Endogenous testosterone does not improve prediction of incident cardiovascular disease in a community-based cohort of adult men: results from the Tehran Lipid and Glucose Study

Abstract

Introduction: To explore the predictive value of testosterone added to the Framingham Risk Score (FRS) for cardiovascular disease (CVD).

Methods: Among 816 men, 30–70 years/old, without prevalent CVD, from a community-based cohort (Tehran Lipid and Glucose Study), we assessed the predictive value of testosterone with incident CVD, using three multivariate Cox proportional-hazards models. Model I: FRS variables; model II: Model I plus total testosterone; model III: Model II plus Systolic blood pressure (SBP) * total testosterone (the best fit interaction-term between testosterone and FRS variables). Discriminations and goodness-of-fit were assessed by the C-statistic and the approach of Grønnesby, respectively. p Value <.05 was significant.

Results: During 12 years of follow-up, 121 CVD events occurred. In all models, age, treated SBP, smoking, and diabetes were associated with increased CVD (p values <.05). Neither testosterone (models II and III), nor SBP * testosterone (model III) were associated with CVD (p values >.05). The C-statistics for models I, II, and III were 0.819, 0.820, and 0.821, respectively, indicating no significant improvement in the discrimination power. The models’ goodness-of-fit did not improve compared with the FRS.

Conclusion: Testosterone could not add to the predictive value of FRS for CVD in men, either directly, or through interactions with FRS variables.

Keywords:

• Endogenous testosterone
• total testosterone
• cardiovascular disease
• Framingham Risk Score
• risk assessment

Introduction

Cardiovascular diseases (CVD), a main cause of death worldwide, is responsible for more than 80% of mortality in low- to middle-income countries [1]. Gender has been known as an important non-modifiable risk determinant, since the lifetime risk of CVD is generally higher in men, compared to pre-menopausal women in the same age [2,3]. This epidemiological concept has drawn attention to the possible modifiable mediators of gender on the cardiovascular system, demonstrating several factors including higher rates of smoking and drinking, ignorance regarding preventive care and healthy life style, and poor metabolic profiles in men compared to women [4]. However, since the gender differences tend to vanish in post-menopausal women, the possible role of sex hormones has been of especial interest in the previous literature [2,3]. Testosterone, the main male sex hormone, declines 1–2% per year since the third decade of life in healthy men [5,6] and free testosterone, the biologically active form, further drops due to decreasing sex hormone binding globulin [7]. Majority of previous observational studies demonstrate negative associations between the level of testosterone in men and cardiovascular risk factors such as dyslipidemia [8], diabetes mellitus (DM) [9], and the metabolic syndrome [10]. Moreover, several studies have indicated that lower levels of testosterone may directly increase the risk of CVD incidence [11,12], severity [12,13], and mortality, especially among older men [14,15]. Although several underlying mechanisms are known to explain the benefits of testosterone for the cardiovascular system including vasodilation, cardio protection, and reducing atherosclerosis and inflammation [16], it has been suggested that the association between testosterone and CVD may be due to poor health among men with lower hormonal levels, rather than being a causal effect [15,17]. Overall, it is yet unclear whether testosterone could be a useful marker in predicting CVD for men.

Among risk assessment models to predict incident CVD, the Framingham Risk Score (FRS) has been applied widely in the last decade, presenting a quantified risk estimation based on a sex-specific multivariable predictive model for CVD [18]. The performance of this model in estimating 5-year CVD risk has been validated in Iran, previously [19]. Previous studies among men have suggested that endogenous testosterone may be associated with the FRS for CVD and its components [20,21]. In this study, we further aimed to assess the predictive value of testosterone and its interactions with other FRS components in a population of generally healthy men, as part of an ongoing prospective study, called the Tehran Lipid and Glucose Study (TLGS).

Methods

Study population

Data were obtained from a subgroup of the TLGS population. The TLGS is a prospective community-based study on non-communicable diseases among 15,005 ethnic Iranian residents aged >3 years, of district 13 of the capital Tehran. Participants were recruited and followed up at three-year intervals; details have been explained, elsewhere [22]. For the current research, after exclusion of subjects with missing data and those for whom blood samples of testosterone were not available, all males aged 30–70 years were recruited. Of the total 831 enrolled men, 15 were further excluded due to prevalent CVD, leaving 816 subjects, who were followed-up until March 20 2012. The proposal of the current study was approved by the ethics committee of the Research Institute for Endocrine Sciences, Shahid Beheshti University of Medical Sciences, and all participants provided written informed consent.

Medical history, clinical examination, and laboratory measurements

Subjects were interviewed by trained nurses according to a questionnaire including information on past medical history, medication and smoking. Blood pressure, weight, and height were measured through a brief physical examination [22] and body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters.

For biochemical and hormonal measurements, blood samples were drawn after a 12–14 h overnight fasting and centrifuged within 30–45 min; fasting plasma glucose (FPG), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and serum creatinine were assayed. Two-hour post-challenge glucose (2-hPCG) was evaluated for those not taking glucose-lowering drugs. Plasma glucose was measured using an enzymatic colorimetric method with glucose oxidase. TC was assayed using enzymatic colorimetric method with cholesterol esterase and cholesterol oxidase. HDL-C was measured after precipitation of the apolipoprotein B containing lipoproteins with phosphotungstic acid. TG was assayed using an enzymatic colorimetric method with glycerol phosphate oxidase. These analyses were performed using commercial kits (Pars Azmoon Inc., Tehran, Iran) and a Selectra 2 auto analyzer (Vital Scientific, Spankeren, The Netherlands). In subjects whose TG concentration was below 4.52 mmol/L, LDL-c was calculated from serum TC, TG, and HDL-C concentration, using the Friedewald formula (24). The intra- and inter-assay coefficients of variations (CVs) were both 2.2% for glucose. For both TC and HDL-C, intra- and inter-assay CVs were 0.5% and 2%, respectively. Intra- and inter-assay CVs were 0.6% and 1.6% for TG, respectively. Total testosterone was measured by enzyme immunoassay (Roche Diagnostics GmbH, Mannheim, Germany) with intra- and inter-assay CVs of 5.7% and 8.4%, respectively at the detection limit of 0.002 ng/mL. All analyses were performed in the TLGS research laboratory.

Cardiovascular disease outcome

As part of the TLGS, the participants were followed-up annually over phone calls by a trained nurse. If medical conditions or mortality were reported, a trained physician would collect further information during a home visit, and if necessary, from the participants’ medical files in hospitals or death certificates. Each medical event was further evaluated by an outcome adjudication committee consisting of a principal investigator, an internist, an endocrinologist, a cardiologist, an epidemiologist, and the physician who collected the outcome data, and if necessary, other specialist consultants. Eventually, the International Statistical Classification of Diseases and Related Health Problems 10th Revision criteria and the American Heart Association classification for cardiovascular events were applied. Accordingly, coronary heart disease (CHD) includes cases of definite myocardial infarction diagnosed by ECG and biomarkers, probable myocardial infarction (positive ECG findings plus cardiac symptoms or signs, but biomarkers showing negative or equivocal results), unstable angina pectoris (new cardiac symptoms or changing symptom patterns and positive ECG findings with normal biomarkers), angiographic proven CHD, and death due to CHD. CVD was specified to any CHD event, stroke, or cerebrovascular death [22].

Definition of terms

Diabetes was defined as FPG ≥ 7 mmol L⁻¹ or 2-h PCPG ≥ 11.1 mmol L⁻¹ or taking anti-diabetic medications. A history of previous CVD reflected any prior diagnosis of CVD by a physician. Cigarette smoking status and antihypertensive medication use was ascertained by self-reporting.

Statistical analysis

Descriptive statistics were presented as mean (SD) for continuous normal variables, frequencies (%) for categorical variables, and skewed variables, for example, total testosterone, were reported as median (inter-quartile range [IQR]). Comparison of baseline characteristics between participants with CVD vs. non-CVD, was performed by Student's t-test for normal continuous variables; otherwise, Mann–Whitney U statistic non-parametric test, and Chi-square test for categorical variables were applied. Multivariate Cox proportional-hazards regression functions were fitted to estimate 5-year risk of CVD in various models. Model I included the same variables as the Framingham general cardiovascular risk profile, that is, age, TC, HDL-C, systolic blood pressure (SBP), antihypertensive medication, current smoking, and diabetes status [18]. Hence, model I was a recalibration of the FRS through our data. In model II, we added the level of total testosterone in order to determine its effect. In addition, several models were fitted considering the interaction effects of testosterone with other risk factors, among which only the best fit, adding the interaction of testosterone and SBP (testosterone * SBP) to model II, has been reported as model III. To improve discrimination and calibration of the models and to minimize the influence of extreme observations, all continuous variables were transformed to natural logarithms.

Hazard ratios (HR) were reported for each 0.1-unit increase in the naturally logarithmically transformed variables and for each 1-unit increase in the other continuous variables.

The Cox proportional hazard assumptions were confirmed by the Schoenfeld residual test. In order to estimate the 5-year risk of CVD, the following equation was adopted from D’Agostino et al. [18]: $$\widehat{p}=1-S_0{\left(t\right)}^{\exp \left(\sum _{i=0}^p{\beta }_i{x}_i-\sum _{i=0}^p{\beta }_i{¯x}_i\right)}$$ where S₀ (t = 5/¯x) is baseline in order to estimate the 5-year risk, β_i is the estimated regression coefficient, ¯x is the naturally log transformed value of the ith risk factor if continuous, ¯x is the corresponding mean, and p is the number of risk factors. The 5-year baseline survival was obtained from a Kaplan–Meier estimate.

Assessment of model performance

In order to assess the models’ performances, we evaluated discrimination and calibration of the models. Discrimination demonstrates the ability of a prediction model to distinguish event (i.e. CVD) from nonevent groups and can be quantified by the Harrell's concordance statistic (C-statistic), which is equal to the area under the receiver operating characteristic (ROC) curve [23]. C-statistic values over 0.7 indicate a good discrimination power and a value of 0.5 indicates no better prediction than by random chance. Calibration is used to assess how close the probabilities predicted by a model are to actual outcomes. For calibration of our models, we applied the Grønnesby and Borgan approach which considered proportional hazards assumption as well, where non-significant results indicated model goodness of fit; otherwise, recalibration was recommended [24].

All analyses were performed using the Stata software (version 12; Stata, INC., College Station, TX, USA), with a two-tailed p values <.05 considered significant.

Results

Among 816 men included for analysis, the mean (SD) age was 46.1 (11.7) years. The median of total testosterone was 4.78 ng/mL in the total population and the 25 and 75 percentiles were 3.95 and 5.54 ng/mL, respectively. A comparison between baseline characteristics of CVD cases and non-CVD subjects is present in Table 1. As expected, CVD patients generally showed more adverse risk factor profiles, compared to non-CVD subjects; they were significantly older, had higher waist circumference, systolic and diastolic blood pressures, LDL, TG, and a greater percent of diabetes and hypertension medication use than those without CVD. However, levels of total and HDL cholesterols, total testosterone, and smoking rates did not differ significantly between the two groups. During a median 12 (IQR = 0.7) years of follow-up, 121 men developed CVD and the incidence was 1297 (confidence interval [CI] 95%: 1107–1487) per 100,000 person-years.

Table 1. Comparison of baseline characteristics between subjects with and without cardiovascular disease event.

	CVD (N = 121)	Non-CVD (N = 695)	Total (N = 816)	p Value*
Continuous variables
Age (year)	56 (9.9)	44.66 (11.33)	46 (12)	.006
Waist circumference	93 (10)	89 (11)	91 (10)	.000
Body mass index (kg/m^2)	26.6 (3.9)	25.9 (4.1)	26.4 (4)	.094
Systolic blood pressure (mm Hg)	133 (24.40)	119 (17.8)	121 (20)	.000
Diastolic blood pressure (mm Hg)	82.31 (12.24)	78.16 (10.87)	79 (11)	.000
Fasting plasma glucose (mmol/L)	6.69 (2.84)	5.30 (1.38)	5.49 (1.78)	.000
2h-PCG (mmol/L)	7.84 (5.09)	5.09 (2.47)	6.05 (3.00)	.000
Total cholesterol (mmol/L)	5.6 (1.1)	5.4 (1.0)	5.40 (1.01)	.587
TG (mmol/L)	2.01 (0.97)	1.50 (0.37)	2.03 (1.15)	.005
HDL-C (mmol/L)	0.99 (0.20)	1.0 (0.25)	1.01 (0.26)	.054
LDL-C (mmol/L)	3.69 (0.94)	3.44 (0.85)	3.50 (0.88)	.005
Total testosterone (ng/mL)	4.7 (1.7)	4.8 (1.6)	4.8 (1.5)	.255
Categorical variables
Current smokers, n (%)	34 (29)	187 (27)	221 (27)	.824
Hypertension medication, n (%)	25 (20)	28 (4)	53 (6)	.001
Diabetes, n (%)	47 (39)	80 (12)	127 (16)	.001
Family history of CVD	7 (8)	29 (6)	36 (4)	.850

Mean (SD) is shown for continuous variables and p values was calculated with t-test; n (%) is shown for categorical variables with p values according to Chi-square or Fisher exact test as appropriate. Total testosterone was shown as median (inter-quartile range) with p values according to Mann–Whitney U test.

CVD: cardiovascular disease; 2h-PCG: 2 h-post-challenge glucose; TG: triglycerides; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol.

* Significance level was considered lower than .05.

Results of the multivariate Cox proportional hazard analyses are shown in Table 2 for the three models. In all models, age, treated SBP, smoking and diabetes were significant risk factors for CVD. As shown in Table 2, when including only the covariates from FRS, age, treated SBP, smoking, and diabetes showed significant associations with CVD by HR (CI 95%) of 2.10 (1.70–2.60), 4.30 (3.90–4.70), 1.70 (1.10–2.50), and 2.20 (1.50–3.20), respectively. After further adjustment for the level of total testosterone in model II, the HRs of smoking and diabetes slightly changed to 1.61 (CI 95%: 1.06–2.44) and 2.11 (CI 95%: 1.42–3.13), respectively, and the HR of treated SBP decreased to 2.72 (CI 95%: 1.21–6.40). We failed to detect any significant associations between testosterone and CVD (HR: 0.49, CI 95%: 0.17–1.4). In model III, after adding the interaction effect of SBP and testosterone, neither testosterone nor SBP * testosterone were associated with CVD.

Table 2. Multivariate Cox models I, II, and III for incident cardiovascular disease.

	Hazard ratio^b (95% CI)
Variables	Model I	Model II	Model III
Age^a	2.10 (1.70–2.60)	2.08 (1.65–2.63)	2.10 (1.65–2.66)
Total cholesterol (mmol/L)	2.00 (0.23–20.40)	1.69 (0.18–16.28)	1.8 (0.19–17.76)
HDL-C (mmol/L)	0.26 (0.04–1.70)	0.25 (0.04–1.62)	0.24 (0.04–1.62)
SBP (mm Hg) if not treated	1.30 (0.99–1.70)	1.30 (0.99–1.71)	1.29 (0.98–1.69)
SBP (mm Hg) if treated	4.30 (3.90–4.70)	2.72 (1.21–6.40)	2.74 (1.77–4.23)
Smoking
Never/past smoker	Reference	Reference	Reference
Current smoker	1.70 (1.10–2.50)	1.61 (1.06–2.44)	1.92 (1.05–2.41)
Diabetes	2.20 (1.50–3.20)	2.11 (1.42–3.13)	2.10 (1.40–3.11)
Total testosterone (ng/mL)	–	0.49 (0.17–1.41)	0.58 (0.19–1.8)
SBP * testosterone (mm Hg ng/mL)	–	–	0.46 (0.60–3.47)

5-year baseline survival: S0 (5) = 0.9734.

Model I: Multivariate Cox model consisting of the Framingham Risk Score variables for cardiovascular disease (CVD). Model II: Multivariate Cox model consisting of the Framingham Risk Score for CVD plus total testosterone. Model III: Multivariate Cox model consisting of the Framingham Risk Score for CVD plus total testosterone and its interaction with systolic blood pressure (SBP * testosterone).

HDL-C: high-density lipoprotein-cholesterol; SBP: systolic blood pressure. Significant values are presented in bold.

^a All continuous variables were transformed to natural logarithms.

^b The beta coefficients and hazard ratios are estimated for each 0.1-unit increase in naturally logarithmically transformed values of age and SBP and each one-unit increase for the other continuous variables.

The C-statistics for models I, II, and III were 0.819, 0.820, and 0.821, respectively, indicating non-significant improvement in model discrimination. There was no statistically significant difference in the goodness of fit between these three models and the FRS [18], judged by the Grønnesby and Borgan’s test (Table 3) and the corresponding ROC curves (Figure 1).

Figure 1. Comparison of discrimination power between models I, II, and III. The ROC curves of models I, II, and III were compared. Model I: Multivariate Cox model consisting of the Framingham Risk Score for cardiovascular disease (CVD). Model II: Multivariate Cox model consisting of model I plus total testosterone. Model III: Multivariate Cox model consisting of model II plus the interaction between total testosterone and systolic blood pressure.

Table 3. Comparison of performances between models I, II, and III and the Framingham Risk Score for cardiovascular disease.

Assessment method	Framingham Risk Score	Model I	Model II	Model III
C statistic (95%CI)^a	0.77 (0.74–0.81)	0.819 (0.78–0.86)	0.820 (0.78–0.86)	0.821 (0.78–0.86)
Grønnesby and Borgan test^b	>0.05	>0.05	>0.05	>0.05
Sensitivity of top quintile (%)	62	62	64	63
Specificity of top quintile (%)	82	82	82	82

Model I: Multivariate Cox model consisting of the Framingham Risk Score for cardiovascular disease (CVD); model II: Multivariate Cox model consisting of the Framingham Risk Score for CVD and total testosterone; model III: Multivariate Cox model consisting of the Framingham Risk Score for CVD plus total testosterone and its interaction with systolic blood pressure.

^a C statistic indicates model discrimination obtained from ROC analysis.

^b Grønnesby and Borgan developed a test for goodness-of-fit under the proportional hazards assumption: non-significant results indicate model’s goodness of fit.

Discussion

In a community-based prospective study on Iranian adult men, we showed that the level of testosterone could not add any predictive value to the FRS, a well-known CVD risk model, neither as a main factor, nor through its interactions with other determinants of incident CVD.

Existing studies on the association of testosterone with CVD have indicated inconsistent results. Previously, Lee et al. showed a negative association between testosterone and the FRS among middle-aged men from East Asia; however, it should be noted that their study population had a median total testosterone of 3.2 ng/mL and had been diagnosed with sexual dysfunction [20]. Moreover, in a retrospective study on veteran males from the USA, both total and free testosterone were negatively associated with the FRS and total testosterone was also associated with several FRS components, showing a positive association with HDL and negative associations with BMI, TC, TG, and hypertension treatments [21]. Although previous observational investigations from the last decade have overall been in favor of a negative association between the level of testosterone and incident CVD in men, these studies are subject to great heterogeneity [15, 25,26]. In 2011, Ruige et al. meta-analyzed 19 prospective studies conducted among men and found a 11% increase in cardiovascular events with each one SD decrease in total testosterone; however, this association was only significant after the age of 70 years and disappeared among healthy middle-aged men [15]. Similarly, a meta-analysis of 12 longitudinal studies including over 11,000 subjects, with mean age of approximately 60 years and mean follow-up of around 10 years, revealed an overall 25% increase of CVD mortality per 2.8 SD decrease in total testosterone levels, again documenting higher relative risk among the elderly; however, the findings of this meta-analysis were also limited by significant heterogeneity in various important factors including age, baseline testosterone levels, and the duration of follow-up [26]. The apparent age-dependence and weakness of association among healthy middle-aged men has led to a hypothesis that the recognized association of testosterone with cardiovascular risk might be solely reflecting low testosterone as a marker of poor health status [15]. This hypothesis could explain why no associations have been found in overall healthy populations including results from the Framingham Heart Study in both middle-aged and elderly men [27,28]. Accordingly, our findings in a generally healthy Caucasian population from the Middle East indicated that testosterone could not add to the predictive value of the FRS; hence, endogenous testosterone may not be a good marker for men’s cardiovascular health.

Although the reviewed literature suggested that the association of low testosterone with cardiovascular events was more dominant among the elderly, studies that have explored cardiovascular risk factors or early stages of disease as their outcome, have also shown associations with testosterone in the middle-aged; in this regard, several previous longitudinal investigations have indicated negative associations between testosterone and poor cardiovascular risk profiles among men who were on average middle-aged, demonstrating higher risk of diabetes [29], incident hypertension and blood pressure changes [30,31], and adverse lipid profiles [8]. Testosterone has also been negatively associated with carotid intimal thickness (as a marker of preclinical atherosclerosis), atherosclerotic plaques, endothelial dysfunction, and hypersensitivity C-reactive protein (as a representative of inflammatory state) among middle-aged diabetic men [32]. However, in the current study, the interactions of testosterone with other risk factors of CVD as indicated by the FRS, could not improve model performance in predicting 5-year risk of CVD. Considering the apparent complexity of the effects of testosterone [5], applying a non-parametric survival model, which can automatically detect interactions between covariates without the need to pre-specify them, may better clarify the relation of testosterone and CVD [33].

The authors acknowledge several limitations in this investigation. Firstly, due to lack of data on the level of free or bioavailable testosterone, analyses were limited to total testosterone. Although a host of related studies have reported the association of total testosterone with CVD [15,29], using the active hormone form is more reliable and studies that have performed analyses based on free or bioavailable testosterone have shown more consistent results [5]. However, the easy measurement of total testosterone in the current clinical setting, could make it a more appropriate biomarker to use in a risk assessment model. Secondly, in this study data on hormone levels were limited to single measurements at baseline; however, for epidemiological purposes, a single value has been declared adequate [34]. Also, to minimize the effect of circadian variations of hormone levels, all blood samples were collected in early morning [22]. Thirdly, similar to many studies on the health effects of testosterone, we have used immunoassay to measure testosterone levels [2,14]. With respect to the importance of hormone assay techniques in the reliability of such studies, applying more accurate methods like liquid chromatography–tandem mass is preferable and may alter the results [35]. Finally, considering the time-assuming nature of the possible underlying mechanisms of testosterone’s role, such as atherogenesis, a longer follow-up period could increase the reliability of our findings. However, it should be noted that the meta-analysis by Ruige et al., revealed no change in the relationship of testosterone with CVD, either due to type of hormone assay, or due to duration of follow-up [15]. Regarding the strengths of this study, besides the prospective and population-based design, this is one of the first investigations of the predictive potential of testosterone and its interactions for incident CVD, to improve a well-known cardiovascular risk assessment model, which has been previously validated and proved to be accurate in our community [19]. Finally, the cardiovascular outcomes were ascertained by an event adjudication committee which adds to the reliability of our results.

Conclusion

In conclusion, our findings indicate that serum total testosterone adds no significant predictive information to the common cardiovascular risk factors and is not an influential added marker to the FRS for CVD prediction in generally healthy men. Further investigations among the elderly could extend this study.

Acknowledgements

This article has been extracted from the thesis written by Donna Parizadeh for fulfillment of a Master of Public Health (MPH) degree, at the School of Public Health, Shahid Beheshti University of Medical Sciences. The authors would like to thank the staff of the Research Institute for Endocrine Sciences and the School of Public health, Shahid Beheshti University of Medical Sciences, for their valuable efforts. The authors also wish to acknowledge Ms. Niloofar Shiva for critical editing of English grammar and syntax of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The TLGS has been funded by the National Research Council of Islamic Republic of Iran [Grant no. 121].