Study population

Leicestershire Health Authority has a comprehensive record linkage system, which is linked to information held at the Office of National Statistics. This allows for follow up of all residents registered with primary care (956 000 in 2001) in terms of events such as hospitalisations and mortality. The system provides details of the dates of events, discharge diagnoses, and the patient’s age, sex, and domicile postcode.

We obtained data on residents aged ≥ 40 years who were admitted for a first heart failure to any of the hospitals serving the population of Leicestershire between 1 April 1993 and 31 March 2001. We excluded all those with a recorded heart failure diagnosis in the five years before the start of, and counted only the first heart failure hospitalisation during, the observation period. A heart failure admission was defined as heart failure (International classification of diseases (ICD) 10th revision or ICD 9th revision code I50* or 428*, respectively) in any discharge coding position. As a relatively small degree of migration occurs in those aged over 40, we included only patients resident in the district for at least five years before the index admission, thus omitting all those who may have had a diagnosis while resident outside the county. Although our data may have omitted a small number of residents, it constitutes a large cohort from a demographically varied population. Incidence data are presented as age standardised annual rates.

Survival

Mortality was identified from death certification records provided by the Office of National Statistics. Survival was measured from the date of admission to the date of death or to the end of follow up (30 September 2001), providing a minimum of six months of follow up for those alive at the end of the period. For patients who migrated from the area before 30 September 2001, the date of migration was taken as the end of follow up. Analysis was undertaken in respect of both all cause and cardiovascular mortality (ICD-9 39–45, diseases of circulatory system, excluding 43, cerebrovascular disease), as defined by the recorded cause of death. Demographic variables potentially affecting survival were age at the date of diagnosis and sex. As a proxy measure of social deprivation we used the index of multiple deprivation (IMD 2000)¹⁹ at the electoral ward level, matched to the patient’s domicile postcode, expressed as quintiles (quintile 5 being most deprived). Variables related to hospitalisation were year of admission, duration of index admission, and diagnoses related to the index and previous hospitalisations.

Previous hospitalisations

We used two measures of comorbidity. Firstly, we used diagnoses related to the index and to previous hospitalisations. In particular we focused on conditions associated with the development of heart failure; myocardial infarction, CHD, other heart disease, hypertension, and diabetes. Secondly, we used average length of hospital stay, by definition for causes other than heart failure, in each of the five years before the index admission. From these same five year data we obtained information on conditions associated with the development of heart failure: AMI (ICD 410/I21), other CHD (ICD 411–414/I20/I22–I25), heart disease other than CHD (ICD 415–429/I26–I52), hypertension (ICD 40/I1), heart valve disease (ICD 39/I0), and diabetes mellitus (ICD 250/E10–E14).

Statistical analysis

We used the χ² test for trend in ordered categories to analyse temporal changes in proportions and evaluated differences between population subsets using normal approximation confidence intervals. Crude survival was estimated using the Kaplan-Meier method, with log rank test to assess temporal (annual) trends in these estimates. Cox proportional hazards modelling was used to investigate the influence on outcome of potential explanatory variables. The stratified Cox procedure was used to analyse survival according to previous and concomitant diagnoses. The strategy for multivariate model selection was that published by Collett,²⁰ with the significance level for inclusion of variables at 10%. Statistical analyses were performed using SPSS software (SPSS Inc, Chicago, Illinois, USA).

Patient characteristics

Between 1 April 1993 and 31 March 2001 a total of 12 220 patients were admitted to hospital for the first time with a diagnosis of heart failure. In 4335 (36%) heart failure was recorded as the primary diagnosis. Demographic features and duration of index admission were similar for patients with heart failure in the primary coding position and in the total cohort (table 1⇓). Half of the patients were women and more than 60% were older than 75 years. The deprivation score, median 16.7, was similar to the national value of 16.9 for all English wards. More than half of the patients came from areas in the lowest two quintiles of deprivation score. With 30% of the population and 42% of cases, patients from the lowest quintile were over represented.

Over the study period there was a significant increase in the number and population rate of admissions. The age of men at presentation increased while that of women was unchanged (table 2⇓). Age and sex adjusted hospital incidence rate among those aged ≥ 40 rose by 62%, from 29/10 000 population in 1993/94 to 47/10 000 in 2000/01, with no sex difference (fig 1⇓). This increase was seen primarily in those aged ≥ 65 years, in whom the rate increased from 75 to 120 cases per 10 000 population. Corresponding trends were observed for heart failure in the primary coding position (fig 1⇓, table 2⇓). Rates also increased almost exclusively in those aged > 65 years, from 27 to 43/10 000. Overall numbers and population rates appeared to plateau after 1998/99 (fig 1⇓, table 2⇓). The median length of hospital stay was unchanged over the observation period (table 2⇓).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Trends in first heart failure admission rates between 1993 and 2001. (A) All diagnostic positions; (B) first diagnostic position. Rates for population older than 40, standardised for age and sex.

Conditions leading to heart failure: previous hospitalisations

Table 3⇓ shows the recorded frequency of comorbid conditions on the index and prior hospitalisations. Overall, in nearly 42% of patients (n = 5098) a form of CHD was recorded either before or during the first heart failure admission. A further 10% of patients (n = 1218) had other heart disease coded. With regard to the index admission, for one third of patients (n = 4195) there was a concomitant diagnosis of CHD, including 13% (n = 1542) with AMI.

Other potential contributory conditions such as hypertension or diabetes were recorded less commonly in the previous five years of hospitalisation data. Heart valve disease was recorded very infrequently (table 3⇑). Nearly half of the patients either had no recorded hospitalisation (n = 3782, 31%) in the prior five years or had been hospitalised only for conditions not normally associated with heart failure (n = 1987, 16%).

Survival

For all 12 220 patients, mean follow up was 651 days (22 months) with a range of 0–3103 days (8.5 years). For those alive at the end of the study, length of follow up was 183–3103 days.

Table 4⇓ presents Kaplan-Meier estimates of survival for all patients and for the subset with heart failure as the primary diagnosis. Overall, 7818 patients (64%) died by the end of follow up, including 5364 (44%) from cardiovascular causes. One month and one year case fatality was 21% and 43%, respectively, the majority being cardiovascular. By six years of follow up case fatality was 75–80%, with slightly worse outcome for those with heart failure in the primary diagnostic position. Hospital mortality remained unchanged at around 20% over the period of observation. Patients dying during the index admission were on average 2–3 years older than the population as a whole (table 2⇑).

Influence of demographic factors

Many of the measured demographic variables were strongly related to the risk of death (table 5⇓). After adjusting for all other factors, there was a 43–45% increase in risk of death for each 10 years of age at the time of admission and a 14–17% increase associated with male sex. There was a clear relation to time spent in hospital in the prior five years, but no apparent influence on survival of social deprivation score.

We observed a number of associations between measures of comorbidity and prognosis. Our general measure of comorbidity—the average number of days in hospital in each of the previous five years—was strongly related to outcome (table 5⇑). More specifically, when adjusted for other factors, prognosis for both all cause and cardiovascular mortality was worse for patients whose first heart failure admission was concomitant with AMI. For these 1542 patients, 30 day cardiovascular survival was 74% (95% confidence interval (CI) 72% to 76%) compared with 86% (95% CI 85% to 87%) in the non-myocardial infarction group. One year survival was also lower at 61% (95% CI 58% to 64%) compared with 70% (95% CI 69% to 71%) in those without concomitant AMI.

We observed no difference in survival between patients with (one month all cause survival 81%, 95% CI 79% to 83%) or without (80%, 95% CI 79% to 81%) diagnosed diabetes. At three years these probabilities were 37% (95% CI 34% to 40%) and 40% (95% CI 39% to 41%), respectively. Similarly, the diagnosis of diabetes did not alter cardiovascular mortality.

Trends in survival

When stratified by the year of index admission there was a clear trend to improvement in the Kaplan-Meyer estimate of survival between 1993/94 and 2000/01 (χ² = 13, p < 0.001). Multivariate modelling confirmed this trend, showing up to 50% reduction in the relative risk of cardiovascular death over the period, most evident in the last three years (fig 2⇓, table 5⇑). Using a stratified model with adjustment for age, sex, and comorbidity, the one month, all cause survival estimates were 72% (95% CI 69% to 75%) in 1993/94 and 82% (95% CI 80% to 84%) in 2000/01. One year survival was 45% (95% CI 42% to 48%) in 1993/94 compared with 62% (95% CI 60% to 64%) in 2000/01. Similar improvements were seen in cardiovascular survival, from 78% to 88% at one month and from 59% to 76% at one year. Trends were similar for heart failure as the primary diagnosis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Estimated hazard ratios and their 95% confidence intervals for all cause and cardiovascular mortality in 12 220 patients, according to the year of first admission, adjusted for age, sex, comorbidity, and social deprivation. Probability (p) values represent the contribution of year of diagnosis to multivariate model. Dotted line indicates linear trend in hazard estimates.

Rates of first hospitalisation

Following steady increases from 1993–98, we observed a plateau in numbers and population rates from 1998–2001. A slowing in admission rates in recent years has been noted in studies from Scotland¹⁵ and the Netherlands.¹⁶ While requiring further observation to clarify its reality, this finding is encouraging. Importantly, the increase in numbers of admissions in the current study was seen largely in those aged 65 years or more. The age of men continued to increase in the latter part of the 1990s, reaching a median of 79 in 2000/01. The median age of women in our study was unchanged at 80, in contrast to the increase in Scotland from 76 in 1986 to 79 in 1995.¹⁷ Such findings have clear implications for increasingly elderly populations.

Influence of age and sex

Our observed association of greater age and male sex with higher mortality are not unexpected. Moreover, the strengths of these associations (hazard ratio of 1.4/10 years of age and 0.87 for female sex) are strikingly similar to those seen in previous studies.^17,21 The differential risk associated with age and male sex has not changed over the period 1985–2001.

Influence of deprivation

In keeping with previous reports from the UK,²⁷ we observed no influence of deprivation on survival. However, a disproportionate number of index patients came from the most deprived areas, again in keeping with previous studies.¹⁷ These areas clearly have a heavy burden of disease associated with heart failure. In Leicestershire these areas are in the vast majority urban, with a high demand for hospital care. A lower threshold for referral to hospital for residents of these areas may explain the apparent lack of effect of social deprivation on mortality but cannot be verified without information on disease severity at diagnosis. More detailed comparison of the characteristics of patients from the various quintiles of deprivation may help clarify this issue.

Influence of comorbidity

When adjusted for demographic factors, comorbidity, and year of diagnosis, mortality was higher in patients for whom heart failure was recorded concomitant with AMI. Once again this is in keeping with data from both large epidemiological²¹ and small cohort²⁷ studies. Patients with AMI are an easily identifiable group at high risk of heart failure and to whom appropriate investigations and treatment should be targeted. Interestingly, only a small proportion of our cohort (7%) had a hospital discharge diagnosis of AMI in the previous five years, compared with 15% of first heart failure admissions in Scotland,¹⁷ perhaps reflecting regional differences in the incidence of CHD.

Although a small percentage of our cohort spent any considerable amount of time in hospital in the previous five years, this very general measure of overall comorbidity associated strongly with outcome. This very simple observation perhaps emphasises the importance of concomitant pathology and interactions between covariables in the natural history of heart failure.²¹

Limitations of the study

Our study is constrained by the limitations inherent in all studies of historical, observational design. Inaccuracies in the diagnosis and coding of heart failure in routine data are well recognised²⁸ and we have of necessity relied on the accuracy of such data. While we identified only hospitalised patients, patients remaining in the community are likely to have a better prognosis. Temporal changes in referral and coding practices, in diagnostic accuracy, and in awareness of heart failure as a diagnostic entity may have influenced our findings. Similar comments can be applied to concomitant diagnoses potentially influencing prognosis such as AMI and diabetes. We have incomplete information on prior diagnoses, disease severity, and drug treatment at presentation. Similarly, we have not assessed the potential impact on trends in outcome of all relevant cofactors, such as renal impairment. These potential criticisms apply equally to previous studies of hospitalised patients, and the demographic features and short term and long term prognosis of our population are very much in keeping with these studies.^17,21 The duration of index admission and inpatient fatality rate did not change during our observation period and are very similar to those reported previously from Scotland for the year 1996.¹⁵ This suggests that the severity of disease was on average similar throughout the study period. We feel that our work stands reasonable comparison with previous large, epidemiological studies of trends in heart failure hospitalisation and prognosis.

Summary

Following many years of increase, numbers of patients with first hospital admission with heart failure reached a plateau after 1998. Clear improvements in survival were observed between 1993 and 2001. However, the prognosis for patients newly admitted with heart failure remains poor, in particular for those of greater age, for men, and for those with heart failure recorded during admission with myocardial infarction. These groups are cohorts in whom screening for heart failure and asymptomatic left ventricular dysfunction are likely to be relatively cost effective. While the plateau in numbers and improvements in survival are welcome, the prevalence of heart failure remains high and the prognosis poor.