction.
The primary metric for the propensity score analysis was the average effect of treatment on the treated (ATT). This is the difference between the outcome for a patient treated by a general surgeon and the outcome for a patient treated by a vascular surgeon. To deal with the uncertainty induced by both the selection process and the data, a standard bootstrapping algorithm was used to compute 95% confidence intervals. Reported inferences for the ATT are based on 50 bootstrap replicates. All statistical analyses were performed using STATA (version 12.1; StataCorp LLP, College Station, TX) and the psmatch2 routines [
]. Statistical significance for all analyses was defined as a
P value < 0.05.