Survival Associated with Two Sets of Diagnostic Criteria for Congestive Heart Failure

American Journal of Epidemiology Copyright 2004 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 160, No. 7 Printed in U.S.A. DOI: 10.1093/aje/kwh268 Survival Associated with Two Sets of Diagnostic Criteria for Congestive Heart Failure Gina D. Schellenbaum 1, Thomas D. Rea 2, Susan R. Heckbert 1, Nicholas L. Smith 1, Thomas Lumley 3, Veronique L. Roger 4, Dalane W. Kitzman 5, Herman A. Taylor 6, Daniel Levy 7, and Bruce M. Psaty 1,2,8 1 Department of Epidemiology, Cardiovascular Health Research Unit, University of Washington, Seattle, WA. 2 Department of Medicine, Cardiovascular Health Research Unit, University of Washington, Seattle, WA. 3 Department of Biostatistics, Cardiovascular Health Research Unit, University of Washington, Seattle, WA. 4 Division of Cardiovascular Diseases and Internal Medicine, Mayo Clinic and Mayo Foundation, Rochester, MN. 5 Section on Cardiology, Department of Medicine, Wake Forest University School of Medicine, Winston-Salem, NC. 6 Department of Medicine, University of Mississippi Medical Center, Jackson, MS. 7 National Heart, Lung, and Blood Institute and the Framingham Heart Study, Framingham, MA. 8 Department of Health Services, Cardiovascular Health Research Unit, University of Washington, Seattle, WA. Received for publication February 3, 2004; accepted for publication April 12, 2004. Congestive heart failure (CHF) definitions vary across epidemiologic studies. The Framingham Heart Study criteria include CHF signs and symptoms assessed by a physician panel. In the Cardiovascular Health Study, a committee of physicians adjudicated CHF diagnoses, confirmed by signs, symptoms, clinical tests, and/or medical therapy. The authors used data from the Cardiovascular Health Study, a population-based cohort study of 5,888 elderly US adults, to compare CHF incidence and survival patterns following onset of CHF as defined by Framingham and/or Cardiovascular Health Study criteria. They constructed an inception cohort of nonfatal, hospitalized CHF patients. Of 875 participants who had qualifying CHF hospitalizations between 1989 and 2000, 54% experienced a first CHF event that fulfilled both sets of diagnostic criteria (concordant), 31% fulfilled only the Framingham criteria (Framingham only), and 15% fulfilled only the Cardiovascular Health Study criteria (Cardiovascular Health Study only). No significant survival difference was found between the Framingham-only group (hazard ratio = 0.87, 95% confidence interval: 0.71, 1.07) or the Cardiovascular Health Study-only group (hazard ratio = 0.89, 95% confidence interval: 0.68, 1.15) and the concordant group (referent). Compared with Cardiovascular Health Study central adjudication, Framingham criteria for CHF identified a larger group of participants with incident CHF, but all-cause mortality rates were similar across these diagnostic classifications. follow-up studies; heart failure, congestive; prognosis; survival Abbreviations: CHF, congestive heart failure; CI, confidence interval; FEV 1, forced expiratory volume in 1 second; HR, hazard ratio. Congestive heart failure (CHF) is especially common among older adults (1 3). Although a standard definition is crucial to understanding the epidemiology of CHF, diagnostic criteria for CHF are difficult to define and apply in population-based settings. No physical sign, blood assay, or imaging test is definitive for the diagnosis of CHF. The primary symptoms of CHF shortness of breath, fatigue, and swelling of the legs, ankles, or feet are not specific to CHF and are prominent in other medical conditions common in older adults (2). In 1971, the Framingham Heart Study published diagnostic criteria for CHF based on signs and symptoms assessed and adjudicated by a physician panel (1). The Cardiovascular Health Study defines CHF by using signs, symptoms, clinical tests, and/or medical therapy via central adjudication by a committee of physicians (4, 5). Differences Correspondence to Gina D. Schellenbaum, 1730 Minor Avenue, Suite 1360, Seattle, WA 98101 (e-mail: ginas@u.washington.edu). 628

Survival Associated with CHF Diagnostic Criteria 629 between these two sets of diagnostic criteria could affect estimates of CHF incidence and prognosis. In the Cardiovascular Health Study, contemporary elements of the Framingham criteria were collected for cardiac hospitalizations of Cardiovascular Health Study participants, and these data enable a direct comparison between the Framingham criteria and the Cardiovascular Health Study central adjudication approach. Given the low survival rates among persons with CHF (1, 6 9), mortality was chosen as an endpoint for comparing the two criteria. The investigation compared survival following the diagnosis of incident CHF by the Framingham criteria, Cardiovascular Health Study adjudication, or both methods. MATERIALS AND METHODS Participants and setting The Cardiovascular Health Study is a prospective, population-based cohort study of 5,888 adults who were at least 65 years of age at baseline. Participants were recruited from a random sample of Medicare-eligible residents from the four study sites: Sacramento County, California; Washington County, Maryland; Forsyth County, North Carolina; and Allegheny County, Pennsylvania. Persons were eligible to participate if they 1) had been sampled, or were living in the household of a person sampled, from the Health Care Financing Administration lists; 2) were at least 65 years of age; 3) were not institutionalized; 4) expected to remain in the area for 3 years; and 5) did not require a proxy respondent. All participants gave informed consent, all relevant institutional committees on human research approved the study, and the procedures followed were in accordance with institutional guidelines. The study enrolled 57 percent of the eligible persons contacted. The original cohort of 5,201 participants was recruited in 1989 1990; 687 additional subjects, most of whom were African American, were recruited in 1992 1993 by using similar methods. TABLE 1. Availability of Framingham Heart Study CHF* criteria (%) among Cardiovascular Health Study participants in the inception cohort at incident CHF hospitalization, the Cardiovascular Health Study, United States, 1989 2000 Original Framingham criteria Present Absent Unknown Major criteria Bilateral moist rales 69 26 6 Paroxysmal nocturnal dyspnea 52 39 10 and/or orthopnea Pulmonary edema by 54 34 11 radiography Neck vein distention in the 28 61 12 semirecumbent position Enlarging heart by radiography 56 30 14 S3 gallop 8 70 22 Hepatojugular reflux 3 69 29 Peripheral venous pressure >16 cm H 2 O Arm-to-tongue circulation time 25 seconds Weight loss of 4.5 kg in 5 days with treatment Minor criteria Ankle edema 49 43 8 Pleural effusion 36 47 17 Hepatomegaly 1 81 18 Tachycardia ( 120 beats per 10 66 24 minute) Dyspnea on exertion 39 22 40 Night cough 6 38 56 33 1/3% decrease in vital capacity * CHF, congestive heart failure. A Framingham diagnosis of CHF requires either two major criteria or one major and two minor criteria (12); autopsy evidence was not considered because only nonfatal CHF cases were included; four criteria were not available. Data collection Patient characteristics were assessed at baseline and annual follow-up clinic visits (10). Prevalent CHF at baseline was ascertained through self-report, with medical record verification; during follow-up, baseline CHF status was updated by using information obtained during investigations for incident events (11). All potential incident cardiovascular events (CHF, myocardial infarction, angina, stroke, transient ischemic attack) and deaths were adjudicated by the events committee (4). Deaths were confirmed by review of medical records and death certificates, as well as the Health Care Financing Administration Medicare health care utilization database for hospitalizations. The present investigation includes morbidity and mortality through June 30, 2000. Framingham diagnostic criteria For all potential incident cardiovascular event hospitalizations, trained medical abstractors reviewed hospital records for diagnostic information, including the Framingham CHF criteria. A Framingham diagnosis of CHF required either two major criteria or one major and two minor criteria (table 1) (12). When no information was available about the presence of a Framingham criterion, this criterion was assumed to be absent. Cardiovascular Health Study diagnostic criteria The events committee a panel of physicians adjudicated CHF by reviewing all pertinent data, including history, physical examination, chest radiography report, and medication use. Physician diagnoses of CHF were confirmed by documentation in the participants medical records of a constellation of symptoms (i.e., shortness of breath, fatigue, orthopnea, paroxysmal nocturnal dyspnea) and physical signs (i.e., edema, pulmonary rales, gallop rhythm, displaced left ventricular apical impulse) or by clinical findings such as evidence of pulmonary edema on chest radiograph.

630 Schellenbaum et al. on the date their matched case entered the analysis. For five of the cases, no exact match could be found, so the age criterion was relaxed to find a match within 3 years. FIGURE 1. Inclusion diagram showing the formation of the congestive heart failure (CHF) inception cohort and diagnostic classifications for Cardiovascular Health Study (CHS) participants, United States, 1989 2000. The no qualifying CHF event box includes participants without CHF and participants whose CHF events were disqualified: 70 who were not hospitalized, six for whom hospital documentation was missing, 123 who died before hospital discharge, and 152 whose events were procedure related. Although ejection fraction was not available for all incident CHF hospitalizations, when it was available it was considered part of the clinical picture. The diagnosis of CHF was also confirmed if, in addition to having a physician diagnosis of CHF, the participant was receiving medical therapy for CHF, including diuretics, digitalis, angiotensin-converting enzyme inhibitors, and beta-blockers. Inception cohort The CHF inception cohort included 875 participants who survived to be discharged from the hospital between July 1, 1989, and June 30, 2000, following an incident, definite episode of CHF (figure 1). CHF cases were classified into three mutually exclusive groups: 1) concordant (fulfilled both the Framingham and the Cardiovascular Health Study criteria for CHF at the incident hospitalization), 2) Framingham only (fulfilled only the Framingham criteria for CHF), and 3) Cardiovascular Health Study only (fulfilled only the Cardiovascular Health Study criteria for CHF). Excluded from the inception cohort were participants with probable or definite heart failure at baseline (n = 273), participants with incident CHF events that were procedure related (n = 152) or were entirely managed outside of the hospital (n = 70), or participants for whom hospital documentation was unavailable (n = 6). Information on clinical conditions and behavioral characteristics was taken from the annual study visit immediately before the qualifying CHF hospitalization (an average of 8 months before the qualifying event), unless otherwise noted. To better understand the survival patterns of the three diagnostic classifications, we created a comparison group without CHF. A random sample of participants (N = 875) who had not developed CHF were individually matched to qualifying CHF cases by sex, age at baseline (year), and recruitment cohort (recruited in 1989 1990 vs. 1992 1993). Each of these participants without CHF entered the analysis Variable definitions Treated hypertension was defined as use of antihypertensive medications and a self-reported history of hypertension. Treated diabetes was defined as use of insulin or oral hyperglycemic agents. Abstracted hospital chart information was used to classify the prevalence of chronic obstructive pulmonary disease during the qualifying CHF hospitalization. Selfreport was used to classify asthma status, as well as kidney disease and heart valve problems. Atrial fibrillation was defined as either self-reported history or electrocardiogram evidence of atrial fibrillation at a Cardiovascular Health Study clinic visit. A hospital discharge diagnosis of heart failure was defined as any International Classification of Diseases, Ninth Revision, code 428, 425, 402.91, 402.11, 402.01, or 398.91. Body mass index calculations were based on annual weight measurement and the most recent height measurement from one of three visits at which height was measured. Forced expiratory volume in 1 second (FEV 1 ) was also measured during only three visits. Percent predicted FEV 1 was considered the measured FEV 1 divided by the calculated FEV 1 for healthy persons of the same age, sex, and height (13). Smoking behavior was reported at each annual visit for the 30 days before the visit, and alcohol consumption was calculated from self-reported typical amounts and frequencies. Statistical analysis Kaplan-Meier curves, log-rank tests, and Cox proportional hazards models with robust variance estimates were used to examine differences in all-cause mortality among the CHF diagnostic classifications and matched participants without CHF. All p values were two-sided. The Kaplan-Meier curves were truncated at 6 years of follow-up because of the small sample size beyond that point: 358 participants (Framingham only: 40, Cardiovascular Health Study only: 17; concordant: 67; no CHF: 234). The proportional hazards assumption was tested and was not rejected for comparisons of survival for the three CHF diagnostic classifications. Restricted subgroup analyses were conducted to explore the possibility that the survival differences might be observed in a distinctive group only, such as those with conditions that make a CHF diagnosis more difficult. Less than 5 percent of the data for variables in this analysis were missing, except for the Framingham criteria. Of all documented hospitalizations, no known values for any of the Framingham diagnostic criteria were available for 15.5 percent. The Framingham diagnostic criteria were more likely to all be missing at one of the clinics (Forsyth, North Carolina: 20.1 percent), for the younger participants (less than 75 years of age at the time of their hospitalized event: 32.6 percent), and early in the study (before the mean hospitalized event date, September 3, 1995: 29.7 percent). Nevertheless, among qualified hospitalizations classified as incident, definite CHF by the Cardiovascular Health Study

Survival Associated with CHF Diagnostic Criteria 631 criteria, data on all of the Framingham criteria were missing for fewer than 2 percent. To address the missing data for individual Framingham criteria, sensitivity analyses were conducted by excluding participants for whom data were missing for most or all of the Framingham criteria and by using a multiple imputation program (NORM software) to assign values for these missing data (14). The multiple imputation analysis used patient characteristics, clinical conditions, available Framingham criteria data, and outcome information to fill in missing Framingham criteria values. Survival analyses were conducted for each of 10 imputed data sets (1,000 iterations each), and the results were recombined. As part of the recombination, the effective amount of missing data was estimated to be approximately 2 percent based on the variations between the 10 imputation analyses. RESULTS The mean age at CHF diagnosis was 80 years, and approximately 50 percent of participants with CHF were female and 85 percent were Caucasian. At their incident event, 31 percent fulfilled only the Framingham criteria for CHF, 15 percent fulfilled only the Cardiovascular Health Study criteria, and 54 percent fulfilled both criteria (figure 1). In general, the three groups defined by diagnostic classification (Framingham only, Cardiovascular Health Study only, and concordant) had similar demographic and clinical characteristics (table 2). The Framingham-only group was less likely than the Cardiovascular Health Study-only group to reside in Forsyth County, North Carolina, or have a discharge diagnosis code for heart failure. Compared with the other two groups, the Framingham-only group was less likely to have a history of myocardial infarction or treated diabetes and was somewhat more likely to have a low body mass index, high percent predicted FEV 1, or prevalent clinical lung disease. Of the original 875 participants in the three diagnostic classification groups, 435 died during a median of 3.7 years (range: 1 day to 10.6 years) of observation following hospital discharge. No significant global difference in survival was seen between the diagnostic classifications (log-rank test: p = 0.36) (figure 2). Compared with those for the concordant group, the all-cause mortality rates in the Framingham-only group (hazard ratio (HR) = 0.87, 95 percent confidence interval (CI): 0.71, 1.07) and the Cardiovascular Health Study-only group (HR = 0.89, 95 percent CI: 0.68, 1.15) were slightly but not significantly lower (table 3). The Framingham-only and Cardiovascular Health Study-only groups had similar survival patterns (HR = 1.02, 95 percent CI: 0.76, 1.35). The survival patterns for the entire Framingham (Framingham only plus concordant) and Cardiovascular Health Study (Cardiovascular Health Study only plus concordant) groups were similar, and the confidence interval excluded a moderate difference in mortality between these groups (HR = 1.02, 95 percent CI: 0.94, 1.11). Fewer than 45 percent of those in the Framingham-only, Cardiovascular Health Study-only, and concordant groups survived for at least 5 years. The survival pattern for 875 age- and sex-matched persons without CHF was significantly better than that for each of the CHF diagnostic classifications (p < 0.001). A higher proportion of the deaths in the Framingham-only group compared with the other two groups were adjudicated to be noncardiovascular related (Framingham only: 57 percent, Cardiovascular Health Study only: 46 percent, concordant: 40 percent). The survival differences among the three main diagnostic classifications were not materially affected by adjustment for the patient characteristics and clinical conditions listed in table 2. In subgroup analyses, Cox models comparing the Framingham-only and Cardiovascular Health Study-only groups with the concordant group suggested that survival was similar across diagnostic classifications or was slightly worse for the concordant group (figure 3). In addition, restricting the analysis of the three main diagnostic classification groups to those with a discharge diagnosis of heart failure did not change the mortality differences substantially (Framingham only vs. concordant: HR = 0.94, 95 percent CI: 0.76, 1.17; Cardiovascular Health Study only vs. concordant: HR = 0.93, 95 percent CI: 0.71, 1.22). In additional analyses, the Framingham CHF definition was modified to require an International Classification of Diseases, Ninth Revision, code for heart failure to emulate the strategy used in several recent epidemiologic studies (15 18). This approach slightly changed the sizes of the groups (Framingham only: n = 216, Cardiovascular Health Study only: n = 156, concordant: n = 447), but the hazard ratios remained consistent with those from the main analysis (Framingham only vs. concordant: HR = 0.94, 95 percent CI: 0.76, 1.17; Cardiovascular Health Study only vs. concordant: HR = 0.87, 95 percent CI: 0.68, 1.12). When the Cardiovascular Health Study definition was modified to include probable cases of CHF, the size of the Framinghamonly and Cardiovascular Health Study-only groups became more similar (Framingham only: n = 238, Cardiovascular Health Study only: n = 165, concordant: n = 495), and survival among the diagnostic classification groups became more similar (Framingham only vs. concordant: HR = 0.91, 95 percent CI: 0.74, 1.13; Cardiovascular Health Study only vs. concordant: HR = 0.94, 95 percent CI: 0.75, 1.18). Sensitivity analyses restricting the cohort to those participants for whom Framingham criteria information was complete (n = 198) or nearly complete (no more than one criterion missing: n = 272; no more than two criteria missing: n = 358) revealed patterns of survival among the diagnostic classifications similar to those seen in the entire cohort. Multiple imputation analyses also were used to look at the sensitivity of the findings to assumptions about missing Framingham criteria data. In each of 10 imputations, the Framingham-only group and the Cardiovascular Health Study-only group had a slightly, but not significantly better survival experience than the concordant group did. Hazard ratios combining the imputation estimates (Framingham only vs. concordant: HR = 0.84, 95 percent CI: 0.68, 1.01; Cardiovascular Health Study only vs. concordant: HR = 0.88, 95 percent CI: 0.65, 1.12) were similar to those from the main analysis. DISCUSSION In this investigation, the Framingham and Cardiovascular Health Study methods of classification produced substan-

632 Schellenbaum et al. TABLE 2. Patient characteristics by CHF* diagnostic classification among Cardiovascular Health Study participants in the inception cohort, United States, 1989 2000 Concordant (n = 472) Framingham Heart Study only (n = 272) Cardiovascular Health Study only (n = 131) Age in years at event (mean (SD*)) 80.2 (6.1) 80.3 (6.0) 80.9 (6.5) Female sex 51 52 50 Caucasian ethnicity 82 88 88 Clinic county and state Forsyth, North Carolina 22 15 35 Sacramento, California 27 29 17 Washington, Maryland 24 29 27 Allegheny, Pennsylvania 27 28 21 ICD-9* code for heart failure 95 82 89 Prevalent clinical disease Myocardial infarction 23 16 23 Angina 34 32 35 Atrial fibrillation 21 25 21 Valve problems 17 18 15 Stroke 15 10 13 Transient ischemic attack 8 8 5 Asthma 10 13 8 Emphysema 9 10 8 COPD* 11 13 6 Kidney disease 11 7 8 Treated diabetes mellitus 22 14 15 Treated hypertension 59 57 50 Body mass index in kg/m 2 (mean (SD)) 27.4 (5.1) 26.1 (4.5) 26.8 (4.5) Percent predicted FEV 1 * (mean (SD)) 79.4 (24.0) 83.2 (23.9) 78.0 (17.9) Smoking status Never 41 44 56 Past 48 45 36 Current 10 11 8 No. of alcoholic drinks per week None 65 60 68 <7 27 33 24 7 <14 5 4 5 14 3 3 4 * CHF, congestive heart failure; SD, standard deviation; ICD-9, International Classification of Diseases, Ninth Revision; COPD, chronic obstructive pulmonary disease; FEV 1, forced expiratory volume in 1 second. Unless otherwise noted, all values are expressed as percentages. COPD prevalence was assessed during the qualifying hospitalization; other clinical conditions, body mass index, smoking status, and alcohol consumption were assessed at the annual visit immediately before the qualifying CHF event; and percent predicted FEV 1 was calculated from the most recent FEV 1 measurement before the qualifying hospitalization, one of three visits at which FEV 1 was measured. tially different incidence estimates: use of the Framingham criteria resulted in an incidence estimate approximately 23 percent greater compared with the Cardiovascular Health Study criteria. The four Cardiovascular Health Study sites differed with respect to the proportions in the Cardiovascular Health Study-only and Framingham-only groups, possibly reflecting differences in populations, documentation patterns, or medical practice or missing data on the

Survival Associated with CHF Diagnostic Criteria 633 FIGURE 2. Kaplan-Meier survival curves by congestive heart failure (CHF) diagnostic classification, the Cardiovascular Health Study (CHS), United States, 1989 2000. Survival estimates for the CHF inception cohort diagnostic classifications, and a group of age- and sex-matched CHS participants without CHF, are compared from the date of hospital discharge (participants without CHF entered on the date that their matched CHF case was discharged from the hospital). Survival curves were truncated at 6 years because of the small sample size beyond this point. Framingham criteria components. Even though the diagnostic criteria classified different persons as having CHF, each diagnostic classification was associated with a similarly poor prognosis in this group of older adults. For each of the CHF diagnostic classifications, mortality rates were higher than those for age- and sex-matched participants without CHF. Several studies of CHF have used modified versions of the Framingham criteria similar to the methods used in the present analysis (15 18). Such studies excluded up to 18 percent of physician-identified cases (15) because of a lack of sufficient, documented Framingham diagnostic criteria. In the present analyses, 22 percent of cases classified as CHF by the Cardiovascular Health Study events committee did not fulfill the Framingham criteria, and 37 percent of cases classified as CHF by the Framingham criteria did not meet the Cardiovascular Health Study definition. Survival in these elderly, community-dwelling adults with CHF was poor; fewer than 45 percent survived for 5 years, a finding congruent with other reports of older adults (6 9). The mortality rate observed in the present study was higher than the rate reported among prevalent CHF cases in the Cardiovascular Health Study (9), probably because of differences in mean age, inclusion criteria, and survival bias. Among those classified as having CHF according to the Framingham criteria alone, the signs and symptoms that make up the Framingham criteria may sometimes have been due to noncardiac conditions. Alternatively, the Framingham-only group may include persons with heart failure who did not meet the Cardiovascular Health Study criteria, perhaps because their condition was ambiguous, atypical, or accompanied by comorbidities that complicated the adjudication for CHF. This possibility is supported by findings that participants who fulfilled the Framingham criteria but not the Cardiovascular Health Study criteria had more pulmonary disease and obesity, had a lower prevalence of diabetes mellitus and myocardial infarction, included a lower proportion with a hospital discharge diagnosis of heart failure, and included a lower proportion of cardiovascular deaths compared with the Cardiovascular Health Study-only TABLE 3. Survival estimates by CHF* diagnostic classification: Kaplan-Meier and Cox proportional hazards model estimates, the Cardiovascular Health Study, United States, 1989 2000 * CHF, congestive heart failure; CI, confidence interval. Deaths (no.) Person-years (no.) Mortality rate/year 5-year survival (%) Hazard ratio 95% CI* p value Concordant 272 1,345 0.20 38 1.00 Framingham Heart 140 794 0.18 44 0.87 0.71, 1.07 0.19 Study only Cardiovascular Health Study only 71 397 0.18 40 0.89 0.68, 1.15 0.36

634 Schellenbaum et al. FIGURE 3. Subgroup hazard ratios and 95% confidence intervals for Framingham-only and Cardiovascular Health Study (CHS)-only congestive heart failure cases compared with the concordant group, the CHS, United States, 1989 2000. In each subgroup, survival of the Framingham-only and CHS-only groups was compared with the concordant group (reference) by using Cox proportional hazards models. and concordant groups. These comorbid conditions would themselves affect survival. Since there is no gold standard for defining CHF, it was not possible to determine which criteria correctly classified CHF. Given the greater number of cases identified by using the Framingham criteria, these criteria may have classified participants without CHF as CHF cases more often than the Cardiovascular Health Study method did. Alternatively, the Cardiovascular Health Study events committee may have excluded CHF cases more often than the Framingham criteria did. The relative value of these and other methods for classifying persons with respect to CHF depends, in part, on the goals of a particular study. Explicit diagnostic criteria, such as the Framingham CHF criteria, may be well suited to comparisons across study populations or over time; central adjudication, as used by the Cardiovascular Health Study, offers the advantage of considering all available information and may be more specific, which can be important in studies of etiology. Indeed, as medical practice evolves, as it did during the course of this study, the relative utility of diagnostic criteria may change. The findings of this analysis are strengthened by sensitivity analyses indicating that the survival analysis results are robust to multivariate adjustment, restriction to subgroups, and missing data assumptions: the Framingham criteria were assumed to be missing because they were absent (main analysis) or missing at random (multiple imputation analysis). Moreover, the approach used in the present analysis to assess Framingham CHF criteria is similar to that used by other studies attempting to find CHF cases by using the Framingham criteria as recorded in medical records. Considerable amounts of missing data for the Framingham criteria may have resulted in incomplete identification of cases meeting these criteria. The number of CHF cases that would have fulfilled the Framingham criteria, and the difference in incidence estimates between the two methods, may have been

Survival Associated with CHF Diagnostic Criteria 635 underestimated in this analysis. Furthermore, the present application of the Framingham criteria was not exactly the same as in the Framingham Heart Study. In that study, the Framingham CHF criteria are applied by using central adjudication by a panel of physicians that considers all hospitalizations with a cardiovascular element. To avoid misclassification, Framingham Heart Study investigators adjudicated CHF cases and excluded those for whom the symptoms and signs of CHF may have been due to another disease process. However, this investigation has taken the approach, used in several recent studies (15 18), of using contemporary components of the Framingham criteria without adjudication. Although the Framingham Heart Study investigators have also proposed diastolic heart failure criteria (19), these criteria were not included in the present investigation. Patients with diastolic heart failure were included in these analyses, but diastolic heart failure was not considered a separate condition. The older age of Cardiovascular Health Study participants may have limited this investigation s ability to examine survival differences between the two sets of diagnostic criteria. However, in the general population, approximately 80 percent of CHF patients are at least 65 years of age (20). The generalizability of this study is limited by the participation rate of 57 percent at baseline in the Cardiovascular Health Study, and the results may not be generalizable to ethnicities that were not represented in the Cardiovascular Health Study cohort. Finally, because of data availability issues, we considered only those hospitalized cases of incident CHF that were not procedure related. Diagnosis and prognosis may differ for outpatients, younger populations, and other groups of CHF patients excluded from this study. For the complex syndrome of CHF, no uniform diagnostic definition has emerged. The method chosen to define CHF cases is likely to affect the observed incidence or prevalence of CHF. In contrast, the current study suggests that survival may be similar for these two diagnostic approaches. Whether this finding applies to other classification approaches or populations requires additional investigation. Diagnostic criteria for CHF should be chosen with care during the design phase of a study, and differences between diagnostic criteria should be considered when interpreting the available literature. ACKNOWLEDGMENTS This research was supported by contracts N01-HC-85079-86, Georgetown Echo RC-HL35129, JHU MRI RC- HL15103, HL43201, and 1-T32-HL07902 from the National Heart, Lung, and Blood Institute and by grant AG09556 from the National Institute on Aging. For a full list of Cardiovascular Health Study investigators and institutions, refer to the following Internet website: http://www.chs-nhlbi.org. REFERENCES 1. McKee PA, Castelli WP, McNamara PM, et al. The natural history of congestive heart failure: the Framingham study. N Engl J Med 1971;285:1441 6. 2. Remes J, Miettinen H, Reunanen A, et al. Validity of clinical diagnosis of heart failure in primary health care. Eur Heart J 1991;12: 315 21. 3. Gottdiener JS, Arnold AM, Aurigemma GP, et al. Predictors of congestive heart failure in the elderly: the Cardiovascular Health Study. J Am Coll Cardiol 2000;35:1628 37. 4. Ives DG, Fitzpatrick AL, Bild DE, et al. Surveillance and ascertainment of cardiovascular events. The Cardiovascular Health Study. Ann Epidemiol 1995;5:278 85. 5. Kitzman DW, Gardin JM, Gottdiener JS, et al. Importance of heart failure with preserved systolic function in patients 65 years of age. Cardiovascular Health Study Research Group. Cardiovascular Health Study. Am J Cardiol 2001;87:413 19. 6. Ho KK, Anderson KM, Kannel WB, et al. Survival after the onset of congestive heart failure in Framingham Heart Study subjects. Circulation 1993;88:107 15. 7. Croft JB, Giles WH, Pollard RA, et al. Heart failure survival among older adults in the United States: a poor prognosis for an emerging epidemic in the Medicare population. Arch Intern Med 1999;159:505 10. 8. Levy D, Kenchaiah S, Larson MG, et al. Long-term trends in the incidence of and survival with heart failure. N Engl J Med 2002; 347:1397 402. 9. Gottdiener JS, McClelland RL, Marshall R, et al. Outcome of congestive heart failure in elderly persons: influence of left ventricular systolic function. The Cardiovascular Health Study. Ann Intern Med 2002;137:631 9. 10. Fried LP, Borhani NO, Enright P, et al. The Cardiovascular Health Study: design and rationale. Ann Epidemiol 1991;1: 263 76. 11. Psaty BM, Kuller LH, Bild D, et al. Methods of assessing prevalent cardiovascular disease in the Cardiovascular Health Study. Ann Epidemiol 1995;5:270 7. 12. Margolis JR, Gillum RF, Feinleib M, et al. Community surveillance for coronary heart disease: the Framingham Cardiovascular Disease Survey. Methods and preliminary results. Am J Epidemiol 1974;100:425 36. 13. Enright PL, Kronmal RA, Higgins M, et al. Spirometry reference values for women and men 65 to 85 years of age. Cardiovascular Health Study. Am Rev Respir Dis 1993;147:125 33. 14. Schafer JL. Imputation of missing covariates under a general linear mixed model. Technical report. University Park, PA: Department of Statistics, Pennsylvania State University, 1997. (http:// www.stat.psu.edu/reports/techrep1997.html). 15. Taffet GE, Teasdale TA, Bleyer AJ, et al. Survival of elderly men with congestive heart failure. Age Ageing 1992;21:49 55. 16. Senni M, Tribouilloy CM, Rodeheffer RJ, et al. Congestive heart failure in the community: trends in incidence and survival in a 10- year period. Arch Intern Med 1999;159:29 34. 17. Diller PM, Smucker DR, David B, et al. Congestive heart failure due to diastolic or systolic dysfunction. Frequency and patient characteristics in an ambulatory setting. Arch Fam Med 1999;8: 414 20. 18. Redfield MM, Jacobsen SJ, Burnett JC Jr, et al. Burden of systolic and diastolic ventricular dysfunction in the community: appreciating the scope of the heart failure epidemic. JAMA 2003;289:194 202. 19. Vasan RS, Levy D. Defining diastolic heart failure: a call for standardized diagnostic criteria. Circulation 2000;101:2118 21. 20. Masoudi FA, Havranek EP, Krumholz HM. The burden of chronic congestive heart failure in older persons: magnitude and implications for policy and research. Heart Fail Rev 2002;7:9 16.