One of the major differences between middle-aged and older populations is in increased prevalence in the older population of poorer health status, which is often the result of co-occurring diseases or comorbidity [12]. Older frail persons with a high burden of disease and resultant acquired low cholesterol levels are more likely to have decreased survival than are those with little or no disease and chronically low cholesterol levels [9, 13, 14]. Given these differences, we hypothesized that in an elderly population, the usual adjustment for the traditional risk factors for coronary heart disease does not account for the possible biological changes associated with frailty, which can affect the relation between cholesterol levels and death from coronary heart disease. Adjustment for factors that reflect frailty [15, 16] and comorbidity should therefore allow us to demonstrate a direct and significant relation (similar to that seen in middle-aged adults) between lipid levels and coronary heart disease in older persons, if such a relation does exist.

We report on the association between total cholesterol level and death from coronary heart disease using prospective data from the EPESE study, a community-based cohort study of men and women 71 years of age and older. In these analyses, clinical and biological markers are used as indicators of frailty and to clarify the role of total cholesterol levels as a risk factor for death from coronary heart disease in older persons.

Methods

Top Methods Results Discussion Author & Article Info References

The data presented here are from the EPESE study, a population-based prospective study of persons who were 65 years of age and older at inception. The study design and the characteristics of the study sample are described in detail elsewhere [15, 17]. The study participants were recruited from three communities-East Boston, Massachusetts; Iowa and Washington counties, Iowa; and New Haven, Connecticut-between 1981 and 1982. In East Boston and the Iowa counties, the study samples represented the total populations 65 years of age and older; in New Haven, the sample was derived from a stratified (by housing type and sex) random sample of residents. Between 80% and 84% of eligible persons in the three communities participated. Participants were followed from 1981 to 1988 with annual in-person and telephone interviews. During the in-person interviews (which were done at baseline and at the third and sixth follow-up points), participants were asked about their health habits, functional status, chronic conditions, and hospitalization history and had their blood pressure measured at home.

At the in-person interview done at the sixth follow-up point (in 1988), which served as the baseline for this study, the surviving participants (6566 persons 71 years of age) were asked to provide a blood specimen for biochemical analyses. Sixty-five percent of participants agreed to have blood drawn. Of the 4271 persons who provided specimens, 4066 had specimens in which total cholesterol, HDL cholesterol, albumin, and iron levels could be measured. Blood was collected at each participant's home while participants were in a nonfasting state. Separated serum samples were refrigerated and sent overnight, by air carrier, to a commercial laboratory (Nichols Institute, San Juan Capistrano, California).

Total cholesterol levels were measured using the standard enzymatic method with a sequential autoanalyzer (SMAC, Technicon, Tarrytown, New York). After precipitation with dextran sulfate, HDL cholesterol level was determined by subtraction on the supernatant of the serum with the standard enzymatic method. Participants were classified into four categories according to total cholesterol level: 4.15 mmol/L or less ( 160 mg/dL) (hypocholesterolemia) [18], 4.16 to 5.19 mmol/L (161 to 199 mg/dL), 5.20 to 6.19 mmol/L (200 to 239 mg/dL) (borderline high cholesterol), and 6.20 mmol/L or higher ( 240 mg/dL) (high cholesterol). They were also classified into three categories according to HDL cholesterol level: less than 0.90 mmol/L (<35 mg/dL), 0.90 to 1.54 mmol/L (35 to 59 mg/dL), and 1.55 mmol/L or higher ( 60 mg/dL). Except for the definition of hypocholesterolemia, all cutpoints were chosen on the basis of National Cholesterol Education Program guidelines [19].

Albumin levels were measured using dye-binding bromocresol green [15]. Serum iron levels were measured with the standard colorimetric method using ferrozine as chromogen [20]. Serum albumin and iron levels were included in the analyses as continuous variables. The presence of chronic conditions was assessed by asking participants whether they had ever been told by a physician that they had high blood pressure or diabetes or had had heart attack or stroke. Blood pressure, measured in participants' homes using the Hypertension Detection and Follow-up Program protocol [21], was the mean of two readings obtained while the participant was seated. Participants were asked about alcohol consumption during the previous year. Smoking status was assessed from self-report of current and previous smoking history.

Participants were followed for death from the time of the 1988 assessment to the end of 1992. Information on vital status was obtained from a seventh follow-up contact (in two sites), contact with proxies, obituaries, and the National Death Index. Death certificates were reviewed and coded by a single nosologist using the International Classification of Diseases, Ninth Revision (ICD-9). Death was defined as attributable to coronary heart disease if ICD-9 codes 410 through 414 were cited as an underlying cause of death. Participants who did not die were censored at the end of 1992.

Crude event rates were obtained by dividing the number of events by the accumulated number of person-years. Summary estimates of the relative risk (RR) for death and death from coronary heart disease, adjusted for potential confounders, were computed from proportional-hazard regression models stratified by community by using the STRATA option of the SAS PHREG procedure [22]. Analyses were further stratified by duration of follow-up when the proportionality of hazard assumptions did not hold for the entire follow-up period. In separate proportional hazard analyses, tests for linear trend were done by entering total cholesterol and HDL cholesterol categories as ordinal variables.

Results

Top Methods Results Discussion Author & Article Info References

The mean age of the participants at baseline was 79.2 years; 5.2% of the participants were black. A total of 42.5% reported having less than 9 years of education, 42.2% reported having 9 to 12 years of education, and 15.3% reported having more than 12 years of education. The percent distribution of total cholesterol categories is shown in Table 1; 9.5% of participants had total cholesterol levels of 4.15 mmol/L or less ( 160 mg/dL). Sixteen percent had HDL cholesterol levels less than 0.90 mmol/L (<35 mg/dL), 60.6% had HDL cholesterol levels between 0.90 and 1.54 mmol/L (35 to 59 mg/dL), and 23% had HDL cholesterol levels of 1.55 mmol/L or more ( 60 mg/dL). The distribution of demographic, clinical, and biochemical characteristics known to be related to coronary heart disease and all-cause mortality differed markedly across total cholesterol categories (Table 1). Low total cholesterol levels were associated, in general, with a poor cardiovascular risk profile. Additionally, with increasing levels of total cholesterol, a consistent decrease was seen in the prevalence of several indicators of poor health, such as low serum iron and albumin levels (Table 1). Only 25 women were receiving hormone replacement therapy; 46 of the participants (1%) were using lipid-lowering drugs (23 of the 46 were in the highest total cholesterol category and only 1 was in the lowest). Because of these low prevalence rates, the use of hormone replacement therapy and the use of lipid-lowering drugs were not investigated as potential confounders or as correlates of total cholesterol levels.

A total of 252 deaths from coronary heart disease were observed during follow-up; 44 of them occurred during the first year. Crude coronary heart disease mortality rates were highest in the lowest total cholesterol category. Table 2 shows a series of multivariate Cox models, all of which included total and HDL cholesterol levels. Unadjusted relative risks are presented in the first model. In subsequent analyses, age, sex, risk factors for coronary heart disease, serum iron levels, and serum albumin levels were added sequentially to the models (Table 2).

In the first unadjusted model, the linear relation between total cholesterol level and death from coronary heart disease was significantly negative (P for trend = 0.04), and total cholesterol levels of 4.15 mmol/L or less ( 160 mg/dL) were significantly associated with a 63% increase in the risk for death from coronary heart disease (RR, 1.63 [95% CI, 1.09 to 2.44]) compared with the reference group (that is, patients with total cholesterol levels between 4.16 and 5.19 mmol/L [161 and 199 mg/dL]) (Table 2). After adjustment for age, sex, and traditional risk factors for coronary heart disease, the relative risk for hypocholesterolemia became nonsignificant, whereas the risk for death from coronary heart disease was moderately increased in patients with borderline (RR, 1.26 [CI, 0.91 to 1.75]) and elevated (RR, 1.28 [CI, 0.89 to 1.84]) total cholesterol levels compared with the reference group (Table 2). With adjustment for serum iron and albumin levels, elevated total cholesterol level was associated with a statistically significant increase in the risk for death from coronary heart disease (RR, 1.45 [CI, 1.06 to 2.10]).

When we analyzed events occurring before and after the first year of follow-up, different hazards related to cholesterol levels were seen. In adjusted analyses limited to events occurring during the first year, a total cholesterol level of 4.15 mmol/L or less ( 160 mg/dL) was associated with a marked increase in the risk for coronary heart disease (RR, 2.29 [CI, 1.01 to 5.20]), whereas the risk in persons with borderline and elevated total cholesterol levels did not differ significantly from the risk in the reference group (RRs, 1.11 [CI, 0.48 to 2.59] and 1.06 [CI, 0.39 to 2.83], respectively [data not shown]).

When coronary heart disease events during the first year of follow-up were excluded, both borderline and elevated total cholesterol levels were associated with a significant increase in the risk for death from coronary heart disease (RRs, 1.45 [CI, 1.02 to 2.08] and 1.57 [CI, 1.06 to 2.34], respectively) (Table 2, last column), and the linear relation between total cholesterol level and death from coronary heart disease became significantly positive (P for linear trend = 0.005) (Figure 1, bottom right). Figure 1 shows the change in the direction of the relation of total cholesterol level and death from coronary heart disease induced by the stepwise cumulative inclusion of adjusting variables in the models and by the exclusion of the first-year events.

View larger version (23K): [in this window] [in a new window]   Figure 1. Relative risks for death from coronary heart disease (CHD) according to total cholesterol level before and after stepwise inclusion of adjustment variables and exclusion of first-year events. Relative risks are from four separate community-stratified proportional-hazard models. Top left. Model 1 is adjusted for high-density lipoprotein (HDL) cholesterol levels only and includes all events. Top right. Model 2 is adjusted for age, sex, and risk factors for coronary heart disease and includes all events. Bottom left. Model 3 is adjusted for age, sex, risk factors for coronary heart disease, and serum iron and albumin levels and includes all events. Bottom right. Model 4 is adjusted for age, sex, risk factors for coronary heart disease, and serum iron and albumin levels and excludes first-year events. Risk factors for coronary heart disease include HDL cholesterol level; isolated systolic hypertension; diastolic hypertension; alcohol consumption; smoking; and history of high blood pressure, diabetes, heart attack, and stroke.

Low HDL cholesterol levels (<0.90 mmol/L [<35 mg/dL]) were consistently associated with an increased risk for death from coronary heart disease before and after adjustment for traditional risk factors and serologic markers of comorbidity (Table 2). In addition, this linear relation remained consistently positive and significant throughout the stepwise addition of variables to the multivariate models (all P values for trend < 0.05; data not shown).

In fully adjusted analyses, no association was found between total cholesterol levels and death from causes other than coronary heart disease (P for trend > 0.2). In addition, no specific relation was seen between low cholesterol levels and death from causes other than coronary heart disease (RR, 0.9 [CI, 0.82 to 1.34]; data not shown).

During follow-up, 950 deaths from all causes were recorded; 162 of them occurred during the first year. In unadjusted analysis, total cholesterol level was strongly and inversely related to all-cause mortality (P for trend < 0.001). In this analysis, hypocholesterolemia was associated with a 52% increase in risk for death (RR, 1.52 [CI, 1.26 to 1.86]), whereas borderline elevated and elevated total cholesterol levels were associated with a significant decrease in the risk for death (RRs, 0.77 [CI, 0.66 to 0.90] and 0.61 [CI, 0.51 to 0.73], respectively).

After adjustment for age and sex, the strength of the association between total cholesterol level and mortality was attenuated (Table 3), but an inverse relation was still present and significant (P for trend = 0.001). After adjustment for traditional risk factors for coronary heart disease, serum iron and albumin levels, and exclusion of first-year events, the inverse association between total cholesterol level and mortality was no longer present, as shown by the absence of any linear relation (P for trend > 0.2) and by the relative risk estimates for specific categories of total cholesterol level, none of which was significantly different from 1 (Table 3).

Discussion

Top Methods Results Discussion Author & Article Info References

Our results indicate that after adjustment for risk factors and indicators of frailty, elevated total cholesterol levels predict increased risk for death from coronary heart disease in older adults. Our findings also support the hypothesis that in populations that are highly heterogeneous in terms of health, the direct relation of total cholesterol levels to death from coronary heart disease can be observed only after adjustment for markers of frailty. In our study sample, these measures of frailty, known to be associated with increased coronary heart disease-related mortality [15, 16], were consistently associated with low cholesterol levels (Table 1), confirming previous findings of a link between poor health status and acquired low cholesterol levels [8, 13, 14]. After we accounted for these indicators of poor health, a positive and significant association between total cholesterol level and death from coronary heart disease was revealed.

Indicators of health status are usually not included in the set of traditional risk factors for coronary heart disease. The adjustment for traditional risk factors is reproduced in the top right panel of Figure 1; neither an association nor a significant trend between total cholesterol levels and coronary heart disease is present. In general, previous studies in older populations have considered only established risk factors for coronary heart disease as possible confounders of the relation between cholesterol levels and coronary heart disease [4-6]. In retrospect, our findings may explain why most previous studies found inconsistent associations between total cholesterol levels and coronary heart disease events. The possible influence of health-related factors that are unmeasured or uncontrolled for in those studies is well simulated by our inclusion of serum iron and albumin levels and our exclusion of first-year events. With the addition of these markers of frailty, a direct and significant relation between total cholesterol levels and death from coronary heart disease was shown (Figure 1; bottom left and bottom right), indicating that the burden of disease is the most likely explanation for an inconsistent association between cholesterol levels and death from coronary heart disease in older populations.

Several observations support our decision to exclude events that occurred during the first year of follow-up, and this exclusion has proven helpful in clarifying relations in observational epidemiologic studies [9, 23]. First, methodologic considerations supported the stratification of analyses by follow-up time [24]. Second, persons in their last year of life may have had several terminal illnesses for which we could not completely adjust. Third, comorbid conditions in persons approaching death could have precipitated acute changes in cholesterol levels and other behavior-related risk factors (such as smoking, alcohol consumption, body weight, and blood pressure), thus increasing the likelihood of misclassification of a person's lifetime exposure to risk factors for coronary heart disease. Excluding first-year events is therefore a way to control for poor health status and allows for the exclusion of persons in whom a major decline in health status before death may directly alter the risk factor under investigation.

Decreased serum iron and albumin levels are nonspecific indicators of poor health and may reflect the overall effect of such factors as malnutrition, chronic inflammatory diseases, infections, hepatic disease, and renal disease [25, 26]. These conditions are generally associated with both low cholesterol levels [8, 27, 28] and increased risk for adverse outcomes [8, 29] and are thus important additional confounders of the association between total cholesterol levels and mortality in a population with a heavy burden of chronic disease. The rationale for the use of biochemical markers of health, such as serum albumin or iron levels, relies on the pathophysiologic characteristics of these markers. Both albumin and iron are negative acute phase reactants that decrease in the presence of a broad spectrum of conditions, including infections, inflammation, weight loss, undernutrition, cancer, and chronic degenerative diseases [25, 26]. Low serum albumin and iron levels are highly sensitive indicators of the presence of disease but have very low diagnostic specificity as markers for particular diseases.

Serious chronic diseases are relatively rare in middle-aged populations. If present, they usually operate as exclusion criteria for enrollment in clinical trials and may reduce participation in population-based studies that require travel to clinical centers. In older populations, however, these conditions tend to be highly prevalent [12], may be present in preclinical stages [30], and often vary in severity [31]. In addition, their presence may not markedly affect participation or retention in a study because most population studies in the elderly are not targeted to disease-free persons and many are designed to examine and follow participants at their homes and nursing homes [17, 30].

Our study has several limitations. Blood samples were obtained from 65% of participants; persons who refused to give blood samples were older, had more disability, and were more likely to have been hospitalized during the previous year. No differences between the group with blood samples and the group without were noted for sex, years of education, current smoking status, or recent diagnosis of cancer requiring hospitalization [6]. Weaker associations between total cholesterol levels and coronary heart disease might be found if older and less healthy persons, for whom data were not available, had been included. Because cholesterol levels were measured only once, within-individual variability could not be considered. Finally, cause-specific mortality was ascertained using death certificates, which may be a source of misclassification. However, misclassification of death both from coronary heart disease and other causes would probably have weakened the association between high cholesterol levels and death from coronary heart disease, unless differential misclassification according to cholesterol levels occurred. This is unlikely.

The observational evidence from this study supports the hypothesis that in the older population, the burden of multiple concomitant conditions can markedly alter associations between certain risk factors and outcomes that are well established in middle-aged populations. As we hypothesized, the consideration of indicators of poor health status and markers of frailty may help to clarify the role of risk factors for coronary heart disease and various other diseases in older persons. Overall, our findings show that the association between low total cholesterol levels and increased mortality seems to be explained by such factors as chronic disease and frailty, which are especially evident during the last year of life. What is not clear is whether a heavy burden of chronic disease and frailty simply hide the effect of lifelong elevated cholesterol levels when they lower cholesterol levels near the end of life or whether they also have a direct pathophysiologic effect on coronary heart disease. For example, in chronically ill persons, production of procoagulant and proinflammatory molecules could lead to increased risk for coronary heart disease events [32].

Our results provide observational evidence to support the hypothesis that it may be useful to screen healthy elderly patients for elevated cholesterol levels and, if levels are elevated, to consider offering cholesterol-lowering therapy [19]. These findings should also be viewed in light of the recently published American College of Physicians guidelines [33] stating that "Screening is not recommended for men and women 75 years of age and older." Considering the probable role of total cholesterol levels as well as the established role of HDL cholesterol levels [6] in defining the coronary heart disease risk profile in the elderly, exclusion of older persons from cholesterol screening and possible treatment [33] could have a more adverse effect in older than in younger populations, given the increased absolute risk for death from coronary heart disease at older ages [34, 35].

Recent studies demonstrate the efficacy of cholesterol-lowering treatment in the primary prevention of death from coronary heart disease among middle-aged men [10] and in the secondary prevention of death from coronary heart disease among men and women 45 to 70 years of age [36, 37]. Results from randomized, controlled clinical trials [38] targeted to older populations are urgently needed to clarify the risks and benefits of cholesterol screening and lipid-lowering therapy in older men and women.

From the National Institute on Aging, National Institutes of Health, Bethesda, Maryland; the U.S. Food and Drug Administration, Rockville, Maryland; the National Institute for Research and Care of the Elderly (INRCA), Fiesole, Italy; and Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts.

Dr. Salive: Epidemiology Branch, Center for Biologics Evaluation and Research, U.S. Food and U.S. Drug Administration, 140 Rockville Pike, HFM-220, Rockville, MD 20852-1448.

Dr. Ferrucci: Geriatric Department, Hospital "I Fraticini," National Institute for Research and Care of the Elderly (INRCA), Via Dei Bosconi, 10, 50014 Fiesole, Italy.

Dr. Glynn: Division of Preventive Medicine, Department of Medicine, Brigham and Women's Hospital, 900 Commonwealth Avenue East, Boston, MA 02215.