| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Division of Geriatrics, Department of Medicine, David Geffen School of Medicine at UCLA, Los Angeles, California (A.S.K., T.E.S.); Office of Population Research, Princeton University, Princeton, New Jersey (B.H.S.).
Address correspondence and reprint requests to Arun S. Karlamangla or Teresa E. Seeman, 10945 Le Conte Avenue #2339, Los Angeles, CA 90095-1687. E-mail: akarlamangla{at}mednet.ucla.edu or tseeman{at}mednet.ucla.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Methods: We measured 10 physiologic parameters at baseline (1988) in a cohort of 171 high-functioning, community-dwelling, 70- to 79-year-old adults. These measurements were repeated 2.5 years later, in 1991. Summary allostatic load scores for 1988 and 1991 were created as the weighted sum of the 10 biological markers and their second-order terms. Mortality status (alive or dead) for participants was determined 4.5 years later, in 1995. The association between change in allostatic load score (1988–1991) and subsequent mortality (1991–1995) was studied using logistic regression.
Results: Compared with participants whose allostatic load score decreased between 1988 and 1991, individuals whose allostatic load score increased had higher risk of all-cause mortality between 1991 and 1995 (15% versus 5%, p = .047). Adjusted for age and baseline allostatic load, each unit increment in the allostatic load change score was associated with mortality odds ratio of 3.3 (95% confidence interval, 1.1–9.8).
Conclusion: Our results suggest that even in older ages, change in risk scores can be followed to improve assessment of mortality risk.
Key Words: risk score • mortality • allostatic load • risk factor change • risk score change
Abbreviations: DHEA-S = dehydroepiandosterone sulfate; HDL = high-density lipoprotein; CVD = cardiovascular disease; ROC = receiver operating curve.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
The concept of allostatic load originates from the idea that the internal physiologic milieu adapts to environmental demands—a phenomenon referred to as allostasis (7). The expected physiologic response to a stressful situation is physiologic arousal, as in the fight-or-flight response. Allostasis is this dynamic regulatory process, with continuous adaptation of physiology in response to stressors. However, allostatic adaptation of the body to repeated stresses can exact a toll. When adaptation efforts are excessive, in terms of frequency, duration, and/or extent, it can lead to gradual loss of the body’s ability to maintain system parameters within normal operating ranges. Frequent or chronic arousal has been associated with ultimate dysregulation of major physiologic systems, including the hypothalamic-pituitary-adrenal axis (8), the sympathetic nervous system (9,10), and the immune system (11). Allostatic load is the total accumulation of such dysregulation across physiologic systems and was hypothesized to mediate the effects of stress on health risks.
A composite allostatic load score constructed from cardiovascular markers and neuroendocrine hormones does indeed predict a variety of health outcomes in high-functioning elderly individuals (12). This score was based on 10 measurements that reflect the resting activity of major physiologic systems. Although none of the 10 markers exhibited substantial individual association with health outcomes, summary measures of allostatic load were found to be significantly associated with incident cardiovascular events, decline in physical and cognitive functioning, and total mortality in high-functioning older adults (12–14). However, it is not clear if risk-score changes in old age have any impact on risk and whether changes in composite risk scores (such as allostatic load) should be followed in older adults. Therefore, our objective was to examine the association between change in allostatic load over 2.5 years and the risk of all-cause mortality over the subsequent 4.5 years in older adults.
The original allostatic load score was a simple count of the number of biological markers that are in the worst quartiles of their distributions. Such a threshold-based scoring system, although adequate for studying population-level associations with health risks, is not appropriate for monitoring change over time in individuals. It will fail to capture small changes over time in individual risk factors, which can add up to meaningful changes in overall health risk. To follow changes in allostatic load over time in individual adults, we need a scoring system that is sensitive to small changes in risk factors; we therefore created a new allostatic load scoring system based on continuous values of risk factors and used weights to combine the risk factors. We used data from the MacArthur Successful Aging cohort of high-functioning 70- to 79-year-old adults who had baseline assessments in 1988, first follow-up in 1991, and second follow-up beginning in 1995. We used risk factor measurements in 1988, in conjunction with 7-year mortality data (1988–1995), to create the allostatic load scoring system based on continuous values of risk factors. We then examined the association between change in this allostatic load score over 2.5 years (1988–1991) and the risk of all-cause mortality in the cohort over the subsequent 4.5 years (1991–1995).
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Baseline demographic information, blood tests, and tests on 12-hour overnight urine collection were obtained in 1988. Complete data for all 10 biological markers (used to create the allostatic load score) were available for 729 of the 1189 individuals in the cohort. These 729 individuals constituted our Study Sample 1 for the development of an allostatic load scoring system. Descriptive analyses suggest that this sample is representative of the full cohort (12).
Repeat data were collected after a mean interval of 28 months, in 1991. Attrition was 177 of 1189, due to 71 deaths (6%), 47 refusals to follow up (4%), and 59 partial or proxy interviews (5%). Owing to financial limitations, repeat overnight urine assays were obtained on only a random subsample. One hundred seventy-one participants had measurements of all 10 biological markers in both 1988 and 1991 and constituted our Study Sample 2 for the allostatic load change analyses. As detailed in the Results section, Study Sample 2 did not differ substantially from Study Sample 1 or the complete cohort in the variables of interest. Mortality status (dead or alive) was assessed at the time of the second follow-up, beginning in 1995, after a mean interval of 57 months from the first follow-up.
Measurements
Biological Measurements
The 10 biological markers used to create allostatic load scores were: 1) waist-to-hip circumference ratio; 2) systolic blood pressure; 3) diastolic blood pressure; 4) urinary cortisol; 5) urinary norepinephrine; 6) urinary epinephrine; 7) serum dehydroepiandrosterone sulfate (DHEA-S); 8) glycosylated hemoglobin; 9) serum high-density lipoprotein (HDL) cholesterol; and 10) total serum cholesterol. Using procedures outlined in the 1988 anthropometric standardization reference manual (16), waist circumference and hip circumference were measured. The ratio of waist circumference to hip circumference was computed from these two measurements. Three seated blood pressure readings were made using the Hypertension Detection and Follow-up Program protocol (17), and average systolic and diastolic blood pressures were computed from the second and third readings.
Participants collected overnight urine from 8 PM to 8 AM the next morning. A primary goal in collecting the urine samples was to obtain integrated estimates of endocrine activity over a resting period. Acidified samples were sent to Nichols Laboratories (San Juan Capistrano, CA) for assays of cortisol, norepinephrine, epinephrine, and creatinine. Determinations were made with high-pressure liquid chromatography (18,19). The 12-hour urinary excretion of cortisol, norepinephrine, and epinephrine were normalized by 12-hour urinary creatinine to adjust for body size and renal function.
Nonfasting blood samples were collected at the participants’ homes at 8 AM and sent to Nichols Laboratories for assays of serum DHEA-S, blood glycosylated hemoglobin, serum HDL cholesterol, and total serum cholesterol. Glycosylated hemoglobin levels were measured by affinity chromatography (20), HDL cholesterol was measured by the direct homogeneous method (Genzyme Diagnostics, Cambridge, MA), and total serum cholesterol was measured by enzymatic colorimetry (21). Both HDL and total cholesterol are relatively insensitive to nonfasting versus fasting state at the time of the blood draw (22). All biology measurements were repeated at first follow-up (1991) for the individuals in Study Sample 2.
The distribution of each biological marker at baseline (1988) was examined for outliers and symmetry. With the exception of waist-hip ratio and diastolic blood pressure, all others needed to be transformed using either natural logarithm (for systolic blood pressure, urinary cortisol, urinary norepinephrine, urinary epinephrine, glycosylated hemoglobin, HDL cholesterol, and total cholesterol) or square-root (for DHEA-S) transformation to achieve symmetry of distribution. The same transformations were applied to the corresponding 1991 measurements.
Mortality Assessment
Deaths were identified through contact with next of kin at the time of the follow-up, ongoing local monitoring of obituary notices, and National Death Index searches. Mortality status (alive or dead) at the time of the second follow-up was available for every participant.
Measurement of Covariates
Gender, ethnicity (white versus black), and age in years were obtained at the time of the baseline examination from participants’ reports. Current smoking information was also collected at baseline. Prevalent health conditions, including diabetes mellitus, previous heart attacks, strokes, cancer, and other chronic diseases, were assessed from baseline self reports. A simple count of chronic conditions (range 0–7) was created. Prevalent cardiovascular disease (CVD) was defined by the presence of at least one of diabetes mellitus, previous myocardial infarction, and previous stroke.
Statistical Analyses
Development of Allostatic Load Scoring System in Study Sample 1
The first proposed allostatic load score was a simple count of the number of biological markers that met a threshold-based high-risk criterion. Because a threshold-based scoring system may fail to capture small changes over time in individual risk factors, here we used the continuous values of each marker to construct the allostatic load score. Also, because increased risk may result from dysregulation in either direction; i.e., from both high and low values of biological parameters, and because quadratic and U-shaped relationships have been documented between cardiovascular risk factors and both coronary heart disease (23) and mortality (24), we created second-order terms (x – 
)2 to capture deviation in either direction from the mean, where x is the (transformed) biological variable and is the sample mean of x. Allostatic load, as originally defined, was meant to capture dysregulation that can occur with deviation from the normal operating range in either direction, high or low. Yet, the initial operationalizations of allostatic load only captured deviations in one direction for each biomarker (e.g., high for blood pressure, low for HDL cholesterol). In the scoring system developed here, we allow for deviation in either direction by including second-order terms for each biomarker. Since different risk factors are measured in different units and contribute differently to health risks, and because not all second-order terms may contribute to health risk, we used weights to combine the risk factors and their second-order terms in the allostatic load score. We selected weights for the 20 allostatic load components (10 risk factors and 10 second-order terms) based on their independent associations with 7-year all-cause mortality.
Using data from the 729 individuals in Study Sample 1, we fit a logistic model for 1988 to 1995 all-cause mortality as a function of the 1988 values of the 20 allostatic load components. Components that did not make a consistent contribution to mortality risk were dropped from the model in a stepwise fashion. To assess a component’s contribution to the model, we examined the following z statistic created from the bootstrap distribution of its model coefficient (based on 200 bootstrap samples):
|
|
In each step, the component with the z statistic of smallest magnitude was eliminated and the model was refitted to 200 new bootstrap samples. This process was terminated when the z statistic for every remaining component was 1 or larger in magnitude. In stepwise prognostic model building, it has been suggested that the selection criterion for candidate predictors should be more liberal than the conventional p < .05 criterion (which is equivalent to |z| > 1.96) (25). Our use of |z| 
1 as the termination criterion is equivalent to selection based on p 
0.3.
The bootstrap means of the coefficients in the final model were used as weights for allostatic load components (biological markers and/or their second-order terms) to create a summary score for every individual in Study Sample 1, as
|
|
Components that had been eliminated from the logistic model got zero weights and did not contribute to the allostatic load score. The 1988 values of biological markers were used to create 1988 allostatic load scores. The constant was chosen to make zero the lowest allostatic load score in 1988.
The resulting allostatic load score was internally validated using bootstrapping, which is felt to be superior to the more traditional methods of validation by sample splitting (26). All associations of the baseline allostatic load score with 7-year mortality (with and without adjustment for potential confounders) were examined on 1,000 new bootstrap samples, and the bootstrap distribution of each odds ratio was used to determine the confidence interval for that odds ratio. We chose to report confidence intervals derived from the bootstrap distribution of odds ratio because parametrically derived confidence intervals for the allostatic load score will be falsely narrow, given that the score was derived from the same data set (27).
For comparison, the 1988 Framingham risk score was also computed for every individual in Study Sample 1 using 1988 values of traditional cardiovascular risk factors: gender, smoking status, blood pressure, prevalent diabetes mellitus, and blood lipid levels (28). The associations of the baseline allostatic load score and the baseline Framingham risk score with 7-year all-cause mortality were compared.
Allostatic Load Change Analyses in Study Sample 2
The weights (and the added constant) used to create the 1988 allostatic load scores were used to also create allostatic load scores in 1991 for the 171 participants in Study Sample 2.
We first compared the 4.5-year (1991 to 1995) all-cause mortality risk in the group of individuals whose allostatic load score decreased between 1988 and 1991 with the risk in the group of individuals whose allostatic load score increased or stayed the same. It is possible that the people who start with high allostatic load at baseline (1988) are also the ones whose allostatic load increases between 1988 and 1991, and the difference in mortality risk between the 2 groups above is actually a reflection of differences in baseline allostatic load. Therefore, in the next step, we used multivariable logistic regression to adjust for the baseline allostatic load score.
Because the magnitude of change in allostatic load (whether big change or small) may also be important, beyond the direction of change (up versus down), we created an allostatic load change score as the 1991 score minus the 1988 score and used the continuous change score as primary predictor in the multivariable models. In these models, the 1991 to 1995 mortality was the outcome, allostatic load change score was primary predictor, and baseline (1988) allostatic load score was included as a potential confounder.
Other covariates (age, gender, ethnicity, prevalent CVD, prevalent cancer, and count of chronic diseases) were introduced in the model one at a time to see if they confounded the association between allostatic load change score and mortality risk. We chose to introduce these potential confounders one at a time into the model because we wanted to avoid problems with overfitting, given the small sample size of 171. To test for effect modification, interaction terms were also introduced one at a time. All p values reported are for two-sided tests. SAS version 8, SAS Institute Inc., Cary, NC, was used for all analyses.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
|
|
Allostatic Load Scoring
One hundred fifty-four of the 729 individuals in Study Sample 1 (21%) died by the time of the second follow-up in 1995. A logistic regression model for 7-year (1988 to 1995) all-cause mortality as a function of 20 allostatic load components (10 biological variables and 10 second-order terms) measured in 1988 was fit to Study Sample 1. The bootstrap mean and SD of the coefficients in the final model are listed in Table 2, labeled weight and standard error, respectively. Waist-to-hip circumference ratio, urine norepinephrine, and glycosylated hemoglobin made positive contributions to mortality risk, whereas HDL cholesterol and total cholesterol were both negatively associated with all-cause mortality. Second-order terms were retained for DHEA-S, systolic and diastolic blood pressure, and glycosylated hemoglobin, suggesting that both low and high values of these biological variables can be deleterious.
|
The 1988 values of biological variables and/or their second-order terms were weighted by the bootstrap means listed in Table 2 and added to the constant 5.6 to construct the 1988 allostatic load score. The constant 5.6 was chosen to make zero the lowest allostatic load score in the study sample. To assess the contribution of various components to the allostatic load score, we examined the Spearman rank correlations between each component and the allostatic load score. As indicated in Table 2, HDL cholesterol had the correlation of largest magnitude (r = –0.52, p < .0001), followed by waist-hip ratio (r = 0.50, p < .0001). Systolic and diastolic blood pressure and serum DHEA-S, which only contributed to allostatic load score via second-order terms, had small correlations with the score (r < 0.07, p > .07); data not shown. However, their second-order terms were significantly correlated with allostatic load score (r > 0.15, p < .0001); see Table 2.
The distribution of baseline (1988) allostatic load scores in Study Sample 1 was fairly symmetric (skew = 0.36), and its mean and SD were 1.3 and 0.5, respectively. The interquartile range was 1.0, 1.6, and the total range was 0, 3.2. Men had higher mean allostatic load score at baseline than women (1.5 versus 1.1, p < .0001). The association between baseline allostatic load score and 7-year mortality in Study Sample 1 is depicted in Figure 1.
|
The area under the receiver-operating curve (ROC) for allostatic load score as a predictor of 7-year mortality was 0.642. This area is a measure of the discrimination ability of the score and has the following interpretation: For any two individuals from the population, the probability that the person with the higher score will also have higher 7-year mortality risk is equal to the area under the ROC. Thus, a score with area under the ROC of 0.5 has no predictive ability, whereas a score with area under the ROC of 1 is a perfect, errorless predictor (25). To get 95% confidence intervals for the area under the ROC, we repeated the ROC area calculation on 1,000 bootstrapped samples, and used the 2.5th and 97.5th percentile values of its resulting distribution. This empirically obtained 95% confidence interval for the area under the ROC for allostatic load score as a predictor of 7-year mortality was 0.597–0.690.
When age (continuous), gender, and baseline allostatic load score were used in a logistic model to predict 7-year all-cause mortality, each of the three predictors made significant contributions to mortality risk. Each additional year of baseline age was associated with adjusted mortality odds ratio of 1.10 (95% bootstrapped confidence interval, 1.03–1.18). Women were less likely to die than were men of the same age and with the same baseline allostatic load; adjusted odds ratio 0.44 (95% bootstrapped confidence interval, 0.29–0.66). After adjusting for age and gender, each unit increment in baseline allostatic load score was associated with mortality odds ratio of 2.25 (95% bootstrapped confidence interval, 1.60–3.32). To test whether the strength of the association between allostatic load score and mortality was different by gender, we introduced a gender x score (interaction) term; the interaction term did not make a significant contribution to the model (p = .9), suggesting that that a unit increment in allostatic load score confers the same increase in mortality odds in both genders.
In contrast to the allostatic load score, the Framingham risk score at baseline was not associated with all-cause 7-year mortality: odds ratio per unit increment in score: 1.04, p = .12, area under ROC 0.526; 95% bootstrapped confidence interval, 0.479–0.570.
Allostatic Load Change Analyses
The weights (bootstrap means) listed in Table 2 and the constant 5.6 were used to also create allostatic load scores in 1991 for the 171 individuals in Study Sample 2. Sixty-three participants had decreases in allostatic load score between 1988 and 1991, whereas 108 participants had increases in allostatic load score. No one had zero change score. The distribution of the allostatic load change score was fairly symmetric (skew = –0.2) with mean 0.11, SD 0.46, interquartile range –0.17, +0.38, and total range –1.67, +1.44. There was no difference in mean change score between men and women (p = .7).
To determine which biological changes were most responsible for the observed changes in allostatic load scores, we examined the Spearman rank correlations between change in individual components (between 1988 and 1991) and the allostatic load change score. Change in waist-hip ratio had the largest correlation with allostatic load change score (r = 0.40, p < .0001) followed by change in norepinephrine level (r = 0.39, p < .0001) and change in glycosylated hemoglobin (r = 0.34, p < .0001). HDL cholesterol, which made the largest contribution to allostatic load score, did not make as large a contribution to allostatic load change score; the Spearman rank correlation between change in HDL and allostatic load change score was –0.18, (p = .01). Compared with participants whose allostatic load score decreased between 1988 and 1991, participants whose allostatic load increased over the same period had larger increases in waist-to-hip circumference ratio (t test, p < .0001), norepinephrine levels (p < .0001), and glycosylated hemoglobin (p = .002) and larger decreases in total cholesterol (p = .0005) and HDL cholesterol (p = .04). There were no significant differences between the 2 groups with respect to changes in the other five biological variables (t test, p > .2).
To assess whether the people who started with high allostatic load at baseline are also the ones whose allostatic load scores increased over time, we examined the Pearson correlation between the baseline allostatic load score and the allostatic load change score. The correlation coefficient was negative (–0.45, p < .0001), implying that more people with low baseline scores experienced increases in allostatic load over time than people with high baseline scores.
Nineteen of the 171 individuals in Study Sample 2 died by the time of the second follow-up in 1995. The mortality rate was higher in participants whose allostatic load score increased between 1988 and 1991 (16 deaths out of 108, or 15%) than in participants whose allostatic load score decreased (3 deaths out of 63, or 5%); p < .05 (Fisher’s exact test). In logistic regression modeling, after adjusting for age (continuous term) and the baseline allostatic load score, each 1-point increase in allostatic load change score was associated with adjusted mortality odds ratio of 3.33 (95% confidence interval, 1.14–9.74), as shown in Table 3. The area under the ROC for the model with age, baseline allostatic load score, and allostatic load change score was 0.75.
|
We tested for confounding and effect modification by the following covariates separately, one at a time: gender, ethnicity, prevalent CVD, prevalent cancer, and count of chronic diseases. None of the covariates confounded the primary association: p > .25 for the covariate term in each case. The covariates also did not modify the primary association between allostatic load change score and subsequent mortality: p > .6 for the interaction term in each case.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
The allostatic load score was better at predicting all-cause mortality risk than the Framingham risk score. Others have found that the Framingham risk score has less predictive value in elderly individuals than in younger adults (5,29–31). This deterioration in the predictive ability of the Framingham risk score with increasing age is a result of the changing role of many traditional risk factors with age (3–5). For instance, several studies have found that in contrast to its role as a risk factor in young and middle-aged adults, high cholesterol may actually be protective in older adults (1,32–34). Consistent with this, we found that total serum cholesterol made a negative independent contribution to all-cause mortality in this cohort: Higher levels of total cholesterol were associated with lower mortality after adjusting for HDL cholesterol and other biological measurements.
Of the various biological variables, HDL cholesterol made the largest contribution to allostatic load score, followed closely by waist-to-hip circumference ratio, a measure of central obesity. This is consistent with previous work that has found high HDL levels to be one of the more reliable predictors of longevity (35,36) and healthy aging (37). Change in waist-to-hip circumference ratio and norepinephrine level made the strongest contributions to allostatic load change score. This is consistent with previous work in this cohort that found elevated norepinephrine levels to be predictive of mortality (38) but is in contrast to studies in less well-functioning and frail older adults where either no association or J-shaped associations have been reported between measures of central adiposity and mortality (39), and weight loss and not weight gain was associated with increased mortality risk (40). Our results suggest that central adiposity continues to be a risk factor for all-cause mortality in high-functioning older adults, even if it is not a risk factor in frail older individuals, and that high-functioning older adults should continue to monitor waist circumference as a marker for health risk.
Differences in the relative importance of different biological markers have been seen in previous allostatic load investigations (13). This study also brings out the importance of considering risk from both high and low values of biomarkers: quadratic terms for DHEA-S, systolic and diastolic blood pressure, and glycosylated hemoglobin played a role in allostatic load. This is consistent with the well-recognized risks associated with low blood pressure and low blood glucose and highlights the importance of considering deviation in both directions from the normal operating range. We are not suggesting that the scoring system developed here for allostatic load is the ultimate scoring system; the best choice of weights for biomarkers may vary with age, outcome of interest, and the inclusion of new biomarkers as they are recognized. Our decision to use continuous values of biomarkers was motivated by the need to be able to detect small changes in multiple markers. Such a scoring system is needed when one is interested in following changes in the score over time in individual patients. Studies that use a single measure of allostatic load can continue to use categorical scoring techniques.
Previous studies have found that allostatic load is positively associated with adverse health outcomes. This is the first study to examine the role of change in allostatic load over time, vis-à-vis the risk of subsequent adverse outcomes. In our study, both baseline allostatic load score and the allostatic load change score were independently and significantly associated with subsequent all-cause mortality. This suggests that both the history of risk factor elevations and the current status of risk factors contribute to mortality risk. Historical elevations in risk factors contribute to future health outcomes by affecting current subclinical disease states such as the extent of atherosclerosis. In fact, data from the Honolulu Heart Program suggested that in older men, remote measurements of serum total cholesterol had a stronger association with coronary heart disease incidence than more recent cholesterol measurements (41). On the other hand, current elevations in risk factors such as catecholamine levels, blood pressure, and blood glucose contribute to risk of adverse health outcomes both by further increasing chronic subclinical pathology and by increasing the likelihood of acute events, such as plaque rupture and thrombus formation (42,43). Our results suggest that both past and current values of risk factors have a role to play in predicting mortality risk and that even in older ages, measurement of risk scores can be repeated to improve assessment of current risk.
In our study, allostatic load scores of 70- to 79-year-old adults changed substantially over a 2.5-year period, the majority of the cohort experienced an increase in allostatic load score, and individuals with low scores at baseline were more likely (than people with high scores) to experience an increase. This suggests that biological dysregulation continues to increase with age, possibly because of continued exposures to life stresses in older ages.
These findings should be interpreted in the context of the study limitations. Our cohort consisted of only high-functioning individuals, so the results may not be generalizable to less well-functioning or frail older adults. However, high-functioning older adults may have the most to gain or lose from preventive interventions and represent an important group in whom the predictive ability of risk factors and changes in risk factors needs to be understood. Second, our sample size for change analyses was small (N = 171) and limited our ability to detect interactions, if present. Though we found that gender, ethnicity, and prevalent CVD status did not modify the association between allostatic load change score and subsequent mortality risk, it is possible that one or more of these variables do modify effect, but our sample size prevented us from detecting it. Third, because the same cohort was used to create the allostatic load scoring system and to test the prediction ability of allostatic load score, there is a certain amount of optimism in our estimate of the association between baseline allostatic load score and mortality (27). However, because the repeated measurements of biomarkers were not used in the development of the scoring system, our primary findings, the results of the change analyses, should be free from this bias. Furthermore, we adjusted for baseline allostatic load score in multivariable models to remove any residual optimism in the change analyses. Finally, this being an observational study, we cannot rule out the possibility that reduction in allostatic load score is simply a marker of some other phenomenon that is more directly responsible for lower mortality risk. However, based on the wealth of evidence of benefits from risk factor modification in younger cohorts, there is good reason to believe that changes in allostatic load are causally responsible for the differences in mortality risk seen in this study.
Several unique strengths of the study also need to be acknowledged. The deliberate choice of a high-functioning sample limited the possibility of confounding by reverse effects of poor health on risk factors. Although others have looked at the health effects of change in individual risk factors, to our knowledge, this is the first study of mortality associations with changes in a composite score constructed from multiple risk factors. In conclusion, this study found that increase in allostatic load score over 2.5 years was associated positively with all-cause mortality over the subsequent 4.5 years in high-functioning 70- to 79-year-old adults. Our results suggest that even in older ages, risk factor changes are associated with mortality risk and that risk assessment can be enhanced by following changes in composite risk scores such as allostatic load.
Work on this article was supported by NIH/NIA Mentored Clinical Scientist Development Award 1K12AG01004, NIA grants AG-17056 and AG-17265, and by the MacArthur Research Network on Successful Aging and the MacArthur Research Network on SES and Health through grants from the John D. and Catherine T. MacArthur Foundation.
|
NOTES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
Received for publication November 27, 2004; revision received December 14, 2005.
DOI:10.1097/01.psy.0000221270.93985.82
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONNOTESREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  J. Maselko, L. Kubzansky, I. Kawachi, T. Seeman, and L. Berkman Religious Service Attendance and Allostatic Load Among High-Functioning Elderly Psychosom Med, June 1, 2007; 69(5): 464 - 472. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
