Hospital readmissions of patients with heart failure from real world: timing and associated risk factors

Study cohort

Sahlgrenska University Hospital/Östra is a university hospital with a catchment area of 260 000 inhabitants. Patient records with a primary discharge diagnosis of HF from the department of internal medicine, including the cardiology ward, were identified from the local hospital discharge registry. The International Classification of Diseases (ICD10) is used when coding diagnosis in Swedish hospitals. In this study, the diagnosis code for HF, I50, was used. The code I42.0 was also screened, and patients were included in the study if, after review of the record, it was evident that the cause for hospitalization was deterioration or new onset of HF. The reason for this was that in our experience, code I42.0 is commonly used as an alternative to I50 among patients with dilated cardiomyopathy (DCM) as primary aetiology for their HF diagnosis. A Swedish study of validity of the HF diagnosis in medical records showed only 6% miscoded cases.¹³

A total of 516 unique patients were identified through the hospital discharge codes, who were admitted in 2016 with a primary diagnoses of HF. Because our aim was to study 1 year outcomes, all patients who died (n = 26) during the index hospitalization were excluded. For the same reason, patients who did not reside in the Västra Götaland region (n = 3) and therefore did not have a follow-up visit within our organization were excluded. Three cases were excluded due to wrong diagnosis. Patients with Grown Up Congenital Heart disease (GUCH) (n = 5) in addition to HF were excluded because they have a special follow-up programme in our department. Finally, patients who were registered in the intensive care unit (ICU) for administrative purposes but in reality treated in the intensive care clinic (n = 3) were excluded, leaving us with a total of 476 patients. In another 28 patients, an echocardiography had never been performed; thus, the HF diagnosis did not fulfil the criteria for HF according to European Society of Cardiology (ESC). Hence, these patients also were excluded from the analysis, leaving 448 patients.

Hospital readmissions

Time to first all-cause or HF readmission at different time periods (1 month, 3 months, 6 months, and 1 year) and number of readmissions during 1 year from day of discharge were recorded. Patients who were readmitted to hospital were compared with patients who had no readmissions.

Collection of data

Collection of data on patient characteristics such as co-morbidities, laboratory parameters, and physical parameters were registered in a pre-specified form (see Appendix A). The type of HF was specified according to ejection fraction (EF) and categorized according to the ESC guidelines as follows: HFrEF (EF < 40%), HFmrEF (EF 40–49%), and HFpEF (EF ≥ 50%). If an exact figure was not given, EF for each patient was recorded as the mean of the range given in the echocardiography report (e.g. 35–45 was recorded as 40%). HF was evaluated according to EF, wide versus narrow QRS-complex, and ischaemic versus non-ischaemic origin of HF. Patients were included in the group of HF with ischaemic origin if it was evident from the medical record that this was the judged reason for HF by the treating physician, or if it was the most likely cause of HF in light of the rest of the medical history. All other reasons for HF, such as hypertension, tachycardia, alcohol abuse, valve disease, and other reasons were included in the group of non-ischaemic HF. Laboratory parameters were recorded as the last value before discharge. Co-morbidities were recorded when occurring in the registered discharge diagnoses in the patients' medical record. Anaemia and renal failure were also recorded as a co-morbidity if the patient fulfilled the criteria according to laboratory reference values even though the diagnoses failed to be mentioned in the medical record. Psychiatric disease, where depression and anxiety were included, was recorded if the patient was treated with antidepressants in addition to ICD code. For thyroid disease, patients treated with thyroid hormone (Levothyroxine) were included even if the diagnosis was not included in the medical record. Poor compliance to medication was recorded if it was evident in the patients' medical record that they were not taking their medication as prescribed by the physician. Alcohol and drug abuse were recorded as a co-morbidity if it was included in the discharge diagnoses or if clearly stated in the discharge medical record. Specific co-morbidities included in the term systemic inflammatory disease were rheumatoid arthritis, systemic sclerosis, lupus erythromatosus, Sjögren syndrome, myositis, and mixed connective tissue disease. The review of the electronic medical records was carried out by two residents in cardiology. Collected data were directly transformed into a specially developed form and thereafter into an electronic file. On matters that were subject to interpretation, such as aetiology of HF, a senior cardiologist and professor was available for consultation.

Mortality

All patients who died during index hospitalization were excluded. Mortality was recorded as date of death according to our administrative database ELVIS, which is linked to the National Patient Register and Cause of Death Register, if the patient died within 1 year.

Statistical analysis

Baseline characteristics were compared between patients with and without all-cause readmissions within the 1 year follow-up. For continuous variables, Wilcoxon rank sum test was used. For categorical variables, χ² test or Fisher exact test was used for non-ordered categorical variables; Wilcoxon rank sum test was used for ordered categorical variables. The all-cause and HF readmission rates from 1 month to 1 year after index discharge and distribution of the first readmission within 1 year follow-up were analysed. Univariable and multivariable Cox regression analyses were used to investigate predictors of short-term (1 month) and long-term (1 year) cause-specific readmissions. Baseline variables in Table 1, except number of co-morbidities and co-morbidities ≤3 vs. >3 (because they would cover up the role of one specific co-morbidity) were all entered into the univariable analysis model, and those with P-value <0.1 in the univariable analysis were further included into the multivariable model. Age and sex were mandatory into all the multivariable regression models. Variables entered into the multivariable analysis model for 1 month all-cause readmission included age, sex, systolic blood pressure (SBP), valvular disease, anaemia, renal dysfunction, psychiatric disease, poor compliance, serum sodium, N-terminal pro-brain type natriuretic peptide (NT-proBNP), and use of angiotensin converting enzyme inhibitors (ACEI)/angiotensin receptor blocker (ARB). For 1 year all-cause readmission, variables included age, sex, SBP, diabetes, atrial fibrillation, pulmonary disease, anaemia, renal dysfunction, systemic inflammatory disease, psychiatric disease, serum potassium, NT-proBNP, use of ACEIs/ARBs, diuretics, long-acting nitrates, and type of follow-up. For 1 month HF readmission, variables included age, sex, SBP, EF, alcohol or drug abuse, poor compliance, and use of β-blockers. For 1 year HF readmission, variables included age, sex, wide QRS-complex, diabetes, anaemia, renal dysfunction, systemic inflammatory disease, poor compliance, serum potassium, NT-proBNP, use of mineralocorticoid receptor antagonists (MRA), diuretics, and long-acting nitrates.

Because co-morbidities were demonstrated to have significant influence on readmission in the aforementioned analyses, we continued to explore the association between number of co-morbidities and readmission rate and association between number of co-morbidities and number of readmissions in 1 year follow-up. P for trend was tested for readmission rate and number of readmissions across the eight co-morbidities groups. Further, the cumulative incidence function curves were performed to compare the risk of 1 month and 1 year all-cause or HF readmission between patients with co-morbidities ≤3 vs. >3. The sub-distribution hazard ratio (sHR) and 95% confidence interval (CI) were calculated by the cumulative incidence function regression model (subdistribution hazard model) in which death was treated as a competing risk and age and sex were both adjusted. Survival data are shown using Kaplan–Meier curves and compared between the two groups with the log rank test. Statistical significance was set to P < 0.05 (two-tailed). All statistical analyses were performed by Stata version 16.0 (StataCorp LLC, College Station, Texas, USA).

Baseline characteristics of the cohort

Baseline characteristics are shown in Table 1. There were no significant differences between patients with or without readmissions regarding demographic variables and data regarding EF. Patients readmitted within 1 year were more likely to have co-morbidities including pulmonary disease, anaemia, renal dysfunction, systemic inflammatory disease, and psychiatric disease. Both the number of co-morbidities and the proportion of patients with more than three co-morbidities were higher in patients readmitted within the 1 year follow-up. Serum potassium was lower while creatinine was higher in readmitted patients. In readmitted patients, NT-proBNP was higher than in never readmitted patients although it fell slightly short of statistical significance (P = 0.065). Patients readmitted were more frequently treated with diuretics and long-acting nitrates and were more often referred to follow-up in primary care than in an HF clinic.

Burden of readmission during subsequent 1 year follow-up after index hospitalization

The 1 month, 3 months, 6 months, and 1 year all-cause readmission rate was 20.3%, 36.6%, 47.1%, and 60.9%, respectively. HF readmissions were 11.4%, 21.0%, 29.2%, and 38.4% for the same time-periods, respectively. All-cause readmissions within the first 3 months after discharge accounted for 60.1% of all readmissions in the 1 year follow-up. The proportion of all-cause readmissions that occurred in the second, third, and fourth quarters was 17.2%, 15.4%, and 7.3%, respectively. HF readmissions in the 1 year follow-up was 54.7% in the first quarter, 21.5% in the second quarter, 14.5% in the third quarter, and 9.3% in the fourth quarter (Figure 1).

Cause-specific readmissions

Psychiatric disease was an independent risk factor for 1 month all-cause readmission, while serum sodium a protective factor (Table 2). Regarding 1 year all-cause readmissions, the independent predictors included pulmonary disease, systemic inflammatory disease, psychiatric disease, NT-proBNP, treatment with long-acting nitrates, and follow-up in primary care. For HF readmissions, the 1 month readmission predictors were poor compliance and use of β-blockers, and the 1 year readmission predictors were systemic inflammatory disease, poor compliance, use of MRA, and nitrates (Table 2).

Association of co-morbidities and readmission

Most HF patients had three to five co-morbidities. The 1 month (P for trend 0.006) and 1 year (P for trend <0.001) all-cause readmission rate and 1 year HF readmission rate (P for trend 0.012) increased significantly with increasing number of co-morbidities (Figure 2). The number of all-cause and HF readmissions also grew with increasing numbers of co-morbidities (Figure 2). Compared with patients with ≤3 co-morbidities, HF patients with >3 co-morbidities had significantly higher risk of cumulative incidence of 1 year all-cause readmission (HR 1.75, 95% CI 1.36–2.27, P < 0.001), 1 month all-cause readmission (HR 2.15, 95% CI 1.33–3.48, P = 0.002), and 1 year HF readmission (HR 1.53, 95% CI 1.10–2.11, P = 0.011) (Figure 3). The 1 year mortality risk was also higher in HF patients with >3 co-morbidities than those with ≤3 co-morbidities.

Mortality

The total 1 month mortality rate was 4.2%. For patients who were never readmitted, the figure was 4.1% and for readmitted patients 4.3%; hence, there was no significant difference between groups. Mortality rate within 1 year was in total 22.8%. There was a significant difference in 1 year mortality rate between patients who were readmitted, where the rate was 29.0%, and patients who were never readmitted, where the mortality rate was 12.8% (P = 0.000).

Strengths and limitations

Major strengths are that we have real-life data extrapolated from medical records and not registries. We have consecutively included all patients hospitalized during 1 year, which avoids seasonal variations, for HF in our institution who were appropriate to study regarding follow up, minimizing selection bias.

Review of records was performed by two residents in cardiology making some of the parameters subject to interpretation bias. Access to a senior cardiologist and professor of cardiology for support and final decision making in unclear cases we believe minimized this risk.

Another limitation is the relatively small size of the study. On the other hand, the strength is our access to complete clinical data through original medical records. This was a retrospective study, and limitations connected to this type of study must be considered.