Secular Decline in Male Testosterone and Sex Hormone Binding Globulin Serum Levels in Danish Population Surveys

Abstract

Context: Adverse secular trends in male reproductive health have been reported to be reflected in increased testicular cancer risk and decreased semen quality in more recently born men. These secular trends may also be reflected by changes in Leydig cell function.

Objective: The objective of the study was to examine whether an age-independent time trend in male serum testosterone levels exists.

Design and Setting: Testosterone and SHBG were analyzed in 5350 male serum samples from four large Danish population surveys conducted in 1982–1983, 1986–1987, 1991–1992, and 1999–2001. Free testosterone levels were calculated. The effects of age, year of birth, and time period on hormone levels were estimated in a general linear statistical model.

Main Outcome Measures: Testosterone, SHBG, and calculated free testosterone levels in Danish men in relation to age, study period, and year of birth were measured.

Results: Serum testosterone levels decreased and SHBG levels increased with increasing age. In addition to this expected age effect, significant secular trends in testosterone and SHBG serum levels were observed in age-matched men with lower levels in the more recently born/studied men. No significant age-independent effect was observed for free testosterone. Adjustment for a concurrent secular increase in body mass index reduced the observed cohort/period-related changes in testosterone, which no longer were significant. The observed cohort/period-related changes in SHBG levels remained significant after adjustment for body mass index.

Conclusions: The observed age-independent changes in SHBG and testosterone may be explained by an initial change in SHBG levels, which subsequently lead to adjustment of testosterone at a lower level to sustain free testosterone levels.

THERE IS GENERAL agreement that male serum testosterone levels decline with age, but the data have not been conclusive. Cross-sectional surveys have demonstrated very different rates of decline, ranging from 0.2 to 0.8%/yr (1–6), and some have even failed to observe a significant decrease in testosterone levels with increasing age (7–11). Furthermore, longitudinal studies have shown more steep age-related declines in serum testosterone than cross-sectional studies (12, 13). A suggested explanation for this apparent discrepancy between cross-sectional and longitudinal studies on age-related declines has been that poor health might accelerate an age-related decline in individual longitudinal testosterone levels (13). Alternatively, cross-sectional data on the older age ranges may be biased toward containing a higher proportion of healthy men, compared with the younger age ranges, implying that unhealthy men are less likely to become old. However, yet another alternative explanation could be that a secular decline in testosterone levels exists, because men studied 20 yr ago had higher serum testosterone levels than men of the same age studied today. The existence of such an age-independent time trend may blunt the effect of increasing age in cross-sectional studies.

If such a secular decline in male testosterone levels exists, this may not only lead to misinterpretation of the effect of age and compromise the durability of normal reference data, but also more importantly, it may be a sign of general changes in male reproductive health. A recently published study showing an age-independent population-level decline in male serum testosterone levels in American men (14) strengthens the suspicion that population trends in male testosterone levels may exist in certain countries, and further scrutiny of these effects is warranted.

SHBG has more conclusively been shown to increase in men with increasing age in both cross-sectional (3, 5, 15, 16) and longitudinal (13) studies. However, as for testosterone, age-independent time trends may exist for SHBG levels and may blunt or even exaggerate the associations between age and hormone levels, depending on the interaction between age and period on SHBG serum levels.

To study age and secular trends in testosterone levels among Danish men, we analyzed serum samples from more than 5300 men participating in four large population-based surveys conducted at Glostrup University Hospital (Glostrup, Denmark) during 1982–2001 for testosterone and SHBG and calculated free testosterone using the equation by Vermeulen et al. (17).

Subjects and Methods

Study population

The present study includes available serum samples from 5350 male participants of four large population-based surveys conducted at the Research Centre for Prevention and Health, Glostrup University Hospital, (Glostrup, Denmark) during 1982–2001, namely the three Danish Monitoring Cardiovascular (MONICA) risk factors cohorts examined in 1982–1983 (MONICA I), 1986–1987 (MONICA II), and 1991–1992 (MONICA III) and the Inter99 cohort examined in 1999–2001 (18, 19). An overview of the different cohorts and the number of serum samples available for this study is given in Table 1. In contrast to the MONICA studies, the Inter99 study was an intervention study aimed to look at effect of changes in lifestyle. However, the Inter99 samples used for this study was from baseline before intervention. Blood samples were taken in the morning after an overnight fast, and serum was stored in aliquots at −20 C. Samples from some of the surveys had been thawed for other analyses before this study, whereas others had been kept in the freezer exclusively. Information on year of birth, age, sex, and body mass index (BMI; weight/squared height) of the participants was available from all surveys.

TABLE 1.

Number of participants in the MONICA I-III and Inter99 surveys from whom serum was available for hormone analysis stratified by age and period of birth

		Age (yr)
Survey (study period)	Birth period	30	40	50	60	70	Total
MONICA I (1982–84)	1952	483
	1942		496
	1932			490
	1922				461
	All						1930
MONICA II (1986–87)	1956	174
	1946		192
	1936			183
	1926				191
	All						740
MONICA III (1991–92)	1961	201
	1951		201
	1941			199
	1931				203
	1921					200
	All						1004
Inter99 (1999–2001)	1969–1970	148
	1959–1960		589
	1949–1950			681
	1939–1940				258
	All						1676
Total		1006	1478	1553	1113	200	5350

Hormone measurements

Testosterone was measured by time-resolved fluoroimmunoassays (DELFIA; Wallac Oy, Turku, Finland) with a detection limit of 0.23 nmol/liter and intra- and interassay coefficients of variation were less than 12%. SHBG was measured by time-resolved immunofluorometric assays (DELFIA; Wallac). The detection limit was 0.23 nmol/liter, and the intra- and interassay coefficients of variation were less than 8%. The detection limit was defined as the concentration corresponding to the signal that is 2 sd above or below the mean of the zero standard measurement in the immunofluorometric assays and competitive immunoassays (fluoroimmunoassays), respectively.

All samples were analyzed during the same period (spring to autumn 2004), and samples from the different surveys were analyzed and mixed in the different assay runs to eliminate any influence of assay variation.

Validation of the integrity of hormone levels in stored samples

At the time of hormone analysis, samples had been stored between 3 and 21 yr at −20 C.

First, because an aliquot of 434 samples (not included in the present study) that had been analyzed in the same laboratory 10 yr previously (in 1994 when the samples were new) was available (11), this allowed a direct comparison of the hormone levels obtained in fresh samples and the same samples after 10 yr of storage at −20 C. This approach allowed us to test whether storage for 10 yr at −20 C resulted in major changes in the levels of the hormones studied, whereas it would not be possible to discriminate whether minor changes were due to storage or assay variation because all immunoassays have a certain day-to-day variation, which may be even higher over a long time span due to the use of different batches of regents over time. For SHBG the same assay format was used 10 yr ago, whereas testosterone in 1994 had been analyzed in a different assay format than used in the present study. For SHBG the median of the ratio between the first and second measure was 1.04 (95% confidence interval 0.88–1.21), and the correlation coefficient between the two measures was 0.98. For testosterone the median of the ratio between the first and second measure was 0.95 (95% confidence interval 0.63–1.47), and the correlation coefficient was 0.83. The difference observed between the two measurements was for both hormones within the expected day-to-day assay variation (see Fig. 1).

Fig. 1.

Comparison of testosterone (left) and SHBG (right) levels measured in the same samples 10 yr apart (measured the first time in 1994 and measured again in 2004). Different testosterone assays were used at the two times of measurement, whereas the same SHBG assay format was used at both time points. The drawn line represents the regression line (and 95% mean confidence interval) between the two measurements. The dotted line represents the identity line.

A previous publication (20) suggested that artificially increased testosterone levels due to a decrease in levels of SHBG with increasing storage time might be observed with testosterone assays susceptible to the concentration of SHBG. Although we did not expect and did not find a decrease in SHBG levels with increasing storage time, we nevertheless tested the susceptibility of the testosterone immunoassay used in this study for variations in SHBG levels. Samples with SHBG concentrations varying from 10 to 193 nmol/liter were spiked with a fixed amount of testosterone and the recovery of testosterone was measured. The recovery of testosterone was not significantly different between samples with low and high concentrations of SHBG [mean recovery (sd) of 110% (12) and 107% (15), respectively]. To test further whether changes in the samples matrix occurring during storage might affect the measured testosterone levels, we also measured the recovery of testosterone spiked to serum samples stored at −20 C for different time lengths (from less than 3 months to 21 yr). No significant difference in recovery was observed.

In the samples from the four population surveys, we generally measured higher hormone levels in the samples that had been stored the longest. Thus, we were concerned about whether any evaporation had occurred during previous handling of the samples, which could lead to concentration of the samples. We therefore measured serum Na⁺ in 25 randomly selected samples from each of the surveys (total n = 100 samples) as an estimate of whether any concentration of samples had occurred. Serum Na⁺ was measured by indirect potentiometry (ISE system; Roche/Hitachi, Basel, Switzerland) at the clinical chemistry department at our hospital. We found a significant negative correlation between year of sampling and the Na⁺ concentration, indicating that evaporation of samples during storage or previous handling may have occurred (Fig. 2). The mean Na⁺ concentrations in the samples from the MONICA I, MONICA II, MONICA III, and Inter99 surveys were 172, 169, 162, and 154 mmol/liter, respectively, which all were above the normal range for serum Na⁺ concentrations. To normalize serum concentration to a normal mean serum Na⁺ concentration of 140 mmol/liter, hormone levels measured in samples from the MONICA I, MONICA II, MONICA III, and Inter99 surveys were multiplied with a correction factor of 0.81, 0.83, 0.86, and 0.91, respectively (and thereby adjusted for any effect of evaporation). The variation in Na⁺ levels was larger than the normal biological variation for Na⁺, indicating that samples from the same sampling period were not identically affected by storage or handling of the samples. Thus, by using a general correction factor for all samples from the same sampling period, some samples may be undercorrected and some may be overcorrected. Unfortunately, serum was not available for Na⁺ measurements in all samples, and thus, individual adjustment was not possible. However, due to the large number of samples available and the fact that we in this study were interested in general trends rather than individual levels, it can be justified to use a general correction factor for each sampling period.

Fig. 2.

Correlation between date of sampling and serum Na⁺ level measured. The shaded area represents the normal range for serum Na⁺.

Statistical methods

Hormone levels were initially corrected for evaporation (see above). Free testosterone was calculated from the testosterone and SHBG concentrations using the method by Vermeulen et al. (17), with the assumption of an average serum albumin concentration of 43 g/liter. Hormone levels of the different age groups were compared using one-way ANOVA with Bonferroni post hoc test.

To disentangle the effects of age, study time (period), and year of birth (cohort), age-period-cohort modeling was used (21). Hormone levels were natural logarithm transformed to obtain approximate homoscedacity and an approximate normal distribution of residuals. The following general linear model was used:  

$$\mathrm{ln(hormone}_{ijk})\ {=}\ {\alpha}_{i}\ {+}\ {\beta}_{j}\ {+}\ {\gamma}_{k}\ {+}\ {\epsilon}_{ijk}$$

where α_i represents the age effect in the i^th age group, β_j the period effect in the j^th period, and γ_k the cohort (birth year) effect of the k^th cohort. The remainder term ε_ijk is the residual representing the noise. The model included age (30, 40, 50, 60, and 70 yr), period (1982–1983, 1986–1987, 1991–1992, and 1999–2001), and year of birth (1921–1926, 1931–1939, 1940–1946, 1949–1952, 1956–1961, and 1969–1970) as categorical variables. An inherent problem in age-period-cohort modeling, stemming from the fact that period can be calculated as the sum of a man’s age and birth year (and vice versa) is that a linear trend in mean level of the response cannot be attributed specifically to the period or the cohort (22). This is the fundamental nonidentifiable problem to which there is no solution. The adequacy of the model can be investigated by introducing a general interaction term. In the presence of a general interaction term, the model does not lend itself to simple interpretation in terms of age, period, or cohort effects. A particular age-cohort interaction is obtained by including period in the model with the restriction that the first and last category are identical. Likewise a particular age-period interaction is obtained by including cohort in the model with the same restriction. This constitutes a main-effects model in which the effects of age, period, and cohort, apart from the linear trend, can be identified. In subsequent analyses, BMI was included in the statistical model as an additional explanatory variable to investigate whether the hormonal changes could partly be explained by a change in BMI. The final models were subjected to standard checks of the residuals.

Results

Median testosterone, SHBG, and free testosterone levels in each age group for all periods together as well as stratified according to study period and year of birth are shown in Figs. 3 and 4. In Fig. 3 the data for testosterone are plotted both without and with adjustment for evaporation.

Fig. 3.

Median testosterone levels plotted in relation to age (A and D) and with stratification by study period (B and E) and year of birth (C and F). A–C, Data without adjustment for evaporation during storage; D–F, data after adjustment for evaporation during storage.

Fig. 4.

Median SHBG (A-C) and free testosterone (D–F) serum levels plotted in relation to age (A and D) and with stratification by study period (B and E) and year of birth (C and F).

Testosterone

When testosterone levels were not stratified, a modest decrease in testosterone levels with increasing age was observed (Fig. 3D). Bonferroni post hoc test showed the levels at 30 yr > 40 yr > 50 yr = 60 yr = 70 yr. However, when data were plotted stratified according to study period or year of birth, a steeper decrease in testosterone levels with increasing age was evident within each category. When stratified according to study period, the age-related changes differed between the different study periods. Thus, whereas the levels and the age-related decline was very similar between the different study periods for the 30- and 40-yr-old men, this was not true for the older men (Fig. 3E). When stratified according to birth cohort, the same age-related decline was observed within all cohorts, revealing a steady decline over all age groups. However, a significant cohort effect was evident with higher levels observed in the oldest cohort (Fig. 3F). This cohort effect was evident between the men born in the 1920s, 1930s, and 1940s, whereas there was no difference between later-born cohorts.

The influence of age, birth year, and study period on serum testosterone levels were further investigated in age-period-cohort general linear models. The parameter estimates of the age-period-cohort models are shown in Table 2. In the model focusing on the age-cohort relation (linear component of period nullified), the estimated age-related decline in serum testosterone on average varied from −0.7 to −1.0% per year of age with the steepest decline from 30 to 40 yr of age. In the age-period model (linear component of birth year nullified), the estimated age-related decline in serum testosterone was slightly smaller, varying from −0.5 to −0.7% per year of age. The age-cohort model also showed a statistically significant decline in serum testosterone with later year of birth. In the age-period model, a declining trend in serum testosterone levels with later study year was found, but this trend was statistically significant only between the oldest and latest study periods.

TABLE 2.

Testosterone as dependent (Ln transformed) variable in age-period-cohort general linear models

	Parameter estimate	Change, %a	95% CIa	P value
Full age-period-cohort model, period restrictedb
Age, yr				< 0.001
70	−0.371	−31	−38%; −24%	< 0.001
60	−0.249	−22	−27%; −17%	< 0.001
50	−0.179	−16	−20%; −13%	< 0.001
40	−0.104	−10	−13%; −6%	< 0.001
30	Ref	Ref
Birth year (cohort)				0.008
1969–1970	−0.149	−14	−22%; −5%	0.003
1956–1961	−0.128	−12	−18%; −6%	< 0.001
1949–1952	−0.126	−12	−17%; −6%	< 0.001
1940–1946	−0.115	−11	−16%; −5%	< 0.001
1931–1939	−0.082	−8	−13%; −3%	0.003
1921–1926	Ref	Ref
Study period (period)				0.792
Full age-period-cohort model, cohort restrictedb
Age, yr				< 0.001
70	−0.252	−22	−29%; 15%	< 0.001
60	−0.157	−15	−19%; −10%	< 0.001
50	−0.123	−12	−15%; −8%	< 0.001
40	−0.071	−7	−11%; −3%	0.001
30	Ref	Ref
Birth year (cohort)				0.163
Study period (period)				0.029
1982–1984	0.053	5	2%; 9%	0.004
1986–1987	0.040	4	−0.2%; 8%	0.064
1991–1992	0.015	2	−2%; −5%	0.438
1999–2001	Ref	Ref

CI, Confidence interval.

^a

Percent change in relation to the reference category.

^b

Age, birth year, and study period are perfectly confounded variables, and the linear component of birth year and study period can therefore not be separated. To account for this problem, models are presented in which either birth year or study period was entered in the model with the restriction that the first and last categories are identical to eliminate the linear component of the restricted variable.

SHBG

The graphical displays for SHBG show a significant increase with increasing age when data were not stratified with median SHBG levels being more than 50% higher in the 70-yr age group, compared with the 30-yr age group (Fig. 4A). When the data were stratified according to study period, a similar age-related increase was observed in all four study periods, although with differences in absolute levels indicating a period effect, with higher levels in the earliest study periods (Fig. 4B). When the data were stratified according to birth cohorts, most of the apparent age-related increase in SHBG levels seemed to be due to a cohort effect, with higher levels in older cohorts (Fig. 4C). However, the age-related trend in SHBG differed between the different cohorts.

The parameter estimates from the age-period-cohort models for SHBG are shown in Table 3. Age, study period, and birth year were all statistically significant confounding variable for SHBG serum levels. The estimates of the age-related increase in serum SHBG levels differed quite significantly between the age-cohort and the age-period models, but in both models the increase seemed to set off between 40 and 50 yr of age and onward. In the age-cohort model, the estimated average increase was 0.6–0.7% per year from 40 yr of age and onward; in the age-period model, the estimated average increase was 1.2–1.5% per year during the same age range. In the age-cohort model, a significant decrease in SHBG levels between the cohort born between 1921 and 1926 and the cohort born between 1931 and 1939, with a further but more moderate decline in more recent cohorts, was seen. In addition, a trend of lower SHBG levels in the more resent study period was found in the age-period model, with the levels obtained in the 1999–2001 study period being 12% lower than the levels obtained in the 1982–1984 study period.

TABLE 3.

SHBG as dependent (Ln transformed) variable in age-period-cohort general linear models

	Parameter estimate	Change, %a	95% CIa	P value
Full age-period-cohort model, period restrictedb
Age, yr				<0.001
70	0.211	23	12%; 36%	<0.001
60	0.181	20	13%; 27%	<0.001
50	0.102	11	6%; 15%	<0.001
40	0.006	1	−3%; 5%	0.759
30	Ref	Ref
Birth year (cohort)				<0.001
1969–1970	−0.306	−26	−33%; −19%	<0.001
1956–1961	−0.284	−25	−30%; −19%	<0.001
1949–1952	−0.258	−23	−28%; −18%	<0.001
1940–1946	−0.208	−19	−24%; −14%	<0.001
1931–1939	−0.161	−15	−19%; −10%	<0.001
1921–1926	Ref	Ref
Study period (period)				<0.001
Full age-period-cohort model, cohort restrictedb
Age, yr				<0.001
70	0.457	58	45%; 72%	<0.001
60	0.371	45	38%; 53%	<0.001
50	0.219	24	19%; 30%	<0.001
40	0.076	8	4%; 12%	<0.001
30	Ref	Ref
Birth year (cohort)				0.003
Study period (period)				<0.001
1982–1984	0.111	12	8%; 16%	<0.001
1986–1987	0.053	5	1%; 10%	0.011
1991–1992	−0.010	−1	−5%; 3%	0.591
1999–2001	Ref

CI, Confidence interval.

^a

Percent change in relation to the reference category.

^b

Age, birth year, and study period are perfectly confounded variables, and the linear component of birth year and study period can therefore not be separated. To account for this problem, models are presented in which either birth year or study period was entered in the model with the restriction that the first and last categories are identical to eliminate the linear component of the restricted variable.

Free testosterone levels

Free testosterone levels decreased with increasing age, and the rate was the same irrespective of whether the data were stratified according to period of sampling, birth cohort, or not stratified (Fig. 4, D and E). Thus, in contrast to testosterone and SHBG, no significant difference between the different study periods or between the different birth cohorts was evident for free testosterone. This was confirmed in the age-period-cohort models in which neither birth year nor study period was a significant confounding variable (Table 4). The average decline in free testosterone varied from −1.0 to −1.2% per year of age.

TABLE 4.

Free testosterone as dependent (Ln transformed) variable in age-period-cohort general linear models

	Parameter estimate	Change, %a	95% CIa	P value
Full age-period-cohort model, period restrictedb
Age, yr				<0.001
70	−0.517	−40	−46%; −34%	<0.001
60	−0.374	−31	−54%; −27%	<0.001
50	−0.256	−23	−26%; −19%	<0.001
40	−0.122	−11	−15%; −8%	<0.001
30	Ref	Ref
Birth year (cohort)				0.897
1969–1970	0.012	1	−8%; 11%	0.794
1956–1961	0.021	2	−4%; 9%	0.541
1949–1952	0.014	1	−5%; 8%	0.654
1940–1946	−0.003	0	−6%; 6%	0.924
1931–1939	0.01	1	−4%; 6%	0.714
1921–1926	Ref	Ref
Study period (period)				0.325
Full age-period-cohort model, cohort restrictedb
Age, yr				<0.001
70	−0.528	−41	−46%; −36%	<0.001
60	−0.382	−32	−35%; −28%	<0.001
50	−0.262	−23	−26%; −20%	<0.001
40	−0.126	−12	−15%; −8%	<0.001
30	Ref	Ref
Birth year (cohort)				0.913
Study period (period)				0.512
1982–1984	−0.006	−1	−4%; 3%	0.733
1986–1987	0.015	2	−2%; 6%	0.460
1991–1992	0.016	2	−2%; 5%	0.397
1999–2001	Ref	Ref

CI, Confidence interval.

^a

Percent change in relation to the reference category.

^b

Age, birth year, and study period are perfectly confounded variables, and the linear component of birth year and study period can therefore not be separated. To account for this problem, models are presented in which either birth year or study period was entered in the model with the restriction that the first and last categories are identical to eliminate the linear component of the restricted variable.

Body composition

BMI was used as a crude index of the body composition. Serum levels of testosterone and SHBG were associated with BMI, even after adjustment for age, period, and cohort reflected in this study by a significant negative correlation between testosterone and SHBG toward BMI (P < 0.01). Because BMI may change with increasing age as well as changes in year of birth, we also included BMI in the statistical model to evaluate how much of the age- and cohort/period-related changes could be explained by concurrent changes in BMI. The estimated changes in hormone levels after adjustment also for BMI are given in Table 5. When adding BMI as a confounding variable in the age-period-cohort models, birth year and study period no longer remained statistically significant confounding variables for serum testosterone levels and were therefore omitted from the final model. In contrast, although the estimated effect of birth year and period on serum SHBG levels diminished when adjusted for changes in BMI, they remained statistically significant confounding variables. Both testosterone and SHBG serum levels were estimated to decrease −4% per increase in BMI. The estimated age-related decline in testosterone levels diminished when the effect of increasing BMI with increasing age was adjusted for. Thus, after adjustment for BMI, the estimated decline in serum testosterone per increasing year varied between −0.3 and −0.5%. In contrast, the estimated age-related effect on SHBG was even more pronounced when the confounding effect of BMI was adjusted for. After adjustment for BMI, a very steady age-related decline in free testosterone of −1% per year was estimated.

TABLE 5.

Effect of BMI in age-period-cohort general linear models of testosterone, SHBG, and free testosterone serum levels

	Parameter estimate	Change, %a	95% CIa	P value
Testosteroneb
Age, yr				<0.001
70	−0.182	−17	−21%; −11%	<0.001
60	−0.083	−8	−11%; −5%	<0.001
50	−0.080	−8	−11%; −5%	<0.001
40	−0.049	−5	−8%; −2%	<0.001
30	Ref	Ref
BMI	−0.037	−4	−4%; −3%	<0.001
SHBG (period restricted)c
Age, yr				<0.001
70	0.399	49	36%; 63%	<0.001
60	0.353	42	35%; 51%	<0.001
50	0.232	26	21%; 31%	<0.001
40	0.084	21	5%; 13%	<0.001
30	Ref	Ref
Birth year (cohort)				<0.001
1969–1970	−0.132	−12	−20%; −4%	0.004
1956–1961	−0.177	−16	−21%; −11%	<0.001
1949–1952	−0.178	−16	−21%; −11%	<0.001
1940–1946	−0.150	−14	−19%; −9%	<0.001
1931–1939	−0.124	−12	−16%; −7%	<0.001
1921–1926	Ref	Ref
BMI	−0.044	−4	−4%; −4%	<0.001
SHBG (cohort restricted)c
Age, yr	0.506	66	53%; 80%	<0.001
70	0.436	55	48%; 62%	<0.001
60	0.283	33	28%; 38%	<0.001
50	0.114	12	8%; 16%	<0.001
40	Ref	Ref
30				<0.001
Study period (period)				0.004
1982–1984	0.049	5%	2%; 9%	0.847
1986–1987	−0.004	0%	−4%; 3%	0.007
1991–1992	−0.049	−5	−8%; −1%
1999–2001	Ref	Ref	Ref	<0.001
BMI	−0.044	−4	−4%; −4%	<0.001
Free testosteroneb
Age, yr				<0.001
70	−0.488	−39	−42%; −35%	<0.001
60	−0.350	−30	−32%; −27%	<0.001
50	−0.227	−20	−23%; −18%	<0.001
40	−0.108	−10	−13%; −7%	<0.001
30	Ref	Ref
BMI	−0.019	−2	−2%; −2%	<0.001

^a

Percent change in relation to the reference age category or percent change per increase in BMI.

^b

Neither birth year nor study period were significantly confounding variables for serum testosterone and free testosterone when BMI was included in the table and were therefore not included in the final models for these two hormones.

^c

Age, birth year, and study period are perfectly confounded variables, and the linear component of birth year and study period can therefore not be separated. To account for this, models are presented where either one of them was entered in the model with the restriction that the first and last category are identical in order to eliminate the linear component of the restricted variable.

Discussion

We found a significant age-independent secular decline in male testosterone and SHBG serum levels in our study of 5350 Danish men from the general population born between 1921 and 1970 and studied between 1982 and 2001. Because factors acting early in life during fetal development as well as contemporary factors may affect reproductive hormone levels in adult life, it is relevant to analyze this observed decline in relation to both birth year and study period. The observation that the age-related changes in testosterone differed between the different periods but not between different birth year groups indicated that the observed age-independent trend in testosterone was related to a birth cohort effect rather than a period effect. The observed differences in the age-related changes in SHBG between different birth cohorts pointed to the existence of a period effect, presumably acting in concert with a birth cohort effect. The age-period-cohort models seem to support that both a period and a birth cohort effect may be in play, although these effects cannot be separated due to the inherent problem that study period and birth year are perfectly confounded when the age is matched.

The secular decline was most pronounced for SHBG and less so for testosterone. The observed age-independent decline in male serum testosterone levels could be explained by a concurrent secular increase in BMI. In contrast, the concurrent increase in BMI explained only part of the secular trend in male SHBG serum levels and the impact of BMI were more pronounced in the later cohorts, whereas changes in BMI seemed to play only a minor role in the changes seen in SHBG between the older cohorts. The fact that the free testosterone level did not seem to be affected by cohort or period effects indicates that whatever causes the secular trend, it apparently is primarily affecting the SHBG serum levels. We speculate that the secular decline in testosterone serum levels could be secondary to the decline in SHBG levels, simply adjusting the pituitary-gonadostat to a lower level to sustain the same level of free testosterone.

The regulation of SHBG is complex and only partly understood. Sex steroids stimulate SHBG production and secretion in vitro (22), and SHBG levels increase during pharmacological oral estrogen treatment (23–25) and decrease during oral androgen treatment (26). However, parenteral administration of sex steroids as well as physiological changes in sex steroids has only moderate effects on serum SHBG levels (26–29), and endogenous sex steroid levels presumably play a minor role in the regulation of SHBG. Correlations between serum sex steroids and SHBG levels are thus more likely due to the indirect regulation of sex steroid levels by serum SHBG levels than vice versa. Serum SHBG levels are negatively associated with obesity and various measures of insulin resistance (30) and has been suggested as a marker of the metabolic syndrome. Furthermore, insulin decreases SHBG production and secretion in vitro (22). It could be speculated that the observed secular decline in serum SHBG levels could be linked to increased incidence of obesity/metabolic syndrome in the later cohorts, although changes in BMI in our study could not explain the observed effects. On the other hand, BMI is a rather crude estimate of obesity. Also IGF-I and thyroid hormones have been indicated as regulators of serum SHBG levels (22, 31), but in the present study, information on these hormones was not available. Thus, the etiology of the observed secular decline in male SHBG serum levels remains unknown.

The secular decline in SHBG was most pronounced for the oldest cohorts, indicating that whatever caused this cohort effect seems to have leveled out in the more recent cohorts. Large changes in lifestyle as well as the environment occurred in Denmark during the 20th century with a general significant increase in standard of life along with an increased industrialization. A declining trend in male reproductive health manifested as an increase in testicular cancer, and a declining sperm quality has been reported for the same period in many Western countries (32–36). These adverse trends in male reproductive health and the observed changes in male reproductive hormone levels presented here could be interrelated.

The secular decline in SHBG and testosterone serum levels did not lead to a change in the level of free testosterone, which often is believed to determine the androgen activity. Nevertheless, a general decreased testosterone production will lead to lower intratesticular testosterone levels, and thus, the paracrine effects of testosterone within the testes may be blunted. This is in line with our previous findings of decreased serum SHBG and testosterone levels as well as decreased sperm concentration in men with BMI greater than 25 kg/m² (37). Furthermore, recently the existence of specific binding sites for SHBG at the cell membrane of steroid-responsive tissues has been shown (38), challenging the dogma that SHBG merely acts as a carrier protein. Binding of SHBG to its cell membrane receptor and subsequently to steroid hormone has been shown to initiate a steroid receptor-independent downstream signal through increased intracellular cAMP (for review see Ref. 39). The free hormone hypothesis, according to which only free steroids are biologically relevant, has been further challenged by the finding that SHBG may play an active role in the cellular uptake of steroids through binding and subsequently internalization of the protein/steroid complex via the cell membrane receptor megalin (40). Thus, the observed decline in serum SHBG levels may in itself have an impact on steroid action.

Studies on long-term trends like ours have inherent challenges related to the integrity of the samples or the methods of measurement over time. In our study design, the integrity of the samples over time was an important issue, which we acknowledge as a limitation of these kinds of studies. However, although the observed magnitude of the decline in SHBG (and testosterone) in later-born cohorts may be slightly affected by storage issues, the overall conclusion that a decline has occurred remains robust. An alternative design could be to measure samples as they are collected to avoid the issue of stability of samples during long-term storages. However, this approach is, on the other hand, vulnerable to the long-term integrity and stability of the methods of measurement over several decades. Thus, there is no simple solution to the design of studies on long-term effects of hormone levels. Nevertheless, taking all these difficulties into consideration, it is interesting that a recent American study adapting the later approach (e.g. measuring samples at different time points over a long period) also found a significant decline in serum testosterone levels in later-born men when matched by age (14). In contrast to our data, a secular decline in calculated free testosterone of similar magnitude to that in total testosterone was also observed in the American study, implying that no secular decline in SHBG had occurred in the American men (14). Thus, although a decline in total testosterone serum levels was observed in both the Danish and American study groups, the etiology for this decline may differ between the two populations. The age-independent changes in male serum testosterone levels observed in our Danish population and the American population may be too small to be of any clinical relevance for the individual man, but in a population-level perspective, it is alarming that changes of this magnitude can be detected over such a relatively short time, evolutionarily speaking. Similar population studies from other countries are much warranted to explore whether the observed trends in male total testosterone levels is a general trend in Western countries. Clarification of the causes for these changes, whether they are related to changes in lifestyle or environment, should provide information on which preventive action can be taken.

Acknowledgments

We thank the personnel at the Research Center for Prevention and Health for making the samples from their large population studies available for this study and skilled technicians Ole Nielsen and Birgitte Schou at the hormone laboratory, Department of Growth and Reproduction, Rigshospitalet, for their assistance and great efforts of running hormone analyses of all the samples.

This work was supported by the European Commission (contracts QLK4–1999-01422 and QLK4-CT-2002-00603, EDEN), the Danish Research Council (Grant 22-03-0198), and the Sven Andersen Foundation.

Disclosure Information: A.-M.A., T.K.J., A.J., J.H.P., T.J., and N.E.S. have nothing to declare.

Abbreviations

 
• BMI

  Body mass index;

 
• MONICA

  Monitoring Cardiovascular.