Context The literature provides no clear answer as to whether low endogenous testosterone increases risk of cardiovascular disease (CVD) in healthy men.
Objective Our purpose was to estimate the predictive value of testosterone for CVD and to identify study features explaining conflicting results.
Data Sources Articles were identified by a Medline and Embase search and citation tracking.
Study Selection Eligible were prospective population-based cohort and nested case-control studies of testosterone and atherosclerosis, stroke, myocardial infarction, ischaemic heart disease, death from coronary heart disease or mortality.
Data extraction Two independent researchers re-expressed associations of testosterone and CVD in a uniform manner to be used in meta-regression analyses for identification of study features explaining conflicting results, and to estimate the predictive value of testosterone for CVD.
Results and Conclusions 19 potentially eligible articles were identified. Overall, a weak independent association was found with an estimated summary RR of 0.89 for a change of one standard deviation in total testosterone level (95% CI 0.83 to 0.96). Age of study population and year of publication modified the relationship between testosterone and CVD. The estimated summary RR was 1.01 (0.95 to 1.08) for studies of men younger than 70 years of age, and 0.84 (0.76 to 0.92) for studies including men over 70 years of age. The latter studies showed a particular pronounced association if published after 1 January 2007. Results were largely confirmed by separate analyses of free- and bioavailable testosterone. The systematic review displayed no association between endogenous testosterone and risk for CVD in middle-aged men. In elderly men, testosterone may weakly protect against CVD. Alternatively, low testosterone may indicate a poor general health.
- cardiovascular diseases
- follow-up studies
Statistics from Altmetric.com
Testosterone concentrations in men decline with ageing and obesity, although it is uncertain whether this hormonal decline itself results in increased risk for cardiovascular disease (CVD).1 Testosterone has been assumed to have an important impact on risk of CVD. This assumption was initially based on differences in CVD between men and women, though at present on differences in testosterone level and presence or absence of CVD in men. The speculations on the impact of testosterone on the cardiovascular system have resulted in various prospective studies for decades,2–20 though a definitive answer as to whether testosterone affects risk for CVD has yet to be provided. One of the problems is that causal inference may be troublesome—that is, studies of patients with systolic chronic heart failure are difficult to interpret for the reason that heart failure as an illness affects sex steroid concentrations.21 Even straightforward intervention studies—for example, testosterone treatment in men with coronary heart disease, have been inconclusive.22 23 Yet, a better knowledge of the role of testosterone on the cardiovascular system is of importance in the light of potential reduction of CVD and the ongoing debate on testosterone supplementation in age- or obesity-related low serum testosterone.1
So far, previous reviews on testosterone and CVD have been narrative in nature.24–26 Therefore, we performed a quantitative approach through a systematic review according to MOOSE guidelines.27 We performed a meta-analysis to estimate the predictive value of endogenous testosterone for CVD, and meta-regression analyses to identify study features that modify the relationship—that is, explain heterogeneity across studies. Study features hypothesised to be potential modifiers were study design, testosterone assay, type of outcome studied, number or type of confounders for which adjustment was made, age of study population, follow-up period, country of study and year of publication, as the last item might, for example, reflect improvement of technology.
Data sources and study selection
Articles were identified by an Index Medicus (Medline) and Embase search, from reference lists and by citation tracking covering the period 1966 to September 2009. Key words were testosterone, androgens, oestradiol, longitudinal studies, prospective studies, cohort studies, CVDs, myocardial infarction, carotid arteries, aorta, mortality and electrocardiography. Articles were eligible if they described prospective population-based cohort or nested case–control studies on the relationship between, on the one hand, testosterone levels (total, free or bioavailable) and, on the other hand, vascular disease, stroke, ECG abnormalities, myocardial infarction, ischaemic heart disease, death from coronary heart disease or mortality. The search was limited to studies in men aged 45+ who were generally healthy. Studies on systolic chronic heart failure, patients undergoing dialysis or patients with HIV were excluded, as interpretation of these studies is complicated because an illness by itself may affect sex steroid concentrations.21
Data were extracted pertaining to testosterone levels (total, free or bioavailable), type of testosterone assay involved, study design (prospective population-based versus nested case–control), type of outcome studied (vascular disease, stroke, CVD, myocardial infarction, ischaemic heart disease, mortality), number and type of confounders for which adjustment was made, age of the study population (mean age, studies with men <70 years of age versus studies with men including those ≥70 years of age), follow-up period, country of study and year of publication. If investigators presented different follow-up periods on the same study population, the study with the follow-up period closest to the mean of other study populations of the meta-analysis was selected. If researchers presented a different outcome, the most specific and most common outcome was used for meta-regression analysis—for example, ischaemic heart disease death14 and CVD15 instead of mortality, and intimal medial thickness of the carotid artery instead of arterial stiffness index.11 Articles had to provide enough information to estimate an RR and a 95% CI or an approximation, such as an OR. We extracted the estimated RR, adjusted for the highest number of potentially confounding variables, from each of the original articles. Quality assessment of individual studies was replaced by exploration of sources of heterogeneity. Searches and analyses were performed independently by two investigators (JBR, AMM).
Summary estimate and standardisation of study-specific associations
The summary estimate of combined study populations was obtained by first estimating a coefficient (b) that represented the relationship between testosterone and CVD for each individual study, and subsequently, by estimating a weighted average of the coefficients. The weight of each study was calculated inversely to the variance estimate of the coefficient after re-expression of the SE in a uniform manner.28 For prospective cohort and nested case–control studies, ‘b’ represents the coefficient for the effect of one standard unit difference (to be specified) in testosterone level in a Cox proportional hazards, logistic regression or Poisson regression model.29 The re-expressed RR of CVD per specified uniform difference in testosterone level is therefore exp(b), assuming that the RR is constant during follow-up and the absolute risk is small. In the same manner, the 95% CI can be calculated as exp(b±1.96 SE). For studies that did not directly supply data that allowed the calculation of b and its SE, the computation methods described by Greenland were used.28 This calculation of the RRs (95% CI) shows an RR for a difference of one SD testosterone (in nmol/l). If a study provided an RR without sufficient specification of the difference in testosterone levels involved, tables from the same paper were used to estimate the difference in testosterone levels at issue. Papers describing the HIMS,19 CHIANTI20 and the MRFIT study2 did not provide enough information in text or tables to identify the difference in testosterone level for which the RRs were calculated. Therefore, data of testosterone distributions of the MrOs18 the Rancho Bernardo16 and a Belgian population study30 were used, respectively, to estimate the difference in testosterone levels of these studies. We postulated that these testosterone distributions could be used to identify the difference in testosterone level, if the mean age of study population, mean testosterone level and SD were largely similar. If the SD of the testosterone distribution was not provided in the original article, the mean SD of all included studies was used. Sensitivity analyses were performed to evaluate the influence of these estimations on final conclusions.31
Sources of heterogeneity
Sources of heterogeneity were identified by meta-regression analyses.28 32 Analyses were performed separately for total testosterone, free testosterone and bioavailable testosterone, thus only one association for each study was used per analysis in meta-regression analyses (and meta-analysis). With this approach, the logarithm of the study RR is regressed on study features of interest.
Subsequently, fixed-effect linear regression models were fitted by weighted least squares.28 The fit of the weighted regression model was evaluated by comparing the residual sum of squares to a X2 distribution. A small probability value indicates a poor fit. The importance of a study feature was evaluated according to the size of the b value as well as its CI. Study features that were consecutively included in meta-regressions as possible sources of heterogeneity were study design, testosterone assay involved, type of outcome studied (vascular disease, stroke, CVD, myocardial infarction, ischaemic heart disease, mortality), number or type of confounders (ie, obesity, smoking, lipids, blood pressure, diabetes, alcohol, physical activity, comorbidity, ethnicity) for which adjustment was made, age of study population, follow-up period, country of study and year of publication.
Subsequently, study features that significantly modified the relationship between testosterone and CVD were used to stratify summary estimates. The importance of a study feature, identified by meta-regression analyses, was confirmed by differences in summary RR (95% CI) between fixed- and random-effect models, and heterogeneity tests. Heterogeneity tests of pooled studies indicate differences in the RRs (95% CI) when p values are small.28 Summary RRs (95% CI) of pooled studies were calculated according to fixed- and random-effects models.33 34 Calculation by fixed-effects models implies that differences in the RRs (95% CI) of pooled studies are due to sampling error, and calculation according to random-effects models makes allowance for unidentified sources of heterogeneity beyond sampling error.34 The incorporation of possible unidentified sources of heterogeneity in the random-effects models results, in general, in a greater contribution of smaller studies to the overall mean than in the fixed-effects models. Differences between summary RRs (95% CI) calculated according to both models indicate unidentified sources of heterogeneity, in which case the RR (95% CI) of the random-effects model is the more appropriate. Analyses were performed using SPSS 12.0 software package (SPSS Inc), and Review Manager (RevMan) (computer program), version 5.0. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008.
Nineteen potentially eligible articles with data on testosterone levels and CVD were identified.2–20 Eighteen studies with total testosterone, seven studies with free testosterone and three studies with bioavailable testosterone were included in separate analyses and are listed in table 1. The coefficient, corresponding SE, and unit increment (specified difference in testosterone level) derived from the original articles, before re-expression, are shown. In the last column, RRs and 95% CIs are presented after re-expression in a uniform manner for a change of 1 SD in testosterone level. The Caerphilly4 35 and Rancho Bernardo16 36 studies provided data on different follow-up periods. The study closest to the mean follow-up period of the included studies was chosen. Sensitivity analyses showed that using a study on any of the other reported follow-up periods had no substantial influence on final conclusions.31 Seven articles did not provide sufficient quantitative information for this meta-analysis.2 11 12 37–40 We tried to obtain additional detailed data by direct correspondence with authors of four articles. The remaining articles did not provide detailed information on testosterone, suggesting that direct correspondence would be pointless, as one cannot expect an extensive re-analysis of a complete dataset. These remaining articles stated that (1) testosterone was not related to stroke40; (2) no relationship was observed between endogenous sex hormone concentrations and incident coronary heart disease in male twins39 and (3); an abnormal neuroendocrine secretory pattern was prospectively associated with an increased incidence of cardiovascular-related events.38 Correspondence with four authors resulted in additional information for two studies.11 12 Data of the remaining two studies could not be recalculated as these studies had been performed too long ago, as replicated by the authors.2 37 A flow diagram of the search strategy is depicted in figure 1.
Overall, a weak independent protective effect of total testosterone was found with an estimated summary RR (95% CI) of 0.89 (0.83 to 0.96), for a change of 1 SD in testosterone level. High testosterone levels were associated with less risk, though with a p value of heterogeneity <0.0001 (table 2).
Sources of heterogeneity
Meta-regression analysis of 18 studies on total testosterone identified ‘age of study population’ and ‘year of publication’ as sources of heterogeneity. Replacing the dichotomous variable age (<70 years vs including ≥70 years of age) by the linear variable (mean age of study population) resulted in reduction of the explained variation. Identification of sources of heterogeneity by separate meta-regression analyses of seven studies on free testosterone or three studies on bioavailable testosterone revealed no study features of interest, probably owing to the limited number of included studies.
Stratification of the total testosterone studies according to sources of heterogeneity—that is, ‘age of study population’ and ‘year of publication’, showed an estimated summary RR of 1.01 (0.95 to 1.08) for studies performed in men younger than 70 years of age, and 0.84 (0.76 to 0.92) for studies performed in men including those over 70 years of age (table 2). Again, the latter results show heterogeneity (p<0.0001). Further stratification of the latter studies according to ‘year of publication’ revealed an estimated summary RR of 0.97 (0.94 to 1.00) for studies published before 1 January 2007, and 0.77 (0.72 to 0.82) for studies published after 1 January 2007. A separate stratification of the seven studies on free testosterone according to ‘age of study population’ and ‘year of publication’ largely confirmed the earlier findings. All studies on bioavailable testosterone (n=3) included ‘men over 70 years of age’ and showed an estimated summary RR of 0.74 (0.62 to 0.88) (table 2). Excluding the study with smallest SE8 resulted in essentially similar findings (data not shown).
Estimated summary RRs were based on individual RRs adjusted for the highest number of potentially confounding variables, though adjustment for confounding varied across individual studies for both type and number for which adjustment was made. We could not discover arguments showing that the presence or absence of a specific confounder (eg, alcohol, physical activity, comorbidity, ethnicity, etc), the total number of confounding variables, or combinations of classic CVD risk factors (smoking, age, obesity, lipids, hypertension and diabetes) modified the relationship between testosterone and CVD. It seems therefore unlikely that summary estimates might have been attenuated by adjustment for ‘so-called’ mediating factors—that is, factors involved in an underlying pathogenic mechanism between testosterone and CVD. Also, neither the type of testosterone assay, country of study, nor the length of follow-up period, nor type of study design modified the relationship between testosterone and CVD.
Publication bias was explicitly investigated by plotting the number of cases versus effect magnitude,41 as well as by the trim-and-fill method.42 The figure resembled a funnel and showed perfect symmetry for the overall estimate, suggesting relative absence of publication bias (figure 2).
The present systematic review does not confirm the ‘widespread idea’ that low endogenous testosterone increases the risk for CVD in middle-aged men, and somehow contrasts with previous narrative reviews.24–26 Yet, in elderly men, a negative association was revealed, which was particularly prominent in studies published after 1 January 2007.
The remarkable finding of an absence of a statistical significant association between endogenous testosterone and CVD in middle-aged men is based on available literature of seven prospective studies.
This is a new finding and suggests that the relationship between testosterone and CVD is more complex than previously expected. Possibly, testosterone displays a combination of positive and negative effects on the cardiovascular system resulting in an overall negligible effect. Or, one may speculate that the overall ‘neutral’ effect is a consequence of testosterone and oestradiol. Testosterone is a pro-hormone, it is partly converted through aromatase (P450a) into oestradiol, and oestradiol also has cardiovascular properties.43
The statistically significant inverse association between endogenous testosterone and CVD/mortality in elderly men is based on 11 prospective studies. The difference of the relationship of testosterone and CVD from that of middle-aged men is also a new finding. At present, different point of views are possible.
Low testosterone might have a distinctly different effect between middle-aged and elderly men with respect to the adaptive response of adipose tissue in cases of severe obesity, with potential consequences for CVD risk.44
Testosterone itself might be weakly protective, though the protective properties of testosterone might turn out to be relevant only after a very long exposure time—that is, at older age. Most studies on exogenous testosterone treatment neither support nor rebut a protective role for testosterone. Yet, application of testosterone gel has recently been associated with increased risk of cardiovascular adverse events in a population of older men with limited mobility and a high prevalence of chronic disease.45
Although previous trials in men with low testosterone are generally of short duration and limited in the number of participants, the absence of unambiguously positive effects on CVD risk suggests an alternative point of view: low endogenous testosterone could be interpreted as ‘marker of poor health’.
Year of publication
Finally, our study identified ‘year of publication’ as source of heterogeneity for the relationship between testosterone and CVD. Studies published after 1 January 2007 showed a more pronounced protective effect of testosterone on CVD/mortality than studies published before 1 January 2007. The possibility of intercorrelation of different study features of interest cannot be excluded, though an alternative explanation is that recent studies may have fewer cases of misclassification owing to improved technology.46
Limitations and future research directions
Results cannot be extrapolated to women, or to non-white populations. Some of the selected articles provided insufficient information or showed presentation bias—that is, some papers provided more information on statistically significant than on statistically non-significant associations.39 40 Procedures to estimate RRs (with 95% CI) could therefore not be performed for all selected studies, despite a partially successful attempt to gather additional information by direct correspondence with authors. We did not search for abstracts (without full text), as it was expected that they would not provide sufficient information for re-expression in a uniform manner.
Although a meta-analysis of observational studies might be no better than the studies upon which it is built, it allows a transparent identification of sources of heterogeneity (study features explaining conflicting results). This identification provides insight into a meaningful direction for future research. Why is there a difference in association between middle-aged and elderly men? What is the interrelationship between testosterone and oestradiol and their effects on precisely defined cardiovascular end points? Is low testosterone indeed a ‘risk factor’ or should it be interpreted as ‘marker of a poor health’? What are the potentially involved mechanisms? What is the meaning of the finding that ‘year of publication’ is a source of heterogeneity?
In summary, our systematic review of testosterone and CVD reveals no arguments for an increase in risk of CVD by low endogenous testosterone in healthy middle-aged men, which might be of importance if considering replacement treatment with side effects in that specific population. In elderly men, testosterone might be weakly protective, though an alternative explanation of the finding is that low testosterone might reflect a decline in health (box 1).
Summary results of the present meta-analysis
In healthy middle aged men, testosterone does not predict cardiovascular disease (CVD).
In elderly men, low testosterone predicts increased risk for CVD and/or mortality. It is at present unclear whether low testosterone has a direct negative effect, or whether it should be regarded as ‘marker of poor health’.
Recent studies on testosterone and CVD/mortality show more pronounced associations than earlier studies.
We are grateful to A Tivesten for providing additional details on the Gothenburg Study and JA Cauley for a kind and informative letter.
See Editorial, p 867
Linked article 217414.
Funding This work was partly supported by the Flemish Fund for Scientific Research (FWO-Vlaanderen Grant G.0662.07). The study sponsor had no role in study design, analysis or conclusions of the report.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.