Indications, Outcomes, and Provider Volumes for Carotid Endarterectomy FREE

DURING THE PAST decade, several controlled trials have provided evidence that carotid endarterectomy is superior to best medical therapy in reducing subsequent stroke rates among important subgroups of persons at risk.^{1 - 4} Studies in North America and Europe^{1 - 2} demonstrated a beneficial effect of carotid endarterectomy among symptomatic persons who have high-grade ipsilateral carotid artery stenosis. Publication of the Asymptomatic Carotid Atherosclerosis Study (ACAS) in 1995⁴ broadened the potential indications for carotid endarterectomy to include asymptomatic patients with high-grade carotid artery stenosis, raising the possibility of a 10-fold increase in the performance of the procedure.^{5 - 6}

As a preventive procedure designed to reduce the risk of future stroke, the appropriateness and outcomes of carotid endarterectomy are nonetheless highly dependent on surgical risk and untreated life expectancy in addition to symptoms and degree of carotid artery narrowing.^{7 - 8} The trials indicating a favorable effect of the procedure were efficacy trials, however, performed under narrow conditions of patient, surgeon, and institutional eligibility.^{9 - 11}

While there are few studies of carotid endarterectomy in more representative settings of patient and provider selection, there is suggestive evidence for higher complication rates in some community settings^{12 - 14} and, at least among Medicare beneficiaries, for an inverse relationship between procedural volumes and complication rates.^{15 - 16} Because studies that have reported the volume-outcome relationship have been unable to accurately measure surgical indications or patients' comorbid illnesses, the relative importance of patient selection vs surgical or institutional proficiency has gone largely unexplored.

Ohio had the third highest rate of carotid endarterectomy in the United States in 1990 among non–health maintenance organization Medicare beneficiaries (Health Care Financing Administration, unpublished data, 1990). During federal fiscal year 1994, carotid endarterectomies were performed in Ohio at 115 acute care hospitals by at least 478 surgeons. Specific objectives of the current study were (1) to profile the use and outcomes of carotid endarterectomy in Ohio's Medicare population over a 1-year period, and (2) to examine the relationships between surgeon-specific and hospital-specific volumes and 30-day stroke or death rate, controlling for surgical indications and other factors predictive of adverse outcomes.

Administrative data sources and chart review were used to provide the study sample and data to address the 2 main study objectives. Medicare Part A claims data (modified Medicare Provider Analysis and Review files) were used to (1) identify all non–health maintenance organization Medicare beneficiaries in Ohio who underwent carotid endarterectomy (International Classification of Diseases, Ninth Revision [ICD-9] 38.12) in Ohio hospitals between July 1, 1993, and June 30, 1994; (2) determine the number of carotid endarterectomies undertaken during the study period by each surgeon who performed at least 1 procedure, and by each hospital in which at least 1 procedure was performed; (3) identify all deaths and hospital readmissions that occurred within 30 days postoperatively; and (4) define index procedures and identify prespecified exclusions. From the universe of all index procedures during the 1-year study period, the medical records of a random statewide sample of 700 patients were selected for detailed review, using a random number generator¹⁷ to refine the study sample by applying further prespecified exclusion criteria and to identify indications for surgery, comorbid patient conditions, and postoperative deaths and new nonfatal strokes that occurred during hospitalization. Records of patients readmitted within the 30-day postoperative period also were reviewed for the occurrence of new nonfatal stroke.

Patients were excluded from the sample if they underwent another major vascular procedure (such as coronary artery bypass graft surgery or sequential bilateral carotid endarterectomies) during the same hospitalization, or if they were undergoing carotid endarterectomy for stroke-in-evolution. If a patient had more than 1 carotid endarterectomy in separate hospitalizations during the study period, only the last procedure was included in the study sample.

Medical records were abstracted by trained nurse-reviewers to classify the indications for surgery, sociodemographic patient information and comorbid conditions, and the outcomes of surgery during hospitalization. Medicare Provider Analysis and Review claims data were used to identify all readmissions within 30 days following surgery, and these records also were reviewed for the occurrence of interval strokes.

Indications were classified into 1 of 4 categories using chart-based data concerning the side of surgery (left or right), and the nature, sidedness, duration of symptoms, and neurological signs. Patients categorized as undergoing surgery for anterior circulation distribution or transient ischemic attack (TIA) had motor or sensory symptoms contralateral to the surgical site, visual disturbances ipsilateral to the surgical site, or reported speech disturbances and had surgery on the left carotid artery. To be classified as TIA, all symptoms and signs had to be recorded as transient and had to be resolved completely prior to admission. Individuals categorized as undergoing surgery for stroke had lateralizing neurological symptoms and signs without resolution prior to hospitalization. Patients categorized as asymptomatic had no record of any neurological symptoms or signs. Patients categorized as having surgery for nonspecific symptoms could have a variety of nonlateralizing symptoms and/or signs (eg, dizziness or dementia) with or without resolution at the time of hospital admission.

Comorbid conditions were identified from admission and progress notes and included the presence or absence of reported hypertension, diabetes mellitus, chronic obstructive pulmonary disease, congestive heart failure, and any history of cigarette smoking, angina, or myocardial infarction. Chronic renal insufficiency was identified from preoperative laboratory data and was defined as a serum creatinine level of at least 177 µmol/L (2.0 mg/dL).

Adverse outcomes included any new stroke or death that occurred within 30 days postoperatively. New stroke was defined as any nonfatal central neurological deficit that was not present preoperatively. To be defined as a stroke during the index hospitalization, the deficit had to be unresolved by the time of patient discharge. Strokes occurring after discharge, but within the 30-day postoperative period, were identified from admission or progress notes for readmitted patients. A random sample of all records, as well as the records of all patients identified as having an adverse outcome, were abstracted by a vascular surgeon in addition to the nurse-reviewer. Interrater agreement for indications for surgery was 77% (κ, 0.69) and for outcomes was 97% (κ, 0.79), suggesting very good reliability of the main records-based items.^{17 - 18}

Medicare Provider Analysis and Review claims data were used to identify the total number of carotid endarterectomies billed to Medicare by surgeons and hospitals during the study period, including procedures for which chart reviews were not performed. Surgeon volumes included procedures undertaken at any Ohio acute care hospital during the study period. The accuracy of surgical unique provider identification numbers was evaluated in the chart-abstracted sample through review of operating room records. The surgical unique provider identification numbers of the principal surgeon, as determined from chart review, was used for all chart-related analyses.

Descriptive statistics were used to profile the indications, comorbid conditions, and 30-day stroke or death rates among the random sample of patients whose records were reviewed. To avoid double-counting, patients classified as having postoperative stroke had to survive for 30 days after surgery. Population-centered statistics were generated to characterize provider volumes (eg, the median provider [surgeon and hospital] volume was defined as the number of provider-specific procedures over the 1-year study period during which 50% of the state's patients underwent carotid endarterectomy). Bivariate analyses used χ²and Fisher exact tests to examine the relationships between categorical variables and outcomes and analysis of variance was used for continuous variables. To examine the relationship between provider volumes and adverse outcomes, logistic regression was used, with adjustments for variables having significant associations (P<.05) in univariate analyses, as well as for variables identified from published work. Patients with missing data were excluded from the logistic regression model. Multivariate odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. All analyses were performed using SAS software.¹⁷

A total of 4120 carotid endarterectomies were performed in the Ohio non–health maintenance organization Medicare population between July 1, 1993, and June 30, 1994. After excluding patients undergoing other major vascular procedures during the same hospitalization as their carotid surgery and selecting the most recent endarterectomy hospitalization among those patients undergoing more than 1 carotid procedure (189 were excluded), a random statewide sample of 700 patients was chosen for medical records review. Of these, medical records were unavailable for 18 patients and an additional 4 were excluded because of stroke-in-evolution as the indication for surgery (n=1) or the performance of bilateral carotid procedures during the same hospitalization (n=3). The remaining 678 patients constitute the sample for full records review and related analyses. The patients whose records were reviewed were nonsignificantly different from the nonsampled group in age (73.11 years vs 73.18 years in nonsampled patients; P=.77), sex (47% female vs 44%, respectively; P=.18), race (6.43% nonwhite vs 5.6%, respectively; P=.39), in-hospital mortality (1.14% vs 0.99%, respectively; P=.72), and 30-day mortality (1.8% vs 1.7%, respectively; P=.78).

Indications, characteristics of the sample, and outcomes of surgery for the 678 patients are listed in Table 1. The most frequent indications for carotid endarterectomy were TIA (43.4%), followed by asymptomatic carotid artery stenosis (24.6%), nonspecific symptoms (22.9%), and completed stroke (9.1%). Seven patients (1.0%) died during their index hospitalizations, and 5 more died in the 30-day postoperative period, for a 30-day mortality rate of 1.8%. Nonfatal stroke was identified within 30 days of surgery in 20 patients (3.0%), including 16 during the initial hospitalization. Overall 30-day stroke or death rate was 4.7% (32 of 678 patients).

During the 1-year study period, carotid endarterectomies were performed by 478 principal (billing) surgeons at 115 acute care hospitals in Ohio. Figure 1 displays the distribution of procedural volumes across hospitals and Figure 2 displays the distribution across surgeons during the study period. In general, many hospitals and surgeons performed relatively low numbers of procedures, and fewer hospitals and surgeons performed relatively high numbers of procedures. Figure 1 and Figure 2 also display the cumulative percentage of patients undergoing carotid endarterectomy at hospitals (Figure 1) or by surgeons (Figure 2) with specified annual procedural volumes. Using these population-based measures, half of the state's procedures were undertaken by surgeons performing 21 or fewer endarterectomies during the 1-year period and at hospitals performing 62 or fewer carotid procedures. Placing the population-centered median values in another perspective, 93 hospitals (81%) performed 62 or fewer carotid endarterectomies during the 1-year period, and 22 hospitals (19%) performed 63 or more. Altogether, 421 surgeons (88%) performed at or below the population-centered median value of 21 procedures during the 1-year study period.

Table 2 summarizes the univariate relationships between 30-day stroke or death rates and indications for surgery and comorbid patient conditions among the 678 patients. By indication, stroke or death rates ranged from 2.4% among the 167 patients operated on for asymptomatic carotid stenosis to 7.1% among the 294 patients undergoing carotid endarterectomy for TIA (P=.06). Among comorbid conditions, only the presence of renal insufficiency was associated with a significantly increased likelihood of stroke or death (14.2%, P=.01), while patients with documented angina or congestive heart failure had nonsignificantly higher rates of adverse outcomes (Table 2). Patient age was not associated with stroke or death rates, whether analyzed as a continuous variable or as a dichotomous variable with a cutoff of 80 years.

Figure 3 displays the relationships between provider volumes and 30-day stroke or death rates. Hospital-specific stroke or death rates were inversely related to the number of procedures performed during the 1-year period, with rates ranging from 7.7% in the hospitals in the lowest quartile of volume to 2.5% in hospitals in the highest quartile (P=.03). When procedural volume was divided into lower- and higher-volume hospitals at the median value of 62 procedures per study period, higher-volume hospitals had significantly lower stroke or death rates than did lower-volume hospitals (2.4% vs 7.1%, respectively; P=.004). This relationship held across all indications for carotid endarterectomy. Patients who underwent surgery in higher-volume hospitals for asymptomatic carotid disease or nonspecific symptoms had no strokes or deaths at 30 days, as compared with rates of 4.9% and 4.6%, respectively, in lower-volume hospitals. Thirty-day stroke or death rates among patients who underwent surgery in higher-volume hospitals for TIA or completed stroke were 4.6% and 2.8%, respectively, as compared with 9.8% and 7.7%, respectively, in lower-volume hospitals. Surgeon-specific procedural volumes had a similar but less striking inverse relationship with stroke or death rates (Figure 3). Dividing procedural volume into lower- and higher-volume surgeons at the median value of 21 procedures per study period revealed a nonsignificantly lower stroke or death rate among higher-volume surgeons (4.3% vs 5.3% among lower-volume surgeons; P=.47); analyses of outcomes by quartiles of surgeon volume were similarly nonsignificant (from lowest- to highest-volume quartile, rates were 6.25%, 4.29%, 3.55%, and 5.0%, respectively; P=.68). Analyses that examined the interaction between hospital and surgeon volume also revealed the greater influence of procedural volumes at the hospital level (Figure 4).

Two additional analyses were undertaken to explore further the observed relationship between hospital-specific procedural volumes and adverse outcomes. To determine whether the higher rates of stroke or death in lower-volume hospitals were related to differences in the distribution of indications for surgery (eg, higher prevalence of TIA as the reason for surgery) or in the prevalence of high-risk comorbid conditions, a direct comparison was made of these features in lower- and higher-volume hospitals (Table 2). Of note, while there were no significant differences by indication, patients with stroke (a relatively higher-risk indication) were somewhat more prevalent in the higher-volume hospitals, while patients with nonspecific symptoms (a relatively lower-risk indication) were somewhat more prevalent in lower-volume hospitals. Among patient comorbidities, only a history of previous myocardial infarction was significantly different in prevalence between the 2 hospital types, and, again, the prevalence of this comorbid condition was higher in the higher-volume hospitals (26.8% vs 18.3% in the lower-volume hospitals; P=.01).

In the second analysis, logistic regression was used to determine whether the association of hospital volume remained after adjusting for other factors known or suspected to be related to stroke or death after carotid endarterectomy. Included as independent variables were patient gender, patient age greater than 80 years, presence of angina, congestive heart failure, and renal insufficiency, as well as hospital volume, surgeon volume, and indication for surgery. When expressed as relative risk reduction (multivariate OR, 1.0), undergoing surgery on in a higher-volume hospital was associated with a 71% reduction in risk of stroke or death at 30 days after adjustments for attributes related to patient selection (OR, 0.29; 95% CI, 0.12-0.69; P=.006). Surgeon volume was not significantly associated with 30-day stroke or death rates, whether examined as a dichotomous (higher- vs lower-volume surgeons) or as a continuous (total number of procedures per surgeon) variable. Factors independently associated with increased risk of 30-day stroke or death rates included TIA as an indication for surgery (OR, 2.9; 95% CI, 1.22-7.85; P=.01), patient history of angina (OR, 2.4; 95% CI, 1.01-5.89; P=.05), and the presence of renal insufficiency (OR, 3.3; 95% CI, 1.35-7.85; P=.05). Analyses that examined as candidate variables only those with univariate significance at the P=.10 level or better showed identical results.

In this random sample of Ohio Medicare beneficiaries undergoing carotid endarterectomy, almost half of the patients were either asymptomatic or had nonspecific symptoms at the time of surgery. While there is a paucity of data on the distribution of indications for endarterectomy in community-based series, recent evidence suggests that between 25% and 60% of endarterectomies are being performed on asymptomatic patients.^{6 ,19} Barnett and colleagues^{5 - 6} have speculated that uncritical acceptance of the reported ACAS findings will lead to a 10-fold increase in the performance of carotid endarterectomy on asymptomatic patients, and caution that population screening studies cannot be justified at present.⁶

Whether justified or not, the preventive use of carotid endarterectomy for asymptomatic carotid artery disease accentuates the long-held concern^{8 ,20 - 21} about the surgical risks and long-term benefits of the procedure, especially among the elderly and persons with major comorbidity. As a preventive operation, the net benefits of endarterectomy are sensitive both to the patient's untreated life expectancy and to the procedural risks of stroke or death.^{7 ,22} Patients over age 80 years and those with life-limiting comorbidity were excluded from the major randomized trials of endarterectomy,^{2 ,4} while approximately 1 in 6 patients in the current investigation was greater than age 80 years, and most had 1 or more major comorbid conditions. Similar results have been reported recently using national Medicare data.²³

The overall 30-day nonfatal stroke or death rate observed in the current study (4.7%) compares favorably with that in other community-based series.^{12 - 14 ,24 - 25} Indication-specific comparisons with the results of North American Symptomatic Carotid Endarterectomy Trial (NASCET) and ACAS are confounded by the older age of the current study population (73 years vs 65 and 67 years in NASCET and ACAS, respectively), making adverse outcomes more likely in the current study, but, because of the shorter life expectancy of study subjects, also less acceptable.^{7 - 8} The 7.1% 30-day stroke or death rate observed among the 294 patients with TIA is higher than the 5.8% rate in NASCET. This unfavorable difference may be underestimated because the ascertainment of postdischarge nonfatal stroke in the current study required rehospitalization; therefore, it is possible that patients with minor postoperative strokes escaped detection in this investigation. Had minor cerebrovascular events been excluded in NASCET, the analogous rate would have been 2.1%, with a fatality rate of 0.6%.²

Patients who were asymptomatic had 30-day stroke or death rates similar to those who underwent carotid endarterectomy for nonspecific symptoms (2.4% vs 2.6%, respectively). While these rates may have underestimated the occurrence of minor cerebrovascular events, as noted above, the overall results compare favorably with both the results from ACAS and published guidelines for maximal acceptable complication rates.^{20 - 21 ,26}

An inverse relationship between a hospital's procedural volume and adverse outcomes from vascular surgery was first reported by Luft et al²⁷ almost 2 decades ago and was used as an argument for the regionalization of major surgical procedures. More recently, others^{15 - 16} have used administrative data sources to demonstrate similar hospital-level, volume-outcome relationships with carotid artery surgery. In general, these studies have been limited by an inability to measure nonfatal stroke outcomes, indications for the procedure, or the extent of comorbid illness among patients in hospitals with different procedure volumes. Studies that do not measure differences in case mix^{15 ,25} are unable to determine whether the poorer performance of lower-volume providers results from operating on sicker patients or from a different distribution of indications for the procedure. While it has been argued that it may be inappropriate to adjust for severity to level the playing field when evaluating the performance of a preventive procedure,²² the absence of detailed data regarding indications and comorbidity has contributed to an impasse in the debate about regionalization.

The current investigation addressed these issues by conducting detailed reviews of records on a random sample of all carotid endarterectomies performed in Ohio during 1993 to 1994. More than 80% of Ohio's hospitals were lower-volume, similar to reports from others in the mid-80s.¹⁵ Higher-volume hospitals had better 30-day outcomes than lower-volume hospitals across all indications, and notably incurred no fatalities or nonfatal strokes among the 152 patients who were asymptomatic or had nonspecific symptoms preoperatively. We found no evidence of adverse selection in the lower-volume hospitals (Table 2). Further, there was a 71% reduction in risk associated with undergoing surgery on in higher-volume hospitals, even after adjusting for multiple potential differences in patient selection.

It was not the intent of this investigation to identify an optimal annual volume or a threshold number of procedures below which a hospital is likely to perform in an unsatisfactory manner. Indeed, the results displayed in Figure 3 are consistent with operations described by Luft et al²⁷ as exhibiting a falling complication rate throughout the measured range.

Likewise, it is not possible from data in this investigation to determine the reasons for the better performance of higher-volume hospitals. Speculation by others has included factors related to better nursing care, anesthetic management, and postoperative care,^{27 - 28} but relevant data were not gathered in the current study.

Finally, it is not possible from this study to determine whether the better performance of higher-volume hospitals was a result of increased experience with carotid endarterectomy (the "practice makes perfect" hypothesis) or whether hospitals with better outcomes attract higher volumes through increased referrals (the "perfect makes practice" hypothesis).^{29 - 30} More complex longitudinal investigations would be required to address these competing hypotheses.

In the face of the strong hospital-level findings, the absence of a significant empirical relationship between an individual surgeon's volume and his or her outcomes deserves further examination. Our investigation's sample had very low power (<10%) to find statistical significance in the observed differences in outcomes between higher- and lower-volume physicians. It also is possible that surgeons who performed a relatively low volume of carotid endarterectomies may have performed a relatively large number of other procedures requiring similar skills, such as other vascular procedures; that is, the volume of interest might be the number of technically similar operations performed in a given time period, which we did not measure in the current study. Further, it is possible that younger (and perhaps better trained) surgeons may have had better outcomes despite the lower volumes associated with their younger practices.²⁸ In the current study, surgeons' age and prior training were not examined. Although similarly negative results have been reported by others,^{31 - 32} an inverse relationship between surgical volume and outcomes in carotid endarterectomy has been reported recently by Ruby and colleagues²⁵ and others.²⁸

In summary, patient and provider response to recent randomized trials may lead to a dramatic increase in the performance of carotid endarterectomy, especially among asymptomatic persons. In the year following the dates of this study, the frequency of carotid endarterectomy in Ohio increased almost 50%, from 4120 to 5997 (Lee Hannah, DVM, MS, MPH, oral communication, November 1996), suggesting that the predictions of Barnett et al⁶ and others may be realized. These observations, and the results of this investigation, highlight the need to identify the determinants of outcomes from the procedure, as well as the need for prospective monitoring of outcomes, as has been proposed by the leadership of the Society for Vascular Surgery.²⁸ Ohio is 1 of 9 states having current initiatives to conduct such prospective monitoring of indications and outcomes (Lee Hannah, DVM, MS, MPH, oral communication, January 1998).