Aging is becoming one of the most substantial social transformations of developed countries in the twenty-first century. Although the average life expectancy in most developed countries has nearly doubled over the past two centuries (1), the increase of health span is more significant and meaningful than simply living longer. Briefly, “health span” is a period of good health during the lifetime and is also termed as healthy and morbidity-free lifespan (2). In many supercentenarians, health span approximates lifespan due to a consistent delay in the occurrence of age-related diseases (3), which has stimulated enormous interest in mysteries behind. Hence, health span becomes a promising phenotype for longevity research.

Interestingly, men usually have a shorter life expectancy than women, suggesting men seem to age slightly faster than women (4). Considering the important effect of sex hormones on maintaining individual muscle mass, bone mass, and physical function, it is rational to speculate that sex hormones have a potential relationship with individual aging. Testosterone is a sex hormone mainly produced by testes in men or synthesized by ovaries and adrenal glands of women in small amounts. Majority of testosterone circulating in the blood is tightly bound to sex hormone-binding globulin (SHBG) or weakly bound to albumin. Approximately only 1%–3% of them is unbound and readily available to the tissues, referred to as free testosterone (FT) (5). It is worth noting that the concentrations of serum testosterone and free testosterone decrease with age in both genders (6), and deficiencies in multiple anabolic hormones have been shown to associate with individual longevity (7). Previous studies have revealed the associations between testosterone and several complex diseases, such as cardiovascular diseases (8), metabolic disease (9), and cancer (10, 11), as well as mortality (12, 13). However, direct evaluation on the effect of testosterone for health span is limited. Additionally, given sex differences in serum testosterone concentrations and biological mechanisms, sex-specific analyses for health span are also necessary.

Recently, a genome-wide association study (GWAS) systematically selected eight health events closely connected with aging to construct a reasonable definition for health span, and further identified 12 single-nucleotide polymorphisms (SNPs) related to health span (14). This provides a great chance to establish polygenic risk score (PRS) for health span. To our knowledge, PRS is an efficient tool estimating the cumulative effect of multiple risk-associated variants to discriminate populations at high risk of complex diseases (15), such as lung cancer (16) and breast cancer (17). Considering that an interplay between serum testosterone and genetic factors may exist, investigation for their joint association with health span is essential as well.

Therefore, based on complete clinical and genetic data from UK Biobank (UKB), we explored the sex-specific associations between endogenous testosterone, genetic factors, and health span by a large-scale sample including 293,214 participants.

Totally, the median follow-up time was 7.70 years (interquartile range: 6.80–8.45 years). Nearly half of the HST events were caused by cancer (46.3%), followed by MI (17.8%) and death (11.4%) ( Supplementary Table 3 ).

The baseline characteristics of the participants by sex and health span are described in Table 1 . There were 26,748 (18.39%) men and 18,963 (12.84%) women with HST during the follow-up. Men with HST were older and had a lower TT and FT level. They were more likely to have higher BMI and Townsend deprivation index. Moreover, they also tended to be current smokers and have a family history of CCVD or cancer. Similar characteristics were observed in women with HST. The majority of these women had reached menopause, were hormone replacement therapy users, and had higher levels of FT than the healthy women. Additionally, characteristics of participants by quintiles of total testosterone were also performed ( Supplementary Table 4 ). The non-normality distribution of TT and FT was exhibited clearly, and the levels of TT and FT were totally different in both genders ( Supplementary Figure 2 ).

The sex-specific associations between TT, FT, and risk of HST are exhibited in Table 2 . For men, those in the second to fourth quintiles (Q2–Q4) of TT all had a lower HST risk in the fully adjusted models with the bottom quintile (Q1) of TT as a reference (Q2, HR: 0.94, 95% CI: 0.91–0.98; Q3, HR: 0.95, 95% CI: 0.91–0.99; Q4, HR: 0.94, 95% CI: 0.90–0.98), and the hazard decreased by 8% per SD (standard deviation) increment of log-TT (HR: 0.92, 95% CI: 0.88–0.97). However, the association of FT in men diminished in the fully adjusted model. Compared with women in the bottom quintile (Q1), only those in the top quintile (Q5) of TT (HR: 1.08, 95% CI: 1.04–1.13) and FT (HR: 1.12, 95% CI: 1.07–1.18) had a significantly increased HST risk.

Non-linear relationships of testosterone and the risk of HST were observed for both genders. In men, an approximate L-shape curvilinear relationship was observed, both in TT and FT, using multivariate cubic models ( Figures 1A, B ). After natural log transformation for TT and FT concentrations, negative associations were observed below the corresponding medians (TT, HR: 0.88, 95% CI: 0.82–0.95; FT, HR: 0.87, 95% CI: 0.81–0.93). Interestingly, TT and FT had an inverse association with HST risk in women. The relationship was linear for TT, while it was more like a J-shape for FT ( Figures 1C, D ). Above the medians, HST risks were positively correlated with TT (HR: 1.13, 95% CI: 1.05–1.21) and FT (HR: 1.16, 95% CI: 1.09–1.22).

Due to the complexity of health span, we further explored the relationship between testosterone and cause-specific incidence. In men, testosterone was observed to have decreased the death, COPD, dementia, diabetes, and lung cancer risk, as well as the overall risk of non-cancer events, but increased the prostate cancer and myocardial infarction risk ( Supplementary Table 5 ). In women, testosterone was positively associated with death, breast cancer, ovary cancer, endometrial cancer, and overall cancer, as well as dementia (total testosterone only) and diabetes (free testosterone only), but negatively related to COPD ( Supplementary Table 6 ). In the stratified analyses, the associations of HST with testosterone were largely consistent across subgroups, except alcohol intake and Townsend deprivation index in men ( Supplementary Table 7 ) and women with a family history of CCVD ( Supplementary Table 8 ). In the joint analyses between TT and SHBG, the HST risk increased from 10% to 25% as the increase of TT level in women with the lowest strata of SHBG ( Supplementary Figure 3 ).

To explore the associations between genetic factors and health span, a polygenetic risk score was constructed by 12 SNPs of health span to evaluate individual genetic risk. As shown by the distribution of polygenetic risk scores, both men and women with HST showed a higher genetic risk ( Supplementary Figure 4 ). Furthermore, we found a positive relationship between PRS and HST risk. Participants from both genders with high genetic risk tended to terminate health span prematurely, when compared with those with low genetic risk (men: HR: 1.19, 95% CI: 1.15–1.24; women: HR: 1.21, 95% CI: 1.16–1.27, Figure 2 ). However, no interaction between PRS and testosterone was observed (P > 0.05), and the associations of testosterone with HST were unchanged after additionally adjusting for PRS ( Supplementary Tables 9, 10 ).

Then, we found a joint effect of genetic risk and TT/FT. Specifically, compared with men with high genetic risk and low level of TT/FT, a 20% lower risk was observed in those with both low genetic risk and high level of TT/FT (TT, HR: 0.80, 95% CI: 0.73–0.88; FT, HR: 0.80, 95% CI: 0.73–0.87, Figures 3A, B ). Besides, the HST risk increased with both genetic risk and testosterone concentration in women. Women with high genetic risk and testosterone level showed the highest risk in subgroups (TT, HR: 1.32, 95% CI: 1.19–1.46; FT, HR: 1.38, 95% CI: 1.25–1.53, Figures 3C, D ).

Finally, sensitivity analyses showed consistent results after excluding participants with HST during the first 2 years of follow-up ( Supplementary Table 11 ). When we excluded the participants with poor self-reported health status at baseline, the associations were also consistent ( Supplementary Table 12 ). Moreover, after excluding participants with outliers of testosterone level (top 1% and bottom 1%), we still observed similar associations ( Supplementary Table 13 ). After additional adjustment for fasting time or menstrual cycle proxy factors, similar associations were also presented ( Supplementary Tables 14, 15 ).

In this large-scale cohort study, comprised of 145,481 men and 147,733 women from the UK Biobank, we observed sex-specific associations between serum testosterone and HST risk. The findings suggested that total testosterone might decrease the risk of health span termination in men. Inversely, women with a high level of TT or FT were at a higher risk of HST. Furthermore, genetic factors increased the risk of premature end of health span to some extent. Although no interplay between testosterone and genetic factors was observed, participants with both high genetic risk and abnormal levels of testosterone (low level in men and high level in women) had the highest HST risk.

Despite the scarcity of direct evidence in the recent two decades relating testosterone and health span, a growing number of epidemiological studies indicated that high levels of testosterone might be protective for all-cause mortality in men (12, 25), but inversely in women (13), which is consistent with our cause-specific findings. Similar to a prospective study from Denmark (10), we observed that the risk of overall cancer for men was not modified by testosterone levels. However, some researchers indicated that high testosterone might be a risk factor for prostate cancer (11). For women, the effects of endogenous testosterone might increase the risk of common gynecological cancer, such as breast (26) and endometrial cancer (27), supporting our cause-specific results. Metabolic syndrome was considered as a critical risk factor for common cancers (28). A high level of testosterone had been proven beneficial for metabolic syndrome risk in men but harmful in women (9, 29, 30), which supported our findings. Additionally, recent researches pointed out that testosterone deficiency in men was associated with increased risk of non-cancer chronic disease, such as cardiovascular disease (8, 31), type 2 diabetes (29, 32) and dementia (33, 34), as well as exacerbated the condition of COPD patients (35). In women, early studies revealed an inverse association of testosterone with increased CVD risk (36, 37), while it was not observed in the recent Rotterdam study (38). Although testosterone may be safe for myocardial function at physiological levels, a toxic effect also exists at supraphysiologic dose (39). As mentioned above, higher levels of testosterone are unfavorable to the metabolism of women and increase their risk of type 2 diabetes (30, 40). Thus, these pieces of evidence support that a sex-specific association indeed exists between testosterone and HST risk.

Except for endocrine factors, genetic variations also play an important role in the determination of individual health span (14). PRS is a simple approach for summing risks across multiple susceptibility loci, aiming at classifying populations with distinct risk levels to drive clinical or personal decision-making (15). Although similar longevity genome-wide association studies (41, 42) exist for a long time, studies that have assessed the predictive performance of PRS on longevity or health span remain scarce. Besides, the small sample size of a former study could affect the strength of the findings (43). In our study, PRS based on 12 SNPs from nearly 300,000 participants stratifies the participants into distinct HST risk strata clearly, indicating GWAS-derived PRS is a promising tool. The effectiveness of PRS still requires independent validation in other populations, as conducted in a large-scale prospective cohort study that applied the PRS to the CKB cohort for re-evaluation (16).

By summarizing the existing evidence related to death, cancer, and other non-cancer events, we indirectly discuss the sex-specific roles of testosterone in maintaining the period of health span. In view of a potential hazard of high concentration in prostate cancer (11) or other diseases in men, the negative association of testosterone with HST we observed in this study might be offset to some degree, accounting for non-significance in quintile 5 of TT. Thus, we conservatively suggest that this association is more stable at an intermediate dose (the second to fourth quintiles, Q2–Q4) in men, which is consistent with the guidelines by the European Endocrine Society that recommended achieving testosterone concentration in the mid-normal range during treatment (44). Previous studies indicated that a loss of testosterone due to aging contributed to the onset of physical frailty, such as triggering muscle loss, muscle weakness, and decreased functional performance, which eventually limited the lifespan in both genders (7, 45). Hence, an appropriate dose of testosterone is not necessarily harmful to women; our results also showed that HST risk only obviously increased at a high concentration of TT or FT (the top quintile, Q5). Besides, individuals with a very high or very low level of testosterone are strongly associated with health status (46, 47). Some diseases, such as pituitary disorders (48), HIV (49), and hyperthyroidism (50), as well as obesity (51), may interfere the testosterone production, increasing the possibility of extreme level of occurrence. It is worth noting that these diseases have a possible influence on personal health span as well. Therefore, we respectively excluded the participants with self-reported poor health status or outliers in the sensitivity analyses to control the potential bias introduced by the extreme values.

Given that both endocrine and genetic factors may contribute to health span collectively, we assess the joint effects of genetic risk and testosterone. Despite the non-significant interaction effect of testosterone and genetic propensity on health span, we still find the combined associations greater than individually. This suggests that those participants both with abnormal testosterone levels (especially high level in women and low level in men) and high genetic risk are in a more dangerous situation and deserve more attention. Therefore, taking appropriate intervention measures early, such as testosterone replacement therapy (TRT) for men with testosterone deficiency or anti-testosterone therapy for women with a supraphysiologic dose of testosterone, is considerable to extend personal health span.

Over the past two decades, TRT has been widely used for low age-related levels of testosterone in men (52) or even for postmenopausal women with low sexual desire (53), so there is a chance for overuse in the population. Therefore, current mainstream guidelines are cautious about such therapy and are critical about the usage and concentration (54, 55). For a better understanding of the roles played by exogenous testosterone in health span and to better assist clinical or personal decision-making before testosterone utilization, it is a wise choice to set endogenous testosterone as an entry point of research. Hopefully soon, we will be able to screen populations suitable for testosterone therapy by utilizing PRS in risk prediction for precise prevention.

Several limitations deserve to be noticed. Firstly, a single measurement for testosterone at baseline may not reflect long-term exposure very well. However, some researchers consider it reliable in categorizing average levels, as it calculates the intraclass correlation coefficients (ICCs) between two measurements of TT that were 4 years apart in a subcohort of UKB (13). Secondly, although the Vermeulen formula is the most widely used among different methods, calculating FT is still possible to be over- or underestimated relative to laboratory measurement (56). Third, given the aromatization of free testosterone into estradiol (57), further adjustment for estradiol is necessary to evaluate the independent association of FT in women. However, it is infeasible to perform at the current phase because most postmenopausal women in the UK Biobank have estradiol levels below the threshold for detection and premenopausal women also have a high missing rate of estradiol. Finally, there is evidence of a “healthy volunteer” selection bias (58), which suggests that it is warranted to replicate our findings in other cohorts to determine the applicability of our findings to other populations.

In conclusion, this prospective cohort study revealed the sex-specific associations that testosterone was negatively associated with HST risk in men and positively associated with HST risk in women. Genetic factors increased the HST risk, which indicated that participants with abnormal testosterone levels (high level in women or low level in men) and high genetic risk should be the target for early prevention. Although our findings highlight the association between testosterone and health span, further mechanistic studies and prospective trials are warranted to explore the causation behind.

Publicly available datasets were analyzed in this study. Datasets are available from the UK Biobank via application to the UK Biobank.

The studies involving human participants were reviewed and approved by the North West Multi-Centre Research Ethics Committee (http://www.ukbiobank.ac.uk/ethics/). The patients/participants provided their written informed consent to participate in this study.

JD supervised the entire project and designed the work. XZ conducted the statistical analysis and wrote the first draft. SL, NW, and TH helped apply for permission to use data and contributed to data management. MS and JF offered statistical support during the study. MZ, CW, DH, and YJ critically revised the manuscript for important intellectual content. All authors contributed to the article and approved the submitted version.

This work was supported by the National Natural Science Foundation of China (Grant ID: 81941020) as part of a project entitled “Identification of longevity-related genetic variations and their biological mechanisms based on a natural population cohort”.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.