Readmission and mortality in patients ≥70 years with acute myocardial infarction or heart failure in the Netherlands: a retrospective cohort study of incidences and changes in risk factors over time

We used the National Medical Registration (LMR) of 2008 (and 2009 for the follow-up) from Statistics Netherlands [11] in which 88% of all hospital admissions in the Netherlands were registered anonymously. The LMR was linked to the Dutch Population Registry (GBA), which contains demographic characteristics. Record linkage was successful in 88.9% of hospital-admitted patients [11]. The dates of death were obtained from the Causes of Mortality registry. The Integrated Income Data of Household registry (IIDH) was used to retrieve additional information about residence, living circumstances and annual income.

Patients with an unplanned hospital admission in 2008 were included. Eligible patients were identified with help of the International Classification of Diseases, 9th revision, Clinical Modification (ICD-9-CM). Patients were eligible if they were ≥70 years old, had a discharge diagnosis of AMI (ICD-9 410) or HF (ICD-9 428) and had a length of stay ≥1 day. The first cardiac admission that met these criteria was considered as the index admission. Transfers to other hospitals or wards during this admission were taken as part of the same admission. No approval of the Medical Ethics Committee was necessary as data were used from national registries with anonymous information.

We examined the cumulative weekly incidence of a first unplanned all-cause readmission and mortality within 6 months. We identified potential risk factors at baseline and examined the extent to which their associations with the outcomes varied over time. An unplanned all-cause readmission was defined as any non-elective admission occurring at least 1 day after discharge from the index admission in any hospital. Risk factors were selected based on availability in the LMR, GBA and IIDH registries (Electronic Supplementary Material: Table S1–S4).

We described data using counts and percentages for categorical variables and means with standard deviations (SD) or medians with interquartile ranges (IQR) as appropriate.

We calculated the cumulative weekly incidence for unplanned all-cause readmission and mortality in AMI and HF patients until 6 months post-discharge. The number of events per week post-discharge was divided by the number of persons at risk at the start of that week. Follow-up ended if patients experienced the event of interest, died (in case of readmission) or at 6 months after the index admission if a target event did not occur.

Then, we examined by extended multivariable Cox regression analyses [12] to what extent the effects of baseline risk factors on unplanned all-cause readmission and mortality until 6 months post-discharge varied across five time points: 3, 7, 14, 30 and 42 days post-discharge. This modified Cox regression analysis is a time-to-event analysis to study if the association of a particular factor with the outcome varies over time. It involves risk factor-time interaction terms into the regression models (dummy variables for time were coded 0 = early and 1 = late) [12]. To reduce the number of statistical tests, we performed chunk tests comparing models with and without all risk factor-time interaction terms based on the likelihood ratio test. Statistically non-significant chunk tests (p ≥ 0.05) were interpreted as an indication that the extended model including the interaction terms did not lead to a better fit and standard multivariable Cox regression analysis was preferred. We took statistically significant chunk tests as an indication that the model with interactions fitted the data better. We performed a stepwise backward procedure with a p-value for entry and removal of 0.05 and 0.10, respectively. Statistically significant risk factor-time interactions were interpreted as risk factors whose effect varied over time. We expressed all hazard ratios (HRs) such that values greater than 1 indicate higher risk at the earlier time point. HRs were displayed on a logarithmic scale to enhance compact visualisation of scattered estimates (Figs. 2, 3 and 4). Analyses were performed with SPSS Statistics 22.0 (SPSS Inc., Chicago, IL, USA).

Fig. 1a shows the cumulative incidences of a first unplanned all-cause readmission within 6 months post-discharge. In total, 20.4% of AMI patients (n = 937) and 24.6% of HF patients (n = 2,103) had been readmitted. The highest incidences were found in week 1: 4.8% (AMI) and 3.7% (HF) were readmitted, respectively. After week 3, the cumulative weekly incidences were lower than 2%.

Fig. 1b shows the cumulative incidences of mortality within 6 months post-discharge. In total, 9.9% of AMI patients (n = 457) and 22.4% of HF patients (n = 1,914) had died. The highest cumulative incidences were found in week 1: 1.4% (AMI) and 2.1% (HF) died, respectively. After week 1, the cumulative incidence of mortality in AMI patients was lower than 1%. In HF patients, a more gradual decline in cumulative incidence was found with incidences between 1.5% and 0.5% from week 4 onward.

In AMI patients, the associations between risk factors and readmission did not vary over time. Therefore, the analyses resulted in the same model for all time windows (Table S1).

In HF patients, a higher Charlson Comorbidity Index (CCI) increased the risk of early readmission within 7, 30 and 42 days, e. g. patients with a CCI ≥3 had a 56% greater risk of readmission within 7 days (HR 1.56, 95% confidence interval (CI) 1.15–2.11) than patients with a CCI of 1 (reference category). Women had a 24% lower risk of readmission within 7 days (HR 0.76, 95% CI 0.60–0.97) than men. Patients with an admission in the previous 6 months before the index hospitalisation had no greater risk of readmission within 30 days (HR 1.23, 95% CI 0.97–1.57) than those without such previous admissions, while a 42% greater risk was found for readmission within 42 days (HR 1.42, 95% CI 1.12–1.80) (Fig. 2, Table S2).

Fig. 3 (and Table S3) shows the extended Cox regression analyses of early mortality post-discharge in AMI patients. Patients living alone had a 57% greater risk of mortality within 14 days (HR 1.57, 95% CI 1.01–2.44). Patients with a CCI ≥3 had a 121% greater risk of mortality within 42 days (HR 2.21, 95% CI 1.22–4.02) than those with a CCI of 1.

In HF patients, risk factor-time interactions were found for early mortality in all time windows (Fig. 4, Table S4). The risk factor-time interaction for readmission indicated an increased risk of mortality in all time windows. Non-native Dutch patients, compared to native Dutch, had a 74% greater risk of early mortality within 14 days (HR 1.74, 95% CI 1.09–2.78). A 15% lower risk of early mortality within 42 days was found for every 10 years of age (HR 0.85, 95% CI 0.72–0.99). Lower risks of early mortality were also found for patients living in an institution or with an admission in the previous 6 months.

One of the strengths of our study is that we used a large nationwide database and had the opportunity to link and combine hospital and sociodemographic data with 1‑year follow-up. This resulted in fairly rich data to examine risk factors for readmission and mortality. To our knowledge, our study is the first to examine change in those effects over time. While 11% of the cases were excluded because no linkage between hospital and sociodemographic data was possible, previous research from Statistics Netherlands showed that the number of linkable admissions were reliable for statistical analyses [21].

This study also has limitations. First, we had access only to the registries of Statistics Netherlands in 2008 and 2009 for the follow-up. Due to national trends, the incidences of readmission and mortality might nowadays have increased in HF patients and decreased in AMI patients [22]. Although the incidences might be different, we expect that the highest incidences are still found in the 1st week post-discharge. Second, the LMR contained only administrative data which precluded adjustment for cardiovascular and geriatric risk factors that are known to increase the risk of readmission and mortality (e. g. history of cardiovascular disease, disability and polypharmacy). Third, we were unable to adjust for competing risk in patients that had died before experiencing an unplanned all-cause readmission which might have resulted in an underestimation for readmission [23]. Finally, the CCIs in our data may be too low because of the underreporting of comorbidities in medical files. This may have caused an underestimation of the effect on readmission and mortality.

Hospitalised high-risk older cardiac patients need to be identified as soon as possible to guide them during care transitions. Instead of single disease-oriented treatments, a broad view on older cardiac patients’ needs is necessary [20]. Around discharge, adequate communication between hospital and community care providers, e. g. accurate and timely discharge letters, and continuity of care after discharge have proven to reduce readmissions [24]. In addition, careful assessment of patients’ readiness for discharge might be needed, as some high-risk patients might even be discharged before stable recovery [25].

While single disease-oriented interventions during hospital admission are not suitable in older cardiac patients [16, 17, 19], disease management interventions might be integrated in TCIs. More disease-specific guidance after discharge, e. g. symptom monitoring, medication reconciliation and specific guidance in medication and lifestyle adherence, might also help to reduce the risk of readmission and mortality [8, 26]. Personalised interventions might be required as HF patients were at higher and prolonged risk compared to AMI patients, and risk factors for readmission and mortality changed over time. Although readmission diagnoses are heterogeneous, early detection and proactive interventions might limit complications [13, 27].