| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
New England Research Institutes, Watertown, Massachusetts 02472
Address all correspondence and requests for reprints to: Dr. Thomas G. Travison, New England Research Institutes, 9 Galen Street, Watertown, Massachusetts 02472. E-mail: ttravison{at}neriscience.com.
 |
Abstract |
|---|
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Objective: The objective of this study was to establish the relative importance of aging, health, and lifestyle in contributing to male T decline.
Design: A prospective cohort study of health and endocrine functioning in randomly selected men with a baseline visit (T1, 1987–1989) and two follow-up visits (T2, 1995–1997; T3, 2002–2004) was conducted.
Setting: An observational study of men residing in greater Boston, Massachusetts, was conducted.
Participants: Participants included 1667 men aged 40 to 70 at baseline; follow-up was conducted on 947 (57%) and 584 (35%) at T2 and T3, respectively.
Main Outcome Measures: Main outcome measures included total serum T, calculated free T (FT), and SHBG.
Results: There were substantial declines in total serum T and FT levels associated with aging alone. However, many health and lifestyle changes were associated with accelerated decline. A 4- to 5-kg/m2 increase in body mass index or loss of spouse was associated with declines in total serum T comparable to that associated with approximately 10 yr of aging. Results were similar for FT, but fewer factors were associated with SHBG after age was taken into account.
Conclusions: Both chronological aging and changes in health and lifestyle factors are associated with declines in serum T. Comorbidities and lifestyle influences may be as strongly associated with declining T levels as is aging itself over the short- to midterm. These results suggest the possibility that age-related hormone decline may be decelerated through the management of health and lifestyle factors.
|
Introduction |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Many authors have distinguished aging itself from para-aging phenomena, the latter encompassing the variety of conditions common among, but not specific to, older adults (13, 14, 15, 16). This distinction is critical in that aging is inexorable, whereas para-aging phenomena offer the possibility of preventative measures to slow health declines.
Longitudinal investigations of hormone decline are rare, and although it is known that factors such as relative weight and smoking may influence serum T levels (17, 18, 19, 20), there have been few comprehensive investigations of age-related T decline in the context of contemporaneous changes in health and lifestyle. The objective of this study was to establish the relative contributions of aging, health, and lifestyle factors to changes in serum total T (TT), calculated free T (FT), and SHBG among community-dwelling older men. To do so, we analyzed data on 1667 subjects enrolled in the Massachusetts Male Aging Study, a longitudinal epidemiological study of men’s health and endocrine functioning.
|
Subjects and Methods |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
All study activities were approved by the Institutional Review Board of the New England Research Institutes. The Massachusetts Male Aging Study design and prior results are described elsewhere (1, 2, 21, 22, 23). Briefly, a random sample of male residents of greater Boston, Massachusetts, was drawn so that within three age strata (40–49, 50–59, 60–69 yr), each resident had an equal probability of selection. There were three data collection waves, which we refer to as T1 (1987–1989), T2 (1995–1997), and T3 (2002–2004). During the baseline period (T1), 1709 subjects were visited in their homes by trained interviewers/phlebotomists. Follow-up data were obtained on 1156 subjects at T2 and on 853 subjects at T3.
Each subject provided information concerning his health and life circumstances as well as a nonfasting blood sample. Subjects were largely married (75%), employed (78%), white (96%), and high school graduates (89%).
Serum hormones may be influenced by elements of experimental design (24). Accordingly, the Massachusetts Male Aging Study took steps to minimize bias and imprecision. Blood samples were drawn within 4 h of subjects’ waking to reduce the impact of diurnal variation in hormone concentrations (25). To counteract episodic hormone secretion (26), two samples were obtained at each visit, 30 min apart, and were pooled in equal aliquots at the time of assay. Blood was kept in an ice-cooled container and centrifuged within 6 h of study visit. Serum was stored in 5-ml scintillation vials at –20 C, shipped to the laboratory within 1 wk by same-day courier, and stored at –70 C until the time of assay. All assays were performed at the Endocrine Laboratory, University of Massachusetts Medical Center, under the direction of Christopher Longcope, M.D.
TT was measured with RIA kits from Diagnostic Products Corporation (DPC; Los Angeles, CA). T1 samples were assayed in 1994, whereas T2 and T3 samples were assayed shortly after collection; validation studies revealed negligible changes in T1 TT concentrations as a result of storage artifact (8). TT interassay coefficients of variation (CV) were 8.0, 9.0, and 8.3% at T1, T2, and T3, respectively. Age-specific TT concentrations were consistent with those obtained in other major epidemiological studies of serum T (21). The proportion of serum FT was calculated using the mass action equations described by Sodergard et al. (27) with association constants taken from Vermeulen et al. (28); FT concentrations were obtained by the computation TT x proportion of serum FT. Serum SHBG was measured using RIA kits at T1 and T2 and at T3 by chemiluminescent enzyme immunometric assay using the DPC Immulite technology; interassay CVs were 10.9, 7.9, and 3.0% at T1, T2, and T3, respectively. LH concentrations were measured by RIA kits from Ciba Corning (Medfield, MA) at T1 (CV = 9.5%), the Abbott Diagnostics (Chicago, IL) IMx system at T2 (CV = 6.9%), and by DPC Immulite at T3 (CV = 3.2%).
Demographic characteristics (age, education, household income, marital status), history of physician-diagnosed chronic illness (cancers, diabetes, heart disease, hypertension, ulcer), self-assessed health, current cigarette smoking, and daily alcohol consumption (29) were obtained by self-report. Self-reported diagnoses of prostate cancer were augmented by examination of available medical records. Dietary intake was measured using the Willett food frequency questionnaire (30). Physical activity and energy expenditure were derived from subjects’ reports of activities over the previous 7 d (31). Anthropometric measures were obtained using methods developed for use in large-scale epidemiological field work (32). Depressive symptoms were measured using the Center for Epidemiologic Studies–Depression (CES-D) scale (33). A comprehensive accounting of the medications in use by subjects was obtained through manual inventory of medication containers.
A subject was considered obese if he had body mass index (BMI) of at least 30 kg/m2, to exhibit sedentary behavior if his energy expenditure did not exceed 200 kcal/d (31, 34), to exhibit heavy drinking if his daily alcohol consumption was in excess of six drinks, and to exhibit depressive symptoms if his CES-D index was greater than 15. In a manner similar to that used in previous investigations (1, 2, 8), a subject was designated apparently healthy if he had no chronic illness, was a nonsmoker, did not report heavy drinking, was nonobese, and used fewer than six medications. Subjects were excluded from analyses if their T concentrations were unavailable, if their apparent health status could not be determined, if they reported the use of T preparations, or if they had been diagnosed with prostate cancer, because its treatment by hormone suppression therapy could not be ruled out as a potential artificial source of T decline.
Analytic sample
Of the 1688 subjects whose T concentrations were available at T1, 13 had a prior diagnosis of prostate cancer, and an additional eight were missing apparent health data, leaving 1667 subjects for analysis. Of these men, 954 and 591 at T2 and T3, respectively, had T and apparent health data and no diagnosis of prostate cancer. At T2, seven of these subjects reported using T preparations; T2 data on these men and data on the four who had a T3 visit were excluded as were T3 data on an additional three subjects who reported use of T preparations at T3 only. We therefore retained T2 data on 947 (57%) and T3 data on 584 (35%) of the 1667 subjects included in the analytic sample. The median time between T1 and T2 observations was 8.9 yr (range, 7.1–10.4 yr) and between T2 and T3 was 6.3 yr (range, 5.6–7.9 yr).
Statistical analysis
As a result of skew in the distributions of hormones, we analyzed their natural (base e) logarithms. Graphic analyses using nonparametric smoothing (35) was used to assess the general trend in log hormone concentrations with age. To formally estimate trends, mixed-effects regression models (36) with subject-level intercepts and slopes were used; this method is a generalized form of linear regression analysis that allows for repeated measures on each subject while accounting for the substantial variation across subjects in both the overall average T level and the trajectory of T concentrations with age. As has previously been reported (2) for data collected at T1 and T2, within-subject T decline with age appears to outstrip cross-sectional age-related decreases in T levels. Consequently, for regression models of TT and FT on covariates, we partitioned subject age into two components: baseline age and "aging," the latter denoting calendar time since study entry. The unadjusted mean decline of serum hormones associated with aging we refer to as the apparent aging effect.
The final statistical models presented subsequently were constructed in three steps. First, covariates were classified into groupings based on whether they more represented aspects of "health," broadly speaking, or rather aspects of "lifestyle." Then, each group of covariates was considered independently to assess the effect of each individual covariate on hormone levels, and those that remained significant when controlling for other members of the group were retained. Finally, these streamlined covariate groups were combined into a larger whole, and those additional variables that were made insignificant by the joining of the groups were removed sequentially until a parsimonious model containing only variables that were significant both within their specific group and in the larger model remained.
The health covariate group was comprised of those variables denoting general well-being (comorbidities, self-reported health, and medications), whereas the lifestyle group was composed of life circumstances and modifiable risk factors (BMI and waist-to-hip girth ratio, smoking, alcohol consumption, diet, energy expenditure, education, household income, and marital status). LH was considered a component of the "health" group. (A third covariate group—"design," which included assay batch, month and time of day of subject interview, and town of residence—was considered to eliminate the potential for systematic bias in results.)
Covariates were allowed to vary with time and were treated as internal time-dependent predictors (37). For a covariate with associated regression estimate ß*, we estimated the corresponding percent change in mean outcome by the monotonic transformation 100 x (eß*–1). Results were considered statistically significant if null hypotheses could be rejected at the 0.05 level. The significance of effects was evaluated using Wald and likelihood ratio tests.
|
Results |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
. The baseline median and interquartile range TT were 504 and 394 to 620 ng/dl (17.5 and 13.7–21.5 nmol/liter). Over the course of study follow-up, there were substantial increases in chronic illness and relative weight. There was also a marked increase in the proportion of subjects reporting polypharmacy and a dramatic decrease in the proportion of subjects who smoked cigarettes.
|
). We observed consistent evidence that accelerations in within-subject declines in T levels were associated with contemporaneous changes in health and lifestyle factors. Several simple examples are provided in Table 2; for instance, those subjects who exhibited no chronic illness both at T1 and T2 had, on average, –7.3% lower FT levels at T2 than at T1. By comparison, those subjects who had no chronic illness at baseline but experienced one or more illnesses at T2 exhibited greater FT decline over the intervening time so that their T2 FT levels were –13.1% lower than at baseline. Although proportionate declines in FT with time appeared to be greater in magnitude than the corresponding changes in TT, the apparent effects of changes in health and lifestyle status on the two quantities were roughly comparable.
|
|
Mixed-effects regression models controlling only for baseline age yielded estimated declines of –14.5% (95% confidence interval, –16.3% to –12.6%) TT and –27.0% (–29.1% to –25.0%) FT per decade of aging (estimates are obtained by transforming regression coefficients from models on the logarithmic scale; see Subjects and Methods). The corresponding estimated trends including only subjects who were apparently healthy were less sharp; among such subjects, we observed a –10.5% (–14.0% to –7.0%) decline in TT and a –22.8% (–26.9% to –18.7%) decline in FT per decade of aging, indicating that a substantial proportion of the apparent aging effect over all subjects was attributable to changes in health status.
Figure 2 provides a pictorial representation of this phenomenon in the context of obesity. Subjects are, by virtue of the inclusion of time-dependent covariates, permitted to "jump the tracks" from one health state to another as time moves forward. A minority of subjects who are nonobese become obese over time and exhibit contemporaneous acceleration in TT decline. Thus, although a naive estimation of the apparent aging effect among subjects whose BMI climbs above 30 kg/m2 would show rapid declines in TT, the multivariate model partitions this decline into components associated with aging itself as well as the additional acceleration that may be associated with weight gain. In this case, in models including only effects for aging and obesity, the model-estimated additional decline in TT levels associated with moving from a nonobese to an obese state (12%) is comparable to that associated with 10-yr aging among subjects whose obesity status is stable (13%).
|
; nonsignificant effects have been omitted as described previously (in particular, covariates in the "design" group displayed little influence in multivariate models and are not included in these results). Even when all other covariate effects were controlled, T declines associated with aging were substantial, roughly –10.1% decline in TT (95% confidence interval, –12.5 to –7.9) and –23.8% (–26.3 to –21.3) decline in FT per decade of aging; these trends are comparable to the estimates obtained from data on apparently healthy subjects reported previously. Substantial changes in health or lifestyle were associated with marked acceleration of T decline. For instance, the estimated decline in TT levels associated with smoking cessation when all other effects were controlled was –8.6% (–11.9 to –5.2) TT and –7.8% (–11.5% to –4.1%) FT; these estimates imply that smoking cessation is, on average, associated with a decline in TT levels equivalent to that associated with roughly 8 yr of aging [the positive association between smoking and T levels has been reported by multiple authors (17, 18, 19)]. Likewise, the onset of diabetes was associated with substantially accelerated T decline, as was polypharmacy, and relative weight gain was associated with a nearly –2% decrease in TT concentrations and a –0.6% decrease in FT per kilogram per square meter increase in BMI.
|
)—and, with the exceptions of age and BMI, the magnitude of the estimated regression effect for any particular covariate was smaller for SHBG than for FT or TT. We observed that although LH was significantly associated with TT at baseline, it was nonsignificant in longitudinal models when age effects were taken into account, and its inclusion had little impact on the other effects presented in Table 3.
To present the estimated relative contributions of the covariate groups to our estimates of TT change over time, we calculated the absolute value of the contributions of each to subject-level changes, from T1 to T3, to changes in the final linear predictor of log TT. The individual covariates contributing to each group are those listed in Table 3. This process was applied to data from all subjects who had complete data at T1 and T3, and the relative contributions were standardized to a percentage scale. The resulting percentages may be displayed in two dimensions using a triangular coordinate system (38); this display is presented in Fig. 3. Each subject is depicted as a single point and the relative contributions of aging, health, and lifestyle to changes in his model-estimated mean TT are visually expressed by that point’s relative proximity to the vertices of the plot. The clustering of points along the age-lifestyle axis (i.e. the line where health has no influence on model fit) is a function of the fact that all subjects age and most experience some lifestyle change (at least in terms of BMI), whereas the health variables tend to remain more stable, in part as a result of their discrete nature. We may surmise from the general scatter of the data that although aging per se remains powerful in predicting TT decline, there are subjects for whom health and/or lifestyle factors exhibit considerable influence, and that for many of these subjects, one or more of those factors are likely acting in concert. Indeed, among the 80 subjects for whom lifestyle factors are assigned at least 50% of model-estimated T decline by this method (i.e. the 80 subjects whose data are closest to the "lifestyle" vertex in Fig. 3), 33 (41%) were smokers at T1 who had quit smoking by T3, 39 (49%) had been employed at baseline but were no longer at T3, and 13 (16%) were married at baseline but later widowed; over the course of the study, these subjects averaged a 3.3 kg/m2 increase in BMI. By comparison, among all subjects who had data at T3, 13% were smokers who quit, 36% were employed and left their jobs, and 4% were married at baseline and widowed at T3; these subjects had a mean increase in BMI of 1.2 kg/m2.
|
|
Discussion |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
We have noted, however, that although the effects presented for FT are similar in magnitude to those presented for TT, the apparent age trend is sharper in FT than in TT, so that the relative contributions of health and lifestyle to FT decline vis-à-vis that of aging appear to be smaller than their contributions to declines in TT. For instance, the decline in TT levels associated with widowhood is comparable to that of 10 yr of normal aging, whereas the corresponding decline in FT levels is closer to that associated with 4 yr of normal aging.
Given the substantial number of effects in even the relatively simple models presented here, mean declines in T levels will likely exhibit great intersubject variation by patterns in covariate data. Although the proportion of subjects who smoke declines rapidly over time, it is not immediately clear how many subjects who stop smoking experience contemporaneous weight gain (or are widowed or take on a new job, and so on), and thus an estimate of the cumulative contribution of health and lifestyle factors to T decline over the entire study cohort is difficult to obtain. A common approach expresses these contributions as the proportion of variance "explained" by covariate clusters using pseudo-R2 statistics. These, however, have troublesome performance properties (39) and are of little use here because they are dominated by aging effects as a result of the fact that every subject ages, although other factors may remain unchanged. We have therefore chosen the graphic summaries presented in Figs. 2 and 3, which have the ability to depict T decline at the subject level. These demonstrate that, although aging effects predominate overall, health and lifestyle outstrip aging effects in a substantial number of subjects.
Some limitations of this investigation should be acknowledged. As noted previously, the technology by which SHBG was measured at T3 differed from that used at T1/T2. Although validation studies indicated that values obtained by the two methods were comparable (8), we cannot completely rule out the potential for design artifacts influencing estimates of trends in SHBG. We have, however, performed sensitivity analyses that indicate that multivariate results as presented in Table 3 would be similar if FT and SHBG data were restricted to T1 and T2. It also bears mentioning that although age has been measured with precision, constructs such as health and lifestyle are by definition incomplete because they are bound by the limitations of an epidemiological study conducted in a population of community-dwelling men; at the same time, emerging evidence of a population-level decline in serum T over calendar time (40) implies that estimates of T declines associated with male aging may themselves be biased. As such, the overall contributions of true health and lifestyle may exceed even the marked effects described in this study.
The results presented here suggest that although hormone declines appear to be an integral aspect of the aging process, rapid declines need not be dismissed as inevitable. Further investigation may reveal opportunities for primary and secondary prevention of T decline focused on modifiable health and lifestyle characteristics.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Disclosure Statement: The authors have nothing to disclose.
First Published Online December 5, 2006
Abbreviations: BMI, Body mass index; CES-D, Center for Epidemiologic Studies–Depression; CV, coefficient of variation; FT, free T; T, testosterone; TT, total serum T.
Received August 23, 2006.
Accepted November 28, 2006.
|
References |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  C. A. Allan, B. J. G. Strauss, H. G. Burger, E. A. Forbes, and R. I. McLachlan Testosterone Therapy Prevents Gain in Visceral Adipose Tissue and Loss of Skeletal Muscle in Nonobese Aging Men J. Clin. Endocrinol. Metab., January 1, 2008; 93(1): 139 - 146. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Endocrinology | Endocrine Reviews | J. Clin. End. & Metab. |
| Molecular Endocrinology | Recent Prog. Horm. Res. | All Endocrine Journals |
