Author + information
- Received January 23, 2001
- Revision received August 8, 2001
- Accepted August 22, 2001
- Published online December 1, 2001.
- Brahmajee K Nallamothu, MD, MPHa,2,
- Sanjay Saint, MD, MPHa,b,
- Scott D Ramsey, MD, PhDc,
- Timothy P Hofer, MD, MSca,b,
- Sandeep Vijan, MD, MSca,b and
- Kim A Eagle, MD, FACCa,* ()
- ↵*Reprint requests and correspondence:
Dr. Kim A. Eagle, Division of Cardiovascular Disease, Department of Internal Medicine, University of Michigan Medical Center, 3910 Taubman Center, Ann Arbor, Michigan 48109-0366, USA.
Objectives The goal of this study was to determine whether outcomes of nonemergent coronary artery bypass grafting (CABG) differed between low- and high-volume hospitals in patients at different levels of surgical risk.
Background Regionalizing all CABG surgeries from low- to high-volume hospitals could improve surgical outcomes but reduce patient access and choice. “Targeted” regionalization could be a reasonable alternative, however, if subgroups of patients that would clearly benefit from care at high-volume hospitals could be identified.
Methods We assessed outcomes of CABG at 56 U.S. hospitals using 1997 administrative and clinical data from Solucient EXPLORE, a national outcomes benchmarking database. Predicted in-hospital mortality rates for subjects were calculated using a logistic regression model, and subjects were classified into five groups based on surgical risk: minimal (<0.5%), low (0.5% to 2%), moderate (2% to 5%), high (5% to 20%), and severe (≥20%). We assessed differences in in-hospital mortality, hospital costs and length of stay between low- and high-volume facilities (defined as ≥200 annual cases) in each of the five risk groups.
Results A total of 2,029 subjects who underwent CABG at 25 low-volume hospitals and 11,615 subjects who underwent CABG at 31 high-volume hospitals were identified. Significant differences in in-hospital mortality were seen between low- and high-volume facilities in subjects at moderate (5.3% vs. 2.2%; p = 0.007) and high risk (22.6% vs. 11.9%; p = 0.0026) but not in those at minimal, low or severe risk. Hospital costs and lengths of stay were similar across each of the five risk groups. Based on these results, targeted regionalization of subjects at moderate risk or higher to high-volume hospitals would have resulted in an estimated 370 transfers and avoided 16 deaths; in contrast, full regionalization would have led to 2,029 transfers and avoided 20 deaths.
Conclusions Targeted regionalization might be a feasible strategy for balancing the clinical benefits of regionalization with patients’ desires for choice and access.
Several studies have demonstrated that hospitals that perform a small number of coronary artery bypass graft (CABG) surgeries have higher in-hospital mortality rates than facilities with greater surgical volumes (1–9). As a result, some authorities have suggested that all cases of nonemergent CABG be referred, or “regionalized,” to high-volume surgical centers (1,10). It is estimated that in 1997 such a referral strategy from low- to high-volume hospitals might have avoided 258 CABG-related deaths in California alone (11). However, implementing regionalization could result in potential adverse outcomes by limiting patient choice and access to care, increasing transfer and travel-related costs and reducing the availability of surgical services in many locations (12). Furthermore, it is uncertain how well high-volume surgical centers could accommodate large numbers of additional cases without creating significant delays or lowering the quality of patient care. Conceivably, a strategy of “targeted regionalization” could balance these risks and benefits, allowing some cases of CABG to be performed at low-volume facilities and transferring only those individuals who would clearly benefit from the specialized services provided by high-volume hospitals.
Implicit in the concept of targeted regionalization is the idea that CABG-related outcomes differ at low- and high-volume hospitals in specific subgroups of patients and not in all individuals. The fact that patients undergoing CABG are heterogeneous and vary considerably in disease severity and surgical risk is consistent with this premise (13). Elective cases in middle-aged men with no preexisting medical conditions, for instance, are associated with an extremely low in-hospital mortality rate (<0.5%); in contrast, urgent cases in 80-year-old women with diabetes mellitus and chronic renal insufficiency have an estimated in-hospital mortality rate of 10% (13). We, therefore, hypothesized that, while CABG-related outcomes such as in-hospital mortality and hospital costs would differ substantially between low- and high-volume facilities in specific subgroups of patients at high surgical risk, minimal differences would exist in subjects at low surgical risk. If true, this would suggest that CABG could be safely performed in low-risk patients at low-volume centers, thereby indicating that targeted regionalization is a viable strategy for policy-makers to consider.
Patients and facilities
We used data from Solucient EXPLORE, a healthcare outcomes benchmarking database from Solucient, LLC for our analysis. Solucient EXPLORE contains administrative and clinical data from more than 150 healthcare facilities across the U.S. and has been used in previous clinical studies evaluating CABG (14,15). Data in Solucient EXPLORE are abstracted from discharge-based claim forms that contain demographic and limited clinical information including age, gender, race, surgical priority, discharge status and up to 40 diagnoses and 20 procedures as identified by International Classification of Diseases, Ninth Edition (ICD-9-CM) codes. Accuracy is ensured through a routine audit of records for internal inconsistencies or coding errors and includes procedures for identifying unmapped charge codes and inconsistencies between physician codes and hospital codes for the same patient admission.
To identify subjects for our analysis, we searched all discharge records from the database between January 1, 1997 and December 31, 1997 for one of eight ICD-9-CM procedure codes for CABG (codes 36.10 to 36.16 and code 36.19). All subjects 35 years or older who had undergone isolated, nonemergent CABG were included in the final data set. Subjects who had undergone emergent CABG, surgeries other than isolated CABG (e.g., concomitant valve or aortic surgery) or coronary angioplasty and CABG during the same hospitalization (presumably for complications of angioplasty) were excluded. After removing potential patient and hospital identifiers, Solucient provided data to researchers at the University of Michigan and University of Washington for unrestricted use and analysis.
Collected data included age, gender, discharge status, surgical priority (elective vs. urgent), severity of illness, hospital costs and length of stay. Information on surgical priority from the claim forms was supplied by the institutions and relied on administrative staff, not the surgeon. Severity of illness was assessed using the All Patient Refined-Diagnosis Related Groups (APR-DRG) (3M/HIS Corporation, Minneapolis, Minnesota), a discharge abstract-based case-mix measure that subclassifies subjects within diagnosis related group (DRG) categories (15). Within each DRG, subjects are categorized into four severity levels (minor, moderate, major or extreme) based on: 1) secondary diagnoses (e.g., diabetes mellitus, chronic obstructive pulmonary disease, hypertension); 2) the complexity of the secondary diagnoses and their interaction with each other; and 3) the interaction between the secondary and principal diagnoses. For our analysis, APR-DRGs subclassified subjects in DRG 106 (hospitalizations of CABG without cardiac catheterization) and DRG 107 (CABG with cardiac catheterization). The APR-DRG has been shown to be useful and valid as a predictor of in-hospital mortality in subjects undergoing CABG (16,17).
Hospital costs for each subject were estimated using a comparative costing methodology from Solucient, which captures both direct and indirect hospital costs from the in-house accounting system of each facility (14). Hospital costs are standardized across facilities using a two-tiered method that allocates costs from: 1) the general ledger to standardized cost centers and then, 2) to standardized transaction codes (based on the Medicare Revenue Code) employing primarily a Relative Value Unit costing methodology. Hospital-specific information in the database included the size and location of a facility and its annual CABG volume.
We used a logistic regression model to calculate predicted risks of in-hospital death for each subject in the database. The risk model contained four known predictors of CABG-related in-hospital mortality: age (less than 50 years old, 50 to 59 years old, 60 to 69 years old, 70 to 79 years old, 80 years or older), gender, surgical priority and severity of illness (using APR-DRG). Subjects were divided into five risk groups based on their predicted likelihood of suffering an in-hospital death: minimal (<0.5%), low (0.5% to <2%), moderate (2% to <5%), high (5% to <20%) and severe (≥20%). The Hosmer-Lemeshow goodness-of-fit statistic was used to assess overall fit in the risk model (18). Model discrimination, or the ability of the risk model to distinguish between subjects who suffered an in-hospital death and those who did not, was measured using the C-index (19).
We classified surgical centers into low- and high-volume CABG facilities based on an annual volume of 200 nonemergent cases. This threshold was chosen after a careful review of the literature and preliminary data analyses. To determine whether our results were sensitive to the choice of a specific threshold, we repeated our analyses varying the threshold from 100 to 300 cases.
We used chi-square tests to assess for differences between in-hospital mortality rates at low- and high-volume facilities and supplemented these tests with odds ratios and 95% confidence intervals. Differences in hospital costs and length of stay were assessed using the Student ttest; because of highly skewed distributions, however, we performed logarithmic transformations on the two variables before performing any statistical tests. Results for the two variables were then retransformed into their natural units using established methods of statistical simulation before final reporting (20). For all statistical analyses, we used robust variance estimates to adjust for the potential effects of clustering at the hospital level (21). All statistical analyses were performed using Stata 6.0 (Stata Corporation, College Station, Texas).
We used results from the above analyses to estimate the number of transfers and avoided deaths that would have occurred if “full” or “targeted” regionalization strategies had been applied to subjects in the Solucient EXPLORE database.
The study population consisted of 13,644 subjects who underwent nonemergent CABG at 56 hospitals in 26 different states. Of these subjects, 7,448 (54.5%) had surgery at a teaching institution and 13,552 (99.3%) at an urban facility. Twenty-five of the facilities had performed less than 200 CABG surgeries in 1997 and were classified as low-volume centers; the remaining 31 facilities were considered high-volume centers (Fig. 1). The mean number of surgeries performed at each hospital was 249 (standard error, 27.4), with 2,029 cases (14.9%) performed at low-volume centers and 11,615 (85.1%) at high-volume facilities. Table 1summarizes the general characteristics of the study population at the different facilities. A significant difference in in-hospital mortality rates was seen between low- and high-volume facilities (3.3% vs. 1.9%; p = 0.001). Compared with high-volume facilities, subjects at low-volume hospitals had trends toward greater hospital costs (low-volume centers, $21,611 vs. high-volume centers, $19,090; p = 0.052) and longer in-hospital lengths of stay (low-volume centers, 8.5 vs. high-volume centers, 7.9; p = 0.09).
Table 2shows results of the logistic regression model for predicting the risk of in-hospital death. Based on the risk model, 7,047 (51.7%) subjects were classified a priori as minimal risk, 4,409 (32.3%) as low risk, 1,273 (9.3%) as moderate risk, 454 (3.3%) as high risk and 461 (3.4%) as severe risk. Predicted in-hospital mortality rates were similar at low- and high-volume hospitals in the different risk groups except for subjects at severe risk (25.5% at low-volume hospitals vs. 28.0% at high-volume hospitals; p < 0.001). Overall, the observed in-hospital mortality rate was 0.2% for those subjects at minimal risk, 1.1% for those at low risk, 2.7% for those at moderate risk, 14.1% for those at high risk and 26.9% for those at severe risk. The Hosmer-Lemeshow goodness-of-fit test for the risk model was nonsignificant, indicating little departure from a “perfect fit” (p = 0.62). The C-index for the risk model was 0.89, which suggests the model had good predictive discrimination.
Table 3shows observed in-hospital mortality rates for subjects at low- and high-volume centers in each of the five risk groups. Significant differences in in-hospital mortality rates were seen between low- and high-volume centers in subjects at moderate (5.3% vs. 2.2%; p = 0.007) and high risk (22.6% vs. 11.9%; p = 0.026) (Fig. 2). While the in-hospital mortality rate for low-risk subjects was slightly higher at low-volume centers, this difference did not reach statistical significance (1.7% vs. 1.0%; p = 0.19). No significant differences in in-hospital mortality rates were found between low- and high-volume centers in those at minimal or severe risk. In addition, no significant differences in hospital costs or length of stay were seen between the low- and high-volume facilities in any of the five risk groups (Table 4). In general, the use of lower or higher volume thresholds for defining centers (e.g., 100 or 300 annual cases) slightly diminished—but did not eliminate—the in-hospital mortality differences between low- and high-volume centers in each of the five risk groups, suggesting that 200 annual cases was a reasonable threshold.
Table 5displays the likely outcomes of implementing different “full” and “targeted” regionalization strategies on subjects in the Solucient EXPLORE database. In calculating these estimates, we assumed that transferring minimal or severe-risk subjects to high-volume centers would have no impact on outcomes since differences in these groups were nonsignificant. However, to favor full regionalization strategies, we assumed that transfers of low-risk subjects to high-volume facilities would result in better outcomes since differences in in-hospital mortality rates were of borderline significance in this group. Sections in Table 5each represent different strategies based on various definitions of a low-volume center (e.g., 100 to 300 annual cases) and different thresholds for transfer based on surgical risk (e.g., low risk and above or moderate risk and above). Transfer of moderate- and high-risk subjects from centers with less than 200 annual cases would have resulted in 370 transfers and 16 “avoided deaths” or, in other words, a number needed to transfer of 23 to avoid a single death. In contrast, full regionalization of all subjects from facilities with less than 200 annual cases of CABG would lead to 2,029 transfers and 20 avoided deaths, yielding a number needed to transfer of 101. When compared with targeted regionalization, however, the incremental number needed to transfer for full regionalization would have been 415 ([2,029 transfers − 370 transfers]/[20 avoided deaths −16 avoided deaths]).
The objective of this study was to determine whether in-hospital mortality, hospital costs and length of stay for CABG varied between low- and high-volume facilities across subjects at different levels of surgical risk. If the beneficial effects of high-volume hospitals are concentrated in an identifiable subgroup of patients, strategies of regionalization could potentially be “targeted” in order to maximize their effectiveness and to minimize possible disadvantages.
Several of our findings are noteworthy. First, we found no significant differences in in-hospital mortality rates between low- and high-volume hospitals in subjects at minimal or low surgical risk. Since 84% of patients at low-volume facilities were classified as either minimal or low risk, nonselectively transferring all individuals to high-volume hospitals, or full regionalization, is likely to have resulted in little to no benefit in most cases. In contrast, we found in-hospital mortality rates at low-volume facilities to be significantly higher in subjects at moderate and high risk, with the adjusted risk of in-hospital death more than twofold greater when compared with high-volume centers. Thus, the association between hospital volume and clinical outcomes is almost entirely due to outcome differences in moderate- and high-risk subjects and suggests that, if feasible, targeted regionalization strategies that focus on identifying and referring such subjects to high-volume hospitals might be as effective as full regionalization strategies. In fact, we estimated that implementing a strategy of targeted regionalization (i.e., transferring subjects at moderate-risk or higher from hospitals with less than 200 annual cases) in our study population would have resulted in 370 transfers and avoided 16 deaths. Full regionalization would have led to 1,659 more transfers to avert four additional deaths or an incremental number needed to transfer of 415 to avoid a single death.
Next, we found no significant differences between low- and high-volume centers in those subjects who were at severe risk for an in-hospital death. We offer three possible explanations for this finding. First, given the small number of severe-risk subjects at low-volume centers (n = 90), this result might just reflect random statistical variation. Second, since we broadly stratified subjects at severe risk—predicted in-hospital mortality rates between 20% and 100%—we might have missed important differences in surgical risk between severe subjects at low- and high-volume hospitals. The fact that we found a statistically significant difference between the predicted in-hospital mortality rates of severe-risk subjects at low- and high-volume facilities (25.5% vs. 28.0%; p < 0.001) supports this hypothesis and suggests that, at high-volume hospitals, these subjects had, on average, higher surgical risk. Finally, we used the APR-DRG, a discharge abstract-based case-mix measure, to adjust for severity of illness in our risk model. Use of the APR-DRG might have resulted in inappropriate adjustment for postprocedural complications, not just pre-existing risk factors, in our risk model and potentially minimized any differences that actually exist between low- and high-volume centers for subjects at severe risk.
Regionalization in previous reports
What are the possible disadvantages of regionalization? First, regionalization largely ignores patient choices and preferences for healthcare, which current evidence suggests could be substantial. A recent study by Finlayson et al. (21), for instance, demonstrated that 100% of subjects given a standardized scenario favored having high-risk surgery performed at a local hospital if outcomes were identical to those at regional centers. Furthermore, nearly 75% of subjects still preferred surgeries at local hospitals when given successive scenarios in which local in-hospital mortality rates were higher than rates at regional centers (22). Second, regionalization could result in considerable travel burdens for patients and their families, particularly for those in distant rural areas. This is potentially important since some evidence suggests that, as distances between patients and providers increase, access to care worsens and healthcare services are underutilized (23).
Third, regionalization could have an impact on the economic viability of small- to moderate-sized healthcare facilities. Without adequate CABG volume, these hospitals would lose a substantial amount of revenue and have a difficult time recruiting and retaining cardiothoracic surgeons (24). Consequently, a facility’s loss of CABG patients would likely extend to other cardiothoracic-related services such as coronary angioplasty or thoracic surgery or to its ability to provide emergency surgical services. While it does not entirely eliminate several of these potential disadvantages, targeted regionalization does minimize them. In addition, targeted regionalization is likely to be a more acceptable alternative for low-volume centers and might avoid the “political firestorm” that others have suggested will occur over policies of full regionalization (25).
We are aware of only one other study examining whether clinical outcomes of CABG differed between low- and high-volume hospitals in a specific subgroup of patients. In 1987, Showstack and colleagues (26)demonstrated that, in comparison with scheduled cases, in-hospital mortality differences between low- and high-volume hospitals were more marked in nonscheduled or emergent surgeries. The authors concluded that the greater skills and experiences of surgical teams at high-volume hospitals were more valuable in CABG during high-risk situations. However, since emergent cases account for a small number of total CABG surgeries and such cases are often difficult to transfer, selectively regionalizing emergent cases alone is unlikely to have much impact as an isolated policy measure.
In contrast with differences in in-hospital mortality rates, we did not find that hospital costs or lengths of stay differed significantly between low- and high-volume centers after subjects were stratified into the five risk groups. While it is assumed that regionalizing CABG to high-volume centers would reduce hospital costs by providing economies of scale, eliminating wasteful duplication and improving clinical outcomes, direct evidence for such benefits is scarce and, at times, conflicting (27,28). Most recently, Menke and Wray (29)found that the long-term cost savings from regionalization of CABG to high-volume hospitals could be substantial; however, their estimates were based on a study of Veterans Affairs hospitals and might not be generalizable to the rest of the healthcare system. Without better data, there is a legitimate concern that regionalization could actually increase overall costs by leading to costly transfers or centralizing services at large teaching hospitals where care may be, in fact, more expensive (30).
Our findings should be interpreted in the context of the following limitations. First, this study was observational in nature and does not establish causal relationships. We can only speculate that higher surgical volumes at institutions lead to better in-hospital outcomes (i.e., the “practice makes perfect” hypothesis) (5). This is important to note since it has been suggested that at least part of the differences in outcomes between low- and high-volume centers might be due to the fact that better hospitals attract more patients (i.e., the “selective referral” hypothesis) (5). Second, we were unable to include all the clinical indexes that have been linked to adverse CABG outcomes in our risk model, such as a history of previous CABG, poor left ventricular function, cerebrovascular disease or peripheral vascular disease (13). Instead, we used the APR-DRG, a discharge abstract-based case-mix measure, to adjust for overall severity of illness and, while it has been established as a useful tool for estimating in-hospital death, length of stay and hospital costs for CABG, it is likely that some degree of residual confounding existed. Moreover, since the APR-DRG uses discharge diagnoses to risk-stratify patients, it is unable to distinguish between preexisting risk factors and postprocedure complications, which can also effect hospital mortality. This last issue is likely to explain, in part, the risk model’s high C-index of 0.89.
Third, we focused on in-hospital mortality rates as our primary clinical outcome. Thus, we could not account for transfers and differential length of stay patterns between facilities or examine the impact of high-volume hospitals on the incidence of important complications such as stroke, renal failure, deep sternal wound infection or long-term mortality and function. Fourth, we were unable to determine the degree to which targeted regionalization already exists among the hospitals in the Solucient EXPLORE database. The distribution of patients based on surgical risk was similar between low- and high-volume centers, however, which suggests that a substantial number of high-risk patients were not being selectively transferred to high-volume centers. Finally, the Solucient EXPLORE database was composed of hospitals interested in providing clinical and economic data for purposes of comparative benchmarking activity; these facilities may not be representative of all hospitals performing CABG in the U.S.
Despite these limitations, our results may have important policy implications. Our finding that in-hospital mortality rates for CABG differed at low- and high-volume hospitals in subjects at moderate and high risk—but not in those at minimal or low risk—suggests that targeted regionalization might be a feasible strategy for effectively balancing the clinical benefits of regionalization with patients’ desires for choice and access. In addition, targeted regionalization is likely to be a more acceptable option for local providers and hospitals, given the realities of the current healthcare system. Of course, further research will be needed to confirm our findings using larger, CABG-specific clinical datasets and possibly through future clinical trials.
The authors thank Keith Kelley, RN, MPH, of Solucient, LLC, for his assistance in data collection and John E. Billi, MD, and Laurence F. McMahon, MD, MPH, for their careful review of the manuscript and helpful comments.
- All Patient Refined-Diagnosis Related Groups
- coronary artery bypass grafting
- diagnosis related group
- International Classification of Diseases, Ninth Edition
- Received January 23, 2001.
- Revision received August 8, 2001.
- Accepted August 22, 2001.
- American College of Cardiology
- Sowden A.J,
- Deeks J.J,
- Sheldon T.A
- Tu J.V,
- Naylor C.D
- Eagle K.A,
- Guyton R.A,
- Davidoff R,
- Ewy G.A,
- Fonger J
- Hosmer D.W,
- Lemeshow S
- Hadorn D.C,
- Keeler E.B,
- Rogers W.H,
- Brook R.H
- King G,
- Tomz M,
- Wittenberg J
- Rogers W.H
- Finlayson S.R,
- Birkmeyer J.D,
- Tosteson A.N,
- Nease R.F