We conducted a retrospective longitudinal study using information provided by the MBDS of the SNHS, which is represented by the full set of public health services in Spain. All episodes with a principal discharge diagnosis of HF from January 1, 2012 to December 31, 2012 were included. Principal diagnoses of HF was identified by the International Classification Diseases Ninth Review, Clinical Modification (ICD-9-CM) codes 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93 and 428.*.9 Patients were identified by their health card code, encrypted by a random number to ensure their privacy. The first discharge for HF in 2012 of a given patient was typified as the “index episode”.

To improve the consistency and quality of the data, we adapted the models developed by the Centers for Medicare and Medicaid Services (CMS).9,10 Thus, we excluded patients aged <35 and> 94 years, those who were discharged alive within 1 day after admission, and those in which the reason for discharge was not clear or discharge was taken against medical advice. We also excluded discharges to other hospitals to avoid duplicates and false readmissions in the same episode.

Risk factors for in-hospital mortality and readmissions were selected according to the CMS.9 All patients were followed up for 1 year after hospital discharge.

The outcome variables were in-hospital mortality at the index episode and in-hospital mortality and readmissions for cardiac diseases readmission at 30-days and 1-year after discharge. For the readmission estimates, in-hospital deaths in the index episode and elective readmissions were excluded. Only the first readmission in the considered period (30 days, 1 year) was taken into account if there were several readmissions. Only readmissions related to “cardiac diseases” as the principal episode were considered. The cardiac diseases considered were rheumatic heart disease (ICD-9-CM codes 390.*-398.*), hypertensive heart disease (ICD-9-CM codes 401.*-405.*), ischemic heart disease (ICD-9-CM codes 410.*-414.*), diseases of pulmonary circulation (ICD-9-CM codes 415.*-417.*), other forms of heart disease (ICD-9-CM codes 420.*-429.*), and aortic aneurysm and dissection (ICD-9-CM codes 441.01, 441.1, 441.2, 444.1).

Hospital characteristics including hospital volume, as well as the availability of cardiologic resources (cardiac surgery, on-site catheterization laboratory, and heart transplant program) were taken into account. Data on structural hospital characteristics were obtained from RECALCAR.11

Continuous variables are presented as mean ± standard deviation or median [interquartile range], and categorical variables are expressed as frequencies and percentages.

Risk-standardized mortality and readmission rates (RSMR and RSRR respectively) were calculated using a multilevel logistic regression model developed by the Medicare and Medicaid Services,9,10 adapted to the database structure of the MBDS of the SNHS, in which hospitals were modeled as random intercept. MBDS secondary diagnoses were grouped in condition categories according to Pope et al.,12 updated by the Agency for Healthcare Research and Quality. The final model was built by means of the backward elimination technique. The adjustment coefficients and the factors ultimately included in the in-hospital mortality and readmission models were derived from our own data. The significance levels for selection and elimination were P <.05 and P ≥ .10, respectively.

In addition to the patients’ demographic and clinical variables, multilevel models of risk adjustment take into consideration specific hospital-level factors.13–15 The RSMR and RSRR were calculated as the ratio of the number of in-hospital deaths and readmissions predicted on the basis of the hospital's performance with its observed casemix to the number of in-hospital deaths and readmissions expected on the basis of the all-hospitals performance with that hospital's casemix, multiplied by the all-hospitals unadjusted in-hospital mortality and readmissions rates. This approach allowed a comparison of a particular hospital's performance given its casemix to an average hospital's performance with the same casemix.9

The discrimination of the model was assessed with receiver operating characteristics (ROC) curves. Multilevel models were used to obtain RSMR and RSRR to compare outcomes among different hospitals. RSRR were not adjusted for death as a competing event because the MBDS of the MBDS does not contain information about out-of-hospital deaths. Data on hospital characteristics were determined by ANOVA or the Student t test, when appropriate.

To discriminate between high- and low-volume centers, a K-means clustering algorithm was used, excluding hospitals with <25 cases of HF (26 exclusions), with the aim of obtaining the maximal intragroup and the minimal intergroup density. The mathematical model was developed with two thirds of the dataset and validated with the remaining one third. The association between volume and RSMR or RSRR was tested with the Pearson correlation coefficient and linear regression models.

All tests were 2-sided, and differences were considered statistically significant at P <.05. Adjusted odds ratios (OR) and their 95% confidence intervals (95%CI) were also obtained. The statistical analysis was performed with Stata software version 13.0 (Stata Corp, College Station, Texas) and SPSS version 20. The present study complies with the Declaration of Helsinki. It was exempt from additional review by the ethics investigation committee because all data were de-identified.

During the year 2012, 86177 index episodes were identified: 8525 (9.9%) were excluded and 77652 patients were finally included in the study. The main causes of exclusion (categories not mutually exclusive) were: same or next day discharges, and discharged alive (35%), age younger than 35 or older than 94 years (32%), and transferred to another hospital (26%).

The mean age was 79.2 (9.9) years and 55.3% were women. The most common comorbidities were peripheral artery disease (64.7%), hypertension (56.3%), chronic kidney disease (33.5%), valvular heart disease (29.1%), dementia (20.9%), and chronic obstructive pulmonary disease (17.7%) (Table 1).

The crude in-hospital mortality rate during the index episode was 9.2%, rising to 14.5% (cumulative rate) throughout the year of follow-up (10.4% at 1 month and 11.8% at 3 months).

Throughout the year of follow-up, 37 247 episodes of cardiac hospitalization were registered, corresponding to 23 661 patients. The overall rate of cardiac readmission reached 32.6% (9.7% during the first month after hospital discharge). The main cause of hospital readmission was decompensated HF (62.8%), followed by ischemic heart disease episodes (6.8%).

Multilevel regression analysis identified the main risk factors for in-hospital mortality at the index episode (Table 2) and for 1-year in-hospital mortality and hospital readmissions (Table 1 of the supplementary data). Stroke, metastatic cancer, cardio-respiratory failure and shock and acute myocardial infarction were found to have the strongest association with in-hospital mortality (Table 2), and also with in-hospital mortality at 1-year, whereas valvular heart disease, diabetes mellitus and myocardial revascularization were consistently related to increased 1-year readmission rates. Renal failure was strongly associated with both in-hospital mortality and hospital readmission (Table 1 of the supplementary data). The area under the ROC curve for in-hospital mortality at the index episode was 0.745 (Figure 1), and was 0.735 and 0.706 at 30 days and 1-year, respectively (Figure 1 of the supplementary data). The median OR for in-hospital mortality at the index episode was 1.49, denoting high interhospital variability. The area under the ROC curve for identifying patients with cardiovascular hospital readmission within 30 days and at 1 year was 0.598 and 0.612, respectively.

Figure 1.

Receiver operating characteristics curve for in-hospital mortality at the index episode. Multilevel adjustment.

(0.07MB).

Information on the number of HF discharges at each hospital was analyzed to classify centers into 2 categories (high- and low-volume hospitals). The cutoff point was set at 340 HF discharges per hospital during 2012. Sixty-four per cent of the patients (49 765) and 35.34% of the hospitals (94) were classified as being in the high-volume category.

RSMR were lower among patients admitted to high-volume hospitals (–16.5% at the index episode,–7.6% at 30 days and–7.1% at 1 year) (Figure 2, Table 3); conversely, high-volume hospitals had higher 1-year cardiac readmission rates (4.2%), (Table 3).

Figure 2.

Risk adjusted in-hospital mortality at the index episode and hospital volume. Blue dots: low-volume hospitals; green dots: high-volume hospitals (≥ 340 heart failure discharges/center).

(0.11MB).

Regarding hospital cardiologic resources, 98 (33.6%) out of the 292 participating hospitals had a cardiac catheterization laboratory. Cardiac surgery was available in 46 centers (15.7%); 15 of them had a heart transplant program. When we analyzed the impact of these resources on outcomes (Table 2 of the supplementary data), we found no association that was statistically significant and independent from volume between their availability and in-hospital mortality (at the index episode, 30 days and 1 year) and 30-day readmissions (P> .05). However, the availability of a structured cardiology department was associated with better outcomes than hospitals without a cardiology department (Table 4).

This study has several limitations. Although it is a retrospective analysis based on MBDS data, the use of administrative information has proven to be valid to estimate outcomes in health services, compared with medical records.29,30 The reliability of this kind of study enables the public comparison of different hospitals.7 The use of the same personal identifiers for recording hospital discharges and in-hospital mortality facilitated the data gathering process during follow-up. Discriminative accuracy for the adjustment model of hospital readmissions in our study was very poor (0.598 and 0.612), but was similar to the CMS model7 and to other population studies on hospital readmissions, whose areas under ROC curve were between 0.55 and 0.65,31,32 their use being less inaccurate than the unadjusted crude readmission rates.

In contrast with the methodology used by the CMS,9,10 we could not analyze all-cause readmissions because the Ministry of Health only provided data on cardiac diseases discharges. In addition, information related to those patients who died outside the hospital could not be obtained. However, based on our 1-year mortality rate, it can be assumed that proportionally there are few patients hospitalized for HF (index episode) who die outside the hospital in the year following discharge, either due to sudden death or other causes. The 1-year in-hospital mortality rate of our series (14.4%) is very similar to the figure reported for 1-year mortality for cardiovascular causes of the ESC-HF Pilot Registry for patients discharged from hospitals in the southern region of Europe (16.8%, with out-of-hospital deaths being included), taking into consideration that our study does not track cerebrovascular deaths.33 Therefore, it is reasonable to conclude that the real HF mortality rate must be somewhat higher than the figure we report.

Another limitation is that there were unidentifiable confounding factors in the adjustment models that could have a significant impact. The secondary diagnoses considered as risk-adjustment variables corresponded to conditions that were present on admission or to complications that might occasionally reflect inadequate treatment. RSRR were not adjusted for death as a competing event. However, our method was similar to previously validated models in terms of its predictive capacity.7,31,32

MBDS data misclassification did not allow us to assess mortality or readmissions based on the current classification of HF. Nevertheless, the impact of ejection fraction on HF outcome remains controversial. The latest data show that mortality in patients with HF is similar regardless of ejection fraction.34