David De Boer, MS

Setting and Subjects

We conducted a retrospective cohort study using a regional Department of Veterans Affairs (VA) administrative and relational database, the Consumer Health Information and Performance Sets (CHIPs), which automatically extracts data from electronic medical records of all facilities in the Veterans Integrated Services Network 20, which encompasses Alaska, Washington, Oregon, and Idaho. CHIPs contains information on both outpatient and inpatient environments, and a record is generated for every contact a patient makes with the VA health care system, which includes picking up prescription medications, laboratory values, demographic information, International Classification of Diseases, 9th Revision (ICD‐9), codes, and vital status. In addition, we used the Beneficiary Identification and Records Locator Subsystem database, which is the national VA death index and includes Social Security Administration data that has been shown to be 90%95% complete for assessing vital status.20

Data for all patients who had vascular surgery at 5 VA medical centers in the region from January 1998 to March 2005 was ascertained. If a patient had a second operation within 2 years of the first, the patient was censored at the date of the second operation. A patient was defined as taking a statin or beta‐blocker if a prescription for either of these medications had been picked up within 30 days before or after surgery. The IRB at the Portland VA Medical Center approved the study with a waiver of informed consent.

Data Elements

For every patient we noted the type of vascular surgery (carotid, aortic, lower extremity bypass, or lower extremity amputation), age, sex, comorbid conditions (hypertension, cerebrovascular disease, cancer, diabetes, hyperlipidemia, chronic obstructive pulmonary disease [COPD], chronic kidney disease [CKD], coronary artery disease [CAD], heart failure), tobacco use, ethnicity, nutritional status (serum albumin), and medication use, defined as filling a prescription within 30 days before surgery (insulin, aspirin, angiotensin‐converting enzyme [ACE] inhibitor, and clonidine). Each patient was assigned a revised cardiac risk index (RCRI) score.21 For each the risk factors: use of insulin, CAD, heart failure, cerebrovascular disease, CKD, and high‐risk surgery (intrathoracic, intraperitoneal, or suprainguinal vascular procedures) 1 point was assigned. These variables were defined according to ICD‐9 codes. CKD was defined as either an ICD‐9 code for CKD or a serum creatinine > 2 mg/dL. Patients were identified by the index vascular surgery using ICD‐9 codes in the CHIPs database, and both inpatient and outpatient data were extracted.

Statistical Analysis

All patients were censored at the point of last contact up to 5 years after surgery to focus on more clinically relevant long‐term outcomes possibly associated with vascular surgery. We conducted 3 separate analyses: (1) statin exposure regardless of beta‐blocker exposure; (2) beta‐blocker exposure regardless of statin exposure, and; (3) combined exposure to statins and beta‐blockers.

Propensity score methods were used to adjust for imbalance in the baseline characteristics between statin users and nonusers, beta‐blocker users and nonusers, and combination statin and beta‐blocker users and nonusers.22, 23 The range of the propensity score distribution was similar in drug users and nonusers in the individual analyses. There was sufficient overlap between the 2 groups in each stratum. To derive propensity scores for the individual drug analyses, statin use and beta‐blocker use were modeled independently with the demographic and clinical variables using stepwise logistic regression with a relaxed entry criterion of = 0.20. Only 1 variable (hyperlipidemia) remained significantly different between statin users and nonusers, and it was included in the subsequent analyses as a potential confounder. The variable albumin had 511 missing values. To keep this variable in the propensity scores, the missing values were replaced by the predicted values of albumin from the multiple linear regression model that included the other demographic variables. The propensity scores were grouped into quintiles and used as a stratification variable in the subsequent analyses. To confirm that the propensity score method reduced the imbalances, the demographic and clinical characteristics of statin and beta‐blocker users and nonusers and combination users and nonusers were compared using Cochran‐Mantel‐Haenzel tests with the respective propensity score as a stratification variable.

For the combined use of both study drugs, we performed univariate analysis with adjustment only for RCRI (as this was a powerful predictor of mortality in our dataset; Table 1) as well as a propensity score analysis in an exploratory manner. There have been limited applications of propensity score methods to multiple treatment groups. Similar to that in the study by Huang et al.,24 we developed a multinomial baseline response logit model to obtain 3 separate propensity scores (statin only vs. none, beta‐blocker only vs. none, and both vs. none). Because of the limited sample size, the data were stratified according to the median split of each propensity score. Each score had similar ranges for each treatment group. All but 5 variables (CAD, hypertension, hyperlipidemia, ACE inhibitor use, and type of surgery) were balanced after accounting for strata. These 5 variables were then included in the final stratified Cox regression model as potential confounders.

To comment on patient‐specific risk by stratification with the RCRI, we used a fixed time point of the 2‐year mortality estimated from the Cox regression model to analyze use of study drugs singly or in combination compared with use of neither.

Chi‐square tests were used to categorize and compare demographic and clinical characteristics of statin users and nonusers, of beta‐blocker users and nonusers, and combination users and nonusers. Survival curves were estimated using the Kaplan‐Meier method and compared using the log‐rank test. Stratified or unstratified Cox regression was used to estimate the hazard ratios of statins and beta‐blockers, with or without adjustment for the propensity score. All analyses were performed using SAS (Statistical Analysis System) software, version 9.1.

Patient Characteristics

The study included 3062 patients whose median age was 67 (interquartile range, 5974; Table 1). Ninety‐nine percent of the study patients were men. Overall, ambulatory use of statins and beta‐blockers was found in 44% and 53% of patients, respectively, and combination use occurred in 30%. Sixty‐one percent of patients had an RCRI of 1 or greater; among them 71% were statin users (Table 2), 68% were beta‐blocker users (Table 3), and 75% were combination users (Table 4). Sixty‐four percent of surgeries were either lower extremity bypass or amputation, 29% were carotid, and 8% aortic. Median follow‐up for all patients was 2.7 years (interquartile range, 1.24.6). Of the whole study cohort, 53% and 62% filled a prescription for a statin or beta‐blocker within 1 year of surgery, respectively, and 58% and 67% filled a prescription within 2 years of surgery, respectively. Overall mortality at 30 days was 3%, at 1 year 14%, and at 2 years 22%.

Univariate Survival Analysis

Univariate Cox regression analysis revealed a strong effect of the composite RCRI, which was predictive of mortality in a linear fashion over the course of the study compared with an RCRI of 0 (Table 1). Univariate analysis showed significant associations with decreased mortality for statins (hazard ratio [HR] = 0.66 [95% CI 0.580.75], P < .0001) and beta‐blockers (HR = 0.74 [95% CI 0.660.84], P = .0001); see Table 1. Of note, compared with that in 1998, mortality did not change for all the years for which data were complete. In addition, compared with taking neither study drug, taking a statin only, a beta‐blocker only, or both was associated with decreased mortality (P = .0002, P = .0079, and P < .0001, respectively; Fig. 1).

Figure 1

Propensity Score Analysis for Single Study Drug

There were significant differences in demographic and clinical characteristics between statin‐users versus statin nonusers, and between beta‐blocker users versus beta‐blocker nonusers. These differences became insignificant after the propensity score adjustment, with the exception of hyperlipidemia for statins, P = .02, which was added to the model as a confounder (Table 2). The distribution of the propensity scores was similar for study drug users and nonusers within each stratum. The association with decreased mortality remained significant after adjusting for propensity score (for statins, HR = 0.78 [95% CI 0.670.92, P = .0050], number needed to treat [NNT] = 22; for beta‐blockers HR = 0.84 [95% CI 0.730.96, P = .0103], NNT = 30; Fig. 2).

Figure 2

Combination Study Drugs and Revised Cardiac Risk Index: Univariate Analysis

We wanted our results to closely model those of combination use of the study drugs by patients in a clinical situation. Therefore, we first examined the effects of ambulatory statins alone, beta‐blockers alone, and a combination of statins and beta‐blockers by univariate analysis. Grouping patients by study drug use has not commonly been reported in the literature. We also examined the statistical interaction between the study drugs and the RCRI. The main‐effects model adequately explained all‐cause mortality, and the statistical interaction between the study drugs and the RCRI was not significant.

The final univariate Cox regression model, which compared use of a statin alone, a beta‐blocker alone, and a statin and beta‐blocker in combination with using neither study drug, demonstrated that the combination of statins and beta‐blockers had an HR over the whole study period of 0.43 (95% CI 0.360.51, P < .0001), statins alone had an HR of 0.59 (95% CI 0.480.72, P < .0001), and beta‐blockers alone had an HR of 0.71 (95% CI 0.610.83, P < .0001).

To clarify the effects of the study drugs on patients at different levels of risk, we stratified patients by the RCRI and evaluated the effects of the study drugs on mortality at 2 years, comparing the results to a referent of taking no study drugs. The use of both a statin and a beta‐blocker consistently produced a relative risk reduction (RRR) of approximately 50% with an NNT of 410, with highly statistically significant results for patients at all levels of risk (Table 5). As patient risk level increased, the NNT decreased, consistent with higher‐risk patients benefiting most from combination therapy with statins and beta‐blockers.

In addition, the range of outcomes can be clearly seen for both patient‐specific risk level and study drug use. For example, overall mortality at 2 years for all patients was 22%. For the study drugs, mortality ranged from 16% for patient who used both a statin and a beta‐blocker to 27% for those patients who used neither study drug. The use of the RCRI showed that the healthiest patients who were taking both a statin and a beta‐blocker did the best, with a 2‐year mortality of 9%, compared with the sickest patients who were taking neither study drug, whose 2‐year mortality was 59%. Use of both study drugs by the sickest patients was associated with a reduction in 2‐year mortality to 32% (P < .0001; Table 5).

Propensity Score Analysis of Use of Combination Study Drugs

Because there was very limited literature to guide us in the use of propensity score analysis of multiple treatment groups, we performed these analyses in an exploratory manner. There were significant differences between combination statin and beta‐blocker users and nonusers. These differences became insignificant after adjusting for propensity score, except for the 5 variables previously mentioned, which were added to the model as potential confounders (Table 4). The propensity‐adjusted Cox regression model comparing use of each study drug alone and in combination with taking neither over the whole study period still showed an association with decreased mortality. The combination of statins and beta‐blockers had an HR of 0.56 (95% CI 0.420.74), P < .0001; statins alone had an HR of 0.79 (95% CI 0.620.99), P = .0472; and beta‐blockers alone had an HR of 0.80 (95% CI 0.670.94), P = .0183.

Combination Study Drugs and Revised Cardiac Risk Index: Propensity Analysis

We performed the stratified Cox regression adjusted for the propensity scores for each level of RCRI and estimated 2‐year mortality. The use of both a statin and a beta‐blocker compared with using none was still consistently statistically significant, with an RRR of approximately 36% and an NNT of 820 for all levels of patient risk (Table 6). Possibly because of the reduced number of patients in each RCRI category, neither single‐agent study drug compared with none showed a statistically significant decrease in mortality at any level of patient‐specific risk (Table 6). Again, higher‐risk patients benefited most from combination therapy.

Study Drug Timing: Subcohort Analysis

A subcohort analysis was performed to clarify the timing of the study drugs. Of the patients taking statins, 69 of 1346 (5.1%) took the drug before surgery only, 119 of 1346 (8.8%) took the drug after surgery only, and 1158 of 1346 (86%) took the drug both before and after surgery. Of the patients taking beta‐blockers, 54 of 1617 (3.3%) took the drug before surgery only, 397 of 1617 (24.6%) took the drug after surgery only, and 1166 of 1617 (72.1%) took the drug both before and after surgery. The use of statins and beta‐blockers had a correlation of 0.29 (contingency coefficient).

Setting and Subjects

We conducted a retrospective cohort study using a regional Department of Veterans Affairs (VA) administrative and relational database, the Consumer Health Information and Performance Sets (CHIPs), which automatically extracts data from electronic medical records of all facilities in the Veterans Integrated Services Network 20, which encompasses Alaska, Washington, Oregon, and Idaho. CHIPs contains information on both outpatient and inpatient environments, and a record is generated for every contact a patient makes with the VA health care system, which includes picking up prescription medications, laboratory values, demographic information, International Classification of Diseases, 9th Revision (ICD‐9), codes, and vital status. In addition, we used the Beneficiary Identification and Records Locator Subsystem database, which is the national VA death index and includes Social Security Administration data that has been shown to be 90%95% complete for assessing vital status.20

Data for all patients who had vascular surgery at 5 VA medical centers in the region from January 1998 to March 2005 was ascertained. If a patient had a second operation within 2 years of the first, the patient was censored at the date of the second operation. A patient was defined as taking a statin or beta‐blocker if a prescription for either of these medications had been picked up within 30 days before or after surgery. The IRB at the Portland VA Medical Center approved the study with a waiver of informed consent.

Data Elements

For every patient we noted the type of vascular surgery (carotid, aortic, lower extremity bypass, or lower extremity amputation), age, sex, comorbid conditions (hypertension, cerebrovascular disease, cancer, diabetes, hyperlipidemia, chronic obstructive pulmonary disease [COPD], chronic kidney disease [CKD], coronary artery disease [CAD], heart failure), tobacco use, ethnicity, nutritional status (serum albumin), and medication use, defined as filling a prescription within 30 days before surgery (insulin, aspirin, angiotensin‐converting enzyme [ACE] inhibitor, and clonidine). Each patient was assigned a revised cardiac risk index (RCRI) score.21 For each the risk factors: use of insulin, CAD, heart failure, cerebrovascular disease, CKD, and high‐risk surgery (intrathoracic, intraperitoneal, or suprainguinal vascular procedures) 1 point was assigned. These variables were defined according to ICD‐9 codes. CKD was defined as either an ICD‐9 code for CKD or a serum creatinine > 2 mg/dL. Patients were identified by the index vascular surgery using ICD‐9 codes in the CHIPs database, and both inpatient and outpatient data were extracted.

Statistical Analysis

All patients were censored at the point of last contact up to 5 years after surgery to focus on more clinically relevant long‐term outcomes possibly associated with vascular surgery. We conducted 3 separate analyses: (1) statin exposure regardless of beta‐blocker exposure; (2) beta‐blocker exposure regardless of statin exposure, and; (3) combined exposure to statins and beta‐blockers.

Propensity score methods were used to adjust for imbalance in the baseline characteristics between statin users and nonusers, beta‐blocker users and nonusers, and combination statin and beta‐blocker users and nonusers.22, 23 The range of the propensity score distribution was similar in drug users and nonusers in the individual analyses. There was sufficient overlap between the 2 groups in each stratum. To derive propensity scores for the individual drug analyses, statin use and beta‐blocker use were modeled independently with the demographic and clinical variables using stepwise logistic regression with a relaxed entry criterion of = 0.20. Only 1 variable (hyperlipidemia) remained significantly different between statin users and nonusers, and it was included in the subsequent analyses as a potential confounder. The variable albumin had 511 missing values. To keep this variable in the propensity scores, the missing values were replaced by the predicted values of albumin from the multiple linear regression model that included the other demographic variables. The propensity scores were grouped into quintiles and used as a stratification variable in the subsequent analyses. To confirm that the propensity score method reduced the imbalances, the demographic and clinical characteristics of statin and beta‐blocker users and nonusers and combination users and nonusers were compared using Cochran‐Mantel‐Haenzel tests with the respective propensity score as a stratification variable.

For the combined use of both study drugs, we performed univariate analysis with adjustment only for RCRI (as this was a powerful predictor of mortality in our dataset; Table 1) as well as a propensity score analysis in an exploratory manner. There have been limited applications of propensity score methods to multiple treatment groups. Similar to that in the study by Huang et al.,24 we developed a multinomial baseline response logit model to obtain 3 separate propensity scores (statin only vs. none, beta‐blocker only vs. none, and both vs. none). Because of the limited sample size, the data were stratified according to the median split of each propensity score. Each score had similar ranges for each treatment group. All but 5 variables (CAD, hypertension, hyperlipidemia, ACE inhibitor use, and type of surgery) were balanced after accounting for strata. These 5 variables were then included in the final stratified Cox regression model as potential confounders.

To comment on patient‐specific risk by stratification with the RCRI, we used a fixed time point of the 2‐year mortality estimated from the Cox regression model to analyze use of study drugs singly or in combination compared with use of neither.

Chi‐square tests were used to categorize and compare demographic and clinical characteristics of statin users and nonusers, of beta‐blocker users and nonusers, and combination users and nonusers. Survival curves were estimated using the Kaplan‐Meier method and compared using the log‐rank test. Stratified or unstratified Cox regression was used to estimate the hazard ratios of statins and beta‐blockers, with or without adjustment for the propensity score. All analyses were performed using SAS (Statistical Analysis System) software, version 9.1.

Patient Characteristics

The study included 3062 patients whose median age was 67 (interquartile range, 5974; Table 1). Ninety‐nine percent of the study patients were men. Overall, ambulatory use of statins and beta‐blockers was found in 44% and 53% of patients, respectively, and combination use occurred in 30%. Sixty‐one percent of patients had an RCRI of 1 or greater; among them 71% were statin users (Table 2), 68% were beta‐blocker users (Table 3), and 75% were combination users (Table 4). Sixty‐four percent of surgeries were either lower extremity bypass or amputation, 29% were carotid, and 8% aortic. Median follow‐up for all patients was 2.7 years (interquartile range, 1.24.6). Of the whole study cohort, 53% and 62% filled a prescription for a statin or beta‐blocker within 1 year of surgery, respectively, and 58% and 67% filled a prescription within 2 years of surgery, respectively. Overall mortality at 30 days was 3%, at 1 year 14%, and at 2 years 22%.

Univariate Survival Analysis

Univariate Cox regression analysis revealed a strong effect of the composite RCRI, which was predictive of mortality in a linear fashion over the course of the study compared with an RCRI of 0 (Table 1). Univariate analysis showed significant associations with decreased mortality for statins (hazard ratio [HR] = 0.66 [95% CI 0.580.75], P < .0001) and beta‐blockers (HR = 0.74 [95% CI 0.660.84], P = .0001); see Table 1. Of note, compared with that in 1998, mortality did not change for all the years for which data were complete. In addition, compared with taking neither study drug, taking a statin only, a beta‐blocker only, or both was associated with decreased mortality (P = .0002, P = .0079, and P < .0001, respectively; Fig. 1).

Figure 1

Propensity Score Analysis for Single Study Drug

There were significant differences in demographic and clinical characteristics between statin‐users versus statin nonusers, and between beta‐blocker users versus beta‐blocker nonusers. These differences became insignificant after the propensity score adjustment, with the exception of hyperlipidemia for statins, P = .02, which was added to the model as a confounder (Table 2). The distribution of the propensity scores was similar for study drug users and nonusers within each stratum. The association with decreased mortality remained significant after adjusting for propensity score (for statins, HR = 0.78 [95% CI 0.670.92, P = .0050], number needed to treat [NNT] = 22; for beta‐blockers HR = 0.84 [95% CI 0.730.96, P = .0103], NNT = 30; Fig. 2).

Figure 2

Combination Study Drugs and Revised Cardiac Risk Index: Univariate Analysis

We wanted our results to closely model those of combination use of the study drugs by patients in a clinical situation. Therefore, we first examined the effects of ambulatory statins alone, beta‐blockers alone, and a combination of statins and beta‐blockers by univariate analysis. Grouping patients by study drug use has not commonly been reported in the literature. We also examined the statistical interaction between the study drugs and the RCRI. The main‐effects model adequately explained all‐cause mortality, and the statistical interaction between the study drugs and the RCRI was not significant.

The final univariate Cox regression model, which compared use of a statin alone, a beta‐blocker alone, and a statin and beta‐blocker in combination with using neither study drug, demonstrated that the combination of statins and beta‐blockers had an HR over the whole study period of 0.43 (95% CI 0.360.51, P < .0001), statins alone had an HR of 0.59 (95% CI 0.480.72, P < .0001), and beta‐blockers alone had an HR of 0.71 (95% CI 0.610.83, P < .0001).

To clarify the effects of the study drugs on patients at different levels of risk, we stratified patients by the RCRI and evaluated the effects of the study drugs on mortality at 2 years, comparing the results to a referent of taking no study drugs. The use of both a statin and a beta‐blocker consistently produced a relative risk reduction (RRR) of approximately 50% with an NNT of 410, with highly statistically significant results for patients at all levels of risk (Table 5). As patient risk level increased, the NNT decreased, consistent with higher‐risk patients benefiting most from combination therapy with statins and beta‐blockers.

In addition, the range of outcomes can be clearly seen for both patient‐specific risk level and study drug use. For example, overall mortality at 2 years for all patients was 22%. For the study drugs, mortality ranged from 16% for patient who used both a statin and a beta‐blocker to 27% for those patients who used neither study drug. The use of the RCRI showed that the healthiest patients who were taking both a statin and a beta‐blocker did the best, with a 2‐year mortality of 9%, compared with the sickest patients who were taking neither study drug, whose 2‐year mortality was 59%. Use of both study drugs by the sickest patients was associated with a reduction in 2‐year mortality to 32% (P < .0001; Table 5).

Propensity Score Analysis of Use of Combination Study Drugs

Because there was very limited literature to guide us in the use of propensity score analysis of multiple treatment groups, we performed these analyses in an exploratory manner. There were significant differences between combination statin and beta‐blocker users and nonusers. These differences became insignificant after adjusting for propensity score, except for the 5 variables previously mentioned, which were added to the model as potential confounders (Table 4). The propensity‐adjusted Cox regression model comparing use of each study drug alone and in combination with taking neither over the whole study period still showed an association with decreased mortality. The combination of statins and beta‐blockers had an HR of 0.56 (95% CI 0.420.74), P < .0001; statins alone had an HR of 0.79 (95% CI 0.620.99), P = .0472; and beta‐blockers alone had an HR of 0.80 (95% CI 0.670.94), P = .0183.

Combination Study Drugs and Revised Cardiac Risk Index: Propensity Analysis

We performed the stratified Cox regression adjusted for the propensity scores for each level of RCRI and estimated 2‐year mortality. The use of both a statin and a beta‐blocker compared with using none was still consistently statistically significant, with an RRR of approximately 36% and an NNT of 820 for all levels of patient risk (Table 6). Possibly because of the reduced number of patients in each RCRI category, neither single‐agent study drug compared with none showed a statistically significant decrease in mortality at any level of patient‐specific risk (Table 6). Again, higher‐risk patients benefited most from combination therapy.

Study Drug Timing: Subcohort Analysis

A subcohort analysis was performed to clarify the timing of the study drugs. Of the patients taking statins, 69 of 1346 (5.1%) took the drug before surgery only, 119 of 1346 (8.8%) took the drug after surgery only, and 1158 of 1346 (86%) took the drug both before and after surgery. Of the patients taking beta‐blockers, 54 of 1617 (3.3%) took the drug before surgery only, 397 of 1617 (24.6%) took the drug after surgery only, and 1166 of 1617 (72.1%) took the drug both before and after surgery. The use of statins and beta‐blockers had a correlation of 0.29 (contingency coefficient).

Setting and Subjects

We conducted a retrospective cohort study using a regional Department of Veterans Affairs (VA) administrative and relational database, the Consumer Health Information and Performance Sets (CHIPs), which automatically extracts data from electronic medical records of all facilities in the Veterans Integrated Services Network 20, which encompasses Alaska, Washington, Oregon, and Idaho. CHIPs contains information on both outpatient and inpatient environments, and a record is generated for every contact a patient makes with the VA health care system, which includes picking up prescription medications, laboratory values, demographic information, International Classification of Diseases, 9th Revision (ICD‐9), codes, and vital status. In addition, we used the Beneficiary Identification and Records Locator Subsystem database, which is the national VA death index and includes Social Security Administration data that has been shown to be 90%95% complete for assessing vital status.20

Data for all patients who had vascular surgery at 5 VA medical centers in the region from January 1998 to March 2005 was ascertained. If a patient had a second operation within 2 years of the first, the patient was censored at the date of the second operation. A patient was defined as taking a statin or beta‐blocker if a prescription for either of these medications had been picked up within 30 days before or after surgery. The IRB at the Portland VA Medical Center approved the study with a waiver of informed consent.

Data Elements

For every patient we noted the type of vascular surgery (carotid, aortic, lower extremity bypass, or lower extremity amputation), age, sex, comorbid conditions (hypertension, cerebrovascular disease, cancer, diabetes, hyperlipidemia, chronic obstructive pulmonary disease [COPD], chronic kidney disease [CKD], coronary artery disease [CAD], heart failure), tobacco use, ethnicity, nutritional status (serum albumin), and medication use, defined as filling a prescription within 30 days before surgery (insulin, aspirin, angiotensin‐converting enzyme [ACE] inhibitor, and clonidine). Each patient was assigned a revised cardiac risk index (RCRI) score.21 For each the risk factors: use of insulin, CAD, heart failure, cerebrovascular disease, CKD, and high‐risk surgery (intrathoracic, intraperitoneal, or suprainguinal vascular procedures) 1 point was assigned. These variables were defined according to ICD‐9 codes. CKD was defined as either an ICD‐9 code for CKD or a serum creatinine > 2 mg/dL. Patients were identified by the index vascular surgery using ICD‐9 codes in the CHIPs database, and both inpatient and outpatient data were extracted.

Statistical Analysis

All patients were censored at the point of last contact up to 5 years after surgery to focus on more clinically relevant long‐term outcomes possibly associated with vascular surgery. We conducted 3 separate analyses: (1) statin exposure regardless of beta‐blocker exposure; (2) beta‐blocker exposure regardless of statin exposure, and; (3) combined exposure to statins and beta‐blockers.

Propensity score methods were used to adjust for imbalance in the baseline characteristics between statin users and nonusers, beta‐blocker users and nonusers, and combination statin and beta‐blocker users and nonusers.22, 23 The range of the propensity score distribution was similar in drug users and nonusers in the individual analyses. There was sufficient overlap between the 2 groups in each stratum. To derive propensity scores for the individual drug analyses, statin use and beta‐blocker use were modeled independently with the demographic and clinical variables using stepwise logistic regression with a relaxed entry criterion of = 0.20. Only 1 variable (hyperlipidemia) remained significantly different between statin users and nonusers, and it was included in the subsequent analyses as a potential confounder. The variable albumin had 511 missing values. To keep this variable in the propensity scores, the missing values were replaced by the predicted values of albumin from the multiple linear regression model that included the other demographic variables. The propensity scores were grouped into quintiles and used as a stratification variable in the subsequent analyses. To confirm that the propensity score method reduced the imbalances, the demographic and clinical characteristics of statin and beta‐blocker users and nonusers and combination users and nonusers were compared using Cochran‐Mantel‐Haenzel tests with the respective propensity score as a stratification variable.

For the combined use of both study drugs, we performed univariate analysis with adjustment only for RCRI (as this was a powerful predictor of mortality in our dataset; Table 1) as well as a propensity score analysis in an exploratory manner. There have been limited applications of propensity score methods to multiple treatment groups. Similar to that in the study by Huang et al.,24 we developed a multinomial baseline response logit model to obtain 3 separate propensity scores (statin only vs. none, beta‐blocker only vs. none, and both vs. none). Because of the limited sample size, the data were stratified according to the median split of each propensity score. Each score had similar ranges for each treatment group. All but 5 variables (CAD, hypertension, hyperlipidemia, ACE inhibitor use, and type of surgery) were balanced after accounting for strata. These 5 variables were then included in the final stratified Cox regression model as potential confounders.

To comment on patient‐specific risk by stratification with the RCRI, we used a fixed time point of the 2‐year mortality estimated from the Cox regression model to analyze use of study drugs singly or in combination compared with use of neither.

Chi‐square tests were used to categorize and compare demographic and clinical characteristics of statin users and nonusers, of beta‐blocker users and nonusers, and combination users and nonusers. Survival curves were estimated using the Kaplan‐Meier method and compared using the log‐rank test. Stratified or unstratified Cox regression was used to estimate the hazard ratios of statins and beta‐blockers, with or without adjustment for the propensity score. All analyses were performed using SAS (Statistical Analysis System) software, version 9.1.

Patient Characteristics

The study included 3062 patients whose median age was 67 (interquartile range, 5974; Table 1). Ninety‐nine percent of the study patients were men. Overall, ambulatory use of statins and beta‐blockers was found in 44% and 53% of patients, respectively, and combination use occurred in 30%. Sixty‐one percent of patients had an RCRI of 1 or greater; among them 71% were statin users (Table 2), 68% were beta‐blocker users (Table 3), and 75% were combination users (Table 4). Sixty‐four percent of surgeries were either lower extremity bypass or amputation, 29% were carotid, and 8% aortic. Median follow‐up for all patients was 2.7 years (interquartile range, 1.24.6). Of the whole study cohort, 53% and 62% filled a prescription for a statin or beta‐blocker within 1 year of surgery, respectively, and 58% and 67% filled a prescription within 2 years of surgery, respectively. Overall mortality at 30 days was 3%, at 1 year 14%, and at 2 years 22%.

Univariate Survival Analysis

Univariate Cox regression analysis revealed a strong effect of the composite RCRI, which was predictive of mortality in a linear fashion over the course of the study compared with an RCRI of 0 (Table 1). Univariate analysis showed significant associations with decreased mortality for statins (hazard ratio [HR] = 0.66 [95% CI 0.580.75], P < .0001) and beta‐blockers (HR = 0.74 [95% CI 0.660.84], P = .0001); see Table 1. Of note, compared with that in 1998, mortality did not change for all the years for which data were complete. In addition, compared with taking neither study drug, taking a statin only, a beta‐blocker only, or both was associated with decreased mortality (P = .0002, P = .0079, and P < .0001, respectively; Fig. 1).

Figure 1

Propensity Score Analysis for Single Study Drug

There were significant differences in demographic and clinical characteristics between statin‐users versus statin nonusers, and between beta‐blocker users versus beta‐blocker nonusers. These differences became insignificant after the propensity score adjustment, with the exception of hyperlipidemia for statins, P = .02, which was added to the model as a confounder (Table 2). The distribution of the propensity scores was similar for study drug users and nonusers within each stratum. The association with decreased mortality remained significant after adjusting for propensity score (for statins, HR = 0.78 [95% CI 0.670.92, P = .0050], number needed to treat [NNT] = 22; for beta‐blockers HR = 0.84 [95% CI 0.730.96, P = .0103], NNT = 30; Fig. 2).

Figure 2

Combination Study Drugs and Revised Cardiac Risk Index: Univariate Analysis

We wanted our results to closely model those of combination use of the study drugs by patients in a clinical situation. Therefore, we first examined the effects of ambulatory statins alone, beta‐blockers alone, and a combination of statins and beta‐blockers by univariate analysis. Grouping patients by study drug use has not commonly been reported in the literature. We also examined the statistical interaction between the study drugs and the RCRI. The main‐effects model adequately explained all‐cause mortality, and the statistical interaction between the study drugs and the RCRI was not significant.

The final univariate Cox regression model, which compared use of a statin alone, a beta‐blocker alone, and a statin and beta‐blocker in combination with using neither study drug, demonstrated that the combination of statins and beta‐blockers had an HR over the whole study period of 0.43 (95% CI 0.360.51, P < .0001), statins alone had an HR of 0.59 (95% CI 0.480.72, P < .0001), and beta‐blockers alone had an HR of 0.71 (95% CI 0.610.83, P < .0001).

To clarify the effects of the study drugs on patients at different levels of risk, we stratified patients by the RCRI and evaluated the effects of the study drugs on mortality at 2 years, comparing the results to a referent of taking no study drugs. The use of both a statin and a beta‐blocker consistently produced a relative risk reduction (RRR) of approximately 50% with an NNT of 410, with highly statistically significant results for patients at all levels of risk (Table 5). As patient risk level increased, the NNT decreased, consistent with higher‐risk patients benefiting most from combination therapy with statins and beta‐blockers.

In addition, the range of outcomes can be clearly seen for both patient‐specific risk level and study drug use. For example, overall mortality at 2 years for all patients was 22%. For the study drugs, mortality ranged from 16% for patient who used both a statin and a beta‐blocker to 27% for those patients who used neither study drug. The use of the RCRI showed that the healthiest patients who were taking both a statin and a beta‐blocker did the best, with a 2‐year mortality of 9%, compared with the sickest patients who were taking neither study drug, whose 2‐year mortality was 59%. Use of both study drugs by the sickest patients was associated with a reduction in 2‐year mortality to 32% (P < .0001; Table 5).

Propensity Score Analysis of Use of Combination Study Drugs

Because there was very limited literature to guide us in the use of propensity score analysis of multiple treatment groups, we performed these analyses in an exploratory manner. There were significant differences between combination statin and beta‐blocker users and nonusers. These differences became insignificant after adjusting for propensity score, except for the 5 variables previously mentioned, which were added to the model as potential confounders (Table 4). The propensity‐adjusted Cox regression model comparing use of each study drug alone and in combination with taking neither over the whole study period still showed an association with decreased mortality. The combination of statins and beta‐blockers had an HR of 0.56 (95% CI 0.420.74), P < .0001; statins alone had an HR of 0.79 (95% CI 0.620.99), P = .0472; and beta‐blockers alone had an HR of 0.80 (95% CI 0.670.94), P = .0183.

Combination Study Drugs and Revised Cardiac Risk Index: Propensity Analysis

We performed the stratified Cox regression adjusted for the propensity scores for each level of RCRI and estimated 2‐year mortality. The use of both a statin and a beta‐blocker compared with using none was still consistently statistically significant, with an RRR of approximately 36% and an NNT of 820 for all levels of patient risk (Table 6). Possibly because of the reduced number of patients in each RCRI category, neither single‐agent study drug compared with none showed a statistically significant decrease in mortality at any level of patient‐specific risk (Table 6). Again, higher‐risk patients benefited most from combination therapy.

Study Drug Timing: Subcohort Analysis

A subcohort analysis was performed to clarify the timing of the study drugs. Of the patients taking statins, 69 of 1346 (5.1%) took the drug before surgery only, 119 of 1346 (8.8%) took the drug after surgery only, and 1158 of 1346 (86%) took the drug both before and after surgery. Of the patients taking beta‐blockers, 54 of 1617 (3.3%) took the drug before surgery only, 397 of 1617 (24.6%) took the drug after surgery only, and 1166 of 1617 (72.1%) took the drug both before and after surgery. The use of statins and beta‐blockers had a correlation of 0.29 (contingency coefficient).