Abstract
One method by which to identify fundamental biological processes that may contribute to age-related disease and disability, instead of disease-specific processes, is to construct endophenotypes comprising linear combinations of physiological measures. Applying factor analyses methods to phenotypic data (2006–2009) on 28 traits representing 5 domains (cognitive, cardiovascular, metabolic, physical, and pulmonary) from 4,472 US and Danish individuals in 574 pedigrees from the Long Life Family Study (United States and Denmark), we constructed endophenotypes and assessed their relationship with mortality. The most dominant endophenotype primarily reflected the physical activity and pulmonary domains, was heritable, was significantly associated with mortality, and attenuated the association of age with mortality by 24.1%. Using data (1997–1998) on 1,794 Health, Aging and Body Composition Study participants from Memphis, Tennessee, and Pittsburgh, Pennsylvania, we obtained strikingly similar endophenotypes and relationships to mortality. We also reproduced the endophenotype constructs, especially the dominant physical activity and pulmonary endophenotype, within demographic subpopulations of these 2 cohorts. Thus, this endophenotype construct may represent an underlying phenotype related to aging. Additional genetic studies of this endophenotype may help identify genetic variants or networks that contribute to the aging process.
Keywords: aging, endophenotypes, longevity, mortality
One of the hypotheses in the field of aging is that fundamental biological processes, in addition to disease-specific processes, may contribute to age-related disease and disability. Identification of environmental factors or networks of genes that influence these fundamental processes of aging could lead to insights or interventions that may promote a long and healthy life for many individuals. One common measure of healthy aging is disease-free survival, usually measured at a specific older age; however, investigators have shown that continuous measures of subclinical disease are more reflective of survival (1). Subclinical measures or biomarkers of disease probably reflect interactions among underlying biological networks; thus, endophenotypes derived from such measures may better characterize fundamental aging processes than any single trait. Recently, Cohen et al. (2) used principal-component analyses on a set of biomarkers and identified an “integrated albuminemia” endophenotype that was replicated across 3 populations and correlated with mortality; thus, it may reflect an underlying physiological process related to aging.
In 2010, Matteini et al. (3) developed 5 endophenotypes comprising linear combinations of 28 traits associated with subclinical disease across 5 health domains using data from 3,224 participants in the Long Life Family Study (LLFS). These endophenotypes were heritable, but their relationship to mortality and validity in other cohorts or subpopulations was not assessed. In the current analysis, we derived endophenotypes for 4,472 participants in the LLFS, assessed their relationship with mortality, and validated these endophenotypes and mortality relationships in the Health, Aging and Body Composition (Health ABC) Study cohort, as well as in demographic subpopulations of both cohorts.
METHODS
The LLFS is a study of a family-based cohort recruited between 2006 and 2009 at 4 study sites across the United States and Denmark. Family ascertainment has been described previously (4). Briefly, sibships were selected on the basis of the exceptionality of a sibship's survival into old age using a family longevity selection score (4). In addition, the offspring and their spouses (as controls) were recruited. For the current study, complete phenotype data were available from 574 families (1,292 probands and siblings, 2,401 offspring, and 779 spouse controls). Prior to recruitment into the study, all participants provided written, informed consent.
Endophenotype development overall and in subpopulations
In the LLFS, we performed factor analyses on phenotypic data for 28 traits from 5 health domains (cognitive, cardiovascular, metabolic, physical, and pulmonary) as described by Matteini et al. (3). Briefly, outliers (4 standard deviations away from the mean) were removed and, to reduce nonnormality, data on 10 traits were transformed by natural logarithm transformation (triglycerides, pulse pressure, creatinine, systolic blood pressure, high-density lipoprotein (HDL) cholesterol, fasting glucose, glycosylated hemoglobin, waist circumference) or square-root transformation (average grip strength and maximum grip strength). Because individuals in the study were related, we obtained an unbiased estimate of the correlation matrix by randomly choosing 1 individual from each family and calculating correlations among variables. This procedure was carried out 1,000 times, and a matrix of average correlations was used to perform factor analysis with principal factor extraction and varimax rotation using the principal function in the R (http://www.R-project.org/) (5) package psych (6). Endophenotype values for 4,472 participants with complete data (including probands, siblings, offspring, and spouse controls) were then calculated for each individual by multiplying the eigenvectors for each factor by their corresponding generation-adjusted, transformed, and standardized trait values. The above process was repeated for each subpopulation.
Heritability and mortality analyses
Heritability, the proportion of phenotypic variance due to additive genetic effects, was estimated for each endophenotype using pedigree-based maximum likelihood methods as implemented in SOLAR (7). Age, sex, and recruitment site were included as covariates in the model. Residual heritability is the proportion of additive genetic variance that remains after the influences of measured covariates have been removed. Associations of endophenotypes with mortality in the proband generation were tested using Cox proportional hazards mixed-effect models with the R (5) package coxme (8), which accounts for the family correlation structure by incorporating a kinship matrix using its varlist option. The proportional hazards assumption was tested in Cox models using standard log t tests. A prediction area under the receiver operating characteristic curve was also calculated for each model to enable a comparison of the mortality predictions of the different models on a common scale (9).
Replication
We used phenotypic data on 1,794 European-American individuals from the Health ABC Study, a longitudinal study of healthy men and women aged 68–80 years at baseline (1997–1998). All participants provided written, informed consent prior to recruitment into the study, and data were available for all traits except those belonging to the cognitive domain. For each trait, outliers (±4 standard deviations away from the mean) were removed, and the same transformations were applied as in the LLFS. The traits were also standardized. Factor analysis was conducted in R (5) using varimax rotation, allowing for 4 factors overall and within subpopulations. Individual endophenotype scores for each factor were calculated by multiplying the eigenvectors by their respective standardized trait values.
We also compared mean differences between the LLFS (proband and offspring generations) and Health ABC cohorts for the quantitative traits in 4 domains using maximum-likelihood–based methods (7) that accounted for the nonindependence of the related individuals in LLFS. Because LLFS participants encompassed a much larger age range than the Health ABC participants, for the comparison of mean values we restricted the age ranges of the LLFS offspring (from 30–88 years to 60–88 years) and proband (from 55–105 years to 55–90 years) cohorts for better overlap with the Health ABC cohort (age range, 68–80 years).
RESULTS
Cohort characteristics
Characteristics of the LLFS and Health ABC participants are summarized in Table 1. The ages of the LLFS participants (including the probands, offspring, and spouses) ranged from 24 years to 105 years overall; mean ages overall and for the proband and offspring generations were 68.7, 88.5, and 60.5 years, respectively. Mean age in the Health ABC cohort was 73.8 years, approximately midway between the proband and offspring generations, with a narrow range (69–80 years). Women comprised 55% of the LLFS cohort and 48% of the Health ABC cohort. Participants were recruited approximately equally from each field center in the LLFS (Boston, Massachusetts: 26%, Denmark: 26%, New York, New York: 21%, Pittsburgh, Pennsylvania: 27%) and Health ABC (Memphis, Tennessee: 52%, Pittsburgh: 48%) cohorts. The specific measures of cognition used in the LLFS were not available for the Health ABC cohort. In general, many of the (unadjusted) mean values in the Health ABC Study were between the means for the LLFS proband and offspring generations, as would be expected for characteristics that have strong correlations with age. However, several of the unadjusted means for Health ABC were lower (e.g., HDL cholesterol) or higher (e.g., gait speed) than those for LLFS. After incorporating covariates into the models and adjusting for multiple tests and relatedness, we determined that the mean values for several traits differed significantly between Health ABC and the age-restricted sample from the LLFS proband and offspring cohorts (see Web Tables 1 and 2, available at http://aje.oxfordjournals.org/). Specifically, compared with Health ABC participants, LLFS probands and offspring had significantly (all P's < 0.003) higher diastolic blood pressure, HDL cholesterol, and body mass index (weight (kg)/height (m)2) but lower pulse pressure, triglycerides, glycosylated hemoglobin, grip strength, and gait speed.
Table 1.
Population Characteristics of Persons With Endophenotype Data in the Long Life Family Study (2006–2009) and the Health, Aging and Body Composition Study (1997–1998)
| Characteristic | Long Life Family Study |
Health ABC Study (n = 1,794) |
||||||
|---|---|---|---|---|---|---|---|---|
| Probands and Siblings (n = 1,292) |
Offspring (n = 2,401) |
Probands/Siblings, Offspring, and Spouse (n = 779) Controls (n = 4,472) |
||||||
| Mean (SD) | % | Mean (SD) | % | Mean (SD) | % | Mean (SD) | % | |
| Age, years | 88.53 (6.47) | 60.50 (8.26) | 68.67 (14.90) | 73.77 (2.86) | ||||
| Female sex | 55 | 58 | 55 | 48 | ||||
| Cognitive Domain | ||||||||
| Animal recall testa | 15.15 (4.92) | 22.16 (5.72) | 20.18 (6.39) | |||||
| Vegetable recall testa | 10.96 (3.94) | 15.27 (4.46) | 13.93 (4.69) | |||||
| Digit span forward testb | 7.78 (2.13) | 8.70 (2.17) | 8.33 (2.20) | |||||
| Digit span backward testb | 5.70 (2.06) | 6.90 (2.34) | 6.46 (2.29) | |||||
| Immediate memory testc | 8.86 (4.27) | 13.31 (3.89) | 12.02 (4.49) | |||||
| Delayed memory testc | 6.77 (4.39) | 11.88 (4.22) | 10.41 (4.87) | |||||
| Cardiovascular Domain | ||||||||
| Hypertensiond (yes/no) | 68 | 42 | 50 | 43 | ||||
| Systolic blood pressure, mm Hg | 139.16 (25.02) | 128.04 (19.49) | 131.49 (21.82) | 133.51 (19.68) | ||||
| Diastolic blood pressure, mm Hg | 73.65 (11.46) | 78.96 (10.61) | 77.50 (11.20) | 69.93 (11.01) | ||||
| Pulse pressure, mm Hg | 65.53 (20.68) | 49.03 (14.26) | 53.98 (17.88) | 63.41 (16.62) | ||||
| Total cholesterol, mg/dL | 188.21 (44.14) | 205.39 (39.59) | 200.39 (41.75) | 201.15 (37.02) | ||||
| HDL cholesterol, mg/dL | 57.09 (16.24) | 60.35 (17.48) | 59.16 (17.06) | 51.64 (15.73) | ||||
| LDL cholesterol, mg/dL | 109.57 (36.53) | 122.21 (34.12) | 118.83 (35.44) | 119.51 (32.85) | ||||
| Triglycerides, mg/dL | 107.18 (53.18) | 109.64 (59.97) | 108.80 (57.69) | 148.98 (76.28) | ||||
| Metabolic Domain | ||||||||
| Diabetese (yes/no) | 9 | 6 | 7 | 11 | ||||
| Body mass indexf | 26.27 (4.16) | 27.52 (4.86) | 27.14 (4.64) | 26.50 (4.06) | ||||
| Creatinine, mg/dL | 1.15 (0.32) | 0.97 (0.20) | 1.03 (0.25) | 1.01 (0.23) | ||||
| Fasting glucose, mg/dL | 95.23 (17.26) | 93.06 (14.81) | 94.02 (15.65) | 98.94 (21.38) | ||||
| Glycosylated hemoglobin, % | 5.75 (0.48) | 5.53 (0.43) | 5.59 (0.46) | 6.09 (0.74) | ||||
| Waist circumference, cm | 94.48 (12.80) | 94.55 (14.27) | 94.62 (13.66) | 98.92 (11.70) | ||||
| Physical Domain | ||||||||
| Average grip strength, kg | 20.26 (7.71) | 32.06 (11.14) | 28.99 (11.67) | 29.32 (9.80) | ||||
| Maximum grip strength, kg | 20.90 (7.85) | 33.00 (11.34) | 29.85 (11.91) | 32.06 (10.37) | ||||
| Gait speed, m/second | 0.78 (0.24) | 1.17 (0.22) | 1.06 (0.29) | 1.25 (0.22) | ||||
| Physical performanceg | 7.56 (3.22) | 11.33 (1.27) | 10.27 (2.64) | 10.35 (1.37) | ||||
| Pulmonary Domain | ||||||||
| FEV1:FEV6, % | 73.93 (7.97) | 78.26 (5.91) | 76.97 (6.90) | 76.55 (7.32) | ||||
| FEV1, mL | 1,707.74 (590.09) | 2,743.28 (770.30) | 2,474.15 (860.98) | 2,284.44 (654.11) | ||||
| FEV6, mL | 2,309.73 (755.42) | 3,503.44 (943.12) | 3,201.03 (1,045.02) | 2,984.02 (800.15) | ||||
| Lung diseaseh (yes/no) | 12 | 13 | 13 | 15 | ||||
Abbreviations: FEV1, forced expiratory volume in 1 second; FEV6, forced expiratory volume in 6 seconds; HDL, high-density lipoprotein; Health ABC, Health, Aging, and Body Composition; LDL, low-density lipoprotein; SD, standard deviation.
a Score: Number of items the participant can name in 60 seconds.
b Source: Wechsler Memory Scales—III (36). Range of possible scores, 0–12.
c Source: Wechsler Memory Scales—Revised (37). Range of possible scores, 0–25.
d Hypertension was defined as a self-report of hypertension confirmed by the use of antihypertensive medication or systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg.
e Diabetes was defined as use of diabetes medication or fasting glucose concentration ≥126 mg/dL.
f Weight (kg)/height (m)2.
g Physical performance was the sum of results from chair stands, a set of balance tests, and walking performance on a short-distance walk (38). Units could range from 0 to 12.
h Lung disease was defined as a self-report of a previous diagnosis of chronic bronchitis, emphysema, or chronic obstructive pulmonary disease.
Estimation of endophenotypes and heritability in the LLFS
Eigenvalues and eigenvectors for the first 5 factors for the LLFS (LLFS factors (LFs)) are shown in Table 2. The first factor (LF1) was dominated by the physical activity and pulmonary domains and explained 13.9% of the variation. The second factor (LF2) was dominated by measures from the metabolic and cardiovascular domains and explained 10.7% of the variation. The third factor (LF3) predominantly comprised measures from the cognitive domain and explained 9.3% of the variation. The fourth factor (LF4) strongly reflected blood pressure–related traits (hypertension, systolic and diastolic blood pressure, and pulse pressure) and explained 9.1% of the variation. The fifth factor (LF5) was dominated by cardiovascular measures and explained 7.7% of the variation.
Table 2.
Results From Factor Analyses of Aging-Related Endophenotypes for the Long Life Family Study (2006–2009) and the Health, Aging and Body Composition Study (1997–1998)
| Trait | Factor Loading |
||||||||
|---|---|---|---|---|---|---|---|---|---|
| Long Life Family Study |
Health ABC Study |
||||||||
| LF1 | LF2 | LF3 | LF4 | LF5 | HF1 | HF2 | HF3 | HF4 | |
| Cognitive Domain | |||||||||
| Animal recall test | 0.18 | −0.12 | 0.53a | −0.07 | 0.06 | ||||
| Vegetable recall test | −0.13 | −0.16 | 0.58a | −0.05 | 0.08 | ||||
| Digit span forward test | 0.03 | 0.05 | 0.42a | −0.09 | −0.18 | ||||
| Digit span backward test | 0.04 | 0.06 | 0.51a | −0.07 | −0.10 | ||||
| Immediate memory test | 0.01 | 0.04 | 0.80a | 0.02 | 0.04 | ||||
| Delayed memory test | 0.01 | 0.01 | 0.80a | 0.01 | 0.06 | ||||
| Cardiovascular Domain | |||||||||
| Hypertension (yes/no) | −0.06 | 0.22 | −0.09 | 0.69a | −0.10 | −0.04 | 0.26 | 0.50a | −0.13 |
| Systolic blood pressure | −0.02 | 0.03 | −0.08 | 0.96a | 0.09 | −0.03 | 0.00 | 0.97a | 0.07 |
| Diastolic blood pressure | 0.22 | −0.01 | −0.04 | 0.65a | 0.16 | 0.20 | −0.01 | 0.51a | 0.23 |
| Pulse pressure | −0.18 | 0.04 | −0.07 | 0.79a | 0.00 | −0.17 | 0.00 | 0.79a | −0.07 |
| Total cholesterol | −0.08 | −0.13 | −0.04 | 0.10 | 0.94a | −0.26 | −0.03 | 0.04 | 0.92a |
| HDL cholesterol | −0.26 | −0.61a | 0.09 | 0.06 | 0.12 | −0.43a | −0.48a | −0.02 | 0.13 |
| LDL cholesterol | 0.03 | −0.03 | −0.08 | 0.06 | 0.93a | −0.05 | 0.02 | −0.01 | 0.91a |
| Triglycerides | 0.02 | 0.56a | −0.05 | 0.09 | 0.44a | −0.09 | 0.44a | 0.16 | 0.24 |
| Metabolic Domain | |||||||||
| Diabetes (yes/no) | −0.14 | 0.45a | 0.03 | −0.00 | −0.17 | −0.08 | 0.58a | 0.00 | −0.16 |
| Body mass index | 0.02 | 0.74a | 0.04 | 0.12 | 0.12 | 0.09 | 0.66a | 0.07 | 0.10 |
| Creatinine | 0.31a | 0.27 | −0.22 | −0.06 | −0.02 | 0.49a | 0.24 | 0.07 | −0.05 |
| Fasting glucose | −0.03 | 0.53a | −0.04 | 0.09 | −0.01 | 0.04 | 0.76a | 0.06 | −0.06 |
| Glycosylated hemoglobin | −0.19 | 0.56a | 0.02 | 0.02 | −0.01 | −0.09 | 0.68a | −0.03 | −0.03 |
| Waist circumference | 0.19 | 0.77a | −0.06 | 0.06 | 0.03 | 0.20 | 0.66a | 0.05 | 0.04 |
| Physical Domain | |||||||||
| Average grip strength | 0.88a | 0.16 | −0.05 | 0.01 | −0.08 | 0.86a | 0.19 | 0.01 | −0.12 |
| Maximum grip strength | 0.88a | 0.17 | −0.06 | 0.01 | −0.08 | 0.86a | 0.19 | 0.00 | −0.12 |
| Gait speed | 0.45a | −0.25 | 0.28 | 0.02 | 0.08 | 0.45a | −0.25 | −0.01 | 0.03 |
| Physical performance | 0.42a | −0.21 | 0.27 | 0.07 | 0.14 | 0.39a | −0.22 | −0.01 | 0.01 |
| Pulmonary Domain | |||||||||
| Lung disease (yes/no) | −0.14 | 0.05 | −0.03 | −0.03 | −0.06 | −0.19 | 0.11 | −0.01 | −0.15 |
| FEV1 | 0.86a | 0.03 | 0.04 | −0.10 | 0.00 | 0.84a | 0.03 | −0.08 | 0.15 |
| FEV6 | 0.88a | 0.00 | 0.02 | −0.10 | −0.03 | 0.87a | 0.03 | −0.09 | 0.08 |
| FEV1:FEV6 ratio | 0.07 | 0.10 | 0.09 | −0.05 | 0.13 | 0.08 | 0.00 | 0.01 | 0.24 |
| Eigenvalue | 4.01 | 3.48 | 2.54 | 2.31 | 1.86 | 4.25 | 2.91 | 2.07 | 1.85 |
| % of variance explained | 13.9 | 10.7 | 9.3 | 9.1 | 7.7 | 18.0 | 13.6 | 9.6 | 9.1 |
Abbreviations: FEV1, forced expiratory volume in 1 second; FEV6, forced expiratory volume in 6 seconds; HDL, high-density lipoprotein; Health ABC, Health, Aging and Body Composition; HF, Health ABC factor; LDL, low-density lipoprotein; LF, Long Life Family Study factor.
a Loading > |0.30|.
Age, sex, and recruitment center explained 53% of the total variation in LF1, which was dominated by the physical and pulmonary domains (Web Table 3). These covariates accounted for less of the total variation in the other factors, ranging from 7% to 16%. After accounting for variation attributable to covariates, genetic factors accounted for a moderate proportion of the residual phenotypic variation of each endophenotype; residual heritability (h2r) for factors 1–5 was 0.51, 0.39, 0.38, 0.21, and 0.23, respectively (Web Table 3).
Endophenotypes in the Health ABC cohort
Eigenvector and eigenvalues of the first 4 endophenotypes for the Health ABC Study are presented in Table 2. The first Health ABC factor (HF1) was dominated by physical and pulmonary function measures, similar to the LLFS results. In addition, the second factor in Health ABC (HF2) was heavily weighted by measures in the cardiovascular and metabolic domains. The cognitive domain measures were not available in Health ABC, so we did not observe the equivalent of LF3. However, factors 3 and 4 in Health ABC (HF3 and HF4) were dominated by blood pressure and lipid measures, respectively, and these results were similar to those observed for LF4 and LF5.
Because the Health ABC Study did not have data on the same cognitive measures as the LLFS, we also derived endophenotypes in LLFS after excluding the cognitive domain variables. Again, the magnitude and direction of the eigenvectors corresponding to the “no cognitive domain” endophenotypes (Web Table 4) were very similar to those of the endophenotypes LF1, LF2, LF4, and LF5 derived over all domains, as well as those from the Health ABC cohorts (HF1, HF2, HF3, and HF4).
Endophenotypes and mortality in the LLFS
We next assessed whether the 5 LLFS endophenotypes were associated with mortality (as of 2014) in the proband generation. Of the 1,292 probands and siblings, 558 (43%) were deceased after a mean follow-up time of 4.0 (standard deviation, 1.6) years. We performed Cox regression analyses to assess whether factors 1–5 improved prediction of mortality as compared with baseline age and sex (Table 3); all models met the proportional hazards assumption. As expected, baseline age and sex were statistically significant predictors of mortality; increasing age was associated with increased mortality (P < 10−10), whereas female sex was associated with lowered mortality (P < 10−4). The area under the (prediction) curve was 0.736, indicating that these 2 variables were able to discriminate between participants who died and those who did not after 4.0 years of follow-up. The addition of LF1 (“physical and pulmonary function”) increased the level of discrimination from 0.736 to 0.749. More interestingly, LF1 was significantly associated with lowered mortality (hazard ratio (HR) = 0.874; P < 10−6) (10, 11), and it attenuated 24.1% of the association with age. Similarly, LF3 (“cognitive”) was significantly associated with decreased mortality (HR = 0.904; P < 10−6), and it attenuated 18.9% of the association of age with mortality. Endophenotypes LF4 (“blood pressure”) and LF5 (“lipids”) also were significantly associated with decreased mortality (P < 10−4 and P < 10−6, respectively) but did not attenuate the association with age. Finally, LF2 (“metabolic”) was not significant and did not contribute to the discriminatory ability of the model. These results indicated that LF1 and LF3 were significant predictors of mortality and that they attenuated the association with age.
Table 3.
Results From Cox Regression Mortality Models of Proband-Generation Endophenotypes Including Baseline Age and Sex, Long Life Family Study, 2006–2014
| Model and Variable | HR | 95% CI | Attenuation of Age HR, % | AUC |
|---|---|---|---|---|
| Model 1 | ||||
| Age | 1.123 | 1.104, 1.142 | Referent | 0.736 |
| Sex | 0.685 | 0.571, 0.822 | ||
| Model 2 | ||||
| Age | 1.093 | 1.072, 1.115 | 24.1 | 0.749 |
| Sex | 0.377 | 0.291, 0.488 | ||
| LF1 | 0.874 | 0.839, 0.911 | ||
| Model 3 | ||||
| Age | 1.123 | 1.103, 1.142 | 0.4 | 0.737 |
| Sex | 0.702 | 0.581, 0.847 | ||
| LF2 | 1.018 | 0.981, 1.056 | ||
| Model 4 | ||||
| Age | 1.100 | 1.081, 1.119 | 18.9 | 0.753 |
| Sex | 0.682 | 0.573, 0.813 | ||
| LF3 | 0.904 | 0.870, 0.938 | ||
| Model 5 | ||||
| Age | 1.124 | 1.105, 1.144 | −0.9 | 0.742 |
| Sex | 0.732 | 0.606, 0.884 | ||
| LF4 | 0.935 | 0.900, 0.971 | ||
| Model 6 | ||||
| Age | 1.118 | 1.100, 1.138 | 3.7 | 0.743 |
| Sex | 0.821 | 0.680, 0.990 | ||
| LF5 | 0.882 | 0.844, 0.921 | ||
| Model 7 | ||||
| Age | 1.079 | 1.059, 1.100 | 35.5 | 0.769 |
| Sex | 0.538 | 0.413, 0.700 | ||
| LF1 | 0.909 | 0.873, 0.946 | ||
| LF2 | 1.001 | 0.964, 1.038 | ||
| LF3 | 0.917 | 0.882, 0.954 | ||
| LF4 | 0.962 | 0.926, 1.000 | ||
| LF5 | 0.904 | 0.864, 0.945 |
Abbreviations: AUC, area under the curve; CI, confidence interval; HR, hazard ratio; LF, Long Life Family Study factor.
Endophenotypes and mortality in the Health ABC Study
Of the 1,794 individuals in the Health ABC cohort (as of 2014), 987 (55%) were deceased after a mean follow-up time of 11.9 (standard deviation, 4.4) years. All models met the proportional hazards assumption except, marginally, HF1 (P =0.03), but a sex-specific baseline hazards model did not produce significant results, and the results for age attenuation and area under the curve were almost identical to those from the model reported. Although the discriminatory abilities of the models in the Health ABC cohort were lower (area under the curve =0.649–0.692), most likely because of the longer follow-up time in the Health ABC cohort, the results (Table 4) were similar to those obtained in the LLFS. The HF1 endophenotype was significantly associated with decreased mortality (HR = 0.906; P < 10−10) and attenuated the association with age by 18.7%. HF2 (“metabolic”) was significantly associated with increased mortality (P < 10−3). In contrast, HF3 (“blood pressure”) was not associated with mortality, whereas HF4 (“lipids”) was significantly associated with decreased mortality (HR = 0.923; P < 10−5). However, none of the latter 3 endophenotypes attenuated a large component of the age association.
Table 4.
Results From Cox Regression Mortality Models of Endophenotypes Including Baseline Age and Sex, Health, Aging and Body Composition Study, 1997–2014
| Model and Variable | HR | 95% CI | Attenuation of Age HR, % | AUC |
|---|---|---|---|---|
| Model 1 | ||||
| Age | 1.120 | 1.095, 1.145 | Referent | 0.649 |
| Sex | 0.689 | 0.604, 0.784 | ||
| Model 2 | ||||
| Age | 1.097 | 1.072, 1.123 | 18.7 | 0.679 |
| Sex | 0.376 | 0.304, 0.467 | ||
| HF1 | 0.906 | 0.881, 0.932 | ||
| Model 3 | ||||
| Age | 1.121 | 1.096, 1.147 | −1.2 | 0.657 |
| Sex | 0.755 | 0.657, 0.868 | ||
| HF2 | 1.048 | 1.022, 1.075 | ||
| Model 4 | ||||
| Age | 1.119 | 1.094, 1.144 | 1.0 | 0.651 |
| Sex | 0.688 | 0.604, 0.783 | ||
| HF3 | 1.016 | 0.984, 1.049 | ||
| Model 5 | ||||
| Age | 1.117 | 1.093, 1.143 | 2.0 | 0.654 |
| Sex | 0.736 | 0.644, 0.841 | ||
| HF4 | 0.923 | 0.891, 0.955 | ||
| Model 6 | ||||
| Age | 1.095 | 1.070, 1.120 | 20.8 | 0.692 |
| Sex | 0.428 | 0.345, 0.532 | ||
| HF1 | 0.896 | 0.871, 0.922 | ||
| HF2 | 1.056 | 1.030, 1.083 | ||
| HF3 | 1.000 | 0.967, 1.034 | ||
| HF4 | 0.910 | 0.879, 0.943 |
Abbreviations: AUC, area under the curve; CI, confidence interval; HF, Health, Aging and Body Composition Study factor; HR, hazard ratio.
Endophenotypes in demographic subgoups of LLFS and Health ABC
For each of the subgroups, we re-derived endophenotypes and performed mortality analyses. The loadings of the re-derived endophenotypes and their associations with mortality for the subpopulations were strikingly similar to those obtained using data on the entire cohort, but there were a few differences (Table 5 and Web Table 5; results for endophenotypes 3–5 were similar but are not shown). For example, traits related to HDL cholesterol, creatinine, and waist circumference contributed minimally to factor 1 in LLFS males and females in both cohorts but had stronger contributions over the entire cohort and for other subgroups. The moderate association (loading = −0.30) between cholesterol traits and HF1 in Health ABC males was probably due to a small sample size; an analysis of males across both cohorts revealed loadings similar to those for LLFS males. Furthermore, although factor 1 explained the largest proportion of variance in each subgroup, the proportion of variance explained by the other endophenotypes varied. For example, in probands, LF3 (“cognitive”) accounted for a larger proportion of the total variance than did LF2 (“metabolic”), whereas in offspring, LF3 (“cognitive”) accounted for a smaller proportion of the total variance than did LF4 (“blood pressure”). These latter results are consistent with our current expectations of the relative impact of different domains of health at different ages.
Table 5.
Factor Loadings for the First Factor (Factor 1) Obtained From Subgroup Analyses of the Long Life Family Study (2006–2009) and the Health, Aging and Body Composition Study (1997–1998)
| Trait | Factor Loading |
||||||||
|---|---|---|---|---|---|---|---|---|---|
| Long Life Family Study |
Health ABC Study |
||||||||
| Subgroup and Age |
Sex |
US Only (Excluding Denmark) (n = 2,735) |
Site |
Sex |
|||||
| Probands (Age ≥76 Years) (n = 1,093) |
Offspring (Age <76 Years) (n = 2,299) |
Male (n = 1,566) | Female (n = 1,952) | Pittsburgh, Pennsylvania (n = 859) |
Memphis, Tennessee (n = 935) |
Male (n = 939) | Female (n = 855) | ||
| Cognitive Domain | |||||||||
| Animal recall test | 0.26 | 0.09 | 0.14 | 0.31a | 0.16 | ||||
| Vegetable recall test | −0.01 | −0.20 | 0.12 | 0.21 | −0.13 | ||||
| Digit span forward test | 0.05 | 0.07 | −0.04 | −0.10 | 0.11 | ||||
| Digit span backward test | 0.04 | 0.07 | −0.02 | −0.05 | 0.08 | ||||
| Immediate memory test | −0.05 | 0.00 | 0.04 | 0.08 | −0.03 | ||||
| Delayed memory test | −0.01 | −0.02 | 0.05 | 0.12 | −0.03 | ||||
| Cardiovascular Domain | |||||||||
| Hypertension (yes/no) | −0.16 | 0.01 | −0.11 | −0.14 | −0.06 | −0.03 | −0.06 | 0.07 | −0.00 |
| Systolic blood pressure | 0.00 | −0.03 | 0.06 | −0.01 | −0.03 | −0.08 | −0.02 | 0.04 | −0.09 |
| Diastolic blood pressure | 0.16 | 0.21 | 0.12 | 0.15 | 0.21 | 0.20 | 0.20 | 0.05 | 0.02 |
| Pulse pressure | −0.09 | −0.22 | −0.02 | −0.11 | −0.19 | −0.21 | −0.15 | 0.01 | −0.12 |
| Total cholesterol | −0.13 | −0.08 | 0.04 | 0.05 | −0.12 | −0.22 | −0.23 | −0.30 | −0.12 |
| HDL cholesterol | −0.22 | −0.37a | −0.13 | −0.01 | −0.29 | −0.35a | −0.49a | −0.26 | −0.08 |
| LDL cholesterol | −0.05 | 0.06 | 0.06 | 0.09 | −0.00 | −0.01 | −0.02 | −0.27 | −0.07 |
| Triglycerides | −0.01 | 0.09 | 0.12 | −0.06 | 0.03 | −0.16 | −0.03 | 0.07 | −0.04 |
| Metabolic Domain | |||||||||
| Diabetes (yes/no) | −0.13 | −0.12 | −0.17 | −0.16 | −0.15 | −0.07 | −0.10 | −0.13 | −0.14 |
| Body mass index | 0.11 | 0.02 | 0.02 | 0.02 | 0.04 | 0.05 | 0.12 | 0.06 | 0.07 |
| Creatinine | 0.23 | 0.48a | −0.08 | −0.03 | 0.31a | 0.46a | 0.50a | 0.16 | 0.01 |
| Fasting glucose | −0.09 | 0.07 | −0.10 | −0.06 | −0.05 | −0.00 | 0.07 | −0.12 | −0.14 |
| Glycosylated hemoglobin | −0.10 | −0.18 | −0.19 | −0.19 | −0.17 | −0.15 | −0.03 | −0.19 | −0.23 |
| Waist circumference | 0.19 | 0.25 | −0.00 | −0.03 | 0.21 | 0.27 | 0.12 | −0.01 | 0.14 |
| Physical Domain | |||||||||
| Average grip strength | 0.83a | 0.91a | 0.82a | 0.83a | 0.88a | 0.87a | 0.86a | 0.82 | 0.77 |
| Maximum grip strength | 0.82a | 0.91a | 0.82a | 0.83a | 0.88a | 0.87a | 0.86a | 0.82 | 0.77 |
| Gait speed | 0.56a | 0.32a | 0.56a | 0.52a | 0.43a | 0.45a | 0.44a | 0.37 | 0.38 |
| Physical performance | 0.58a | 0.30 | 0.53a | 0.46a | 0.41a | 0.38a | 0.40a | 0.39 | 0.29 |
| Pulmonary Domain | |||||||||
| Lung disease (yes/no) | −0.18 | −0.10 | −0.18 | −0.13 | −0.13 | −0.14 | −0.23 | −0.10 | −0.22 |
| FEV1 | 0.82a | 0.87a | 0.71a | 0.77a | 0.86a | 0.84a | 0.85a | 0.51 | 0.75 |
| FEV6 | 0.84a | 0.89a | 0.68a | 0.77a | 0.87a | 0.87a | 0.87a | 0.52 | 0.75 |
| FEV1:FEV6 ratio | 0.07 | 0.04 | 0.27 | 0.21 | 0.09 | 0.03 | 0.12 | 0.18 | 0.23 |
| % of variance explained | 13.4 | 14.5 | 11.4 | 12.1 | 13.9 | 18.0 | 18.3 | 11.7 | 12.7 |
Abbreviations: FEV1, forced expiratory volume in 1 second; FEV6, forced expiratory volume in 6 seconds; HDL, high-density lipoprotein; Health ABC, Health, Aging and Body Composition; LDL, low-density lipoprotein; LLFS, Long Life Family Study.
a Loading > |0.30|.
DISCUSSION
In the current study, we re-derived 5 healthy-aging endophenotype constructs using a larger cohort from the LLFS and also obtained similar endophenotype constructs for the Health ABC cohort, as well as within demographic subgroups of these 2 cohorts. Furthermore, we have shown that the most dominant endophenotype for both cohorts was significantly associated with decreased mortality and that it attenuated the association with age by 24.1% and 18.7% in LLFS and Health ABC, respectively.
We performed factor analyses to develop 5 endophenotypes comprising 28 trait measures from 5 health domains (cognitive, cardiovascular, metabolic, physical, and pulmonary) using data from the LLFS. The dominant traits and eigenvectors of the first 5 endophenotypes were strikingly similar to previously reported results based on a subset of these individuals (Web Table 4). The first factor predominantly reflected the physical activity and pulmonary domains, whereas the second factor reflected the cardiovascular and metabolic health domains. LLFS factors LF3, LF4, and LF5 predominantly reflected the cognitive and cardiovascular (blood pressure traits and lipid traits) domains, respectively. When we performed factor analyses using data on the same traits from 4 of the health domains in the Health ABC cohort (data on the cognitive domain traits were not available), both the composition and eigenvectors of the 4 Health ABC endophenotype constructs were very similar to those obtained in LLFS (Table 2). Furthermore, factor analyses performed by generation, sex, and site within each cohort also revealed similar endophenotype constructs (Table 5 and Web Table 5).
The results of the factor analyses on data from the LLFS and Health ABC cohorts indicated that these endophenotypes might be capturing an underlying phenotype. These results were even more intriguing given that the (covariate-adjusted) mean values for many traits used in the development of the endophenotypes differed significantly between the LLFS and Health ABC cohorts (Web Table 1). Similarly, covariate-adjusted means between subpopulations differed. For example, 10 out of 19 trait means differed significantly between the Memphis and Pittsburgh sites in the Health ABC Study (results not shown), but the endophenotype constructs were similar among subpopulations and similar to those from the cohort overall. This reproducibility across cohorts and subpopulations is consistent with a recent study in which Cohen et al. (2) performed principal-component analyses on 43 biomarkers with data available from 3 different cohorts (2 from the United States and 1 from Italy). These investigators showed that their first 2 axes (endophenotypes) were reproducible across cohorts and within subpopulations (e.g., age group, sex) and were associated with mortality. They also proposed that their endophenotypes might be capturing underlying aging phenotypes. Their endophenotype constructs differed from ours because the traits differed (i.e., biomarkers only vs. biomarkers plus physical and functional measures).
We also assessed whether the endophenotypes were predictive of mortality and determined that the endophenotypes reflecting physical and pulmonary function (LF1 and HF1) were significantly associated with mortality in both cohorts. Higher LF1 and HF1 values were independently and significantly associated with lower mortality (hazard ratios < 1.0), and they also attenuated 18%–24% of the association of increased age with mortality in their respective cohorts.
The association between LF1 (and HF1) and mortality is consistent with current knowledge. Pulmonary function and grip strength are both considered indicators of overall health and vitality and hence are strongly associated with each other (12, 13). Both physical function and pulmonary health decline at older ages, possibly because of overall age-related decline in skeletal muscle strength (14). Decline in muscle strength has been shown to be associated with disability, morbidity, and mortality (15–18). Grip strength, which is a surrogate measure for overall muscle strength (19), has been proposed as a predictor of frailty (20) and has been shown to be associated with long-term survival in middle age (21) and in older years (22). Pulmonary function also has been associated with decline in mobility in older individuals (23) and is a risk factor for mortality (24, 25). Thus, this endophenotype may reflect an underlying phenotype associated with exceptional aging, perhaps related to skeletal muscle strength.
The significant association between LF3 (the cognitive domain) and decreased mortality is consistent with other studies of the association between cognitive function and mortality (26–28). Because the same cognitive tests were not conducted in the Health ABC cohort, we could not assess this specific endophenotype. The relationships between mortality and endophenotypes reflecting the metabolic syndrome (LF2), blood pressure (LF4), and lipids (LF5) were less consistent between the 2 cohorts, perhaps because of a more adverse metabolic profile in the Health ABC cohort than in the LLFS cohort (Web Tables 1 and 2). Despite these differences in mean values, which probably reflect the prevalence of lipid-related and metabolic conditions, the correlation structures of the endophenotypes were very similar for the 2 cohorts (Table 2 and Web Table 5). In addition, the relationship of the metabolic endophenotype to mortality was also similar in both cohorts (HR = 1.018 and HR = 1.048 for LLFS and Health ABC, respectively) (Table 3). Nevertheless, the difference in the association of the metabolic endophenotype to mortality may be related to the difference in the age distributions in the 2 cohorts. Mortality risks associated with diabetes, body mass index, and lipid levels have all been found to vary by age. Diabetes-related mortality is lower in persons diagnosed at late ages than in those diagnosed in middle age (29, 30). High levels of total cholesterol and low-density lipoprotein cholesterol are associated with greater cardiovascular and all-cause mortality in middle-aged populations than in the elderly (31), while lower HDL cholesterol levels are a stronger predictor of mortality in the elderly than is total cholesterol (32–34). Low body mass index is associated with increased mortality risk in the elderly, independent of many comorbid conditions (35).
There were multiple limitations of the current study, including differences in age structure, length of follow-up, and selection of subjects in our 2 cohorts, that may have affected results of the mortality analyses. Additionally, a potential bias due to the Danish national health-care system, in which there are no out-of-pocket costs for primary health care or hospitalization, could be that socioeconomically disadvantaged families genetically enriched for longevity had a better chance to survive to older ages than similar US families. However, because the selection of the Danish families was based on their better survival compared with other Danish families and not US families, this is unlikely to have affected the analyses. Furthermore, our analyses of subgroups indicated that no measureable bias occurred for constructing endophenotypes. Finally, replication of these findings in additional cohorts, perhaps including an expanded set of traits and biomarkers, is required to confirm that these endophenotypes represent robust endophenotypes that may facilitate identification of fundamental aging pathways. For example, inclusion of different traits and biomarkers will result in different endophenotypes, as evidenced by a comparison of our study with that of Cohen et al. (2), although both groups identified slightly different endophenotypes that appear to reflect the metabolic syndrome. One way to test this latter hypothesis would be to perform genetic association studies to determine whether similar genes were implicated in both endophenotypes.
In addition to their relationship to mortality, these endophenotypes were heritable; thus, we are currently performing linkage and association analyses of these endophenotypes. These analyses may reveal variants or genes or pathways that influence fundamental processes of aging. These genetic results can be followed up in other cohorts, even if all of the endophenotype traits are not present. In addition to genetic applications, because these endophenotypes may be more reproducible across subgroups than are the individual components, they can be used in assessments of comorbidity across subgroups. Furthermore, we plan to assess their relationship with various chronic diseases and measures of frailty.
Supplementary Material
ACKNOWLEDGMENTS
Author affiliations: Department of Human Genetics, Graduate School of Public Health, University of Pittsburgh, Pennsylvania (Jatinder Singh, Candace M. Kammerer); Taub Institute for Research on Alzheimer's Disease and the Aging Brain, New York, New York (Nicole Schupf); Department of Epidemiology, Graduate School of Public Health, University of Pittsburgh, Pennsylvania (Robert Boudreau, Tanushree Prasad, Anne B. Newman); Center on Aging and Health, Johns Hopkins Medical Institutions, Baltimore, Maryland (Amy M. Matteini); Department of Epidemiology and Prevention, Division of Public Health Sciences, Wake Forest University, Winston-Salem, North Carolina (YongMei Liu); Danish Aging Research Center, University of Southern Denmark, Odense, Denmark (Kaare Christensen); and Department of Clinical Biochemistry and Pharmacology and Department of Clinical Genetics, Odense University Hospital, Odense, Denmark (Kaare Christensen).
The Long Life Family Study was supported by the National Institute on Aging, US National Institutes of Health (grants U01AG023712, U01AG023744, U01AG023746, U01AG023749, and U01AG023755). The Health, Aging and Body Composition Study was supported in part by the Intramural Research Program of the National Institute on Aging (grants AE-2-1024, N01-AG-6-2101, N01-AG-6-2103, and N01-AG-6-2106).
Conflict of interest: none declared.
References
- 1.Newman AB, Boudreau RM, Naydeck BL, et al. A physiologic index of comorbidity: relationship to mortality and disability. J Gerontol A Biol Sci Med Sci. 2008;636:603–609. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cohen AA, Milot E, Li Q, et al. Detection of a novel, integrative aging process suggests complex physiological integration. PLoS One. 2015;103:e0116489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Matteini AM, Fallin MD, Kammerer CM, et al. Heritability estimates of endophenotypes of long and health life: the Long Life Family Study. J Gerontol A Biol Sci Med Sci. 2010;6512:1375–1379. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sebastiani P, Hadley EC, Province M, et al. A family longevity selection score: ranking sibships by their longevity, size, and availability for study. Am J Epidemiol. 2009;17012:1555–1562. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2012. [Google Scholar]
- 6.Revelle W. psych: Procedures for Personality and Psychological Research. http://cran.r-project.org/web/packages/psych/psych.pdf Published 2013. Accessed February 26, 2013.
- 7.Almasy L, Blangero J. Multipoint quantitative-trait linkage analysis in general pedigrees. Am J Hum Genet. 1998;625:1198–1211. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Therneau T. coxme: Mixed Effects Cox Models. http://cran.r-project.org/web/packages/coxme/coxme.pdf Published 2012. Accessed May 15, 2012.
- 9.Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics. 2005;611:92–105. [DOI] [PubMed] [Google Scholar]
- 10.Vickers AJ, Cronin AM, Begg CB. One statistical test is sufficient for assessing new predictive markers. BMC Med Res Methodol. 2011;11:13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Chen W, Samuelson FW, Gallas BD, et al. On the assessment of the added value of new predictive biomarkers. BMC Med Res Methodol. 2013;13:98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Burchfiel CM, Enright PL, Sharp DS, et al. Factors associated with variations in pulmonary function among elderly Japanese-American men. Chest. 1997;1121:87–97. [DOI] [PubMed] [Google Scholar]
- 13.Cook NR, Albert MS, Berkman LF, et al. Interrelationships of peak expiratory flow rate with physical and cognitive function in the elderly: MacArthur Foundation studies of aging. J Gerontol A Biol Sci Med Sci. 1995;506:M317–M323. [DOI] [PubMed] [Google Scholar]
- 14.Goodpaster BH, Park SW, Harris TB, et al. The loss of skeletal muscle strength, mass, and quality in older adults: the Health, Aging and Body Composition Study. J Gerontol A Biol Sci Med Sci. 2006;6110:1059–1064. [DOI] [PubMed] [Google Scholar]
- 15.Rantanen T. Muscle strength, disability and mortality. Scand J Med Sci Sports. 2003;131:3–8. [DOI] [PubMed] [Google Scholar]
- 16.Newman AB, Kupelian V, Visser M, et al. Strength, but not muscle mass, is associated with mortality in the Health, Aging and Body Composition Study cohort. J Gerontol A Biol Sci Med Sci. 2006;611:72–77. [DOI] [PubMed] [Google Scholar]
- 17.Metter EJ, Talbot LA, Schrager M, et al. Skeletal muscle strength as a predictor of all-cause mortality in healthy men. J Gerontol A Biol Sci Med Sci. 2002;5710:B359–B365. [DOI] [PubMed] [Google Scholar]
- 18.Sayer AA, Dennison EM, Syddall HE, et al. Type 2 diabetes, muscle strength, and impaired physical function: the tip of the iceberg? Diabetes Care. 2005;2810:2541–2542. [DOI] [PubMed] [Google Scholar]
- 19.Innes E. Handgrip strength testing: a review of the literature. Aust Occup Ther J. 1999;463:120–140. [Google Scholar]
- 20.Syddall H, Cooper C, Martin F, et al. Is grip strength a useful single marker of frailty? Age Ageing. 2003;326:650–656. [DOI] [PubMed] [Google Scholar]
- 21.Rantanen T, Harris T, Leveille SG, et al. Muscle strength and body mass index as long-term predictors of mortality in initially healthy men. J Gerontol A Biol Sci Med Sci. 2000;553:M168–M173. [DOI] [PubMed] [Google Scholar]
- 22.Ling CHY, Taekema D, de Craen AJM, et al. Handgrip strength and mortality in the oldest old population: the Leiden 85-Plus Study. CMAJ. 2010;1825:429–435. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Buchman AS, Boyle PA, Leurgans SE, et al. Pulmonary function, muscle strength, and incident mobility disability in elders. Proc Am Thorac Soc. 2009;67:581–587. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Schünemann HJ, Dorn J, Grant BJ, et al. Pulmonary function is a long-term predictor of mortality in the general population: 29-year follow-up of the Buffalo Health Study. Chest. 2000;1183:656–664. [DOI] [PubMed] [Google Scholar]
- 25.Beaty TH, Cohen BH, Newill CA, et al. Impaired pulmonary function as a risk factor for mortality. Am J Epidemiol. 1982;1161:102–113. [DOI] [PubMed] [Google Scholar]
- 26.Smits CH, Deeg DJ, Kriegsman DM, et al. Cognitive functioning and health as determinants of mortality in an older population. Am J Epidemiol. 1999;1509:978–986. [DOI] [PubMed] [Google Scholar]
- 27.Schupf N, Tang M-X, Albert SM, et al. Decline in cognitive and functional skills increases mortality risk in nondemented elderly. Neurology. 2005;658:1218–1226. [DOI] [PubMed] [Google Scholar]
- 28.Bassuk SS, Wypij D, Berkman LF. Cognitive impairment and mortality in the community-dwelling elderly. Am J Epidemiol. 2000;1517:676–688. [DOI] [PubMed] [Google Scholar]
- 29.Nwaneri C, Cooper H, Bowen-Jones D. Mortality in type 2 diabetes mellitus: magnitude of the evidence from a systematic review and meta-analysis. Br J Diab Vasc Dis. 2013;134:192–207. [Google Scholar]
- 30.Roglic G, Unwin N, Bennett PH, et al. The burden of mortality attributable to diabetes: realistic estimates for the year 2000. Diabetes Care. 2005;289:2130–2135. [DOI] [PubMed] [Google Scholar]
- 31.Stamler J, Stamler R, Neaton JD, et al. Low risk-factor profile and long-term cardiovascular and noncardiovascular mortality and life expectancy: findings for 5 large cohorts of young adult and middle-aged men and women. JAMA. 1999;28221:2012–2018. [DOI] [PubMed] [Google Scholar]
- 32.Chyou PH, Eaker ED. Serum cholesterol concentrations and all-cause mortality in older people. Age Ageing. 2000;291:69–74. [DOI] [PubMed] [Google Scholar]
- 33.Corti MC, Guralnik JM, Salive ME, et al. HDL cholesterol predicts coronary heart disease mortality in older persons. JAMA. 1995;2747:539–544. [PubMed] [Google Scholar]
- 34.Schupf N, Costa R, Luchsinger J, et al. Relationship between plasma lipids and all-cause mortality in nondemented elderly. J Am Geriatr Soc. 2005;532:219–226. [DOI] [PubMed] [Google Scholar]
- 35.Luchsinger JA, Patel B, Tang M-X, et al. Body mass index, dementia, and mortality in the elderly. J Nutr Health Aging. 2008;122:127–131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Wechsler D. WMS-III Administration and Scoring Manual. San Antonio, TX: The Psychological Corporation; 1997. [Google Scholar]
- 37.Wechsler D. WMS-R: Wechsler Memory Scales—Revised. San Antonio, TX: The Psychological Corporation; 1987. [Google Scholar]
- 38.Guralnik JM, Ferrucci L, Simonsick EM, et al. Lower-extremity function in persons over the age of 70 years as a predictor of subsequent disability. N Engl J Med . 1995;3329:556–561. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.