Long-term mortality in older patients discharged after acute decompensated heart failure: a prospective cohort study

ELISA is a prospective longitudinal cohort study of 680 older patients with ADHF seen at the emergency departments of five French hospitals [21], between October 2004 and December 2007. The ELISA cohort was established in compliance with good clinical practice guidelines and was approved by the appropriate ethics committee (institutional review board of the Henri-Mondor Teaching Hospital, Créteil, France). Cohort participants or their relatives gave their written informed consent for the collection of personal data to be used in further analyses. The present report complies with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [22].

We estimated 1-year and 2-year mortality rates and we looked for associations linking admission characteristics, including geriatric findings and co-morbidities, to 2-year mortality among hospital survivors.

As previously described [21], consecutive patients aged 75 years or older and admitted to an emergency department with a diagnosis of ADHF during a 1-year study period were included in the ELISA cohort if they met Framingham criteria for HF (at least two major criteria or one major criterion plus two minor criteria) [23]; dyspnoea at rest or with minimal exertion; and an expected hospital stay duration ≥24 h. Exclusion criteria were ventricular arrhythmia at admission and transfer to another hospital after the initial evaluation. For the present study focusing on survival after hospital discharge, we excluded the patients who died in the hospital.

Included patients underwent a standardised clinical evaluation at the emergency department. However, some parameters such as the get-up-and-go test were collected as early as possible after the relief of ADHF symptoms. Baseline data included socio-demographic characteristics, medical history, clinical characteristics, and laboratory test results. Trained clinical research assistants recorded geriatric parameters using validated tools. Nutritional status was assessed using the Mini Nutritional Assessment-Short Form [24] to classify patients into three nutritional risk categories (≥12, well-nourished; 8-11, at risk; and <8, malnourished) [25]. Weight loss over the last three months (none, 1 to 3 kg, or >3 kg) was recorded from the patients or relatives then validated by medical record review. For cognitive status, assessed using the MMSE, the cut-off of 17 indicating severe cognitive impairment was chosen [26‐28]. Mood was assessed using the 15-item Geriatric Depression Scale (GDS), with scores ≥5 indicating a risk of depression [29, 30]. Functional status was assessed using the Katz activities of daily living scale (ADL) with six items (bathing, dressing, toileting, transferring, continence, and feeding) scored from 2 (able to perform the activity) to 0 (unable to perform the activity) [31]. The number of impaired ADL items was recorded, as well as the presence of functional impairment defined as a need for assistance for at least one ADL (i.e., ADL score <12). Impaired mobility was defined as a get-up-and-go test time >20 s or an inability to perform the test [32]. Glomerular filtration rate was calculated using the abbreviated Modification of Diet in Renal Disease.

Patients were monitored by telephone calls at three-month intervals for 2 years after discharge or until death. Vital status was collected from the next of kin, usual physician, or patient’s residential institution. The primary outcome for the present analysis was overall 2-year mortality after discharge.

Sample size was estimated for the main objective of the ELISA cohort, namely, the identification of factors associated with in-hospital mortality [21].

Quantitative data are described as mean ± SD or median [25^th-75^th centiles], as appropriate, and categorical variables as n (%). Overall survival was assessed from discharge to death from any cause, 2 years after hospital discharge or last follow-up for censored patients. We used the non-parametric Kaplan-Meier method to estimate 1-year and 2-year survival rates with their 95% confidence intervals (95% CIs).

Characteristics of survivors and non-survivors were compared using Cox proportional hazards regression models with estimation of hazard ratios (HRs) and their 95% CIs. Log-linearity was tested for quantitative variables. Analyses were routinely adjusted for age dichotomised based on the median value (>85 versus ≤85 years). Variables associated with P values <0.15 were selected for multivariable analyses. Confounders and interactions were tested in bivariate models. We also investigated a potential centre effect. To avoid introducing correlated variables into multivariable models, correlations were assessed using Cramer’s V for categorical variables [33] and Spearman’s non-parametric rank correlation (Rho) [34] for quantitative variables; values above 0.50 were considered to indicate correlations. The most relevant variables were identified based on clinical relevance, number of missing values, and the Akaike information criterion (AIC) [35, 36]. The proportional hazards assumption was assessed both graphically and statistically using the Schoenfeld residuals test. This assumption was met for all variables in the final models. Calibration and discrimination of the final multivariable model were assessed using the Gronnesby-Borgan goodness-of-fit test and Harrell’s c-index, respectively [37, 38]. Model robustness was tested using a non-parametric bootstrap resampling procedure. The final Cox model was refit for the 500 bootstrap samples thus created [39].

By univariate analysis (Table 2), factors significantly associated with mortality were older age (Fig. 3); male sex; anaemia; lower systolic blood pressure; renal failure; lower body mass index, malnutrition, and recent weight loss; and impairments in functional status and mobility. Trends (P < 0.15) were observed for history of myocardial infarction and severe cognitive impairment. Therefore, these variables were potential candidates for multivariable analysis. Neither other cardiovascular co-morbidities nor polypharmacy were associated with 2-year mortality. No centre effect or effect modification by centre was demonstrated.

Age, sex, creatinine clearance, anaemia, systolic blood pressure, myocardial infarction, weight loss, body mass index, MNA-SF score, ADL score, Timed Get-Up-and-Go score, and MMSE score were available for the multivariable Cox model. Mobility impairment, lower body mass index, and MNA-SF category were not introduced into multivariable models, as they correlated with functional impairment and recent weight loss, respectively (correlation index >0.5, P < 0.05). Because of their association with other parameters, anaemia, myocardial infarction, and cognitive impairment were not independently associated with death in the multivariable model (P > 0.15). No significant interactions were found between variables associated with 2-year mortality. By multivariable analysis (Table 3), five factors were independently associated with 2-year mortality, namely, male sex, age older than 85 years, higher number of impaired ADL items, recent weight loss, and lower systolic blood pressure. A trend was noted for renal failure. The final model had good calibration (P value of the goodness-of-fit test >0.20) and acceptable discrimination (Harrell’s c-index, 0.64). The HRs estimated after bootstrap resampling were close to those of the original model, suggesting excellent internal validity. All five factors were also independently associated with 1-year mortality (Additional file 1: Table S1).

The inclusion in a vast multicentre cohort of unselected elderly patients with successful in-hospital treatment for ADHF and the low lost-to-follow-up rate over 2 years (7.5%) support the validity of our results. Furthermore, the inclusion of consecutive patients limited the risk of selection bias, and the main endpoint (overall mortality) is a robust criterion that leaves little room for classification bias. Finally, we routinely adjusted all analyses for age as a potential confounder, and we took interactions and confounding into account in the analyses. The similar HR values obtained after bootstrapping resampling procedures further support the robustness of our findings. Moreover, few data exist on long-term outcomes of patients older than 75 years after successful inhospital ADHF management. A limitation of our study is the absence in the analysis of several factors having previously reported associations with increased mortality, such as, ventricular function, BNP or NT-proBNP concentration, and treatments. Several studies suggest that a score including NT-proBNP may predict post-discharge 1-year mortality in ADHF populations with a mean age of about 60-75 years; therefore, in our population of very elderly patients, using this score to adjust the fragility indicators associated with post-discharge mortality would have been of interest [15]. However, our findings demonstrate the prognostic value of geriatric characteristics in elderly patients with ADHF.