OUP user menu

Prognosis of all-cause heart failure and borderline left ventricular systolic dysfunction: 5 year mortality follow-up of the Echocardiographic Heart of England Screening Study (ECHOES)

F.D. Richard Hobbs , Andrea K. Roalfe , Russell C. Davis , Michael K. Davies , Rachel Hare
DOI: http://dx.doi.org/10.1093/eurheartj/ehm102 1128-1134 First published online: 25 April 2007

Abstract

Aims Heart failure (HF) is reported to have an essentially malignant prognosis that can be modified by several interventions. Most outcome data on HF are available from randomized controlled treatment trials and longitudinal epidemiological studies. However, for a number of reasons, neither type of study have, to date, provided generalizable data on HF mortality. Furthermore, data on the prognosis of borderline left ventricular systolic dysfunction (LVSD) are even more limited.

Methods and results ECHOES (Echocardiographic Heart of England Screening Study) screened a total of 6162 patients from a total of 10 161 invited (61% response rate). Patients were randomly selected from four pre-specified cohorts: the general population, diuretic users, those with a prior clinical label of HF, and a population with risk factors for HF, to identify the prevalence of HF and LVSD based on clinical assessment, ECG, and echocardiography. Causes of death during a 5–9 year follow-up period were recorded from routine mortality statistics. The 5-year survival rate of the general population was 93%, compared with 69% of those with LVSD without HF, 62% with HF and no LVSD, and 53% with HF plus LVSD. Survival improved significantly with increasing ejection fraction (EF) (log rank test for trend, χ2 = 534.5, 1, P < 0.0001).

Conclusion The ECHOES mortality data confirm the poor prognosis of patients suffering prevalent HF across the community with a mortality risk estimate of 9% per year. Borderline systolic dysfunction (EF 40–50%) on echocardiography carries a poor prognosis.

  • Heart failure
  • LVSD
  • Mortality
  • Prognosis
  • Community
  • ECHOES
See page 1047 for the editorial comment on this article (doi:10.1093/eurheartj/ehl573)

Introduction

Recent epidemiological studies using reproducible objective criteria have shown the prevalence of heart failure (HF) of all causes in the population over 45 years of age is at least 2.3% and more probably 3.1%.1 The prevalence of left ventricular systolic dysfunction (LVSD) in those over 45 years of age in the ECHOES study was 1.8%,1 compared with 2.9% of patients aged 25–742 in a higher cardiovascular risk population. Around 50% of people with LVSD will be symptomatic and therefore classifiable as suffering HF.1,2 The prevalence of borderline systolic dysfunction is not as well elucidated, but was 0.8% (95% CI 0.6–1.1%) in ECHOES.1 Both HF and LVSD are considered to have a very poor prognosis.3 HF is also one of the most costly conditions to manage in most healthcare systems,4 due to high admission rates, and results in gross impairment of quality of life.5 The natural history of HF and LVSD is not well described and follow-up of borderline states of conditions that might lead to HF is therefore important. Moreover, despite therapeutic advances (such as initiation of ACE-inhibitors,6 and ß-blockers7 in most patients with systolic dysfunction, spironolactone8 in severe HF, or surgery9 for significant valve disease), the morbidity and mortality associated with HF remains high.

Most information on the prognosis of patients with HF comes from the large randomized therapy mortality trials and epidemiological follow-up studies.3 Both types of studies inevitably have limitations, for example, some epidemiological studies use less rigid criteria than trials for establishing HF10,11 and trials usually recruit younger patients, those with less co-morbidity, and mainly investigate only HF secondary to LVSD (the underlying cardiac abnormality in perhaps half of HF cases).12,13 These study differences result in varied mortality outcomes. The prognosis of mild-to-moderate HF in the placebo arms of the early pre-ACE inhibitor trials showed annual mortality rates of 12–18%.14,15 Annual mortality in trial placebo arms, against the background of ACE-inhibitors, have ranged from 7%16 in mild HF, 11–13%12,17 in moderate cases, and 20–28%,5,18,19 in severe HF. In comparison, the Framingham cohort showed an overall 1 year HF mortality rate of 17%, 2 year mortality rate of 30%, and a 10 year mortality of 78%.20 Mortality data from more recent epidemiological studies provide more reliable case definition, but mainly report on only LVSD HF, younger patients only,21 or patients presenting to hospital with incident decompensated HF.22,23

There are few data on the prognosis of prevalent, all-cause, all-stage, well-defined HF and LVSD and even fewer data on mortality outcomes in borderline systolic dysfunction. All these mortality outcomes are available from the ECHOES study cohort, with initial follow-up mortality data (minimum 5 years) presented here.

Methods

This is a prospective follow-up study of the ECHOES (Echocardiographic Heart of England Screening) study carried out in 16 randomly selected primary care practice populations in England following stratification for age and socio-economic status. Methods are described previously,1 but briefly patients were randomly sampled from each of four discrete population cohorts, identified from computerized practice registers, comprising: randomly sampled patients from the general population aged 45 and older (general population screen, n = 3960); patients with a clinical (not validated) diagnostic label of HF (n = 782); patients prescribed diuretic medication (n = 928); and patients at high risk of HF (history of previous myocardial infarction, angina, hypertension, or diabetes, n = 1062). The four cohorts were identified for differing study objectives, for example the main study objective to determine the prevalence of HF was sought from the 3960 general population cohort. For this analysis, however, the total 6162 (some individuals appeared in more than one of the four groups) patients screened out of 10 161 invited (61% response) were followed-up for mortality.

All patients were screened by history, New York Heart Association (NYHA) functional class, clinical examination, quality of life (SF-36 health status questionnaire), spirometry, resting 12-lead ECG, and echocardiography, including Doppler studies. LVSD was defined as an ejection fraction (EF) under 40% by an area–length method from the apical four-chamber view.24 HF was diagnosed on standard guideline criteria,6,7 namely relevant symptoms plus objective evidence of cardiac dysfunction. Blinded adjudication of clinical records by three experienced cardiovascular clinicians was conducted in equivocal cases. Aetiologies for HF included LVSD, atrial fibrillation, and significant valve disease, but other causes of HF were not explored. A random sample of 100 echocardiogram videotapes from the study was reviewed by a blinded senior cardiologist. There was disagreement between the reporters in only two cases (both were EF reported over 50% by one reviewer but EF 40–50% by the other). In no case was there disagreement on cases of definite LVSD.

All persons included in the ECHOES study were flagged by the Office for National Statistics Central Register Office and notifications of deaths were received on a quarterly basis. The analysis presented here includes notifications of deaths up to 1 November 2004. Causes of death were categorized using WHO ICD-10 classification and summarized by ECHOES cohort. The survival time was defined as time in days from screening to death from any cause. Persons who had not died by November 2004 were censored at that time. These data provide reliable estimates on the prognosis of prevalent, not incident, HF.

Statistical methods

Univariate exploratory analyses comparing survival times of patients within various strata were demonstrated by Kaplan–Meier curves and two-sided log-rank tests. The strata examined included case definition (previous clinical label of HF, diagnosed HF); diagnosis (Definite HF, LVSD); cause (LVSD, AF, valve disease, multiple cause); and severity (EF). Results from the multiple paired comparisons were adjusted using the Bonferroni method of correction.

Multivariable analysis was performed to confirm whether risk factors identified by the univariate analyses were still important when adjusted for the effects of age, gender, morbidity, and medication. Weibull proportional hazards survival analysis was carried out using the combined data from all four cohorts (n = 6162). Variables entered into the model included general practice, age, gender, smoking, body mass index (BMI), systolic and diastolic blood pressure, drugs (ACE-inhibitor, angiotensin-receptor blocker, beta-blocker), shortness of breath, tiredness and NYHA classification. Some adjustment for the analysis of the aggregated cohorts was made by the inclusion of dummy variables identifying the cohort characteristics [label of HF, taking diuretics, high risk (patient history of hypertension, diabetes, angina or myocardial infarction)]. All two-way interaction terms were considered and the final model was achieved using the backward elimination method using P < 0.05 as deletion criteria. An additional model, constructed on the same basis, examined the differences in risk between causes of HF. The proportional hazards assumption was examined by log–log survival plots. The linearity assumption was tested for age and BMI by categorization into deciles and plotting against the hazard ratio. Validation of the parsimonious model was made by data splitting, where the survival analysis was repeated on two randomly split subsets of the data. A likelihood ratio test was then performed comparing the difference in log-likelihood of the original analysis to the sum of the log-likelihoods obtained from analysis of the subsets. A P-value < 0.05 was considered statistically significant.

Results

The characteristics of the ECHOES populations and selected diagnostic groups are presented in Table 1, 2.3% (92) patients from the general population were diagnosed with HF and 1.8% (72) with LVSD. The median length of follow-up of subjects was 7 years 7 months (range 5 years 7 months to 9 years 7 months), with 20% (1231/6162) dying (2.7 deaths per 100 patient years). Deaths by cause during the first 5 years of follow-up are presented by cohort and selected diagnoses in Table 2. The 5 year mortality rates for each of the four populations sampled were 7% (274) general population; 37% (287) previous clinical label of HF; 20% (189) on prescribed diuretics; and 15% (159) at high risk (some patients fall into more than one of these populations). Most deaths (46%) were cardiovascular (cerebrovascular or coronary heart disease), 34% due to definite or possible HF, 21% due to cancer, and 13% due to pneumonia. Of those patients with a diagnosis of HF at screening, 43% had died at 5 years; the mortality rate from cardiovascular or cerebrovascular disease was 26% and the mortality rate due to HF was 12% at 5 years.

View this table:
Table 1

Characteristics of ECHOES cohorts and diagnostic groups

View this table:
Table 2

Cause of death at 5 years by ECHOES cohorts and diagnostic groups

The 5 year survival rate of the general population was 93%, compared with 69% of those with LVSD without HF, 62% with HF and no LVSD, and 53% with HF and LVSD.

Figure 1A and B show the survival curves of patients diagnosed with HF and LVSD, respectively. The median survival time of patients shown to have definite HF was 6 years. A significant difference was found between the survival curves of those with and without a diagnosis of HF (log-rank test, χ2 = 615.7, 1, P < 0.0001). Figure 2 shows the survival curves of those patients with and without HF split by LVSD. Persons with neither HF nor LVSD had the best survival (log-rank paired comparisons P < 0.0001), no significant differences were found between those with either HF or LVSD.

Figure 1

(A) Kaplan–Meier curves showing effect of definite heart failure on survival. (B) Kaplan–Meier curves showing effect of left ventricular systolic dysfunction on survival.

Figure 2

Kaplan–Meier curves showing the effect of heart failure and left ventricular systolic dysfunction on survival.

Survival curves comparing screened patients with a prior clinical label of HF and those with/without a diagnosis of HF are shown in Figure 3. Those persons with a prior HF label who were confirmed as having HF at screening had significantly poorer survival than those with a prior HF label without a confirmed diagnosis (5 year survival 52% vs. 69%, P < 0.001), and also in comparison with those without a prior label of HF but with a diagnosis of HF (5 year survival 52% vs. 65% P < 0.05). Patients with a prior HF diagnosis shown not to have HF at screening had similar survival rates to patients with HF but no prior diagnostic label (5 year survival 69% vs. 65% P = 0.36).

Figure 3

Survival curves comparing patients with a previous clinical heart failure label and those with/without a diagnosis of definite heart failure.

Figure 4 shows the survival curve of patients with differing EFs. Survival improved significantly with increasing EF (58% (EF < 40%) vs. 74% (EF 40–50%) vs. 90% (EF > 50%); log-rank test for trend, χ2 = 534.5, 1, P < 0.0001). Figure 5 presents the survival of persons by singular cause of HF: LVSD, valve disease or AF, and a combination of these. Multiple causes of HF had the poorest survival (log-rank tests χ2valve dis = 12.9, χ2AF = 14.3, χ2LVSD = 13.9: P < 0.001).

Figure 4

Kaplan–Meier curves showing the effect of EF on survival.

Figure 5

Kaplan–Meier survival curves by cause of heart failure.

The final primary model of the Weibull proportional hazards survival analysis is given in Table 3. It suggests that after allowing for age, sex, and some adjustment for the aggregation of cohorts, the risk of death for someone with significant valve disease is 1.32 (95% CI, 1.15 to 1.51). The risk of death for a person with an EF < 40% is 1.61 (95% CI, 1.41 to 1.84) times greater than that of a person with an EF higher than 50%. The risk also increases with HF symptoms and limitations of physical activity (NYHA class 4 vs. 1 is 1.64 (95% CI, 1.36 to 1.97). BMI did not have a linear effect on mortality, only persons with a BMI below the lower quartile of 23.75 kg/m2 were found to have a raised risk of death [hazard ratio = 1.15 (95% CI 1.05–1.26)]. The additional Weibull model (not presented) examining adjusted risk differences for causes of HF confirmed that multiple causes of HF had the poorest survival [multiple causes vs. LVSD only: hazard ratio = 1.64 (1.22–2.17); multiple causes vs. AF only: hazard ratio = 1.96 (1.45–2.63); multiple causes vs. valve disease; hazard ratio = 2.04 (1.49–2.78)]. No difference was found between the log-likelihoods of the full data set and the sum of the two subset models, confirming the model validity (χ2 = 29.1, 32 df, P = 0.61).

View this table:
Table 3

Hazard ratios and 95% confidence intervals from Weibull model for time to death

Discussion

These data demonstrate that prevalent HF carries a high mortality rate, over 40% at 5 years (9% per year)—a six-fold adverse rate compared with the general population. Overall, these ECHOES data represent valid mortality estimates for the prevalent HF population in the community (at least in countries with high background rates of cardiovascular disease) by evaluating people with HF due to most causes, across the full age range and levels of symptoms, and including unrecognized as well as diagnosed disease.

Not surprisingly, cardiovascular disease was the commonest causes of death among these patients, with over 30% of deaths due to myocardial infarction or worsening HF. Indeed, these figures may be under-estimated since a proportion of the 13% of deaths assigned to pneumonia may have been actually due to, or complicated by, worsening HF. Surprisingly, only 2% of deaths were attributed to sudden death, but since these causes were derived from routine death certificates, they may represent reluctance by physicians to ascribe an ‘uncertain cause’ label to deaths in the community.

A further interesting observation is the significantly worse mortality rates seen among the patients with ‘borderline’ EF levels of between 40–50%, suggesting these actually represent ‘borderline abnormal’ levels. Indeed, people identified with this degree of systolic impairment suffered rates of 1.3 times higher than people with EFs over 50%. An EF of 40–50% is potentially significant, bearing in mind the results of studies of 5 year survival following myocardial infarction25,26 showing survival of 95% in those with EF >50%, 83% in those with EF 41–50%, and 65% in those with EF <40%. Our study found similar increases between the EF categories. Further evidence from a study of patients with chronic HF27 has demonstrated a 14% increase in risk of death for every 5% decrease in EF below 45%. Given these incremental observational data and extrapolating from the HF prevention trials, there are reasonable grounds for believing that standard HF interventions are also indicated in this borderline systolic dysfunction population. Definitive trials are therefore worthwhile.

HF mortality rates observed among the people who also had an existing clinical label of HF at the time of the screening were greater than mortality rates of HF cases defined from the general population. Interestingly, this observation should also be viewed alongside the overall unreliability of prior clinical HF labels, with only around 50% of these people being confirmed as actually suffering HF. Despite this scale of misdiagnosis, persons labelled as suffering HF by their physicians showed a significantly worse prognosis than the general public, whether or not they actually had HF confirmed. Indeed, persons thought by their doctors to have HF, which was not confirmed on screening, suffered a mortality rate which was identical to patients with HF diagnosed for the first time by the ECHOES investigators. These data indicate that family doctors can identify individuals who are at much accelerated mortality risk, even though they may incorrectly ascribe HF as the cause of this accelerated risk in half of cases. This finding reinforces the importance of providing enhanced diagnostic access to doctors where they suspect HF; all such patients have a poor prognosis and determining the cause of symptoms more accurately should enhance secondary triage of appropriate treatments.

The relationship found between low body mass and risk of death is similar to that reported in CHARM.27

An interesting finding, at variance to existing data, was the observation that the mortality of HF at 5 years in the ECHOES cohort was similar across the varied aetiologies of the disease, even after adjusting for baseline characteristics. Only the 9% of patients with multiple causes for their HF showed a significantly worse prognosis. This finding reinforces the importance of trials for new interventions in this common chronic syndrome should enrol all causes of HF and not limit entry to systolic dysfunction alone, a common criteria in most trials to date.

In summary, these ECHOES mortality data confirm the poor prognosis of patients suffering HF across the community and provide a generalizable mortality risk estimate of 9% per year. This rate is lower than the rates suggested from hospital diagnosed HF populations or LVSD alone. Importantly, borderline systolic dysfunction carries a poor prognosis as well and HF prognosis is worse in those patients with a correct clinical label of HF.

Conflict of interest: F.D.R.H. is a member of the European Society of Cardiology (ESC) Heart Failure Association and is a Board Member of the British Primary Care Cardiovascular Society (PCCS). M.K.D. is past Chairman of the British Society for Heart Failure. F.D.R.H. and M.K.D. have received research grants, travel sponsorship, and honoraria from biotechnology and pharmaceutical companies with cardiovascular products to support research, talks, and attending scientific congresses.

References

View Abstract