In Spain and the majority of the developed countries, heart disease1 constitutes a considerable health care burden as it is responsible for 22% of overall mortality and is the most common cause of in-hospital morbidity.1–3 Improvements in its treatment have contributed to the increase in life expectancy, with returns on the investment made in terms of cost-effectiveness.4 These achievements are due in large part to new therapies and management strategies and to good clinical administration practices.5,6

In contrast to the efforts devoted to investigating the efficacy of technological innovations,7 the relevance of the organization and administration of clinical services to their effectiveness has warranted less attention.8 However, the study of the outcomes in health care services is gaining in interest, especially with regard to the provision of cardiology services.9–13

The objective of the RECALCAR project is to investigate the relationship between organizational and administrative aspects of the cardiology units of the Spanish National Health Service (NHS) and patient outcomes. Among other activities, RECALCAR has analyzed the episodes of discharges from NHS hospitals recorded in the minimum basic data set in 2009-2010. In this article, we report the results obtained from the use of the minimum basic data set concerning the relationship between the characteristics of the hospital, the care provided (clinical services and procedures), and the in-hospital mortality due to acute myocardial infarction (AMI).

For a more detailed description of the methodology used, the reader can consult the supplementary material.

Risk-adjustment of In-hospital Mortality Due to Acute Myocardial Infarction

To improve data consistency and eliminate entries with problems in terms of the quality of the diagnosis,15,16 we excluded patients under 35 years or over 94 years of age, those whose stay was less than 1 day, and those treated in hospitals with fewer than 25 episodes a year (1319, 3445, and 1061 cases, respectively). This resulted in a sample of 95 177 episodes, with a crude in-hospital mortality rate of 7.26%.

The in-hospital mortality has been adjusted for risk, according to the model published15 by the Institute for Clinical Evaluative Sciences (ICES) of Ontario, Canada, in 1999, using SPSS 15.0 and considering the risk factors described in Table 1. The adjustment coefficients and the factors ultimately included in the model were derived from our own data and were used to estimate the expected mortality and to calculate the risk-adjusted in-hospital mortality ratio as the ratio of observed to expected mortality. However, the actual probability of a patient dying is considered to be a combination of his or her individual risk factors (case history) and the quality of the care provided (hospital-specific functionality).17–19 In addition to the patients’ demographic and clinical variables, hierarchical models of risk adjustment (multilevel models)20–22 take into consideration a specific effect at the “hospital” level. Applying this approach, we applied a multilevel logistic regression model using MLwiN 2.25 and the same variables as the 1999 ICES model.

Table 1.

Description of the Variables (Institute for Clinical Evaluative Sciences, 1999)15

Variable	Description	ICD-9-CM codes	% variable/universe
In-hospital mortality (observed)	Dependent variable.1=death during the hospitalization episode; 0=discharge for a cause other than death
Age	Discrete quantitative (years).34
Sex	Dichotomous nominal qualitative.1=woman; 0=man (reference category)		29.27
Shock	Dichotomous nominal qualitative* 1=secondary diagnosis code present in the MBDS registry;0=secondary diagnosis code absent from the MBDS registry	785.5	3.75
Diabetes mellitus with complications		250.1-250.9	5.02
Congestive heart failure		428.x	18.11
Malignant tumor		140.0-208.9	2.77
Cerebrovascular disease		430.0-438.x	4.26
Pulmonary edema		518.4, 514.x	0.60
Acute renal failure		584.x, 586.x, 788.5	5.64
Chronic renal failure		585.x, 403.x, 404.x, 996.7, 394.2, 399.4, v451	9.06
Arrhythmia		427.0-427.9	19.30

ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; MBDS, minimum basic data set.

*

The chi-square test demonstrates the existence of an association between the observed mortality and the dependent variables in every case (P<.001).

Using the multilevel model, the in-hospital mortality ratio (risk-standardized mortality ratio [RSMR]) was calculated as the ratio of the predicted mortality (which considers, on an individual basis, the functionality of the hospital in which the patient is being treated) to the expected mortality (which considers a standard functionality according to the average of all the hospitals), multiplied by the crude mortality rate.18,23 Thus, if the RSMR of a hospital is higher than the crude mortality rate, the probability of a patient dying in that hospital is greater than the average of the hospitals considered.

The results obtained show high correlation between the different estimations of mortality: expected (ICES, 1999), predicted (multilevel), and expected (multilevel), with Pearson correlation coefficients between 0.990 and 1 (P<.01) (Fig. 1), as well as good fit and discrimination, both in the ICES model of 1999 (Table 2) and with the multilevel model (Table 3), although the results are better with the latter.

Figure 1.

Correlation between mortality estimates (P<.01). A, expected mortality (Institute for Clinical Evaluative Sciences, 1999). B, predicted mortality (multilevel). C, expected mortality (multilevel). ML, multilevel. aPearson correlation coefficient=0.99. bPearson correlation coefficient=0.99. cPearson correlation coefficient=1.

(0.24MB).

Table 2.

Adjustment of the Logistic Regression Model for In-hospital Mortality Due to Acute Myocardial Infarction in 2009-2010 (Institute for Clinical Evaluative Sciences, 1999)15

	B	OR (95%CI)
Age	0.0656	1.0678 (1.0646-1.0710)
Sex	0.1251	1.1333 (1.0673-1.2033)
Shock	2.9963	20.0120 (18.4151-21.7475)
Diabetes mellitus with complications	0.2500	1.2840 (1.1569-1.4251)
Congestive heart failure	0.6973	2.0083 (1.8920-2.1318)
Malignant tumor	0.8132	2.2552 (1.9933-2.5516)
Cerebrovascular disease	0.8535	2.3480 (2.1341-2.5832)
Pulmonary edema	1.0567	2.8768 (2.3267-3.5569)
Acute renal failure	1.0309	2.8036 (2.5992-3.0242)
Arrhythmia	0.5678	1.7645 (1.6622-1.8730)
Constant	–8.3500	0.0002

95%CI, 95% confidence interval; B, regression coefficient; OR, odds ratio.

The variable “chronic renal failure” is not significant.

Calibration: Hosmer-Lemeshow chi-square=92.558; degrees of freedom=8; P<.001.

Discrimination: area under the receiver operating characterisctic curve=0.864 (95%CI, 0.860-0.868); P<.001.

Table 3.

Adjustment of the Multilevel Logistic Regression Model for In-hospital Mortality Due to Acute Myocardial Infarction in 2009-2010 (Institute for Clinical Evaluative Sciences variables 1999)15

Variable	B	OR (95%CI)
Age	0.065	1.0672 (1.0630-1.0714)
Sex	0.120	1.1275 (1.0610-1.1981)
Shock	3.061	21.3489 (19.6233-23.2262)
Diabetes mellitus with complications	0.275	1.3165 (1.1843-1.4635)
Congestive heart failure	0.723	2.0606 (1.9391-2.1897)
Malignant tumor	0.816	2.2614 (1.9948-2.5637)
Cerebrovascular disease	0.859	2.3608 (2.1446-2.5988)
Pulmonary edema	1.079	2.9417 (2.3666-3.6567)
Acute renal failure	1.076	2.9329 (2.7171-3.1659)
Arrhythmia	0.584	1.7932 (1.6875-1.9055)
Constant	–8.377	0.0002

95%CI, 95% confidence interval; B, regression coefficient; OR, odds ratio.

The variable “chronic renal failure” is not significant.

Intercept variance corresponding to the “hospital” level=0.102; intrahospital correlation coefficient=0.0301; median OR=1.3561.

Discrimination: area under the receiver operating characteristic curve for predicted mortality=0.871 (95%CI, 0.866-0.875); P<.001 and for expected mortality=0.864 (95%CI, 0.864-0.869); P<.001.

The weights of the adjustment variables are nearly the same in both models and similar to those of the adjustment model used as a reference.15 The variable with the greatest weight as a predictor of mortality is the presence of shock, followed by others related to heart failure (acute pulmonary edema, acute renal failure). In contrast to the reference model, “chronic renal failure” is not significant in terms of the adjustment.

The estimation according to the multilevel model indicates that approximately 3% of the total variance is explained by the differences between hospitals (intrahospital correlation coefficient=0.0301) and that the characteristics of the hospitals in which the patients are being treated help to explain the variation in the individual probability of dying from AMI (median odds ratio=1.3561).

Significant differences were observed between the means of the RSMR related to cluster type (Table 7), to the services responsible for discharging the patients (Table 8), and to the type of procedure performed (Table 9). The F statistic provided by analysis of variance was significant in every case (P<.05), and when differences in the error variances were detected, the Kruskall-Wallis test was applied and the null hypothesis rejected. Tables 7–9 show the results obtained for the risk-adjusted mortality ratio, as additional information to enable comparison with other studies that use this method of adjustment.

The analysis according to clusters (Table 7) showed that the hospitals in cluster 3 and, especially, cluster 4 had lower mortality (RSMR, 7.22% and 6.74%, respectively) than those in clusters 1 (8.49%), 2 (7.53%), and 5 (7.84%). The association between the RSMR and the clusters follows the same sequence as the crude mortality rate: higher in the less complex hospitals (approximately 8% for clusters 1 and 2). No revascularization procedure was carried out in 47.6% of the cases, whereas angioplasty was performed in 43.5% and fibrinolysis in 4.5% (Table 7). There is a clear association between complexity and the use of an interventional approach, and most of the angioplasties reported for the hospitals in clusters 1 and 2 were probably performed in a referral hospital with greater complexity.

The analysis according to the services responsible for patient discharge (Table 8) excludes those carried out by intensive care units (6.8% of all discharges), as this is an intermediate service. In all, 69% of the discharges were carried out by the cardiology services, which obtain better outcomes than all other services, excluding intensive care (RSMR, 7.22% and 7.84%, respectively) (Table 8).

The analysis according to the procedures performed shows better outcomes for angioplasty (RSMR, 7.19% for angioplasty and 7.83% for fibrinolysis) (Table 9). The group that did not undergo a revascularization procedure had more advanced mean age and higher proportions of women and comorbidities (Table 4).

This study demonstrates that, in the Spanish NHS, the probability of in-hospital survival after AMI is associated with the characteristics of the hospital, as well as with whether the discharge is carried out by the cardiology service and whether angioplasty has been performed (Fig. 2).

Figure 2.

Risk-standardized mortality ratio according to groups of procedures performed, services responsible for patient discharge, and hospital clusters (P<.01). CL1, cluster 1; CL2, cluster 2; CL3, cluster 3; CL4, cluster 4; CL5, cluster 5; GP1, no revascularization procedure; GP2, fibrinolysis; GP3, angioplasty; GP4, angioplasty+fibrinolysis; S1, cardiology; S2, other services.

(0.1MB).

The recommendations of clinical practice guidelines24–26 and national agencies27–30 have promoted the development of health care networks to guarantee access of patients with AMI to angioplasty as the treatment of choice for acute coronary syndrome with ST segment elevation. In non-ST-segment elevation acute coronary syndrome, the guidelines recommend an invasive strategy in patients with moderate to high risk.31,32 Thus, AMI patients can be expected to have greater possibilities of survival if they have rapid access to centers at which interventional techniques can be performed.

The studies carried out in Spain reveal substantial differences between regions and between hospitals in the management of patients with AMI.33–40 The PRIAMHO I study found no significant differences in the mortality associated with access to a catheterization laboratory,41 whereas the PRIAMHO II study demonstrated that the improvement in 1-year mortality was related to better access, a more widespread use of reperfusion, and better therapeutic management at discharge.42 The IBERICA study reported a lower 28-day mortality rate among the AMI patients who were admitted to coronary care units,43 and a lower mortality due to AMI 30 days after hospital admission has been associated with the opening of a catheterization laboratory.44 The probability of an AMI patient dying during the hospital stay has been estimated to be 25% higher in hospitals that do not have an intensive care unit.35 Studies carried out by the Spanish Agencia de Evaluación de Tecnologías Sanitarias (Agency for the Evaluation of Heath Care Technologies) found a higher in-hospital mortality rate in AMI patients who did not undergo an interventional coronary procedure,45 but they did not report differences in mortality related to the technological level of the centers.16 The GYSCA study found that the treatment received by patients with non-ST elevation acute coronary syndrome admitted to hospitals with no catheterization laboratory differed more widely from that recommended by the guidelines; these investigators observed no significant differences in in-hospital mortality, but the incidence of readmission was significantly higher.38 The MASCARA study found no relationship between the performance of primary percutaneous coronary intervention in patients with ST-elevation acute coronary syndrome and survival at 6 months.36

The analysis of the RSMR estimated according to clusters results in a curve showing that mortality is significantly reduced as the size and complexity of the hospital increases; cluster 4, which corresponds to hospitals of over 500 beds and with a high level of complexity, obtains the best outcomes, followed by cluster 3 (more than 500 beds, intermediate complexity). The lack of association between complexity and better outcomes observed in cluster 5 may be due to several causes. Hospital complexes constitute a highly heterogeneous group, occasionally made up of several hospitals of different complexity (Table 4), a circumstance that makes it difficult to compare them with the other facilities. Use of a more specific grouping based on the characteristics of the units in which AMI is treated might produce different results in terms of the association between complexity and the RSMR. The size of the hospital may not be directly related to its technological capabilities, and the organizational complexity could result in poorer outcomes in hospitals with a large number of beds, having the same negative effect as it has on their efficacy.46

The study demonstrates an association between the cardiology unit as the service that discharges the patient and lower risk-adjusted in-hospital mortality, as well as between angioplasty and lower mortality, findings that corroborate the relationship between structure35,43 and processes45 and mortality. The variable that best discriminates the standardized mortality ratio is the hospital typology (cluster); in addition to the higher probability of care being provided in a cardiology service and including angioplasty, this classification probably captures other factors that this study does not analyze—such as the availability of an on-call cardiologist, a critical care unit,35,43 and a catheterization laboratory41,42—and other unknown aspects.

The findings indicate that a policy based on the regionalization of services,1,29,47 concentrating services and patients in hospitals with a certain complexity in which primary angioplasty can be performed,29,30,42 may reduce mortality without an increase (or even with a reduction) in costs. Thus, it is necessary to collect data on the structure and processes that are more specific to the complexity and the quality of cardiology care than the generic classification of hospitals according to clusters offered by the Spanish Ministry of Health, Social Services, and Equality.

In contrast to certain findings from prospective studies,36,38,41,42 we found a robust association between indicators of structure and process and the in-hospital mortality. In comparison with other studies, this is the first in Spain to consider all discharges of AMI patients from the Spanish NHS, employing a multilevel approach. The greater number of centers and cases considered and the use of validated risk adjustment methods may explain the differences in the findings.

The most relevant conclusion of this study is probably that, in the Spanish NHS, the differences between hospitals help to explain the variation in the individual probability of dying from AMI. The type of hospital, the provision of care by a cardiology service, and the performance of a percutaneous coronary intervention are variables that, in this study, are independently and significantly associated with the survival of AMI patients admitted to NHS hospitals. We recommend the creation of care networks that favor percutaneous coronary intervention and the participation of cardiology services in the management of patients with AMI.

This work for this report was funded by an unconditional grant from Menarini.

None declared.