- Email: [email protected]
Variation in Hospital Mortality Rates for Patients With Acute Myocardial Infarction
Variation in Hospital Mortality Rates for Patients With Acute Myocardial Infarction Elizabeth H. Bradley, PhDa,c,*, Jeph Herrin, PhDb, Leslie Curry, PhDa,c, Emily J. Cherlin, PhDa, Yongfei Wang, MSb,d, Tashonna R. Webster, MPH, MSa, Elizabeth E. Drye, MD, SMd, Sharon-Lise T. Normand, PhDe, and Harlan M. Krumholz, MD, SMa,b,c,d Hospitals vary by twofold in their hospital-specific 30-day risk-stratified mortality rates (RSMRs) for Medicare beneficiaries with acute myocardial infarction (AMI). However, we lack a comprehensive investigation of hospital characteristics associated with 30-day RSMRs and the degree to which the variation in 30-day RSMRs is accounted for by these characteristics, including the socioeconomic status (SES) profile of hospital patient populations. We conducted a cross-sectional national study of hospitals with >15 AMI discharges from July 1, 2005 to June 20, 2008. We estimated a multivariable weighted regression using Medicare claims data for hospital-specific 30-day RSMRs, American Hospital Association Survey of Hospitals for hospital characteristics, and the United States Census data reported by Neilsen Claritas, Inc., for zip-code level estimates of SES status. Analysis included 2,908 hospitals with 513,202 AMI discharges. Mean hospital 30-day RSMR was 16.5% (SD 1.7 percentage points). Our multivariable model explained 17.1% of the variation in hospital-specific 30-day RSMRs. Teaching status, number of hospital beds, AMI volume, cardiac facilities available, urban/rural location, geographic region, ownership type, and SES profile of patients were significantly (p <0.05) associated with 30-day RSMRs. In conclusion, substantial variation in hospital outcomes for patients with AMI remains unexplained by measurements of hospital characteristics including SES patient profile. © 2010 Elsevier Inc. All rights reserved. (Am J Cardiol 2010;106:1108 –1112) Hospitals across the country have a greater than twofold difference in 30-day risk-standardized mortality rates (RSMRs) in patients with acute myocardial infarction (AMI), with RSMRs of 10.9% to 24.9% using 3 years of experience.1 Previous research has identified teaching status,2 AMI volume,3 urban location of hospital,4 and geographic location5,6 as correlates of lower 30-day mortality rates for patients after AMI. More recent studies using the risk-adjustment method endorsed by the National Quality Forum have identified hospital urban location,1,7 teaching status,1 geographic region,1 and safety net status.8 None of the studies, however, has assessed how much of the variation in RSMR can be accounted for by these and other hospital characteristics including the socioeconomic status (SES) profile of hospital patient populations. Identifying
a
Section of Health Policy and Administration, Yale School of Public Health, New Haven, Connecticut; bSection of Cardiovascular Medicine and cRobert Wood Johnson Clinical Scholars Program, Department of Medicine, Yale University School of Medicine, New Haven, Connecticut; d Center for Outcomes Research and Evaluation, Yale–New Haven Hospital, New Haven, Connecticut; and eDepartment of Health Care Policy, Harvard Medical School, and Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts. Manuscript received April 26, 2010; revised manuscript received and accepted June 2, 2010. This work was supported by Grant RO1-HS0-16929-1 from the Agency for Healthcare Quality and Research, Maryland, Delaware, the United Health Foundation, Chicago, Illinois, and the Commonwealth Fund, New York, New York. *Corresponding author: Tel: 203-785-2937; fax: 203-785-6287. E-mail address: [email protected] (E.H. Bradley). 0002-9149/10/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.amjcard.2010.06.014
hospital-level correlates of 30-day RSMR can contribute evidence about types of hospitals where improvement efforts are most needed and lay the foundation for future research to uncover processes of care that may be improving survival rates at higher-performing hospitals. Methods We conducted a cross-sectional study using the hospital as the unit of analysis. The sample included all short-term acute and critical access nonfederal hospitals that submitted ⬎15 inpatient Medicare claims to the Centers for Medicare and Medicaid Services for discharges from July 1, 2005 to June 30, 2008 for Medicare fee-for-service beneficiaries with a principal discharge diagnosis of AMI.1,9 We used the 2006 American Hospital Association Survey of Hospitals for data on hospital characteristics and data from 2009 Population Facts (Nielsen Claritas, Inc., Los Angeles, California), previously used in cardiology outcomes research,10,11 to characterize the SES profile of hospitals’ patient population. Outcome was hospital 30-day RSMR for July 2005 through June 2008, calculated with the model used by the Centers for Medicare and Medicaid Services for public reporting of 30-day RSMRs.1,9 The 30-day RSMR is calculated for each hospital using a hierarchical generalized linear model. For each hospital, the RSMR is estimated by dividing the predicted number of deaths within 30 days of admission by the expected number of deaths within 30 days of admission and then multiplying this ratio by the national unadjusted 30-day mortality rate. The national rate is obtained using data on deaths from the Medicare beneficiary www.ajconline.org
Coronary Artery Disease/Variation in Hospital 30-Day Mortality Rates
1109
Table 1 Hospital characteristics of hospitals (n ⫽ 2,908) Variable Type of hospital Council of Teaching Hospitals member Has residency programs Nonteaching Number of staffed beds ⱕ50 51–100 101–200 201–300 ⬎300 Acute myocardial infarction annual discharges 15–50 51–180 181–500 ⬎500 Cardiac facilities Open heart surgery capacity Catheterization laboratory only No catheterization laboratory Urban/rural location Division Metropolitan Micropolitan Rural Ownership type Government For profit Geographic region New England Mid Atlantic South Atlantic East north central East south central West north central West south central Mountain Pacific United States Territories Cardiac rehabilitation services No Yes Tobacco treatment services Hospice beds No Yes Percent acute myocardial infarction—low socioeconomic status ⱕ2% 3–10% 11–30% 31–65% 66–100%
Number (%)
30-Day RSMR, Mean ⫾ SD
p Value*
242 (8.3%) 480 (16.5%) 2,186 (75.2%)
15.6 ⫾ 2.0 16.2 ⫾ 1.8 16.7 ⫾ 1.7
reference ⬍0.001 ⬍0.001
17.0 ⫾ 1.3 16.9 ⫾ 1.6 16.6 ⫾ 1.8 16.4 ⫾ 1.8 16.0 ⫾ 1.9
reference 0.029 0.111 0.001 ⬍0.001
884 (30.4%) 1,013 (34.8%) 802 (27.6%) 209 (7.2%)
17.0 ⫾ 1.2 16.8 ⫾ 1.8 16.0 ⫾ 1.9 15.4 ⫾ 1.8
reference 0.097 ⬍0.001 ⬍0.001
1,033 (35.5%) 643 (22.1%) 1,232 (42.4%)
16.0 ⫾ 1.9 16.8 ⫾ 1.8 16.9 ⫾ 1.4
reference ⬍0.001 ⬍0.001
480 (16.5%) 1,425 (49.0%) 595 (20.5%) 408 (14.0%)
16.0 ⫾ 1.8 16.4 ⫾ 1.8 17.0 ⫾ 1.6 17.1 ⫾ 1.4
reference ⬍0.001 ⬍0.001 ⬍0.001
515 (17.7%) 1,969 (67.7%)
17.0 ⫾ 1.7 16.4 ⫾ 1.7
reference ⬍0.001
163 (5.6%) 318 (10.9%) 496 (17.1%) 535 (18.4%) 218 (7.5%) 308 (10.6%) 391 (13.4%) 152 (5.2%) 314 (10.8%) 13 (0.4%)
15.8 ⫾ 1.7 16.2 ⫾ 1.7 16.5 ⫾ 1.6 16.4 ⫾ 1.7 16.9 ⫾ 1.6 16.6 ⫾ 1.7 17.0 ⫾ 1.8 16.5 ⫾ 1.7 16.6 ⫾ 1.8 18.9 ⫾ 1.6
reference ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001
792 (27.2) 2,116 (72.8) 1,955 (67.2%)
16.9 ⫾ 1.6 16.5 ⫾ 1.8 16.4 ⫾ 1.7
reference ⬍0.001 ⬍0.001
964 (33.1%) 1,944 (66.9%)
16.7 ⫾ 1.7 16.4 ⫾ 1.8
reference ⬍0.001
468 (16.1%) 506 (17.4%) 783 (26.9%) 485 (16.7%) 666 (22.9%)
504 (17.3%) 650 (22.4%) 813 (28.0%) 571 (19.6%) 370 (12.7%)
16.5 ⫾ 1.4 16.3 ⫾ 1.7 16.5 ⫾ 1.6 16.7 ⫾ 1.6 17.0 ⫾ 1.3
F Test
R2
⬍0.001
0.061
⬍0.001†
0.037
⬍0.001†
0.088
⬍0.001
0.042
⬍0.001
0.038
⬍0.001
0.015
⬍0.001
0.056
⬍0.001
0.013
⬍0.001 ⬍0.001
0.013 0.007
⬍0.001†
0.038
reference 0.452 0.034 ⬍0.001 ⬍0.001
* For variation in RSMR over categories based on analysis of variance, weighted by number of cases used to estimate RSMR for each hospital. † For trend test.
denominator file. The approach12–14 accounts for sampling variability due to differences in hospital AMI volume and for lack of statistical independence in patients treated in the same hospital. Specifications for RSMR, including patient-level variables (e.g., medical history, clinical co-morbidities, age, and gender) used for standardization, have been previously described.1,9
Independent variables, selected based on previous literature and our hypotheses, included hospital teaching status. We categorized teaching status in 3 levels: (1) membership in the Council of Teaching Hospitals (COTH), (2) having residency programs accredited by the Accrediting Commission for General Medical Education but not being a member of COTH, and
1110
The American Journal of Cardiology (www.ajconline.org)
Table 2 Multivariable weighted least squares regression (n ⫽ 2,908 hospitals)
Intercept Type of hospital Council of Teaching Hospitals member Has residency programs Nonteaching Number of staffed beds ⱕ50 51–100 101–200 201–300 ⬎300 Acute myocardial infarction annual discharges 15–50 51–180 181–500 ⬎500 Cardiac facilities Open heart surgery capacity Catheterization laboratory only Neither Urban/rural location Metropolitan division储 Metropolitan Micropolitan Rural Geographic region New England Mid Atlantic South Atlantic East north central East south central West north central West south central Mountain Pacific United States Territories Cardiac rehabilitation services No Yes Tobacco treatment services No Yes Hospice beds No Yes Percent acute myocardial infarction—low socioeconomic status ⱕ2% 3%–10% 11%–30% 31%–65% 66%–100%
Coefficient ⫾ SE
p Value
14.667 ⫾ 0.324
⬍0.001
reference 0.500 ⫾ 0.104 0.628 ⫾ 0.106
⬍0.001 0.002
Wald p Value ⬍0.001
0.001 reference ⫺0.059 ⫾ 0.237* 0.367 ⫾ 0.239† 0.548 ⫾ 0.248 0.626 ⫾ 0.254
0.804 0.124 0.027 0.014 ⬍0.001
reference ⫺0.183 ⫾ 0.189 ⫺0.946 ⫾ 0.204 ⫺1.318 ⫾ 0.219
0.335 ⬍0.001 ⬍0.001
reference 0.417 ⫾ 0.105 0.286 ⫾ 0.137
⬍0.001 0.037
⬍0.001
0.002 reference 0.305 ⫾ 0.092 0.466 ⫾ 0.134 0.492 ⫾ 0.234
0.001 0.001 0.035
reference 0.674 ⫾ 0.155 0.837 ⫾ 0.151 0.680 ⫾ 0.147 0.883 ⫾ 0.188 0.671 ⫾ 0.175 1.304 ⫾ 0.173 0.752 ⫾ 0.206 1.036 ⫾ 0.175 2.961 ⫾ 0.701
⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001
⬍0.001
0.543 reference ⫺0.65 ⫾ 0.106
0.543 0.241
reference ⫺0.102 ⫾ 0.087
0.241 0.641
reference 0.040 ⫾ 0.079 reference 0.065 ⫾ 0.112‡ 0.274 ⫾ 0.112§ 0.529 ⫾ 0.127 0.752 ⫾ 0.183
0.641 ⬍0.001 0.562 0.014 ⬍0.001 ⬍0.001 R2 17.1
* p ⬍0.01 for 51- to 100-bed category versus each larger category of number of staffed beds variable. p ⫽ 0.029 for 101- to 200-bed category versus ⬎500-bed category. ‡ p ⬍0.05 for 3% to 10% of AMI discharges in low SES versus each of the larger categories of percent AMI discharged in low SES. § p ⬍0.01 for 11% to 30% of AMI discharges in low SES versus each of the larger categories of this variable. 储 Counties or group of counties with ⬎2.5 million. †
(3) being nonteaching. Other variables included number of staffed beds (ⱕ50, 51 to 100, 101 to 200, 201 to 300, ⬎300 beds), AMI volume per year (16 to 50, 51 to 180, 180 to 500, and ⬎500 Medicare discharges with AMI), cardiac facilities (no catheterization laboratory, catheterization laboratory but no
open heart surgery, open heart surgery), urban/rural level (division, metropolitan, micropolitan, and rural as defined by the United States Census),15 ownership type (government-owned, nonprofit, and for-profit), and census region of hospital. We also examined the reported presence of cardiac rehabilitation
Coronary Artery Disease/Variation in Hospital 30-Day Mortality Rates
programs, tobacco treatment programs, and hospice beds. We included hospice beds because some have argued that hospitals have higher mortality rates if hospice care is available as part of the hospital. We measured the SES profile of patients with AMI as an additional hospital characteristic using the SES scale developed by Nielsen Claritas, Inc., in 2009 Population Facts and used in previous studies,10,11 which we linked to the Medicare claims data. The SES scale is based on a zip-code algorithm in which each zip code is given a score for 2009 using the most recent United States Census data and derived from a weighting of household income, educational level, occupation, and home value, projected forward to reflect economic and population growth for each region. We classified each patient discharge into a quintile according to the SES score assigned to the patient’s zip code. A patient discharge was classified as “low SES” if the patient lived in a zip code that had a lowest quintile SES score of all Medicare patients hospitalized with AMI. For each hospital, we then calculated the percentage of Medicare AMI discharges that were categorized as low SES. We categorized this variable in quintiles, which corresponded to hospitals with ⱕ2%, 3% to 10%, 11% to 30%, 31% to 65%, and 66% to 100% of patients with AMI who were from low SES zip codes. We summarized all independent variables and estimated a weighted analysis of variance model for each using RSMR as the dependent variable, weighted for the number of AMI admissions included in the RSMR calculation. For each bivariate model, we reported the p values for each level of categorical variables, the overall p value based on the F test, and the proportion of variance in RSMR explained by each independent variable, as measured by R2 statistics. Because the 30-day RSMR incorporates patient-level factors related to medical history, clinical co-morbidities, age, and gender, the R2 statistics refer to the proportion of the variation remaining after accounting for these patient-level factors that is explained by the independent variables. To assess the association of each independent variable with RSMR, we estimated a multivariable weighted least squares model using RSMR as the dependent variable, weighted for number of AMI admissions included in RSMR calculations. We calculated Wald p values for each categorical independent variable. All analyses were performed using SAS 9.1 (SAS Institute, Cary, North Carolina) and STATA 11 (STATA Corp., College Station, Texas). Results From the initial sample of 4,601 hospitals and 587,779 AMI discharges, 143 hospitals were excluded because they did not have a match in the American Hospital Association Survey of Hospitals, resulting in 4,458 hospitals with 582,949 AMI discharges. We then excluded 1,094 hospitals because they had ⬍15 AMI discharges during the study period. We eliminated 456 hospitals with 62,773 AMI discharges due to missing data, resulting in a final sample of 2,908 hospitals, representing 513,202 discharges. Hospitals excluded due to missing data had fewer beds, lower AMI volume, and more patients with from low SES zip codes than included hospitals but did not differ significantly (p ⬎0.05) in RSMRs.
1111
Mean of RSMR was 16.5% with an SD of 1.7 percentage points. The range was 10.9% to 24.9%. Unadjusted analysis (Table 1) indicated that lower RSMRs (i.e., better hospital performance) were significantly (p ⬍0.05) correlated with COTH teaching status (compared to having Accrediting Commission for General Medical Education residencies only or being a nonteaching hospital), more staffed beds, greater AMI volume, having open heart surgery facilities (vs having catheterization laboratories only or having neither open heart or catheterization laboratory facilities), more urban locations, for-profit or nonprofit ownership (compared to government-owned), New England region, presence of cardiac rehabilitation services, existence of tobacco treatment programs, presence of hospice beds, and having lower percent AMI discharges from lower SES areas. The SES profile of patients with AMI explained 3.8% of the variation in RSMR, and the variable with the highest R2 was AMI volume with an R2 of 8.8% (Table 1). Several hospital characteristics were significantly associated with lower 30-day RSMRs in the multivariable analysis (Table 2) including COTH teaching hospitals (compared with non-COTH hospitals with residency programs and compared with nonteaching hospitals), greater AMI volume, larger number of beds, having open heart surgery capability (compared with having a catheterization laboratory only and compared with having no catheterization laboratory), more urban location, nonprofit versus government ownership, and New England region. In this multivariable model including AMI volume and number of staffed beds, AMI volume was negatively associated with 30-day RSMR, suggesting that for hospitals with a given number of beds, those with higher volume have lower RSMRs, and given a fixed volume, hospitals with a larger number of staffed beds had higher RSMRs. Percent AMI discharges from low SES zip codes was also significantly associated with RSMR. Having larger proportions of AMI discharges classified as low SES was generally associated with higher RSMR. Presence of cardiac rehabilitation, tobacco treatment services, and hospice beds was not significantly associated with RSMR. The proportion of variance in RSMR explained by multivariable regression was 17.1%. Discussion We found that several hospital characteristics are associated with hospital performance as measured by 30-day RSMR for patients with AMI. Together these factors explained 17.1% of the variation in this outcome, which reflects mortality rates adjusted for differences in patient medical history, clinical co-morbidities, age, and gender. The finding that hospital characteristics (such as teaching status, AMI volume, and geographic region) leave substantial variation in hospital-specific 30-day RSMR unexplained is understandable because these factors are likely poor markers for quality of inpatient care and postdischarge processes that may more strongly influence survival 30-day after admission for AMI. We also found limited evidence to support the assertion by some hospitals that higher 30-day RSMR may be explained by having patient populations with lower SES or by having larger hospice services, where the goal of care is not survival. In our analysis, although hospitals with more patients from low SES areas had higher 30-day RSMRs, this factor explained only 3.8% of overall variation in outcomes. Furthermore, we found no sig-
1112
The American Journal of Cardiology (www.ajconline.org)
nificant association between having hospice beds and 30-day RSMR in multivariable analysis, and having hospice beds was associated with lower, not higher, RSMRs in unadjusted analysis. Although having hospice beds is an incomplete proxy for a hospital’s approach to terminal care at the hospital, there was no indication that hospitals with hospice beds had higher mortality rates than those without such beds. The primary implication from these analyses is that determinants of hospital performance in 30-day RSMR for patients with AMI are likely more complex than easily observed hospice characteristics. Based on our previous research16,17 and evidence from studies in organizational theory,18,19 hospital processes of care, infrastructure for quality improvement, and interdepartmental and interdisciplinary co-ordination and communication may be key elements in decreasing hospital 30-day RSMR. Our findings regarding correlates of 30-day mortality rates were consistent with our hypotheses and previous literature showing better performance associated with greater volume,3 teaching status,2 urban location,4,9 and geographic location.5,6,9 Furthermore, our findings demonstrate a link between lower SES patient profiles and poorer hospital performance; however, the magnitude of the association was modest with hospitals with ⱖ66% versus ⱕ2% of their patients from the lowest SES quintile zip codes having ⬍1 percentage point difference in 30-day RSMR. Our results should be interpreted in light of their limitations. The data are cross sectional, and we were unable to determine causal relations. Our goal was, however, to understand how much of the variance in outcomes could be explained by these hospital characteristics, rather than establish evidence of causality. Furthermore, our findings identify markers rather than determinants of RSMR. Therefore, although the findings can be used in targeting hospitals for intervention, our findings do not provide insight into what kinds of interventions might be most effective. We also did not examine the effect of hospital treatment patterns on RSMR. Also, approximately 15% of hospitals were excluded for missing data, although having missing data was not significantly associated with RSMR and we believe the bias from their exclusion is limited. In addition, our census data were not contemporary with American Hospital Association and Centers for Medicare and Medicaid Services data used, although we used the most recent census data available. We used a measurement of SES based on patients’ zip codes rather than their individual circumstances, because individual-level SES measurements do not exist in our national dataset. We did examine alternative approaches such as using average SES score based on patients’ zip code to characterize patient SES within hospitals and this did not change the results substantially. Nevertheless, our SES measurement encompasses multiple aspects of SES including education, income, occupation, and housing value and has been used in previous efforts in cardiology literature.10,11 Despite substantial investment in public reporting and increasing interest in hospital outcomes in risk-standardized mortality, we have little understanding of what influences these outcomes. Our study suggests that targeting certain types of hospitals (e.g., lower volume, rural, or nonteaching hospitals) for improvement efforts may have limited impact, because these characteristics are only modestly correlated with overall variation in mortality rates. Furthermore, our findings highlight the importance of direct measure-
ment of outcomes compared to inferences based on hospital type. Better evidence regarding the internal processes of hospitals including interdepartmental and interdisciplinary co-ordination may be needed to identify key determinants of performance and potential levers of change. 1. Krumholz HM, Merrill AR, Schone EM, Schreiner GC, Chen J, Bradley EH, Wang Y, Wang Y, Lin Z, Straube BM, Rapp MT, Normand ST, Drye EE. Patterns of hospital performance in acute myocardial infarction and heart failure 30-day mortality and readmission. Circ Cadiovasc Qual Outcomes 2009;5:407– 413. 2. Allison J, Kiefe C, Weissman N, Person S, Rousculp M, Canto J, Bae S, Williams O, Farmer R, Centor R. Relationship of hospital teaching status with quality of care and mortality for Medicare patients with acute MI. JAMA 2000;284:1256 –1262. 3. Thiemann DR, Coresh J, Oetgen WJ, Powe NR. The association between hospital volume and survival after acute myocardial infarction in elderly patients. N Engl J Med 1999;340:1640 –1648. 4. Baldwin LM, MacLehose RF, Hart LG, Beaver SK, Every N, Chan L. Quality of care for acute myocardial infarction in rural and urban US hospitals. J Rural Health 2004;20:99 –108. 5. Krumholz HM, Chen J, Rathore SS, Wang Y, Radford MJ. Regional variation in the treatment and outcomes of myocardial infarction: investigating New England’s advantage. Am Heart J 2003;146:242–249. 6. Normand ST, Glickman ME, Sharma RG, McNeil BJ. Using admission characteristics to predict short-term mortality from myocardial infarction in elderly patients. Results from the Cooperative Cardiovascular Project. JAMA 1996;275:1322–1328. 7. Ross JS, Normand SL, Wang Y, Nallamothu BK, Lichtman JH, Krumholz HM. Hospital remoteness and thirty-day mortality from three serious conditions. Health Aff (Millwood) 2008;27:1707–1717. 8. Ross JS, Cha SS, Epstein AJ, Wang Y, Bradley EH, Herrin J, Lichtman JH, Normand SL, Masoudi FA, Krumholz HM. Quality of care for acute myocardial infarction at urban safety-net hospitals. Health Aff (Millwood) 2007;26:238 –248. 9. Krumholz HM, Wang Y, Mattera JA, Han LF, Ingber MJ, Roman S, Normand SL. An administrative claims model suitable for profiling hospital performance based on 30-day mortality rates among patients with an acute myocardial infarction. Circulation 2006;113:1683–1692. 10. Denton TA, Luevanos J, Matloff JM. Clinical and nonclinical predictors of the cost of coronary bypass surgery: potential effects on health care delivery and reimbursement. Arch Intern Med 1998;158:886–891. 11. Rathore SS, Masoudi FA, Wang Y, Curtis JP, Foody JM, Havranek EP, Krumholz HM. Socioeconomic status, treatment, and outcomes among elderly patients hospitalized with heart failure: findings from the National Heart Failure Project. Am Heart J 2006;152:371–378. 12. Daniels MJ, Gatsonis C. Hierarchical generalized linear models in the analysis of variations in health care utilization. J Am Stat Assoc 1999;94:29–41. 13. Krumholz HM, Brindis RG, Brush JE, Cohen DJ, Epstein AJ, Furie K, Howard G, Peterson ED, Rathore SS, Smith SC Jr, Spertus JA, Wang Y, Normand SL. Standards for statistical models used for public reporting of health outcomes: an American Heart Association scientific statement from the quality of Care and outcomes research. Circulation 2006;113:456–462. 14. Normand SL, Glickman ME, Gatsonis CA. Statistical methods for profiling providers of medical care: issues and applications. J Am Stat Assoc 1997;92:803– 814. 15. US Census Bureau. Metropolitan and Micropolitan Statistical Areas. Maryland: Suitland, 2000. 16. Bradley EH, Curry LA, Webster TR, Mattera JA, Roumanis SA, Radford MJ, McNamara RL, Barton BA, Berg DN, Krumholz HM. Achieving rapid door-to-balloon times: how top hospitals improve complex clinical systems. Circulation 2006;113:1079 –1085. 17. Bradley EH, Holmboe ES, Mattera JA, Roumanis SA, Radford MJ, Krumholz HM. A qualitative study of increasing beta-blocker use after myocardial infarction: why do some hospitals succeed? JAMA 2001;285:2604–2611. 18. Alexander JA, Weiner BJ, Shortell SM, Baker LC, Becker MP. The role of organizational infrastructure in implementation of hospitals’ quality improvement. Hosp Top 2006;84:11–20. 19. Gittel JH. Coordinating mechanisms in care provider groups: relational coordination as a mediator and input uncertainty as a moderator of performance effects. Manag Sci 2002;48:1408 –1426.
a
Section of Health Policy and Administration, Yale School of Public Health, New Haven, Connecticut; bSection of Cardiovascular Medicine and cRobert Wood Johnson Clinical Scholars Program, Department of Medicine, Yale University School of Medicine, New Haven, Connecticut; d Center for Outcomes Research and Evaluation, Yale–New Haven Hospital, New Haven, Connecticut; and eDepartment of Health Care Policy, Harvard Medical School, and Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts. Manuscript received April 26, 2010; revised manuscript received and accepted June 2, 2010. This work was supported by Grant RO1-HS0-16929-1 from the Agency for Healthcare Quality and Research, Maryland, Delaware, the United Health Foundation, Chicago, Illinois, and the Commonwealth Fund, New York, New York. *Corresponding author: Tel: 203-785-2937; fax: 203-785-6287. E-mail address: [email protected] (E.H. Bradley). 0002-9149/10/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.amjcard.2010.06.014
hospital-level correlates of 30-day RSMR can contribute evidence about types of hospitals where improvement efforts are most needed and lay the foundation for future research to uncover processes of care that may be improving survival rates at higher-performing hospitals. Methods We conducted a cross-sectional study using the hospital as the unit of analysis. The sample included all short-term acute and critical access nonfederal hospitals that submitted ⬎15 inpatient Medicare claims to the Centers for Medicare and Medicaid Services for discharges from July 1, 2005 to June 30, 2008 for Medicare fee-for-service beneficiaries with a principal discharge diagnosis of AMI.1,9 We used the 2006 American Hospital Association Survey of Hospitals for data on hospital characteristics and data from 2009 Population Facts (Nielsen Claritas, Inc., Los Angeles, California), previously used in cardiology outcomes research,10,11 to characterize the SES profile of hospitals’ patient population. Outcome was hospital 30-day RSMR for July 2005 through June 2008, calculated with the model used by the Centers for Medicare and Medicaid Services for public reporting of 30-day RSMRs.1,9 The 30-day RSMR is calculated for each hospital using a hierarchical generalized linear model. For each hospital, the RSMR is estimated by dividing the predicted number of deaths within 30 days of admission by the expected number of deaths within 30 days of admission and then multiplying this ratio by the national unadjusted 30-day mortality rate. The national rate is obtained using data on deaths from the Medicare beneficiary www.ajconline.org
Coronary Artery Disease/Variation in Hospital 30-Day Mortality Rates
1109
Table 1 Hospital characteristics of hospitals (n ⫽ 2,908) Variable Type of hospital Council of Teaching Hospitals member Has residency programs Nonteaching Number of staffed beds ⱕ50 51–100 101–200 201–300 ⬎300 Acute myocardial infarction annual discharges 15–50 51–180 181–500 ⬎500 Cardiac facilities Open heart surgery capacity Catheterization laboratory only No catheterization laboratory Urban/rural location Division Metropolitan Micropolitan Rural Ownership type Government For profit Geographic region New England Mid Atlantic South Atlantic East north central East south central West north central West south central Mountain Pacific United States Territories Cardiac rehabilitation services No Yes Tobacco treatment services Hospice beds No Yes Percent acute myocardial infarction—low socioeconomic status ⱕ2% 3–10% 11–30% 31–65% 66–100%
Number (%)
30-Day RSMR, Mean ⫾ SD
p Value*
242 (8.3%) 480 (16.5%) 2,186 (75.2%)
15.6 ⫾ 2.0 16.2 ⫾ 1.8 16.7 ⫾ 1.7
reference ⬍0.001 ⬍0.001
17.0 ⫾ 1.3 16.9 ⫾ 1.6 16.6 ⫾ 1.8 16.4 ⫾ 1.8 16.0 ⫾ 1.9
reference 0.029 0.111 0.001 ⬍0.001
884 (30.4%) 1,013 (34.8%) 802 (27.6%) 209 (7.2%)
17.0 ⫾ 1.2 16.8 ⫾ 1.8 16.0 ⫾ 1.9 15.4 ⫾ 1.8
reference 0.097 ⬍0.001 ⬍0.001
1,033 (35.5%) 643 (22.1%) 1,232 (42.4%)
16.0 ⫾ 1.9 16.8 ⫾ 1.8 16.9 ⫾ 1.4
reference ⬍0.001 ⬍0.001
480 (16.5%) 1,425 (49.0%) 595 (20.5%) 408 (14.0%)
16.0 ⫾ 1.8 16.4 ⫾ 1.8 17.0 ⫾ 1.6 17.1 ⫾ 1.4
reference ⬍0.001 ⬍0.001 ⬍0.001
515 (17.7%) 1,969 (67.7%)
17.0 ⫾ 1.7 16.4 ⫾ 1.7
reference ⬍0.001
163 (5.6%) 318 (10.9%) 496 (17.1%) 535 (18.4%) 218 (7.5%) 308 (10.6%) 391 (13.4%) 152 (5.2%) 314 (10.8%) 13 (0.4%)
15.8 ⫾ 1.7 16.2 ⫾ 1.7 16.5 ⫾ 1.6 16.4 ⫾ 1.7 16.9 ⫾ 1.6 16.6 ⫾ 1.7 17.0 ⫾ 1.8 16.5 ⫾ 1.7 16.6 ⫾ 1.8 18.9 ⫾ 1.6
reference ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001
792 (27.2) 2,116 (72.8) 1,955 (67.2%)
16.9 ⫾ 1.6 16.5 ⫾ 1.8 16.4 ⫾ 1.7
reference ⬍0.001 ⬍0.001
964 (33.1%) 1,944 (66.9%)
16.7 ⫾ 1.7 16.4 ⫾ 1.8
reference ⬍0.001
468 (16.1%) 506 (17.4%) 783 (26.9%) 485 (16.7%) 666 (22.9%)
504 (17.3%) 650 (22.4%) 813 (28.0%) 571 (19.6%) 370 (12.7%)
16.5 ⫾ 1.4 16.3 ⫾ 1.7 16.5 ⫾ 1.6 16.7 ⫾ 1.6 17.0 ⫾ 1.3
F Test
R2
⬍0.001
0.061
⬍0.001†
0.037
⬍0.001†
0.088
⬍0.001
0.042
⬍0.001
0.038
⬍0.001
0.015
⬍0.001
0.056
⬍0.001
0.013
⬍0.001 ⬍0.001
0.013 0.007
⬍0.001†
0.038
reference 0.452 0.034 ⬍0.001 ⬍0.001
* For variation in RSMR over categories based on analysis of variance, weighted by number of cases used to estimate RSMR for each hospital. † For trend test.
denominator file. The approach12–14 accounts for sampling variability due to differences in hospital AMI volume and for lack of statistical independence in patients treated in the same hospital. Specifications for RSMR, including patient-level variables (e.g., medical history, clinical co-morbidities, age, and gender) used for standardization, have been previously described.1,9
Independent variables, selected based on previous literature and our hypotheses, included hospital teaching status. We categorized teaching status in 3 levels: (1) membership in the Council of Teaching Hospitals (COTH), (2) having residency programs accredited by the Accrediting Commission for General Medical Education but not being a member of COTH, and
1110
The American Journal of Cardiology (www.ajconline.org)
Table 2 Multivariable weighted least squares regression (n ⫽ 2,908 hospitals)
Intercept Type of hospital Council of Teaching Hospitals member Has residency programs Nonteaching Number of staffed beds ⱕ50 51–100 101–200 201–300 ⬎300 Acute myocardial infarction annual discharges 15–50 51–180 181–500 ⬎500 Cardiac facilities Open heart surgery capacity Catheterization laboratory only Neither Urban/rural location Metropolitan division储 Metropolitan Micropolitan Rural Geographic region New England Mid Atlantic South Atlantic East north central East south central West north central West south central Mountain Pacific United States Territories Cardiac rehabilitation services No Yes Tobacco treatment services No Yes Hospice beds No Yes Percent acute myocardial infarction—low socioeconomic status ⱕ2% 3%–10% 11%–30% 31%–65% 66%–100%
Coefficient ⫾ SE
p Value
14.667 ⫾ 0.324
⬍0.001
reference 0.500 ⫾ 0.104 0.628 ⫾ 0.106
⬍0.001 0.002
Wald p Value ⬍0.001
0.001 reference ⫺0.059 ⫾ 0.237* 0.367 ⫾ 0.239† 0.548 ⫾ 0.248 0.626 ⫾ 0.254
0.804 0.124 0.027 0.014 ⬍0.001
reference ⫺0.183 ⫾ 0.189 ⫺0.946 ⫾ 0.204 ⫺1.318 ⫾ 0.219
0.335 ⬍0.001 ⬍0.001
reference 0.417 ⫾ 0.105 0.286 ⫾ 0.137
⬍0.001 0.037
⬍0.001
0.002 reference 0.305 ⫾ 0.092 0.466 ⫾ 0.134 0.492 ⫾ 0.234
0.001 0.001 0.035
reference 0.674 ⫾ 0.155 0.837 ⫾ 0.151 0.680 ⫾ 0.147 0.883 ⫾ 0.188 0.671 ⫾ 0.175 1.304 ⫾ 0.173 0.752 ⫾ 0.206 1.036 ⫾ 0.175 2.961 ⫾ 0.701
⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001
⬍0.001
0.543 reference ⫺0.65 ⫾ 0.106
0.543 0.241
reference ⫺0.102 ⫾ 0.087
0.241 0.641
reference 0.040 ⫾ 0.079 reference 0.065 ⫾ 0.112‡ 0.274 ⫾ 0.112§ 0.529 ⫾ 0.127 0.752 ⫾ 0.183
0.641 ⬍0.001 0.562 0.014 ⬍0.001 ⬍0.001 R2 17.1
* p ⬍0.01 for 51- to 100-bed category versus each larger category of number of staffed beds variable. p ⫽ 0.029 for 101- to 200-bed category versus ⬎500-bed category. ‡ p ⬍0.05 for 3% to 10% of AMI discharges in low SES versus each of the larger categories of percent AMI discharged in low SES. § p ⬍0.01 for 11% to 30% of AMI discharges in low SES versus each of the larger categories of this variable. 储 Counties or group of counties with ⬎2.5 million. †
(3) being nonteaching. Other variables included number of staffed beds (ⱕ50, 51 to 100, 101 to 200, 201 to 300, ⬎300 beds), AMI volume per year (16 to 50, 51 to 180, 180 to 500, and ⬎500 Medicare discharges with AMI), cardiac facilities (no catheterization laboratory, catheterization laboratory but no
open heart surgery, open heart surgery), urban/rural level (division, metropolitan, micropolitan, and rural as defined by the United States Census),15 ownership type (government-owned, nonprofit, and for-profit), and census region of hospital. We also examined the reported presence of cardiac rehabilitation
Coronary Artery Disease/Variation in Hospital 30-Day Mortality Rates
programs, tobacco treatment programs, and hospice beds. We included hospice beds because some have argued that hospitals have higher mortality rates if hospice care is available as part of the hospital. We measured the SES profile of patients with AMI as an additional hospital characteristic using the SES scale developed by Nielsen Claritas, Inc., in 2009 Population Facts and used in previous studies,10,11 which we linked to the Medicare claims data. The SES scale is based on a zip-code algorithm in which each zip code is given a score for 2009 using the most recent United States Census data and derived from a weighting of household income, educational level, occupation, and home value, projected forward to reflect economic and population growth for each region. We classified each patient discharge into a quintile according to the SES score assigned to the patient’s zip code. A patient discharge was classified as “low SES” if the patient lived in a zip code that had a lowest quintile SES score of all Medicare patients hospitalized with AMI. For each hospital, we then calculated the percentage of Medicare AMI discharges that were categorized as low SES. We categorized this variable in quintiles, which corresponded to hospitals with ⱕ2%, 3% to 10%, 11% to 30%, 31% to 65%, and 66% to 100% of patients with AMI who were from low SES zip codes. We summarized all independent variables and estimated a weighted analysis of variance model for each using RSMR as the dependent variable, weighted for the number of AMI admissions included in the RSMR calculation. For each bivariate model, we reported the p values for each level of categorical variables, the overall p value based on the F test, and the proportion of variance in RSMR explained by each independent variable, as measured by R2 statistics. Because the 30-day RSMR incorporates patient-level factors related to medical history, clinical co-morbidities, age, and gender, the R2 statistics refer to the proportion of the variation remaining after accounting for these patient-level factors that is explained by the independent variables. To assess the association of each independent variable with RSMR, we estimated a multivariable weighted least squares model using RSMR as the dependent variable, weighted for number of AMI admissions included in RSMR calculations. We calculated Wald p values for each categorical independent variable. All analyses were performed using SAS 9.1 (SAS Institute, Cary, North Carolina) and STATA 11 (STATA Corp., College Station, Texas). Results From the initial sample of 4,601 hospitals and 587,779 AMI discharges, 143 hospitals were excluded because they did not have a match in the American Hospital Association Survey of Hospitals, resulting in 4,458 hospitals with 582,949 AMI discharges. We then excluded 1,094 hospitals because they had ⬍15 AMI discharges during the study period. We eliminated 456 hospitals with 62,773 AMI discharges due to missing data, resulting in a final sample of 2,908 hospitals, representing 513,202 discharges. Hospitals excluded due to missing data had fewer beds, lower AMI volume, and more patients with from low SES zip codes than included hospitals but did not differ significantly (p ⬎0.05) in RSMRs.
1111
Mean of RSMR was 16.5% with an SD of 1.7 percentage points. The range was 10.9% to 24.9%. Unadjusted analysis (Table 1) indicated that lower RSMRs (i.e., better hospital performance) were significantly (p ⬍0.05) correlated with COTH teaching status (compared to having Accrediting Commission for General Medical Education residencies only or being a nonteaching hospital), more staffed beds, greater AMI volume, having open heart surgery facilities (vs having catheterization laboratories only or having neither open heart or catheterization laboratory facilities), more urban locations, for-profit or nonprofit ownership (compared to government-owned), New England region, presence of cardiac rehabilitation services, existence of tobacco treatment programs, presence of hospice beds, and having lower percent AMI discharges from lower SES areas. The SES profile of patients with AMI explained 3.8% of the variation in RSMR, and the variable with the highest R2 was AMI volume with an R2 of 8.8% (Table 1). Several hospital characteristics were significantly associated with lower 30-day RSMRs in the multivariable analysis (Table 2) including COTH teaching hospitals (compared with non-COTH hospitals with residency programs and compared with nonteaching hospitals), greater AMI volume, larger number of beds, having open heart surgery capability (compared with having a catheterization laboratory only and compared with having no catheterization laboratory), more urban location, nonprofit versus government ownership, and New England region. In this multivariable model including AMI volume and number of staffed beds, AMI volume was negatively associated with 30-day RSMR, suggesting that for hospitals with a given number of beds, those with higher volume have lower RSMRs, and given a fixed volume, hospitals with a larger number of staffed beds had higher RSMRs. Percent AMI discharges from low SES zip codes was also significantly associated with RSMR. Having larger proportions of AMI discharges classified as low SES was generally associated with higher RSMR. Presence of cardiac rehabilitation, tobacco treatment services, and hospice beds was not significantly associated with RSMR. The proportion of variance in RSMR explained by multivariable regression was 17.1%. Discussion We found that several hospital characteristics are associated with hospital performance as measured by 30-day RSMR for patients with AMI. Together these factors explained 17.1% of the variation in this outcome, which reflects mortality rates adjusted for differences in patient medical history, clinical co-morbidities, age, and gender. The finding that hospital characteristics (such as teaching status, AMI volume, and geographic region) leave substantial variation in hospital-specific 30-day RSMR unexplained is understandable because these factors are likely poor markers for quality of inpatient care and postdischarge processes that may more strongly influence survival 30-day after admission for AMI. We also found limited evidence to support the assertion by some hospitals that higher 30-day RSMR may be explained by having patient populations with lower SES or by having larger hospice services, where the goal of care is not survival. In our analysis, although hospitals with more patients from low SES areas had higher 30-day RSMRs, this factor explained only 3.8% of overall variation in outcomes. Furthermore, we found no sig-
1112
The American Journal of Cardiology (www.ajconline.org)
nificant association between having hospice beds and 30-day RSMR in multivariable analysis, and having hospice beds was associated with lower, not higher, RSMRs in unadjusted analysis. Although having hospice beds is an incomplete proxy for a hospital’s approach to terminal care at the hospital, there was no indication that hospitals with hospice beds had higher mortality rates than those without such beds. The primary implication from these analyses is that determinants of hospital performance in 30-day RSMR for patients with AMI are likely more complex than easily observed hospice characteristics. Based on our previous research16,17 and evidence from studies in organizational theory,18,19 hospital processes of care, infrastructure for quality improvement, and interdepartmental and interdisciplinary co-ordination and communication may be key elements in decreasing hospital 30-day RSMR. Our findings regarding correlates of 30-day mortality rates were consistent with our hypotheses and previous literature showing better performance associated with greater volume,3 teaching status,2 urban location,4,9 and geographic location.5,6,9 Furthermore, our findings demonstrate a link between lower SES patient profiles and poorer hospital performance; however, the magnitude of the association was modest with hospitals with ⱖ66% versus ⱕ2% of their patients from the lowest SES quintile zip codes having ⬍1 percentage point difference in 30-day RSMR. Our results should be interpreted in light of their limitations. The data are cross sectional, and we were unable to determine causal relations. Our goal was, however, to understand how much of the variance in outcomes could be explained by these hospital characteristics, rather than establish evidence of causality. Furthermore, our findings identify markers rather than determinants of RSMR. Therefore, although the findings can be used in targeting hospitals for intervention, our findings do not provide insight into what kinds of interventions might be most effective. We also did not examine the effect of hospital treatment patterns on RSMR. Also, approximately 15% of hospitals were excluded for missing data, although having missing data was not significantly associated with RSMR and we believe the bias from their exclusion is limited. In addition, our census data were not contemporary with American Hospital Association and Centers for Medicare and Medicaid Services data used, although we used the most recent census data available. We used a measurement of SES based on patients’ zip codes rather than their individual circumstances, because individual-level SES measurements do not exist in our national dataset. We did examine alternative approaches such as using average SES score based on patients’ zip code to characterize patient SES within hospitals and this did not change the results substantially. Nevertheless, our SES measurement encompasses multiple aspects of SES including education, income, occupation, and housing value and has been used in previous efforts in cardiology literature.10,11 Despite substantial investment in public reporting and increasing interest in hospital outcomes in risk-standardized mortality, we have little understanding of what influences these outcomes. Our study suggests that targeting certain types of hospitals (e.g., lower volume, rural, or nonteaching hospitals) for improvement efforts may have limited impact, because these characteristics are only modestly correlated with overall variation in mortality rates. Furthermore, our findings highlight the importance of direct measure-
ment of outcomes compared to inferences based on hospital type. Better evidence regarding the internal processes of hospitals including interdepartmental and interdisciplinary co-ordination may be needed to identify key determinants of performance and potential levers of change. 1. Krumholz HM, Merrill AR, Schone EM, Schreiner GC, Chen J, Bradley EH, Wang Y, Wang Y, Lin Z, Straube BM, Rapp MT, Normand ST, Drye EE. Patterns of hospital performance in acute myocardial infarction and heart failure 30-day mortality and readmission. Circ Cadiovasc Qual Outcomes 2009;5:407– 413. 2. Allison J, Kiefe C, Weissman N, Person S, Rousculp M, Canto J, Bae S, Williams O, Farmer R, Centor R. Relationship of hospital teaching status with quality of care and mortality for Medicare patients with acute MI. JAMA 2000;284:1256 –1262. 3. Thiemann DR, Coresh J, Oetgen WJ, Powe NR. The association between hospital volume and survival after acute myocardial infarction in elderly patients. N Engl J Med 1999;340:1640 –1648. 4. Baldwin LM, MacLehose RF, Hart LG, Beaver SK, Every N, Chan L. Quality of care for acute myocardial infarction in rural and urban US hospitals. J Rural Health 2004;20:99 –108. 5. Krumholz HM, Chen J, Rathore SS, Wang Y, Radford MJ. Regional variation in the treatment and outcomes of myocardial infarction: investigating New England’s advantage. Am Heart J 2003;146:242–249. 6. Normand ST, Glickman ME, Sharma RG, McNeil BJ. Using admission characteristics to predict short-term mortality from myocardial infarction in elderly patients. Results from the Cooperative Cardiovascular Project. JAMA 1996;275:1322–1328. 7. Ross JS, Normand SL, Wang Y, Nallamothu BK, Lichtman JH, Krumholz HM. Hospital remoteness and thirty-day mortality from three serious conditions. Health Aff (Millwood) 2008;27:1707–1717. 8. Ross JS, Cha SS, Epstein AJ, Wang Y, Bradley EH, Herrin J, Lichtman JH, Normand SL, Masoudi FA, Krumholz HM. Quality of care for acute myocardial infarction at urban safety-net hospitals. Health Aff (Millwood) 2007;26:238 –248. 9. Krumholz HM, Wang Y, Mattera JA, Han LF, Ingber MJ, Roman S, Normand SL. An administrative claims model suitable for profiling hospital performance based on 30-day mortality rates among patients with an acute myocardial infarction. Circulation 2006;113:1683–1692. 10. Denton TA, Luevanos J, Matloff JM. Clinical and nonclinical predictors of the cost of coronary bypass surgery: potential effects on health care delivery and reimbursement. Arch Intern Med 1998;158:886–891. 11. Rathore SS, Masoudi FA, Wang Y, Curtis JP, Foody JM, Havranek EP, Krumholz HM. Socioeconomic status, treatment, and outcomes among elderly patients hospitalized with heart failure: findings from the National Heart Failure Project. Am Heart J 2006;152:371–378. 12. Daniels MJ, Gatsonis C. Hierarchical generalized linear models in the analysis of variations in health care utilization. J Am Stat Assoc 1999;94:29–41. 13. Krumholz HM, Brindis RG, Brush JE, Cohen DJ, Epstein AJ, Furie K, Howard G, Peterson ED, Rathore SS, Smith SC Jr, Spertus JA, Wang Y, Normand SL. Standards for statistical models used for public reporting of health outcomes: an American Heart Association scientific statement from the quality of Care and outcomes research. Circulation 2006;113:456–462. 14. Normand SL, Glickman ME, Gatsonis CA. Statistical methods for profiling providers of medical care: issues and applications. J Am Stat Assoc 1997;92:803– 814. 15. US Census Bureau. Metropolitan and Micropolitan Statistical Areas. Maryland: Suitland, 2000. 16. Bradley EH, Curry LA, Webster TR, Mattera JA, Roumanis SA, Radford MJ, McNamara RL, Barton BA, Berg DN, Krumholz HM. Achieving rapid door-to-balloon times: how top hospitals improve complex clinical systems. Circulation 2006;113:1079 –1085. 17. Bradley EH, Holmboe ES, Mattera JA, Roumanis SA, Radford MJ, Krumholz HM. A qualitative study of increasing beta-blocker use after myocardial infarction: why do some hospitals succeed? JAMA 2001;285:2604–2611. 18. Alexander JA, Weiner BJ, Shortell SM, Baker LC, Becker MP. The role of organizational infrastructure in implementation of hospitals’ quality improvement. Hosp Top 2006;84:11–20. 19. Gittel JH. Coordinating mechanisms in care provider groups: relational coordination as a mediator and input uncertainty as a moderator of performance effects. Manag Sci 2002;48:1408 –1426.