Customize your JAMA Network experience by selecting one or more topics from the list below.
Roger VL, Weston SA, Redfield MM, et al. Trends in Heart Failure Incidence and Survival in a Community-Based Population. JAMA. 2004;292(3):344–350. doi:10.1001/jama.292.3.344
Context The epidemic of heart failure has yet to be fully investigated, and
data on incidence, survival, and sex-specific temporal trends in community-based
populations are limited.
Objective To test the hypothesis that the incidence of heart failure has declined
and survival after heart failure diagnosis has improved over time but that
secular trends have diverged by sex.
Design, Setting, and Participants Population-based cohort study using the resources of the Rochester Epidemiology
Project conducted in Olmsted County, Minnesota. Patients were 4537 Olmsted
County residents (57% women; mean [SD] age, 74  years) with a diagnosis
of heart failure between 1979 and 2000. Framingham criteria and clinical criteria
were used to validate the diagnosis
Main Outcome Measures Incidence of heart failure and survival after heart failure diagnosis.
Results The incidence of heart failure was higher among men (378/100 000
persons; 95% confidence interval [CI], 361-395 for men; 289/100 000 persons;
95% CI, 277-300 for women) and did not change over time among men or women.
After a mean follow-up of 4.2 years (range, 0-23.8 years), 3347 deaths occurred,
including 1930 among women and 1417 among men. Survival after heart failure
diagnosis was worse among men than women (relative risk, 1.33; 95% CI, 1.24-1.43)
but overall improved over time (5-year age-adjusted survival, 43% in 1979-1984
vs 52% in 1996-2000, P<.001). However, men and
younger persons experienced larger survival gains, contrasting with less or
no improvement for women and elderly persons.
Conclusion In this community-based cohort, the incidence of heart failure has not
declined during 2 decades, but survival after onset of heart failure has increased
overall, with less improvement among women and elderly persons.
The burden of heart failure and its societal cost are staggering. Approximately
4.9 million Americans have been diagnosed with the disease.1 Hospital
discharges for heart failure increased by 155% during the last 20 years, and
heart failure is the most frequent cause of hospitalization in persons aged
65 years or older. Within this context, heart failure constitutes a public
health problem1 singled out as an emerging
epidemic.2 Although the clinical and public
health importance of heart failure is undisputed, this epidemic is not adequately
understood. Data on the incidence of heart failure are relatively sparse and
lack consistency. Most data are derived from hospital discharge records or
self-report and thus do not reflect incidence, have uncertain validity caused
by documented shifts in coding practices because of reimbursement incentives,3 and cannot fully capture the burden of the disease
because of the shift of care toward outpatient settings. Thus, little is known
about temporal trends in the incidence of heart failure and on survival after
In the Framingham Heart Study, the incidence of heart failure between
1950 and 1999 changed little among men but declined in women, whereas survival
improved in both sexes.4 These findings require
replication in other settings because the design of the Framingham Heart Study,
as acknowledged by the authors, may impose a "healthy participant" effect.5 Furthermore, although using consistent research standards
is necessary, such as the Framingham criteria to assess secular trends, it
is important to examine their relationship to clinical diagnoses.
This study was designed to estimate the secular trends in the incidence
of heart failure and in survival after its onset in a geographically defined
community by using 2 sets of criteria: the Framingham criteria and clinical
criteria. We hypothesized that, over time, the incidence of heart failure
increased and that survival after its onset improved but that the trends in
incidence and survival differed by age and sex.
The study was conducted among the population of Olmsted County, Minnesota,
the characteristics of which are similar to those of US non-Hispanic whites.
The Mayo Clinic and Olmsted Medical Center provide the majority of medical
care for this population. Both organizations use a unit medical record system
in which information is collected by health care clinicians in a single record,
regardless of site of care. These records are easily retrievable because the
Mayo Clinic has maintained extensive indices of diagnoses and procedures,
which were extended through the Rochester Epidemiology Project to the records
of other clinicians to county residents, resulting in the linkage of all medical
records from all sources of care through a centralized system.6,7
All persons with a first diagnosis of heart failure were identified
with codes from the International Classification of Diseases,
Ninth Revision, Clinical Modification (ICD-9-CM),
similar to those used in other studies.8 These
codes included 428 (heart failure), 402.01 (hypertensive heart disease malignant
with congestive heart failure), 402.11 (hypertensive heart disease benign
with congestive heart failure), 425 (cardiomyopathy), 429.3 (cardiomegaly),
and 514 (pulmonary congestion). Codes 402.91 (hypertensive heart disease unspecified
with congestive heart failure), 404.01 (malignant hypertensive heart and renal
disease with heart failure), 404.11 (benign hypertensive heart and renal disease
with congestive heart failure), and 404.91 (unspecified hypertensive heart
and renal disease with congestive heart failure) were queried but not used
as part of the coding practices during the study.
The codes, which are not based on hospital billing codes, were assigned
by coders primarily according to physician diagnoses for outpatient visits
and on discharge diagnoses for hospitalizations. Samples of each code were
reviewed to determine their respective yield toward validated heart failure.
Because of the predominant use of code 428 and its frequent association with
other codes, other codes were reviewed only if not associated with code 428.
Nonresidency in Olmsted County was an exclusion criterion. The yield of each
code toward a validated diagnosis of heart failure was examined in aggregate,
as well as according to time, age, and sex by using logistic regression. The
resulting model was used to construct the validated heart failure incidence
cohort, with 2 sets of criteria.
The Framingham criteria,9 which have
been used consistently in various settings and can provide qualitative comparisons
with results from other studies, were used as 1 standard (Box). A second criterion, defined by a physician's diagnosis
of heart failure, was used to assess these trends in clinical practice. Three
experienced abstractors reviewed the records, masked to the study hypothesis.
Clinical characteristics, including comorbidity defined by the Charlson index,10 were collected in a stratified random sample of 11%
of code 428 and 8% of each of codes 425, 429.3, and 514. Overt coronary disease
was defined as history of myocardial infarction, angiographic coronary disease,
or coronary artery bypass graft surgery. Myocardial infarction was defined
by epidemiologic criteria.11,12 Angiographic
coronary disease was defined as stenosis greater than 75% of the left anterior
descending, left circumflex, or right coronary artery or 50% of the left main
coronary artery.13 Clinical diagnoses of valvular
diseases as documented by attending physician were also recorded.
Paroxysmal nocturnal dyspnea or orthopnea
Neck vein distension
Acute pulmonary edema
Increased venous pressure ≥16 cm H2O
Circulation time ≥25 seconds
Dyspnea on exertion
Vital capacity decreased one third from maximum
Tachycardia rate ≥120/min
Major or Minor Criterion
Weight loss ≥4.5 kg in 5 days in response to treatment
Heart failure present with 2 major or 1 major criterion plus 2 minor criteria
The feasibility of applying the Framingham criteria was assessed by
determining the frequency with which the required components were characterized
in the medical record. Missing values were rare, and the Framingham criteria
could be applied in 98% of cases. The records of 30 patients were reabstracted
to examine interabstractor variability and assess the reliability of ascertainment
of heart failure. There was 100% interabstractor agreement for both classifications,
indicating that these ascertainment methods are highly reproducible.
Follow-up was performed through 2003 by using the inpatient and outpatient
medical records. The ascertainment of death includes several procedures. In
addition to the deaths noted during clinical care, all death certificates
for Olmsted County residents are obtained every year from the county office.
The Mayo Clinic registration office records the obituaries and notices of
deaths in the local newspapers. Finally, data on all Minnesota deaths are
obtained from the State of Minnesota every year. The county population is
overall stable, and health status has little influence on migration.14
Characteristics of patients with validated heart failure are presented
as frequencies or mean values with SDs. Differences between men and women
were tested with χ2 tests for categorical variables and t tests for continuous variables.
Age-, sex-, and year-specific incidence rates of validated heart failure
were calculated. The counts of validated heart failure were used as the numerators,
and the denominators were the Olmsted County population, as determined by
census data for 1970, 1980, 1990, and 2000, with linear interpolation for
the intracensus years.15 The rates were adjusted
directly to the age distribution of the 2000 US total population. Standard
errors and 95% confidence intervals (CIs) around the point estimates were
calculated assuming a Poisson error distribution.
A Poisson regression model (SAS GENMOD procedure, version 8; SAS Institute,
Inc, Cary, NC) was used to examine the temporal trends in the incidence of
validated heart failure, with categorical year variables and adjustment for
age. Results are summarized by presenting the relative risk (RR) of validated
heart failure for men and women in each year group, using 1979 to 1984 as
Proportional hazards modeling was used to examine the association between
year of heart failure diagnosis and survival while adjusting for age. Results
are summarized by presenting the RR of validated heart failure in the various
year groups compared with the 1979 to 1984 period for men and women at different
ages. Sex-specific 30-day, 1-year, and 5-year mortality rates were estimated
from the proportional hazards regression models and are presented for patients
75 years of age.
All analyses were stratified by sex. In the modeling, year was modeled
categorically, and a nonlinear effect of age was assessed by testing the quadratic
term. Comparison of time trends across age groups was accomplished by including
interaction terms between year groups and age. P =
.05 was selected for the threshold of statistical significance, except when
an interaction was tested for, when P = .10 was used.
All analyses were replicated in 1000 random samples to ensure that results
Between 1979 and 2000, 7298 cases of first diagnosis of heart failure
identified by code 428 and 1877 cases of first diagnosis of heart failure
identified by other codes in isolation without a code 428 occurred. Code 428
constituted the majority (80%) of heart failure codes. This proportion increased
over time (77% in 1979-1984 vs 83% in 1996-2000; P for
Eighty-two percent of the cases coded as 428 met Framingham criteria
for heart failure, whereas cases with other codes used in isolation without
a code 428 met Framingham criteria in 14% to 30% of the cases. The proportion
of code 428 cases meeting Framingham criteria did not change over time: 80%
in 1979-1984 vs 79% in 1996-2000 (P for trend = .69).
The validation rate for each of the target codes adjusted for age, sex, and
time were combined with residency status in Olmsted County and incident nature
of the event, allowing an incidence cohort of 4537 cases of validated heart
failure to be assembled. Among these, 57% were women. The mean (SD) age at
the diagnosis of heart failure was 74 (14) years, and 58% were aged 75 years
According to the Framingham criteria, the age-adjusted incidence of
heart failure was higher among men (378/100 000 persons; 95% CI, 361
to 395) than women (289/100 000 persons; 95% CI, 277 to 300) (P<.001). The incidence of heart failure did not change over time
(Table 1). When analyzed with
year as a continuous variable, the estimated annual percent increase was 0.15%
(95% CI, –0.55 to 0.85) or 3% (95% CI, –11 to 20) for 1979 to
2000 for men and 0.37% (95% CI, –0.25 to 0.98) annually or 8% (95% CI,
–5 to 23) for 1979 to 2000 for women.
Patients with validated heart failure diagnosis were elderly, particularly
women who were also less likely to be overweight but more likely to be hypertensive
than men (Table 2). Men were more
likely to be current or past smokers. Forty-two percent of the cases were
diagnosed in the outpatient setting. Among these cases, 74% were hospitalized
within 1.7 (3.1) years of the diagnosis, whereas 26% were never hospitalized.
The use of angiotensin-converting enzyme inhibitors and β-blockers
at diagnosis increased significantly between the first and last time period,
from 0% to 51% for angiotensin-converting enzyme inhibitors and from 10% to
30% for β-blockers (both P for trend <.001).
After a follow-up of 4.2 years (range, 0-23.8 years), 3347 deaths occurred,
1930 among women, 1417 among men, 1127 among patients younger than 75 years,
and 2220 among patients aged 75 years or older. Mortality rates after the
onset of heart failure, adjusted for age, was higher among men, irrespective
of the period (RR for men vs women, 1.33; 95% CI, 1.24-1.43, P<.001) (Table 3).
Mortality declined over time, with an overall improvement from 57% in
1979 to 1984 to 48% in 1996 to 2000 for the age-adjusted 5-year mortality
estimates (P<.01) (Figure 1). However, there were age and sex differences in the degree
of improvement in survival (P<.001 for year-by-age
interaction and year-by-sex interaction) (Table 4). Men in their 60s experienced a 52% improvement in survival
between the first and the last period. Survival also improved, although to
a lesser degree, among older men. Among women, survival improved in younger
ages but to a lesser extent than it did among men and did not change in older
All analyses were repeated with the clinical diagnosis. The yield of
code 428 toward clinical heart failure was 90%, whereas that of other codes
used in isolation without a code 428 ranged from 14% to 36%. According to
the clinical diagnosis, 4962 cases of heart failure were identified, representing
a 14% increment over those identified by the Framingham criteria. This increment
was constant over time (P for trend = .80). The age
and sex distributions of the clinical cases were similar. Compared with patients
who met Framingham criteria, those who did not were more likely to be overweight
or obese and to have had a previous myocardial infarction and were less likely
to be hospitalized at diagnosis (P<.05 for all
comparisons). Although the incidence of clinical heart failure was higher
than that defined by the Framingham criteria, survival was similar. The secular
trends in incidence and survival were similar, irrespective of the criteria
Analyses replicated in 1000 samples yielded similar results.
In this large geographically defined community, the incidence of heart
failure has remained stable during the past 2 decades in men and women. The
incidence of heart failure is higher among men, and survival after its onset
is worse among men. Although survival after the onset of heart failure improved
over time, there were disparities in the magnitude of the improvement, which
was greater among men and younger individuals.
The clinical diagnostic criteria identified slightly more cases of heart
failure than the Framingham criteria, but the secular trends were similar,
irrespective of the criteria, attesting to their robustness.
Information on the incidence of heart failure and how it may have changed
over time is limited. Studies of hospital discharges are event based, not
person based, with multiple hospitalizations for the same individuals counted
so that incidence cannot be measured. Furthermore, standardized diagnostic
criteria are seldom used, and case ascertainment is affected by shifts in
hospital discharge diagnoses preferences.3,16,17 Patients
diagnosed and treated for heart failure in the outpatient setting constitute
an increasingly important component of the cases but are seldom included.
Given these limitations, it should not be unexpected that published data on
the incidence of heart failure lack consistency. For instance, using Medicare
hospital claims in 1986 and 1993, Croft et al18 reported
an increase in the initial hospitalization for heart failure. Conversely,
Stewart et al19,20 suggested that
trends in hospitalization in Scotland in the 1990s had "leveled off."
Although these data are limited by the lack of validation and restriction
to inpatient data, they prompt the question of whether the stabilization of
heart failure hospitalization rates could be offset by increasing outpatient
care patterns. Few studies included outpatient data.21-24 The
present study underscores the importance of doing so because 42% of the cases
were diagnosed as outpatients, and among these, 26% were never hospitalized
and thus would not have been identified by using hospital-based surveillance.
Data from the Henry Ford Health system, a managed care organization,23 indicated that the prevalence of heart failure was
increasing over time but did not report a secular change in the incidence
of heart failure or mortality after its onset. In addition, the Framingham
Heart Study recently reported a decline in the incidence of heart failure
in women but not among men and an improvement in survival in both sexes.24 As acknowledged by the authors, these data could
be affected by the healthy participant effect.5
The absolute magnitude of the incidence of heart failure cannot be compared
qualitatively between the 2 studies because of different age-adjustment strategies.
However, the use of identical criteria and the inclusion of outpatient cases
in both studies facilitate the qualitative comparison of trends. The Framingham
Heart Study reported no change in the incidence of heart failure among men
but found a decline in women, primarily in the earlier periods. Conversely,
our study did not detect a change in either sex. Our sample size, larger than
that of the Framingham Heart Study, provided power to detect a change (either
increase or decrease) in the incidence of heart failure of 0.8% per year,
or 18% during the 22 years of the study. Thus, although we cannot exclude
smaller changes, the size of our cohort was adequate to detect changes of
clinical and public health significance.
Our study supports and extends the concept brought forth by Framingham
investigators that the heart failure epidemic is not related to an increase
in incidence but also challenges the notion that further progress is being
made in preventing heart failure in the community. Because heart failure is
a disease of the elderly, the stagnation of incidence rates among an aging
and growing population will increase the number of cases of heart failure
and its public health burden.
Estimates of the prevalence of coronary disease in studies of heart
failure vary from 68% among clinical trial enrollees25 to
36% in a population-based study, relying on physician adjudication.26 Our study reports a prevalence of coronary disease
of 28% by using rigorous criteria for overt disease that are conceivably conservative.
However, determining the etiology of heart failure is complex, particularly
in the community, given the nonuniform use of coronary angiography and that
several of the putative etiologies of heart failure, such as hypertension,
diabetes, and coronary disease, may coexist. Thus, their respective responsibility
in the genesis of heart failure may be difficult to establish.
Several studies have reported on improved inhospital mortality of heart
failure.27,28 However, these studies
are fraught with biases related to lack of consistent criteria over time and
the ensuing differences in case mix; decreasing length of hospital stay, which
in turn confound trends in outcomes; and lack of inclusion of outpatient cases.
These factors limit generalizability of these data to a large number of persons
with heart failure.
Among studies including outpatient cases, previous data from Olmsted
County reported no temporal trends in heart failure survival but were powered
to detect only large trends.22 The Framingham
Heart Study4 reported improvement in the survival
of heart failure in both sexes. Because this improvement applies to a closed
cohort, it is unclear how much of this change reflects survivor bias vs true
improvement in outcome.30
The present results from a larger community-based population indicated
a substantial improvement in survival after heart failure in younger men,
contrasting with less improvement in women and the elderly.
The hypothetical explanations for these trends are likely multifactorial.
Treatment effectiveness may play a role. Trials that demonstrated the efficacy
of angiotensin-converting enzyme inhibitors were published in the late 1980s.31 The ensuing clinical practice changes could have
participated in the improved survival after the onset of heart failure noted
in most recent years. Our observed increase in use of angiotensin-converting
enzyme inhibitors at heart failure diagnosis would support this construct.
The improvement in heart failure survival could also be related to earlier
diagnosis as the use of cardiac diagnostic procedures increases over time,
resulting in lead-time bias.30 While this factor
may play a role, it could not by itself explain contrasting sex-specific trends
in survival. The mechanism of heart failure may differ by sex and could affect
outcomes. A greater proportion of women with heart failure also have hypertension,
suggesting that diastolic heart failure may play a greater role in women,
which in turn may explain why less improvement in outcome was noted among
women because there is no effective therapy for diastolic heart failure. Irrespective
of these hypothetical explanations, these differences in improvement in survival
according to age and sex call for additional studies to identify the reasons
for these disparities.
Potential limitations of the study include the racial and ethnic composition
of Olmsted County, which limits the generalizibility of these data to groups
underrepresented in the population. Although no single community can completely
represent the nation as a whole, studies of chronic diseases in Olmsted County
indicate that results from the county can be extrapolated to a large part
of the population, and the characteristics of the Olmsted County population
are similar to those of US whites, with the exception of slightly higher income
and education.6 This study should, however,
be replicated in other racial and ethnic groups.
Medication use in heart failure in Olmsted County has been reported32,33 to be similar to that in other settings.
Providing detailed information on medications used in this cohort is beyond
the scope of this article; however, we observed an increase over time in the
use of β-blockers and angiotensin-converting enzyme inhibitors at diagnosis. β-Blockers
were not recognized as improving survival in heart failure until the late
1990s, so the period covered by the cohort largely precedes this evidence.
Furthermore, to gain relevant insight into the use of medications, data on
the introduction or withdrawal of drugs during follow-up would be needed but
are unavailable for this cohort. Finally, because this is an observational
study, adjustment for medication use would not indicate causality.
Our study, conducted in a community-based population, has several important
strengths. We used 2 sets of criteria to classify heart failure, adding clinical
ascertainment to the Framingham criteria and thereby providing a robust connection
with clinical practice. Second, we include outpatient data. Third, the size
of our cohort confers additional power to detect effect modification and thus
identify disparities such as the ones presented here. Finally, our study reports
on the experience of an open cohort and thus reflects the experience of a
community without the survivor bias that might affect results in a closed
This community-based cohort study demonstrates no substantial reduction
in the incidence of heart failure during more than 2 decades. Moreover, women
and the elderly experienced less improvement in survival after the onset of
heart failure, suggesting that the apparent gains in secondary prevention
have not been achieved equally. These findings highlight the increasing population
burden of heart failure mediated by stagnating incidence and unequal improvement
in survival within the context of an aging and growing population.