The Impact of Hospital’s Cardiac Specialization on Outcomes after Coronary Artery Bypass Graft Surgery: Analysis of Medicare Claims Data

Abstract

Background

Hospital volume has been widely embraced as a proxy measure for hospital quality; little attention has been focused on an alternative quality measure-hospital specialization. Even though specialization occurs on a continuum, previous studies have only focused on a small number of highly specialized hospitals (single-specialty hospitals). Studies on the broad relationship between hospital specialization and outcomes after coronary artery bypass grafting (CABG) are limited.

Methods and Results

We conducted a retrospective cohort study of 705,084 Medicare patients (1130 hospitals) who underwent CABG during 2001-2005. We stratified hospitals into quintiles based on their degree of cardiac specialization (proportion of a hospital’s Medicare discharges classified as Major Diagnostic Category 5 – cardiovascular diseases). We compared patient and hospital characteristics and outcomes across quintiles of cardiac specialization. Patient characteristics were generally similar across quintiles, but mean annual CABG volume increased progressively from quintile 1 (least-specialized) to quintile 5 (most-specialized). Unadjusted 30-day mortality was similar at hospitals in quintiles 1-4 (4.8%), except quintile 5 where mortality was lower (4.3%). A strong inverse association was seen between hospital cardiac specialization and 30-day mortality after adjustment for patient characteristics (P _trend=0.001). However, this was no longer significant after additional adjustment for CABG volume (P _trend=0.65). Results were similar for other mortality outcomes and length of stay.

Conclusions

After accounting for patient characteristics and CABG volume, greater cardiac specialization was not associated with clinically significant improvement in patient outcomes. This study calls into question the benefit of cardiac specialization for the vast majority of CABG-performing U.S. hospitals.

Introduction

Nearly 450,000 patients undergo coronary artery bypass graft surgery (CABG) surgery each year in the United States.¹ Innovations in surgical technique, anesthesia, peri-operative care and quality improvement initiatives have led to a significant decline in mortality after CABG.² Numerous studies have found a strong association between higher hospital volume and improved patient outcomes after CABG.^3-5 Based on these studies, an array of public and private coalitions including the Leapfrog Group have promoted regionalization whereby patients requiring CABG are directed to higher volume hospitals (i.e. > 450 CABG per year) with the hope of improving patient outcomes.^{6, 7}

However, CABG volume has been shown to be an imperfect measure of hospital quality with presence of low-quality/high-volume hospitals and high-quality/low-volume hospitals.^{8, 9} Also, volume-based referrals could limit patient access and adversely impact those low-volume hospitals that are currently delivering high-quality CABG services. Hospital specialization may be an alternative measure of quality, but has received far less attention. Specialization may be conceptualized as the degree to which a given hospital focuses its resources on specific diagnoses (e.g., orthopedic diseases) or procedures (e.g., CABG) and may be quantified as the proportion of a hospital’s total admissions falling within a single disease category or undergoing a specific procedure.^10-13 Thus, a hospital may be low-volume but highly specialized if it concentrates resources in select areas.

The majority of previous studies on hospital specialization have been in the context of physician-owned specialty hospitals, which represent the most extreme example of specialization. While these studies have demonstrated 10%-20% better risk-adjusted outcomes at specialty hospitals compared to competitors,^{10, 12, 14} by focusing on a very small number of specialty hospitals (fewer than 50), these studies have overlooked the fact that specialization occurs on a continuum and applies broadly to all hospitals. Recent studies using more general definition of specialization have found hospital specialization to be associated with improved outcomes after primary percutaneous coronary intervention (PCI)¹³ and major joint replacement surgery.¹¹ However, studies investigating the broad relationship of hospital specialization with CABG outcomes remain limited.

This study examines the overall relationship of hospital cardiac specialization with patient mortality and length of stay (LOS) after CABG. We hypothesized that hospitals with greater cardiac specialization would have improved outcomes after adjusting for patient characteristics, hospital characteristics and procedural volume.

Methods

Data Sources

1. Medicare Provider and Analysis Review (MedPAR) Part A, 2001-2005

2. American Hospital Association Survey (AHA), 2001-2005

3. US Census Tract Files, 2000

Hospital-Level Data

Using MedPAR Part A data, we first identified all hospitals (n=1194 hospitals) that performed any CABG on fee-for-service Medicare beneficiaries during January 1, 2001-December 31, 2005 (procedure codes 36.10-36.19, International Classification of Diseases, 9^th Clinical Modification (ICD-9-CM)). For each hospital we determined the average Medicare CABG volume by dividing the total number of CABG procedures performed at the hospital during the study period by the number of years each hospital performed any CABG. We calculated each hospital’s average Medicare total volume (all discharges) and average MDC5 volume (MDC5: Major Diagnostic Category 5, diseases of cardiovascular system, ICD9-CM). MDC5 constitutes an amalgamation of ICD9-CM codes that relate to cardiovascular diseases. We excluded hospitals that performed < 10 CABG procedures per year to prevent inclusion of CABG procedures that may have reflected data coding errors (n=64 hospitals). For the remaining 1130 hospitals, we obtained additional hospital-level information such as teaching status (membership in the council of teaching hospitals), number of beds, and ownership (for profit/not for profit) by linking the hospital level MedPAR data to the AHA Annual Survey (2001-2005).¹⁵ Successful linkage was performed on 1121 hospitals (99%). Since we did not include any of the AHA variables in our models, hospitals that couldn’t be matched to the AHA dataset (n=9 hospitals) were not excluded from our study. Lastly, we categorized each hospital’s geographic location into census regions (Northeast, South, Midwest and West) and urban or non-urban using rural-urban commuting area codes obtained from the 2000 United States Census.¹⁶

Cardiac Specialization

We define hospital cardiac specialization as the degree to which a hospital concentrates its resources in treating patients with cardiovascular diseases relative to other diseases. Specifically, we measured cardiac specialization as the proportion of all Medicare discharges during the study period classified as MDC5. This approach has been previously used to not only identify specialty hospitals^{10, 17} but also to categorize hospitals as more or less specialized.^{11, 18} We used graphical methods to examine the distribution of the cardiac specialization index. We then stratified hospitals into quintiles of specialization for the primary analysis. We also considered an alternative specialization measure defined as the proportion of all Medicare surgical discharges identified using surgical diagnosis-related groups within MDC 5 (i.e. all cardiovascular surgery). Our study findings were unchanged and therefore, we report findings using only the first measure.

Patient-Level Data

We identified all patients who underwent CABG at the 1130 hospitals described above. The study cohort was restricted to patients aged 66 or older to ensure at least one year of Medicare enrolment prior to surgery. We excluded patients discharged alive, within 24 hours of admission and not against medical advice (n=3088 patients) as such patients are unlikely to have undergone CABG and may represent coding errors. We also excluded 535 patients who underwent CABG at very low-volume centers (< 10 CABG/year) as described above. Since patients are frequently transferred from one acute care hospital to another expressly for the performance of CABG, transfers were included in our study. Transferred-in patients were identified using methodology that has been previously used for identifying transfers in administrative datasets (see Supplemental Methods).¹⁹ We restricted transfers to those patients who carried a primary discharge diagnosis of “ischemic heart disease” from the transferring facility (ICD-9-CM codes 410.xx-414.xx, n =92,289 patients).

Co-morbid illnesses were identified using algorithms described by Elixhauser²⁰ and updated by Quan.²¹ We also obtained zip code level median household income for all patients from the 2000 US Census data.¹⁶ Additional high-risk conditions specific to CABG (emergent surgery, same day surgery as PCI, use of intra-aortic balloon pump (IABP)/ventilator on admission) were identified using algorithms used in prior studies using administrative data (see Supplemental Methods).^{10, 19}

Outcomes & Statistical Analyses

The primary outcome was death due to any cause within 30 days of CABG. Secondary outcomes were in-hospital mortality, 1 year mortality and LOS.

We began by comparing trends in patient characteristics (e.g., age, race, sex, and presence of specific co-morbid conditions) across quintiles of cardiac specialization. For performing test of trend, we ranked quintiles in ascending order (lowest = 1, highest = 5) and report the p value associated with this rank variable in a logistic regression model when comparing categorical variables (for continuous variables, we used linear regression). We also compared the characteristics of less and more specialized hospitals using similar methods. Next, we compared unadjusted risk of primary and secondary outcomes across quintiles. We used multivariable generalized linear mixed models to account for differences in patient characteristics and within hospital clustering of patients. Covariates were entered as fixed effects with a random hospital intercept. Mortality models were estimated using a binary distribution and a logit link function. LOS models were estimated using a gamma distribution and a log link function; to reduce the impact of outliers, we truncated LOS at 30 days. Patient who died during the hospital stay were also included in LOS analysis. We report odds ratios (OR) with 95% confidence interval (CI) from our models and p values for trend (as described above). Finally, we tested for interaction between hospital specialization and CABG volume.

Covariates

Candidate variables for inclusion into the models were demographic factors, co-morbidities, transfer status, surgical procedure (concurrent valve surgery), variables that reflect clinical severity or urgency of CABG (e.g. primary diagnosis of acute myocardial infarction (AMI), CABG on the same day as PCI, use of IABP and ventilator on admission, etc), census region and Medicare CABG volume. Initial models were unadjusted; models were sequentially adjusted for patient risk and hospital CABG volume. Variables for inclusion into the model were selected using a combination of clinical judgment and statistical criteria. Variables that were deemed important confounders were included regardless of statistical significance. Stepwise selection was used to identify additional variables for inclusion in the hierarchical models; variables were entered if they satisfied the entry criteria of p < 0.10 and were retained if they satisfied the stay criteria of p < 0.05. Based on this approach, 31 variables were selected for inclusion in the final model for 30-day mortality (see Supplemental Table 1 for the list of all included variables). We assessed model discrimination using c-statistic and calibration using Hosmer-Lemeshow statistic from the logistic model. For LOS models, we estimated R-squared from the linear model. Finally, using the variance of the hospital intercepts, we estimated the intrahospital correlation (IHC), which is the ratio of the between hospital variation to the total variation in outcomes.²²

Since volume is an important covariate in these analyses, we examined the relationship between hospital specialization and CABG volume using Spearman correlation coefficient. We analyzed the effect of volume on the association between specialization and outcomes in several ways (for details see Supplemental Material). Since the results were not markedly different, here we only present results from models that included CABG volume as a continuous variable.

Sensitivity Analyses

We performed a series of sensitivity analyses to assess the robustness of our findings. First, we compared patient outcomes at the most specialized hospitals (quintile 5) with patients at all other hospitals (quintiles 1-4). Second, we repeated our analyses after stratifying hospital specialization using quartiles and deciles rather than quintiles. Third, we compared outcomes across categories of hospital specialization defined as follows (<25%, 25%-30%, 30%-35%, 35%-60%, and > 60% specialized). Fourth, in order to study the impact of specialization outside the realm of single-specialty cardiac hospitals, we repeated our analysis after excluding all physician-owned cardiac specialty hospitals (n=21 hospitals, 17,286 patients). Fifth, we repeated our analyses excluding transferred-in patients (n = 92,289 patients) and excluding patients undergoing valve replacement with CABG (n=98,535 patients).

All analyses were performed using SAS version 9.2 (SAS Institute Inc., Cary, NC). The authors had full access to the data and take responsibility for the integrity of the results. All authors have read and agree to the manuscript as written. The study was approved by the Institutional Review Board at University of Iowa.

Results

Our study included 1130 hospitals that performed CABG on 705,084 Medicare beneficiaries during 2001-2005. Table 1 shows the patient characteristics across quintiles of cardiac specialization (quintile 1: least-specialized, quintile 5: most-specialized). Most comparisons of patient demographics and clinical characteristics across quintiles reached statistical significance though few clinically meaningful differences were observed. In general, patients undergoing CABG at more specialized hospitals had slightly higher prevalence of hypertension, diabetes, and admission with index AMI (Table 1). Likewise, more specialized hospitals performed a greater proportion of CABG with concurrent valve replacement and treated more patients admitted after transfer. In contrast; patients who underwent CABG at less specialized hospitals were more likely to require ventilator support or intra-aortic balloon pump on admission and to undergo CABG emergently, though the absolute differences were quite small.

Table 1.

Baseline Characteristics of Medicare Patients Who Underwent CABG Surgery 2001-2005

	Increasing Specialization
Characteristic	Quintile 1 n=62,167 No. (%)	Quintile 2 n=96,096 No. (%)	Quintile 3 n=129,014 No. (%)	Quintile 4 n=170,785 No. (%)	Quintile 5 n=247,022 No. (%)	P trend
DEMOGRAPHICS
Age, years (Mean (SD))	74.9 (5.6)	74.8 (5.6)	74.9 (5.5)	74.8 (5.5)	75.0 (5.6)	< 0.001
Female Sex	20,498 (33.0)	32,813 (34.1)	44,803 (34.7)	60,462 (35.4)	85,832 (34.8)	< 0.001
Black Race	2,560 (4.1)	4,508 (4.7)	6,140 (4.8)	7,641 (4.5)	10,047 (4.1)	< 0.001
Median Income, USD (Mean (SD))	45,397 (10,428)	45,950 (9,358)	44,552 (8,961)	44,393 (9,343)	46,240 (10,782)	< 0.001
CO-MORBIDITIES
Congestive Heart Failure	14,981 (24.1)	24,257 (25.2)	34,178 (26.5)	43,410 (24.2)	62,900 (25.5)	< 0.001
Hypertension	37,703 (60.7)	59,036 (61.4)	80,058 (62.1)	105,437 (61.7)	154,419 (62.5)	< 0.001
Valve Disease	14,455 (23.3)	22,658 (23.7)	31,549 (24.5)	42,529 (24.9)	65,307 (26.4)	< 0.001
Peripheral Vascular Disease	6,460 (10.4)	10,741 (11.2)	14,323 (11.1)	18,938 (11.1)	28,754 (11.6)	< 0.001
Cerebrovascular Disease	4,747 (7.6)	7,385 (7.7)	10,064 (7.8)	14,110 (8.3)	20,253 (8.2)	< 0.001
Diabetes	16,544 (26.6)	25,684 (26.7)	35,994 (27.9)	47,093 (27.6)	68,352 (27.7)	< 0.001
Renal disease	2,420 (3.9)	4,366 (4.5)	5,621 (4.4)	7,461 (4.4)	10,505 (4.3)	< 0.001
COPD*	12,714 (20.5)	20,670 (21.5)	27,362 (21.2)	36,752 (21.5)	51,643 (20.9)	0.89
3 or more co-morbidities	40,577 (65.3)	63,672 (66.3)	87,080 (67.5)	116,189 (68.0)	171,759 (69.5)	< 0.001
DISEASE SEVERITY
Index AMI	13,202 (21.2)	19,878 (20.7)	27,179 (21.1)	38,006 (22.3)	53,230 (21.6)	< 0.001
Emergent Surgery	1,783 (2.9)	1,952 (2.0)	2,484 (1.9)	3,378 (2.0)	4,402 (1.8)	< 0.001
Balloon pump on admission	2,310 (3.7)	3,319 (3.5)	5,236 (4.1)	5,338 (3.1)	7,378 (3.0)	< 0.001
Ventilator on admission	1,628 (2.6)	2,339 (2.4)	3,011 (2.3)	4,076 (2.4)	4,205 (1.7)	< 0.001
PCI on same day as CABG	569 (0.9)	845 (0.9)	1,083 (0.8)	1,385 (0.8)	1,795 (0.7)	< 0.001
Transferred patient	5151 (8.3)	8199 (8.5)	13,613 (10.6)	22,231 (13.1)	43,095 (17.5)	< 0.001
CABG with valve replacement	8314 (13.4)	12,331 (12.8)	17,521 (13.6)	23,120 (13.5)	37,249 (15.1)	< 0.001

^*COPD indicates Chronic Obstructive Pulmonary Disease.

The median cardiac specialization for all hospitals was 29.7% (IQR 26.5-34.0%) implying that roughly 30% of all Medicare admissions to study hospitals were for cardiac-related diagnoses (Figure 1). Compared to less specialized hospitals, more specialized hospitals were more likely to be for-profit, teaching, and were predominantly located in the north-eastern and southern regions of the U.S. (Table 2). Also, more specialized hospitals had higher CABG volume (Table 2 & Figure 2). There was a modest correlation between CABG volume and cardiac specialization index (ρ=0.52, P < 0.001). Other hospital characteristics (teaching status, for-profit ownership and urban location) were only weakly correlated with specialization (ρ < 0.10).

Figure 1.

Range of Cardiac Specialization Among U.S. Hospitals Performing Coronary Artery Bypass Graft Surgery on Medicare Patients 2001-2005

Table 2.

Characteristics of Hospitals Performing CABG on Medicare Beneficiaries

	Increasing Specialization
Characteristic	Quintile 1 ≤ 25.8% n = 226 Mean (SD)	Quintile 2 25.8%-28.6% n = 226 Mean (SD)	Quintile 3 28.6%-31.1% n = 226 Mean (SD)	Quintile 4 31.1%-35.4% n = 226 Mean (SD)	Quintile 5 > 35.4% n = 226 Mean (SD)	P trend
Mean Specialization Index	23.2 (2.3)	27.2 (0.7)	29.7 (0.7)	33.0 (1.2)	45.2 (14.6)	NA
Annual Medicare MDC5 Volume	968 (507)	1,394 (712)	1,670 (980)	1,983 (1,268)	2,223 (1,425)	< 0.001
Annual Medicare Total Volume	4,131 (2036)	5,124 (2608)	5,619 (3292)	5,984 (3729)	5,359 (3608)	< 0.001
Annual Medicare CABG Volume	63 (46)	94 (64)	125 (92)	162 (125)	239 (177)	< 0.001
Hospital Beds	335 (189)	357 (193)	390 (238)	403 (251)	347 (210)	0.13
Teaching Hospitals, no. (%)†	51 (22.6)	25 (11.3)	51 (22.7)	47 (20.8)	57 (25.7)	0.07
Location, Urban, no. (%)†	194 (86.7)	197 (88.7)	202 (90.2)	208 (92.0)	209 (95.0)	0.007
For-Profit Ownership, no. (%)†	52 (23.0)	48 (21.4)	25 (11.1)	37 (16.4)	63 (28.1)	0.04
Census Region, no. (%)						< 0.001
Midwest	57 (25.2)	63 (27.9)	71 (31.4)	74 (32.7)	41 (18.1)
North-East	12 (5.3)	21 (9.3)	31 (13.7)	36 (15.9)	56 (24.8)
South	80 (35.4)	94 (41.6)	82 (36.3)	85 (37.6)	92 (40.7)
West	77 (34.1)	49 (21.7)	42 (18.6)	31 13.7)	37 (16.4)

^†Data are missing for teaching status (9 hospitals), urban location (14 hospitals), and ownership status (4 hospitals)

Figure 2.

Distribution of Average Hospital Annual Medicare CABG Volume According to Quintiles of Hospital Specialization

^*Leapfrog Volume Threshold (> 450) roughly corresponds to a Medicare CABG volume of 250 (based on the assumption that 57% of all CABG surgeries is performed on Medicare beneficiaries - 450×0.57 ~ 250)

Unadjusted 30-day mortality was generally similar for quintile 1 through quintile 4. In contrast, mortality was roughly 12% lower in quintile 5 compared to quintile 1 (4.3% vs. 4.9%, Table 3). Similar findings were noted for other mortality outcomes and LOS. After adjusting for patient characteristics, odds of 30-day mortality decreased with increasing hospital cardiac specialization (P _trend=0.001). However, this difference was no longer significant after further adjustment for CABG volume P _trend=0.65). Likewise, differences in in-hospital, 1-year mortality and LOS were not significantly different between quintiles of cardiac specialization in fully adjusted analyses (Table 4). The c-statistic for the fully adjusted mortality models ranged from 0.75-0.77 for mortality models. For LOS models, the adjusted R-squared was 0.25. The IHC for the mortality models was quite low (0.02-0.03) and that for the LOS model was 0.06.

Table 3.

Unadjusted Outcomes in Medicare Patients Who Underwent CABG

	Increasing Specialization
Outcome	Quintile 1 (n= 62167) No. (%)	Quintile 2 (n=96096) No. (%)	Quintile 3 (n=129014) No. (%)	Quintile 4 (n=170785) No. (%)	Quintile 5 (n=247022) No. (%)	P trend
	30-Day Mortality
Overall	3,033 (4.9)	4,612 (4.8)	6,042 (4.7)	8,109 (4.8)	10,619 (4.3)	< 0.001
CABG only	2,357 (4.4)	3,596 (4.3)	4,577 (4.1)	6,249 (4.2)	7,878 (3.8)	< 0.001
Excluding transfer patients	2,735 (4.8)	4,135 (4.7)	5,272 (4.6)	6,853 (4.6)	8,338 (4.1)	< 0.001
In-Hospital Mortality	2,725 (4.4)	4,131 (4.3)	5,612 (4.4)	7,486 (4.4)	9,786 (4.0)	< 0.001
1 year Mortality	6,470 (10.4)	10,053 (10.5)	13,557 (10.5)	17,909 (10.5)	24,783 (10.0)	< 0.001
Mean LOS, days (SD)	10.0 (6.1)	10.1 (6.2)	10.2 (6.2)	10.4 (6.3)	10.2 (6.2)	< 0.001

Table 4.

Unadjusted & Adjusted Risk of Primary & Secondary Outcomes among Medicare Beneficiaries who underwent CABG

	Increasing Specialization
Outcome	Quintile 1	Quintile 2	Quintile 3	Quintile 4	Quintile 5
	Odds Ratio (95% CI)					P trend
30-day Mortality
Unadjusted	1.14 (1.06-1.23)	1.08 (1.01-1.16)	1.06 (0.99-1.13)	1.09 (1.02-1.17)	1.00	0.003
Risk Adjusted	1.15 (1.07-1.24)	1.08 (1.01-1.16)	1.06 (0.99-1.13)	1.09 (1.02-1.16)	1.00	0.001
Risk & Volume Adjusted	1.05 (0.97-1.14)	1.00 (0.93-1.08)	0.99 (0.92-1.07)	1.04 (0.97-1.11)	1.00	0.65
In-hospital Mortality
Unadjusted	1.13 (1.05-1.23)	1.07 (1.00-1.16)	1.08 (1.00-1.16)	1.10 (1.03-1.19)	1.00	0.01
Risk Adjusted	1.17 (1.08-1.27)	1.09 (1.00-1.17)	1.09 (1.01-1.18)	1.11 (1.03-1.20)	1.00	0.002
Risk & Volume Adjusted	1.11 (1.02-1.22)	1.04 (0.95-1.13)	1.05 (0.97-1.14)	1.08 (1.00-1.17)	1.00	0.15
1 year Mortality
Unadjusted	1.05 (0.98-1.12)	1.03 (0.97-1.10)	1.03 (0.97-1.10)	1.04 (0.98-1.10)	1.00	0.22
Risk Adjusted	1.10 (1.03-1.17)	1.06 (1.00-1.12)	1.05 (0.99-1.11)	1.05 (0.99-1.11)	1.00	0.005
Risk & Volume Adjusted	1.03 (0.96-1.10)	1.00 (0.94-1.07)	1.01 (0.95-1.07)	1.02 (0.96-1.08)	1.00	0.66
Length of Stay
Unadjusted	1.02 (0.99-1.04)	1.02 (0.99-1.04)	1.03 (1.00-1.06)	1.04 (1.01-1.07)	1.00	0.70
Risk Adjusted	1.05 (1.02-1.07)	1.03 (1.01-1.06)	1.04 (1.02-1.07)	1.05 (1.02-1.07)	1.00	0.004
Risk & Volume Adjusted	1.04 (1.01-1.07)	1.03 (1.00-1.06)	1.04 (1.01-1.06)	1.05 (1.02-1.07)	1.00	0.045

No significant interaction was noted between cardiac specialization and hospital CABG volume in analyses of 30-day mortality (P for interaction= 0.89, Supplemental Table 6). Finally, other hospital characteristics (teaching status, for-profit ownership, urban location) were not significantly associated with patient outcomes (results not shown).

Results from sensitivity analyses are included in the Supplemental Material. Specifically, results were unchanged when we used alternative categorization of specialization Supplemental Tables 2A-2D). Also, exclusion of physician-owned cardiac specialty hospitals did not change our findings; neither did exclusion of transfer-in patients or exclusion of patients undergoing valve replacement surgery (Supplemental Tables 3-5). Finally, adjustment of CABG volume as a categorical variable did not materially change our study findings either (Supplemental Table 7).

Discussion

In analysis of over 700,000 Medicare patients undergoing CABG at over 1,100 U.S. hospitals, we found little evidence that greater cardiac specialization is associated with reductions in mortality or hospital LOS after adjusting for patient co-morbidity and hospital procedural volume.

Hospital specialization is theoretically appealing and has been advocated by champions drawn from industry and corporate strategy.²³ Supporters commonly assert that specialization fosters greater efficiency in care delivery that leads to improvement in quality and reduction in cost by focusing resources on a single disease or procedure. Also, physicians increasingly frustrated with a lack of control over the operational aspects of their hospital have found physician-ownership and investment in hospitals as an appealing model for having greater control over their work environment. At the same time, inequities in the DRG-based payment system which reimburses procedures (e.g. PCI, CABG, joint arthroplasty) more generously than cognitive services have created strong financial incentives for hospitals to specialize in “profitable” areas (e.g. cardiovascular, orthopedics). While specialty hospitals with physician-ownership offer one model, general hospitals have learned that they can develop similar delivery systems through the growth of specialty service lines (“hospitals within hospitals”) suggesting that the phenomenon of specialization is a more pervasive development.²⁴

Despite its theoretical appeal, empirical data on the relationship of hospital specialization and patient outcomes suggest a more complex picture.^{10, 12, 14, 25} While one study showed lower risk standardized 30-day mortality at cardiac specialty hospitals for AMI (15.0% versus 16.2%, p < 0.001) and CHF (10.7% vs. 11.3%, p < 0.001) compared to competitor hospitals,¹⁴ another study found risk and volume-adjusted outcomes after PCI and CABG at specialty cardiac hospitals to be similar.¹⁰ In sharp contrast, studies among orthopedic surgery patients have consistently shown improved patient outcomes at orthopedic specialty hospitals compared to their competitor hospitals.^{12, 26} Thus, these studies might suggest that the benefit of specialization depends on the disease under study. However, one needs to exercise caution before drawing firm conclusions since these studies did not so much examine specialization but rather physician-owned specialty hospitals. While physician-owned specialty hospitals have been a focus of intense scrutiny by regulatory agencies and policy makers,^{23, 27} the fact remains that these hospitals are small in number and actually represent the most extreme example of hospital specialization.

Far fewer studies have examined specialization in the broader context of all acute care hospitals. Nallamothu et al. showed that greater hospital specialization with primary PCI was associated with shorter door-to-balloon times and lower mortality among patients with ST elevation myocardial infarction; the effect was seen equally among low-volume and high-volume hospitals.¹³ Likewise, Hagen et al found a very strong inverse relationship between increasing hospital orthopedic specialization and reduced risk of adverse outcomes among Medicare patients undergoing major joint replacement surgery even after adjusting for patient co-morbidity and hospital surgical volume.¹¹ On the other hand, using the 5% inpatient Medicare sample 2001-2003, Hwang et al classified all CABG-performing hospitals into least specialized (<40%), moderately specialized (40-60%) and cardiac specialty hospitals (>60%) and found no difference in outcomes after CABG surgery across the groups.¹⁸ However, this study used a relatively coarse gradation with over 80% of the hospitals categorized as having low specialization. Our study, using a different approach and a much larger sample showed a very modest association between increased specialization and improved CABG outcomes after adjusting for patient co-morbidity that was largely limited to the highest quintile of specialization; this effect was eliminated after accounting for hospital volume. Our findings did not change after using an alternative definition of specialization, excluding physician-owned specialty hospitals or using alternative study outcomes.

Thus, while some studies suggest that hospital specialization may be associated with improved patient outcomes, there are others, including ours that have found little or no association. What constitutes this heterogeneity is harder to discern and remains unclear. It is also possible that the strong association between hospital specialization and patient outcomes noted in some studies is partly explained by reverse causation. Hospitals that achieve improved patient outcomes in a given disease area due to better quality care may become progressively more specialized as patients select that hospital for treatment of that specific condition.²⁸ Future studies examining the underlying mechanism of the association between specialization and outcomes might help clarify this issue.

Lastly, our negative findings should be interpreted in the light of the potential negative consequences of hospital specialization. Hospital specialization may lead to increasing health care costs by driving utilization of costly but discretionary procedures such as PCI, joint arthroplasty and back surgery.^29-31 Also, by cherry picking low-risk wealthy patients, highly specialized hospitals may adversely impact the financial health of the competitor general hospitals and prevent them from cross-subsidizing necessary but unprofitable care (e.g. emergency care).³² Finally, there are reports that highly specialized hospitals may be ill-equipped to handle emergencies arising out of the narrow domain of their expertise.³³

Limitations

In interpreting our findings, it is important to consider the following limitations. First, our risk adjustment model was based on administrative claims data. All the patient level information that may contribute to CABG mortality may not be coded in the data and therefore potential for residual confounding remains. Second, important measures like patient satisfaction, quality of life or functional outcome are not recorded limiting our ability to examine the influence of specialization on these important outcomes. Likewise, our inability to differentiate in-hospital events/complications from pre-existing conditions prevented us from examining the effect of specialization on the risk of non-fatal complications. Third, our study included Medicare patients aged 66 or older and therefore, our findings may not be generalizable to other age groups. Unfortunately, there is lack of national data on CABG on younger patients that includes post discharge follow up and Medicare patients account for the vast majority of patients that undergo CABG. Fourth, since volume is endogenous to specialization to some extent, we were unable to fully disentangle its effect from the relationship between specialization and outcomes. Fifth, one might argue that inclusion of physician-owned specialty hospitals in our study might have influenced our results. However, our sensitivity analyses excluding these hospitals do not support that conclusion. Lastly, only 2 to 3% of the variation in total mortality was explained by between hospital differences as evidenced by low values of IHC. Overall, significant variability in mortality and LOS remains unexplained and that is indicative of the limitation of risk-adjustment models in general.

Conclusion

Among Medicare patients undergoing CABG, higher hospital specialization was not associated with clinically significant reduction in patient mortality or length of stay after accounting for CABG volume. This study calls into question the importance of cardiac specialization for the vast majority of US hospitals performing CABG.

Abbreviations

CABG

Coronary Artery Bypass Graft

PCI

Percutaneous Coronary Intervention

AMI

Acute Myocardial Infarction

IABP

Intra-aortic Balloon Pump

MDC 5

Major Diagnostic Category 5: diseases of cardiovascular system

AHA

American Hospital Association

MedPAR

Medicare Provider & Analysis Review

ICD-9-CM

International Classification of Diseases, 9^th Clinical Modification

OR

Odds Ratio

CI

Confidence Interval

LOS

Length of Stay

DRG

Diagnosis-Related Group

IHC

Intrahospital Correlation