In this retrospective cohort study, we obtained data from the database of the MJ Health Research Foundation, which is a large private membership medical clinic in Taiwan (Republic of China). The database covers four health screening centers (in Taipei, Taoyuan, Taichung, and Kaohsiung), where members can receive regular medical examinations based on their preference. Participants underwent medical checkups on own will from January 2007 to December 2016. The age of participants at the first examination was 30 −79 years. The exclusion criteria were as follows: female sex, age younger than 30 years at the time of the first examination or older than 79 years at the time of the second examination, lack of data required for risk score calculation, such as age, sex, race, history of diabetes mellitus, hypertension and the use of antihypertensive drugs, blood pressure, total testosterone, triglyceride, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and glucose levels, body mass index (BMI), and smoking status. Personal data of participants in this study were anonymous. All participants in this cohort had provided consent by signing consent forms to authorize the data analysis before the health examination. All data used in this study were authorized by the Taichung Armed Forces General Hospital (Taichung, Taiwan) and obtained from the MJ Health Research Foundation (authorization code: MJHRF2019016A) (10). This study was approved by the Tri-Service General Hospital Institutional Review Board (TSGHIRB number: A202005160) and the procedures were performed according to the principles stated in the Declaration of Helsinki.

All participants underwent physical examination, body measurements, and laboratory tests of blood samples under fasting state. They also answered a detailed questionnaire about their medical history, including hormone therapy. Systolic and diastolic blood pressure was measured after 10-min rest. The BMI was calculated as weight in kilograms divided by square of height in meters. Total testosterone was expressed as ng/mL; triglycerides, glucose, uric acid, as well as high-density lipoprotein cholesterol and LDL-C, as mg/dL; estimated glomerular filtration rate (eGFR), as mL/min/1.732 m²; and prostate-specific antigen, as μg/L. A reduction in testosterone level indicated that the serum level of total testosterone at the second medical checkup was lower than that at the first medical checkup. Data on medical history such as the use of antihypertensive and oral hypoglycemic drugs, family history of chronic diseases, and smoking status (current and former smoking vs. nonsmoking) was listed as one of the data obtained via the questionnaire completed by participants at the time of the examination. Data standardization was achieved by the use of identical screening procedures with the same model of instruments, as well as formal training of health practitioners at all clinics.

In this study, two risk scores were used to predict the 10-year CV risk of participants: the FRS (2008) (11) and ASCVD Risk Estimator from the ACC/AHA (2013) (12). The formulas of both calculators reference from website of UpToDate online (13), for the FRS and (14) for the ASCVD Risk Estimator. Both risk scores include age, sex, systolic blood pressure, total cholesterol, high-density lipoprotein cholesterol, use of antihypertensive drugs, smoking, and presence of diabetes, while the ASCVD Risk Estimator additionally includes race. We chose to compare these two risk scores because they are widely used tools that were published at a similar time. Moreover, all data required for the calculation of the risk scores were available in our database, and all the tests and examinations required to obtain data for the risk score calculation were cost effective and easily accessible.

Both risk scores can predict the 10-year CVD risk, which are normally presented as percentage. To apply this to clinical use, we divided the estimated CV risk into four different risk categories according to previous studies. For the FRS, the results were classified as very low risk (≤1%), low risk (>1% to <10%), intermediate risk (10% to <20%), and high risk (≥20%) (15, 16). For the ASCVD Risk Estimator, the categories were as follows: low risk (<5%), borderline risk (≥5% to <7.5%), intermediate risk (≥7.5% to <20%), and high risk (≥20%) (17).

In this study, we designed two types of outcomes. The first one is to assess whether overall CV risk increases at the second medical checkup or not. If the result of CV risk estimated at the second time is higher than the first medical checkup, the participants would be classified into “increased group” (difference > 0%), and the others would be classified into “control group” (difference ≤ 0).

The second outcome is to assess whether CV risk categories increase at the second medical checkup or not. If the CV risk category at the second medical checkup is higher than the first time, the participants would be classified into the “increased group”, and the others would be classified into “control group”. For example, one participant was assessed as low CV risk category at the first medical checkup, but the category elevated to intermediate risk at the second medical checkup, he or she would be classified into “increased group”, if the participant is assessed as low CV risk category at the second time, he or she would be classified into “control group”.

Data from two subsequent medical checkups were collected, and 10-year CV risk was estimated by the FRS and ASCVD Risk estimator. The results between risk scores were compared using the Pearson χ² test. Categorical variables were analyzed using the McNemar test.

Men who were free of CV disease at age of 50 have remarkably longer survival and very low remaining risk for CV disease (18), and people aged older than 65 were defined as the elderly by World Health Organization (WHO). Thus, we divided participants into three groups by age, including age 30–49, age 50–64, and age ≥65.

A logistic regression analysis was performed to assess a correlation between the increase in CV risk estimated by the FRS and the reduction in total testosterone. The reduction in total testosterone indicated that the concentration of serum total testosterone at the second checkup was lower than at the first checkup. The definition of outcomes we used in logistic regression are mentioned in “Outcome” section. Some factors had already been included in FRS and ASCVD Risk Estimator, including age, sex, habit of smoking, total cholesterol, high-density lipoprotein cholesterol, blood pressure, and diagnosis of diabetes. In logistic regression analysis, we adjusted BMI, LDL-C, eGFR, and uric acid as confounders, for both of them are not included in CV prediction tools yet are commonly considered to be related to CV risk. Considering the influence of interval of medical checkups (follow year), we also adjusted it as a confounder in the logistic regression analysis (Table 2).

In addition, we also analyzed young participants who aged ≤ 49 based on different follow years. According to the interval of twice medical checkups, participants were divided into three groups, including one-year apart, two-year apart, and 3 and more years apart. Logistic regression analysis adjusted to BMI, LDL-C, eGFR, and uric acid were also performed (Table 3). A p < 0.05 was considered significant. All analyses were conducted at the Department of Medical Education and Research, Taichung Armed Forces General Hospital.

A total of 125,414 participants underwent medical checkup at the MJ Health Screening Center during the study, including 52,945 male participants aged 30–79 years. However, only 1,260 participants who met the criteria of the FRS and ASCVD Risk Estimator and had their serum total testosterone tested were included in this cohort study. After the exclusion of seven participants who reached the age of 80 years by the second visit, the final study sample included 1,253 male participants (Figure 1). There was no evidence indicating that any of the final participants received hormone therapy. Age, testosterone levels, blood pressure, and other clinical and demographic characteristics of participants are presented. The mean age of participants was 47.3 (SD = 9.5) years, and the mean time between the checkups was 1.5 (SD = 0.9) years. Of this cohort, 728 participants were younger than 49 years; 478 were between 50 and 64 years, and 47 were 65 years or older at the first medical checkup (Table 1). Diastolic blood pressure differed significantly between visits in participants aged 30 to 49 years, but no differences were noted in the remaining age groups. Serum total testosterone and prostate-specific antigen levels were significantly higher at the second visit in participants aged 30–49 years and 50–64 years, but no differences between visits were noted among participants aged 65 years or older. Cardiovascular risk assessed by both calculators was higher at the second visit in participants younger than 65 years (p < 0.05), but no differences were noted in older individuals (p > 0.05) (Figure 2). A detailed comparison of study group characteristics according to age and medical checkup is presented in Table 1.

Changes in overall CV risk and by individual risk categories of the FRS and ASCVD Risk Estimator among participants divided into three age groups (30–49 years, 50–64 years, and ≥65 years). The increase in CV risk estimated by two risk scores was consistent in all participants (61.2% by the FRS and 67.6% by ASCVD Risk Estimator) and age groups. About the changes of CV risk categories, 167 (13.3%) of all participants elevated CV risk categories by FRS, and 170 (13.6%) of all participants elevated CV risk categories by ASCVD Risk Estimator at the second medical checkup.

The McNemar test revealed that both the FRS and the ASCVD Risk Estimator predicted categories of CV risk with a similar accuracy of 0.872 (Pearson's correlation r = 0.927, p < 0.001). This indicates a significant correlation between the risk scores not only in terms of predicting the overall CV risk but also in terms of individual risk categories.

In participants aged 30–49 years, a reduction in the total testosterone level(Δ Testosterone) was significantly negatively correlated with the increase in CV risk (OR = 0.804, 95% CI = 0.711–0.909, p < 0.01). A reduction in the total testosterone level elevated the odds of increased estimated CV risk by 19.6%. However, no significant results were noted for participants older than 49 years (Age 50–64, OR = 0.914, 95% CI = 0.806–1.038. Age ≥ 65, OR = 0.987, 95% CI = 0.654–1.490). Moreover, the baseline level of total testosterone was not significantly associated with the overall CV risk in the whole study cohort. In participants aged 30–49 years, BMI at baseline had a significant protective effect on cardiovascular risk (OR = 0.934, 95% CI = 0.892-−0.979, p < 0.01) (Table 2). Baseline LDL-C levels also had a protective effect in this age group (OR = 0.988, 95% CI = 0.983–0.993, p < 0.001). Both eGFR and uric acid did not have influence on CV risk significantly in this study. In participants aged 30-49, the group of follow year = 2 years and follow year ≥ 3 years showed significantly higher odds of increased CV risk while the group of follow year = 1 year being the reference, which means the interval of medical checkups is one of risk factors of increased CV risk in young male people. (OR = 1.597, 95% CI = 1.113-2.295, p < 0.05; OR = 2.983, 95% CI = 1.556–5.717, p < 0.01, respectively) (Table 2). After classification into four risk categories, no significant correlations between CV risk and reductions in the total testosterone level were shown (Age 30-49, OR = 1.039, 95% CI = 0.878-1.230. Age 50–64, OR = 0.861, 95% CI = 0.717–1.033. Age ≥ 65, OR = 1.464, 95% CI = 0.659–3.251. Total participants, OR = 0.958, 95% CI = 0.851–1.077). Detailed data are also shown in Table 2. This indicates that reductions in testosterone levels affect CV risk in the young male population, but without difference in individual risk category between medical checkups with a mean interval of 18 months apart. This might be due to the fact that it takes 2.5–12.5% of elevation in CV risk to increase one grade in CV risk category. The results above revealed that although a reduction of total testosterone level increases CV risk, the impact is not strong enough to cause actual elevation in CV risk category in the short term. Thus, the result of CV risk categories is not significant. However, further studies are needed to confirm this finding.

The logistic regression analysis of correlations between increase in CV risk estimated by the ASCVD Risk Estimator and reductions in total testosterone, baseline testosterone, BMI, and LDL-C provided similar results to those observed for the CV risk assessed by the FRS. The reduction in total testosterone levels was correlated with a significantly higher odds of increased CV risk of participants aged 30–49 years (OR = 0.841, 95% CI = 0.742–0.953, p < 0.01) (Table 2). A reduction in the total testosterone level elevated the odds of increased estimated CV risk by 15.9%. No significant correlations were noted for participants older than 49 years (Age 50–64, OR = 0.906, 95% CI = 0.794–1.034. Age ≥ 65, OR = 0.864, 95% CI = 0.561–1.330). In participants aged 30–49, the group of follow year = 2 years and follow year ≥ 3 years showed significantly higher odds of increased CV risk while the group of follow year = 1 year being the reference (OR = 1.769, 95% CI = 1.211–2.584, p < 0.05; OR = 3.944, 95% CI = 1.871–8.316, p <0.001, respectively) (Table 2). After classification into four risk categories, no significant correlations between CV risk and the above dependent variables were shown (Age 30–49, OR = 0.990, 95% CI = 0.810–1.209. Age 50–64, OR = 0.938, 95% CI = 0.805–1.097. Age ≥ 65, OR = 1.410, 95% CI = 0.845–2.351. Total participants, OR = 1.004, 95% CI = 0.894–1.128).

We independently analyzed correlations between the increase in CV risk and selected variables depending on the time between medical checkups in participants aged 30–49 years. The increase in CV risk calculated by the FRS was significantly associated with the reduction in testosterone levels both in participants who underwent medical checkups one-year apart (OR = 0.893, 95% CI: 0.802–0.995, p < 0.05) and those with checkups 2 years apart (OR = 0.805, 95% CI: 0.680–0.954, p < 0.05) (Table 3), but there is no significant association in the group of follow year 3 ≥ years (OR = 0.833, 95% CI = 0.654-1.061). For CV risk estimated by the ASCVD Risk Estimator, only participants who underwent medical checkups 2 years apart showed significant correlations (OR = 0.707, 95% CI: 0.585–0.855, p < 0.001) (Table 3).

Baseline LDL-C levels were significantly correlated with lower or no change in CV risk both for one-year and two-year interval between checkups irrespective of the risk score used (Table 3).