In the community, all subjects with HF experience high mortality irrespective of EF and the frequency of non-cardiovascular deaths is high. Subjects with preserved EF have less documented coronary disease. Accordingly, cardiovascular deaths were less frequent among subjects with preserved EF. Age, male sex, diabetes, smoking, and kidney disease were important indicators of an increased risk of overall and cardiovascular death while reduced EF was associated with an increased risk of cardiovascular death but not all cause death.
Few studies reported on cause-specific deaths in HF. Studies that did underscored that non-cardiac causes of death were frequent in HF,30
estimated at nearly one-third in a cohort of hospitalized patients from Canada.
The present community-based findings support and extend previous findings by demonstrating that HF patients have a poor survival and that the frequency of non-cardiovascular deaths in this cohort, including both outpatients and hospitalized subjects, is higher than previously reported,30
accounting for nearly half of all deaths. Factors associated with worse survival include advanced age, male sex, preexisting diabetes, smoking history, and chronic kidney disease.16,31,32
In our cohort, heart failure patients were becoming older and had increasing comorbidity over the study period. This underscores the importance of the identification and management of comorbid diseases among all HF patients. Indeed, as noncardiac comorbidities are highly prevalent in patients with HF in the community, further improvement in the survival of patients with HF may be hindered by comorbid conditions, which interfere with HF management strategies and adversely affect outcomes.33
HF is a disease of the elderly, typically with a similar distribution across sexes or a slight female preponderance. Community studies have consistently indicated a high prevalence, even predominance, of preserved EF among subjects with HF.7
As HF is a syndrome, its pathogenesis differs by EF34
and the mechanisms of HF with preserved EF, while remaining controversial, are likely related to impaired myocardial relaxation and reduced LV compliance, leading to impaired left ventricular filling.34–36
Within this context, examining the cause of death in HF can enhance our understanding of the pathophysiology of the disease.7
Herein, patients with preserved EF were less likely to have a history of diabetes, smoking, or documented coronary disease, compared to those with reduced EF. Over a long follow-up period, EF was not associated with mortality. These data extend a prior report from our group showing no difference in short-term mortality according to EF.18
Additionally, the SENIORS study of older adults with heart failure showed no difference in all cause mortality between patients with preserved or reduced EF.37
Other studies, however, indicated that preserved EF was associated with better survival.7,8,14,15
These conflicting results likely reflect differences in study design, sample size and cause of death ascertainment and distributions. These methodological considerations are important as they determine the applicability of these results to different populations and thus the clinical usefulness of such data.
To this end, in the Digitalis Investigation Group (DIG) trial, EF was associated with increased mortality.32
These data present some striking differences compared to the present community study. Indeed, participants to the DIG trial were more than 10 years younger than the present community population, included one-fourth of women as opposed to half in the present study and 63% of trial enrollees had a history of MI compared to 23% herein. Accordingly, 78% of deaths were cardiovascular in the DIG trial compared to 57% in the present study. Similarly, participants in the Candesartan in Heart Failure: Assessment of Reduction in Mortality and Morbidity (CHARM) trial,38
were younger, more likely to be male with a preponderance of coronary disease and most deaths (85%) of cardiovascular cause. Thus, in CHARM like in DIG, the association between EF and overall mortality reflects the characteristics and outcomes of selected clinical trial participants that differ markedly from that of community populations. The discrepancies between clinical trial findings and the present community-based data illustrate the limitations of extrapolating the observations made in clinical trials to the community.39
Among hospitalized patients in Ontario, no association between EF and survival was detected.40
This study however included only subjects who underwent an assessment of EF, which represented only 42% of all patients with HF hospitalized during the time period. Thus, this by design led to a substantial selection bias, which may impact on the external validity of these results amplified by the fact that outpatient subjects with HF were excluded. This underscores in turn the importance and relevance of the present data to the community practice. These findings contribute to resolving the aforementioned controversy on the impact of EF on death in HF by indicating that preserved EF carries a lower risk of cardiovascular, but not overall, death. Patients with preserved EF have fewer comorbid cardiovascular conditions than their counterparts with reduced EF, thus deaths from non-cardiac causes predominate among subjects with preserved EF. Further, the present study indicates that the proportion of cardiovascular deaths has decreased overtime among subjects with HF and preserved EF, a finding previously not reported, which should be interpreted in light of a previous report from our group indicating that the prevalence of HF with preserved EF increased overtime with no improvement in survival among these patients.8
This may help to explain findings such as those, in the PEP-CHF study, in which older patients with preserved EF had no 1-year mortality benefit with use of perindopril therapy.41
Indeed, the present findings extend data by indicating that, within the context of stable overall survival, the distribution of the causes of death is shifting towards less cardiovascular causes, which has important implications for the understanding of secular trends in HF, and for therapeutic trials for this condition.
Limitations and strengths
As no study will be generalizable to the entire US population, the racial and ethnic composition of the present population may impact the extrapolation of the data to under-represented populations. While the population of the present study consists mainly of white Caucasian subjects, the value of Olmsted County studies lies in the ability to measure in one population the occurrence of disease and subsequent outcomes and provide benchmarks for needed comparisons to other populations. Ascertainment of the cause of death relied on death certificates. The procedure for death certificate completion, as indicated in the method section, is quite standardized. The validity of death certificate to diagnose deaths due to coronary disease in the outpatient setting is quite robust.42
While we cannot exclude that some deaths could be misclassified, it seems unlikely however that misclassification would differ appreciably according to EF such that it should not affect the primary findings of the study. Further, misclassification tends to be less problematic for broad categories of death causes, such as were used herein. We acknowledge that there are limitations to the Framingham criteria for the diagnosis of heart failure.43,44
Our community-based study has notable strengths. The HF cohort, validated using standardized criteria18
includes both inpatient and outpatient data. Our findings address the stated need for more data on the cause of death among subjects with preserved EF7
in a community cohort, which optimizes its applicability to clinical practice. Ejection fraction was directly measured in a larger proportion of subjects than most previous reports.45
Further, multiple imputation was used to impute EF values when EF was not directly measured, thereby enabling to report on the entire experience of all subjects with HF in the community. This methodology provides unbiased estimates and therefore is a better approach to handling missing data than using an indicator variable to represent missing EF which has known biases.46