Should Volume Standards for Cardiovascular Surgery Focus Only on High-Risk Patients?
Background— Payers and policy makers are attempting to concentrate selected cardiovascular procedures in high-volume centers. A recent analysis of coronary artery bypass grafting (CABG), however, suggests that volume-based referral initiatives should focus only on high-risk patients.
Methods and Results— Using the national Medicare database (1994 to 1999), we studied the operative mortality in patients undergoing 4 cardiovascular procedures (CABG, aortic valve replacement, mitral valve replacement, and elective abdominal aortic aneurysm repair). We defined 2 categories of patient risk: high-risk (patients in the highest 25th percentile of predicted risk on the basis of a logistic regression model) and low-risk (patients in the lowest 75th percentile). We then compared operative mortality in patients undergoing surgery at very-high volume hospitals (VHVH, highest 20th percentile of procedure volume) and very-low volume hospitals (VLVH, lowest 20th percentile of procedure volume). Absolute differences in operative mortality between VLVH and VHVH were somewhat larger in high-risk patients. However, volume-related differences in mortality were also significant for low-risk patients undergoing one of the 4 procedures. In relative terms, the effect of hospital volume was similar in both high- and low-risk patients. For high- and low-risk patients, the relative risk (RR) of mortality between VHVH and VLVH were nearly equal for CABG (RR=0.78 for low-risk patients, RR=0.77 for high risk patients), aortic valve replacement (0.73 versus 0.76), mitral valve replacement (0.73 versus 0.74), and abdominal aortic aneurysm repair (0.51 versus 0.54).
Conclusions— Although the merits of volume-based referral initiatives can be debated on many grounds, there seems to be little rationale for restricting these initiatives to high-risk patients.
Received September 25, 2002; revision received November 11, 2002; accepted November 21, 2002.
Prompted by numerous studies describing an inverse relationship between volume and outcome in cardiac surgery,1–4 several policy efforts are attempting to concentrate complex cardiovascular procedures in high-volume hospitals. The Leapfrog group, a coalition of over 100 public and private employers representing 30 million patients, has recommended to its members that they undergo coronary artery bypass grafting (CABG) in hospitals that perform at least 500 of these procedures per year.5 The Center for Medicare and Medicaid services is also considering whether they should inform Medicare patients about the importance of volume in several high-risk surgical procedures and publish hospital-specific volume data. If fully implemented, volume-based referral initiatives could potentially prevent thousands of surgical deaths.5,6 Achieving this goal would require moving hundreds of thousands of patients, however, which could have many undesirable effects on the healthcare system.7–9
Restricting these initiatives to high-risk patients may be one way to significantly reduce mortality while minimizing how many patients have to move. A recent analysis of 13 644 patients undergoing CABG reported a substantial volume-outcome relationship in high-risk patients, but little effect (in absolute or relative terms) in low-risk patients.10 The authors concluded that restricting volume-based referral initiatives to high-risk patients might dramatically reduce the number of patients moved, yet still save almost as many lives as with full implementation of regionalization.
Because this finding could have important implications for the ongoing debate about volume-based referral initiatives, we sought to replicate this CABG study using the much larger national Medicare database. To explore its generalizability to other procedures, we also examined relationships between volume and mortality with aortic valve replacement (AVR), mitral valve replacement (MVR), and abdominal aortic aneurysm (AAA) repair.
Using data from the national Medicare database (MEDPAR, 1994 to 1999), we used International Classification of Diseases (ICD)-9 codes to identify patients undergoing 1 of 4 cardiovascular procedures.1 Patients undergoing concomitant valve replacement were excluded from the CABG cohort. However, the valve replacement cohorts included patients undergoing concomitant CABG. We excluded patients undergoing AAA repair if they had a diagnosis or procedure code suggesting rupture or a thoracoabdominal aneurysm. Patients under the age of 65 years and over the age of 99 years were excluded.
Defining Patient Risk Groups
As described in detail elsewhere,1 we used multiple logistic regression models to estimate the predicted risk of death for each patient on the basis of demographic characteristics, comorbidity, and urgency of admission. Covariates in the predicted risk model included patient age (65 to 69 years, 70 to 74 years, 75 to 79 years, 80 to 84 years, and 85 to 99 years), sex, race (black, non-black), mean social security income, admission acuity (elective or urgent/emergent), and Charlson comorbidity score (0,1,2, and 3 or more).11 Comorbidities comprising the Charlson score include congestive heart failure, prior myocardial infarction, peripheral vascular disease, diabetes mellitus, chronic obstructive pulmonary disease, and 14 other conditions. Comorbidities were identified from prior hospitalizations, as well as from the admission of the index procedure, and were weighted according to their relative effects on mortality.11,12 We also included 2- and 3-way interactions among age, sex, and race, as well as their main effects. For AVR and MVR, we also included concurrent CABG as a covariate. We stratified patients into two levels of operative risk, low and high. The low-risk group contained patients in the lowest 75th percentile of predicted mortality. The high-risk group contained patients in the highest 25th percentile of predicted mortality. We repeated our analyses using additional patient risk strata. Because these analyses did not alter our main conclusions, however, here we present only analyses based on the binary risk definition.
Defining Volume Categories
We determined the average annual number of procedures that most closely sorted the patients in each procedure cohort into 5 evenly sized groups (quintiles). To simplify the presentation of our results, we decided to focus on mortality at very low-volume hospitals (VLVHs, those in the lowest 20th percentile of volume) and very high-volume hospitals (VHVHs, those in the highest 20th percentile).
Our main outcome measure was operative mortality, which was assessed at the patient level. Operative mortality was defined as death before hospital discharge or within 30 days after the index procedure. Because patients were already stratified by predicted risk, we compared observed (unadjusted) mortality rates at VLVHs and VHVHs in our primary analysis. To account for residual confounding within patient risk groups, we repeated the analysis while adjusting for patient characteristics. These results were essentially identical to those from the unadjusted analysis.
As expected, patients in the high-risk group, when compared with low-risk patients, were significantly older, more likely to be female, had higher Charlson scores, and were more likely to be admitted urgently or emergently (Table). As a result, both predicted and observed mortality rates were nearly twice as high for high-risk patients than for low-risk patients for all procedures. Within risk categories, however, patient characteristics (and thus predicted mortality rates) were very similar for patients at VLVHs and VHVHs.
For all 4 procedures, observed operative mortality rates were significantly lower at VHVHs than at VLVHs (Figure). Absolute differences in observed operative mortality between VLVHs and VHVHs were somewhat larger in high-risk patients. However, volume-related differences in mortality were also significant for low-risk patients in all 4 procedures. In terms of relative risks, the effect of hospital volume was similar in both high- and low-risk patients. The relative risk of mortality between VLVHs and VHVHs were nearly equal for CABG (0.78 versus 0.77, respectively), AVR (0.73 versus 0.76), MVR (0.73 versus 0.74), and AAA repair (0.51 versus 0.54).
Like many previous studies, our study describes a significant volume-outcome effect for 4 cardiovascular procedures. Unlike recent work by Nallamothu et al,10 however, we did not find the volume effect to be restricted to high-risk patients. Whereas absolute differences in observed operative mortality were larger in high-risk patients for all 4 procedures in our study, relative risks of mortality between VHVHs and VLVHs were nearly identical for high-risk and low-risk patients.
There are several possibilities for the discordant findings between our study and that of Nallamothu et al. First, because the study by Nallamothu et al was relatively small (n=13 644), its results are likely less precise than our findings, which were based on over 800 000 patients. Second, the database used for the study by Nallamothu et al was limited to 150 hospitals, which limits the generalizability of their findings. In contrast, our study is based on essentially all hospitals performing cardiac surgery in the United States. Moreover, our findings were consistent across 4 common cardiovascular procedures.
Our study has 2 important limitations. First, given the well-described limitations of claims data for classifying illness severity,13,14 our risk prediction models were limited in their ability to discriminate high- and low-risk patient groups. Nonetheless, the 2-fold differences in predicted and observed mortality rates between high- and low risk patients with each procedure suggest at least face validity for our risk prediction methods. Second, because we relied on Medicare data (patients over 65 years of age), our analysis does not include the very low-risk subgroup of younger patients. Although the generalizability of our findings to younger patients is unknown, it is important to note that Medicare patients comprise a majority of patients undergoing cardiovascular procedures and an even larger majority of those who die perioperatively.15
The merits of volume-based referral initiatives like the Leapfrog Group effort are debated hotly.7,16,17 Pointing to the magnitude of observed volume-outcome associations, proponents argue that these initiatives could potentially avert thousands of surgical deaths and that, at the very least, patients should be informed about the importance of volume for some procedures. Opponents point out that the potential benefits of volume-based referral initiatives come at the cost of disrupting traditional referral lines for tens or even hundreds of thousands of patients. Some of indirect effects of these policies could be harmful for both patients and small hospitals. Our study does not address these basic tradeoffs. However, it does suggest that restricting volume-based referral initiatives to high-risk patients is unlikely to be an easy solution to this policy dilemma.
This study was supported by a grant (R01-HS10141-01) from the Agency for Healthcare Research and Quality.
The views expressed herein do not necessarily represent the views of the United States Government.