| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
1 Office of Population Research, Princeton University, New Jersey.
2 Cedeplar, Federal University of Minas Gerais, Brazil.
3 Department of Demography, University of California, Berkeley.
4 Center on Aging and Health and 5 Department of Population, Family, and Reproductive Health, Johns Hopkins University, Baltimore, Maryland.
6 Center for Population and Health Survey Research, Bureau of Health Promotion, Department of Health, Taichung, Taiwan.
7 Center for Population and Health, Georgetown University, Washington, D.C.
Address correspondence to: Noreen Goldman, DSc, Office of Population Research, Princeton University, 243 Wallace Hall, Princeton, New Jersey 08544-2091. E-mail: ngoldman{at}princeton.edu
 |
A |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Data come from the 2000 Social Environment and Biomarkers of Aging Study, a national sample of Taiwanese persons aged 54 years or older; 1497 persons were interviewed in their homes, and 1023 participated in a hospital examination. The analysis is based on 927 respondents with complete information. Logistic regression models describe the association between biomarkers and the 3-year probability of dying.
Results. Although both groups of biomarkers are significantly associated with mortality, the model with neuroendocrine and immune biomarkers has better explanatory and discriminatory power than the one with clinical measures. The association between these nonclinical measures and mortality remains strong after adjustment for the clinical markers, suggesting that the physiological effects of the nonclinical biomarkers are broader than those captured by the cardiovascular and metabolic system measures included here.
Conclusions. Nonclinical markers are likely to provide warning signs of deteriorating health and function beyond what can be learned from conventional markers. Our findings are consistent with those of recent studies that (i) demonstrate the importance of neuroendocrine and immune system markers for survival, and (ii) indicate that standard clinical variables are less predictive of mortality in older than in younger populations.
In this article, we focus on the link between biological risk factors and mortality for an older population. We examine all-cause mortality because, as a consequence of multiple comorbidities, prediction of specific diseases is especially problematic at these older ages (11). Rather than concentrating on individual biological risk factors, we focus on two clusters. The first comprises standard clinical risk factors related to cardiovascular and metabolic function: obesity, blood pressure, lipids, and glucose metabolism. The second cluster—which we denote "nonclinical" measures—comprises cortisol, dehydroepiandrosterone-sulfate (DHEA-S), epinephrine, norepinephrine, interleukin-6 (IL-6), insulin-like growth factor 1 (IGF-1), and dopamine. Biomarkers in this latter group are not usually measured in medical examinations; most do not have well-established clinical cutoffs.
The two clusters are motivated by a theoretical framework (12,13) that hypothesizes that chronic exposure to stressors results in prolonged dysregulation of "primary mediators" (represented by the nonclinical cluster), leading to abnormal values for "secondary outcomes" (represented by the standard clinical cluster) and ultimately, disease and death. We compare the performance of the two clusters in predicting mortality over a 3-year follow-up period.
|
METHODS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
On a scheduled day several weeks after the in-home interview, participants collected a 12-hour overnight urine sample, fasted overnight, and visited a nearby hospital the following morning where medical personnel drew a blood specimen and took blood pressure and anthropometric measurements. Compliance was extremely high (i.e., 95.7% fasted overnight and provided a urine specimen deemed suitable for analysis).
Among the 1713 respondents selected for SEBAS, 1497 provided interviews (92% of survivors) and 1023 participated in the physical examination (68% of those interviewed). Disproportionately high nonparticipation rates were found among the healthiest and least healthy respondents. Overall, persons who received the medical examination reported the same average health status (on a 5-point scale) as those who did not. Although respondents older than 70 years were less likely than younger persons to participate, sex and measures of socioeconomic status were not significantly related to participation. These results suggest that, in the presence of controls for age, estimates based on the biomarkers are unlikely to be seriously biased (14).
Survival status was ascertained in 2003 by linking to the Household Registration file of the Taiwanese Ministry of Interior. Among the 1023 examination participants, there were 72 verified deaths by 2003 and 14 respondents with unknown vital status in 2003. After excluding persons with missing vital status (14), missing data on explanatory variables (65), and proxy interviews (17), the analysis sample comprised 927 respondents (866 survivors and 61 deaths).
Explanatory Variables
The 13 biomarkers examined in this study were included in SEBAS because of their hypothesized association with stressful experiences and chronic disease. They comprise six standard clinical indicators of cardiovascular risk and metabolic activity and seven nonclinical biomarkers related to hypothalamic–pituitary–adrenal (HPA) axis activity, sympathetic nervous system (SNS) activity, and inflammatory response.
Blood and urine specimens were analyzed at Union Clinical Laboratories (UCL) in Taipei. In addition to routine standardization and calibration tests performed by the laboratory staff, during the early stages of fieldwork nine individuals outside of the sample contributed triplicate sets of specimens: Two sets were submitted to UCL, and a third set was sent to Quest Diagnostics in the United States for the purpose of checking laboratory reliability.
Systolic and diastolic blood pressures (in mmHg) were measured as the average of two seated blood pressure readings taken with a mercury sphygmomanometer at least 20 minutes after the respondent arrived at the hospital. The body mass index (BMI) was calculated as weight divided by height squared (kg/m2). Measures of IGF-1, IL-6, DHEA-S, total cholesterol, high density lipoprotein (HDL) cholesterol, and glycosylated hemoglobin (HbA1c) were obtained from the fasting blood sample. Epinephrine, norepinephrine, cortisol, and dopamine measurements were based on the 12-hour overnight urine sample to provide integrated values of basal operating levels during a time that most participants were resting; they are measured in micrograms per gram (µg/g) creatinine to adjust for body size. The assays used to measure the biomarkers derived from the blood and urine samples are described elsewhere (15).
We included age in 2000 (which ranges between 54 and 91), sex, and urban or rural residence as demographic control variables (and to adjust for the sampling design). Because of potential reverse causality (i.e., the possibility that prior health may influence biomarkers), we incorporated extensive controls for health in 2000. These comprise (i) measures of chronic conditions, mobility limitations, global self-rated health, depressive symptoms, and cognitive function; (ii) a 5-point measure of the respondent's typical level of pain; and (iii) a dummy variable indicating whether the respondent smoked in the past 6 months.
The measure of current illness counts the number (0–12) of chronic conditions reported by the respondent including: high blood pressure, diabetes, heart disease, cancer or malignant tumor, lower respiratory tract disease, arthritis or rheumatism, gastric ulcer or stomach ailment, liver or gall bladder disease, cataracts, kidney disease, gout, and spinal or vertebral spurs. The measure of mobility limitations counts the number (0–9) of physical tasks that the respondent reported difficulty performing without aid including: standing continuously for 15 minutes and for 2 hours, squatting, raising both hands over the head, grasping or turning objects with the fingers, lifting or carrying an object weighing 11–12 kg, running a short distance (20–30 meters), walking 200–300 meters, and climbing two or three flights of stairs. Global self-rated health is described by a three-category reformulation of the conventional 5-point ordinal scale (excellent/good; average; not-so-good/poor). Depressive symptoms are measured by a 10-item short-form of the Center for Epidemiologic Studies Depression scale (CES-D), coded according to standard practice (potential range 0–30). Previous studies have demonstrated that a shortened form of the CES-D yields similar internal consistency, factor structure, and accuracy in detecting depressive symptoms as the full 20-item CES-D among elderly Chinese people (16). Cognitive function is a count of cognitive tasks completed correctly, including basic orientation questions, a series of four subtractions, and immediate memory recall (potential range 0–24).
Analytic Strategy
We estimated a series of four logistic regression models to describe the associations between the biomarkers and the probability of dying over a 3-year period (2000–2003). Because the clustered sampling design may lead to underestimates of standard errors, we incorporated random effects for the primary sampling units. A baseline model includes the demographic and health control variables. The standard clinical markers are added in Model 1; the nonclinical biomarkers replace the clinical measures in Model 2. Model 3 includes both sets of biomarkers. All biomarkers are specified as continuous variables. Because outliers can have a substantial influence on the parameter estimates, we recoded 25 biomarker values in the study sample that were larger than five standard deviations from the mean to equal this cut point.
We explored the inclusion of linear and quadratic terms in the models, because both low and high values of some measures (e.g., BMI, cortisol, diastolic blood pressure, and epinephrine) have been shown to be associated with adverse health outcomes (17). We included a quadratic term if two conditions were satisfied: (i) the quadratic term was significant (p <.05) in a model that included only that biomarker along with age and sex; and (ii) the quadratic term remained marginally significant (p <.10) in a model that included all control variables, all biomarkers, and the quadratic terms satisfying the first condition. We excluded the quadratic term for systolic blood pressure because of its high correlation (0.7) with diastolic blood pressure, and because the literature suggests that the relationship between systolic blood pressure and mortality is monotonic (18). Because of the limited sample size, we do not include interaction terms in these models—for example, to capture potentially different effects of the biomarkers on mortality by sex or age.
In addition to identifying significant coefficients, we provide statistical comparisons of Model 1 (the standard clinical model) and Model 2 (the nonclinical biomarker model). We use likelihood-ratio tests for nested models to ascertain the joint significance of the set of standard or nonclinical markers. In addition, we calculate the receiver operating characteristic (ROC) curve to evaluate the accuracy of the models in discriminating between decedents and survivors (19). For a given model, the ROC curve compares the probability that the regression equation correctly predicts death for persons who died (sensitivity) with the probability of an incorrect prediction among survivors (1 – specificity) across the entire range of possible cut points. We use the area under the ROC curve (AUC) to summarize the performance of a model (higher values indicate better accuracy) and compare AUC values between models based on a chi-square test (20). All analyses were performed using Stata 8.2 (21).
|
RESULTS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
|
|
Likelihood ratio tests suggest that both sets of biomarkers are significantly related to survival. Comparisons of Models 1 and 2 with the baseline model indicate that inclusion of either set of biomarkers significantly improves the model (p =.04 for the standard clinical markers and p <.001 for the nonclinical markers). Similarly, in comparison with the full model (Model 3), removal of either set of markers results in a significantly poorer fit (p =.04 for the clinical measures and p <.001 for the nonclinical ones).
Two comparative assessments provide evidence that the nonclinical biomarker model is superior to the standard clinical one. First, the nonclinical biomarker model has a substantially larger pseudo R2 value (0.26 vs 0.21) despite having the same number of parameters, suggesting better explanatory power (19). Second, the ROC curves depicted in Figure 1 illustrate the better discriminatory power of the nonclinical biomarker model. Chi-square tests indicate that the nonclinical model has a significantly higher AUC than the baseline model (0.852 vs 0.790, p <.01), whereas the difference between the standard clinical and baseline models is not significant (0.808 vs 0.790, p =.23). The difference in AUC between the clinical and nonclinical models is marginally significant (p <.09). Although it is plausible that the superior performance of the neuroendocrine and immune measures results from correlation between the clinical markers and self-reported health measures, the results remain essentially unchanged when we exclude controls for health in 2000.
|
|
DISCUSSION |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Although there are few, if any, population-based studies in East Asian populations to corroborate our results, several recent community-based studies in the United States underscore the potential importance of markers of immune activity and inflammation for the functioning and survival of older persons. The Framingham Heart Study revealed that high levels of circulating IL-6 and low levels of IGF-1 were associated with increased mortality after adjustment for important clinical conditions (9); similar results were obtained for a cohort of older women in the Women's Health and Aging Study (4). The Iowa 65+ Rural Health Study indicated that higher IL-6 levels were associated with substantially greater risk of death (5).
Estimates from Western populations also suggest that HPA-axis and SNS activity are related to survival. Although most studies document linkages between cortisol levels and the presence of chronic conditions and illnesses, cortisol levels have been shown to be predictive of mortality among adults who suffered an acute myocardial infarction (22) or a stroke (6,23). DHEA-S, which is believed to counterbalance the effects of cortisol, is associated with mortality risks among men (7,10). Older persons with high baseline urinary excretion of epinephrine or norepinephrine in the MacArthur Studies of Successful Aging had higher risks of dying (8).
The greater importance of nonclinical as compared with standard risk factors for older-age mortality is consistent with findings that conventional clinical measures are less predictive of morbidity and mortality among elderly persons than among younger persons. For example, higher levels of cholesterol and blood pressure are generally related to higher adult mortality, but at older ages the relationship has been U-shaped or sometimes inverse (24–26). Research also suggests that the relationship of BMI with mortality weakens at older ages (27).
A limitation of the present study is the small number of deaths. The low statistical power makes it impossible to rule out associations between specific biomarkers and survival or to assess the relative importance of different individual markers. A second limitation pertains to nonresponse: The selective exclusion of the healthiest and most disabled respondents from the medical examination portion of the study may lead to biased estimates. In addition, because of both genetic and environmental variation, there are questions regarding the generalizability of our results to Western populations. For example, previous studies have documented that: (i) the distributions of the biomarkers included in this analysis, especially the clinical measures, differ between Taiwan and the United States (28), and (ii) the effects of clinical biomarkers on mortality are likely to vary across populations (29).
Nonetheless, the findings of this study are sufficiently robust to suggest that, taken as a group, the neuroendocrine and immune markers are important predictors of future survival. They are likely to provide additional warning signs of deteriorating health and function above and beyond what can be learned from standard clinical measures. That is, these markers may indicate risk even when the clinical markers do not, suggesting that they may represent a direct pathway to mortality. At the same time, the results are not sufficient to warrant the inclusion of these nonclinical measures in a standard diagnostic toolkit. We have little information regarding what constitutes normal or abnormal levels of these markers, a limited understanding of the extent to which changes over time reflect normal processes of aging or are causally linked to health and survival, and few notions about how to modify levels of these markers (or, indeed, whether modification would ultimately enhance survival). With additional research, we are virtually certain to witness an expansion of the list of clinically assessed measures that are predictive of future survival.>
|
A |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
The survey procedures were approved by the institutional review boards at Princeton University, Georgetown University, and the Bureau of Health Promotion, Department of Health, Taiwan and conformed to the principles embodied in the Declaration of Helsinki.
We thank Germán Rodríguez, Bruce McEwen, and Douglas Ewbank for their comments.
|
F |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Received November 22, 2005
Accepted April 18, 2006
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  N. Goldman, D. A. Glei, Y.-H. Lin, and M. Weinstein Improving Mortality Prediction Using Biosocial Surveys Am. J. Epidemiol., January 15, 2009; (2009) kwn389v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. B. Dowd, M. N. Haan, L. Blythe, K. Moore, and A. E. Aiello Socioeconomic Gradients in Immune Response to Latent Infection Am. J. Epidemiol., January 1, 2008; 167(1): 112 - 120. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
