University of Birmingham Serum testosterone, sex hormone-binding globulin and sex-specific risk of incident type 2 diabetes in a retrospective primary care cohort

Citation for published version (Harvard): O'reilly, M, Glisic, M, Kumarendran, B, Subramanian, A, Manolopoulos, K, Tahrani, A, Keerthy, D, Muka, T, Toulis, K, Hanif, W, Thomas, GN, Franco, OH, Arlt, W & Nirantharakumar, K 2019, 'Serum testosterone, sex hormone-binding globulin and sex-specific risk of incident type 2 diabetes in a retrospective primary care cohort', Clinical Endocrinology, vol. 90, no. 1, pp. 145-154. https://doi.org/10.1111/cen.13862


INTRODUCTION
hormones such as androgens may mediate these differences, but the association between androgens and 58 metabolic dysfunction is complex and sex-specific 2 . Androgen excess has recently been identified as an 59 independent risk factor for non-alcoholic fatty liver disease (NAFLD) in women 3 , and promotes lipid 60 accumulation in female adipose tissue as well as systemic lipotoxicity 4 . Female-to-male gender 61 reassignment patients undergoing androgen therapy develop dyslipidemia and abnormal body 62 composition 5,6 . Mirroring this, the adverse metabolic phenotype of male androgen deficiency bears a 63 striking similarity to that of female androgen excess; lower testosterone levels in men are associated with 64 impaired glucose homeostasis, hepatic steatosis and coronary artery disease [7][8][9] . A number of metaanalyses support a sex-specific relationship between androgens and the risk of metabolic dysfunction, and 66 suggest that low circulating sex hormone-binding globulin (SHBG) concentrations may be metabolically 67 harmful in both sexes 9,10 .

68
Delineating an independent role for androgens in the pathogenesis of T2DM is confounded by 69 changes in body composition, body mass index and lean mass observed in disorders of androgen excess 70 and deficiency 11 . Against the background of a global epidemic of T2DM 12 , there is an urgent health need 71 to understand the sexually dimorphic role played by androgens in the pathogenesis of hyperglycemia. The 72 shared constellation of risk factors observed in women with androgen excess and men with androgen 73 deficiency suggests that circulating androgen concentrations common to both disorders may be 74 metabolically disadvantageous 2 . To our knowledge, however, no large longitudinal studies have 75 specifically examined the association between circulating androgen exposure per se and risk of T2DM in 76 a sex-specific context.

77
The aim of this study was to investigate the independent sex-specific association between serum 78 testosterone concentrations and the risk of hyperglycaemia in men and women by undertaking a 79 retrospective cohort study in a large and diverse UK population base.

124
Townsend quintiles and smoking status. In women, an additional model included polycystic ovary 125 syndrome (PCOS) as a covariate to explore if the risk of T2DM in women was independent of a diagnosis 126 of PCOS. In an additional sensitivity analysis, when adjusting for PCOS. we accepted the presence of hirsutism and anovulation as indicative of PCOS given that the diagnosis is underreported in primary <25.0kg/m 2 , 25-29.0kg/m 2 and >30kg/m 2 . All analyses were performed in STATA 14.0.

133
In women, we performed stratified analysis by age (<50 years and 50 years and above) to explore 134 if the association was similar before and after the average age of menopause. A similar age-stratified 135 analysis was also carried out in men. In addition to this, in those patients with simultaneous 136 measurements of testosterone and SHBG, a free androgen index (FAI) was calculated [(Tx100)/SHBG], 137 and risk of T2DM calculated to control for the confounding effect of low SHBG levels.

139
This study used routinely collected, anonymised primary care data. Patients were not involved in 140 the study, and therefore no consent was required. Research using THIN data was approved by the NHS  (Table 1). A diagnosis 158 of PCOS was only documented in 6.3% (N=5,136) and 7.9% (N=3,303) of the female testosterone and 159 SHBG cohorts, respectively. However, clinical features suggestive of PCOS, anovulation and clinical 160 evidence of hirsutism, were documented in 25.8% and 11.2% of the female testosterone cohort, 161 respectively, and in 26.9% and 12.1% of the female SHBG cohort, respectively.

213
Subgroup analyses 214 Subgroup analysis stratified by age (<50 and 50 years) did not alter the observed associations. In independent sex-specific effect on the risk of incident T2DM. To our knowledge, this is the largest study,

231
This finding is in agreement with observations from some previous studies, which demonstrated a 232 stronger inverse association between SHBG levels and risk of T2DM in women compared to men 10,24 .

233
This inverse relationship with T2DM appears to be particularly strong in postmenopausal women 25 . A 234 2011 meta-analysis, however, found that higher SHBG levels were equally associated with a reduced risk 235 of metabolic syndrome in both sexes 26 .

236
A systematic review and meta-analysis, which included a total of 3825 men and 4795 women in 237 36 cross-sectional studies, as well as 368 cases from 7 prospective study populations, previously 238 demonstrated that increased serum testosterone was associated with a 60% higher risk of T2DM in 239 women; higher testosterone levels in men reduced the risk of T2DM by 42% 10 . Goodman-Gruen et al 240 also observed sex differences in the association between serum androgens and glucose tolerance status in 241 an older community cohort of 775 men and 633 women above the age of 55 27 . Men with impaired fasting glucose, impaired glucose tolerance and T2DM had significantly lower levels of serum testosterone, 243 while women with T2DM had significantly higher levels of bioavailable testosterone, independent of total 244 body fat, fat distribution, physical activity and smoking. However, our study is the only longitudinal 245 retrospective analysis to comprehensively evaluate these associations.

256
Male androgen deficiency occurs as a consequence of primary testicular pathology, 257 hypothalamic-pituitary disorders, obesity or as part of the ageing process in older men 32,33 . Additionally, 258 iatrogenic hypogonadism due to androgen deprivation therapy is observed in men with prostate cancer 34 .

259
Whilst the relationship between obesity and hypogonadism in men is complex and bidirectional 35 , data 260 from male cohorts treated with short term androgen deprivation therapy show that hypogonadism directly 261 induces metabolically deleterious changes in body composition, with increases in weight and in 262 percentage fat body mass 36 . However, studies of androgen deprivation therapy, which result in significant 263 hypogonadism, are not an ideal model to compare to the relatively modest reductions in testosterone 264 observed in community-dwelling older men. The results of our study are particularly surprising, given 265 that an increased risk of T2DM was apparent at circulating testosterone concentrations considered physiologically normal, but below the reference group of 20nmol/L, independent of age, obesity and other must be interpreted with caution in both men and women, and is particularly inaccurate in women when 295 the SHBG concentration falls below 30nmol/l 47 .

296
This study has a number of important limitations, not least its retrospective nature. Detailed 297 clinical phenotyping in studies of this type are not possible. There are also no detailed data available on 298 laboratory assays used to measure serum testosterone. This is not of particular concern in men, as 299 physiologically higher testosterone concentrations do not represent a challenge for quantification by either 300 radioimmunoassay (RIA) or tandem mass spectrometry techniques. In women, however, where low 301 circulating concentrations pose significant analytical and quantification difficulties for standard RIAs, the 302 consensus is that today measurements should be performed by liquid chromatography-tandem mass 303 spectrometry to improve quantification and avoid cross reactivity 48 . Furthermore, we have no information 304 on the time of day blood sampling for serum testosterone took place; in men, Endocrine Society 305 guidelines emphasize that morning samples are crucial to accurately diagnose hypogonadism 49 . Lastly,

306
we must assume that testosterone data were obtained from men and women with a clinical indication for 307 serum measurement; this introduces a potential bias by indication. However, we believe that these 308 limitations are ameliorated by the large patient numbers and the clear and potent gradient towards sex-309 specific T2DM risk in the study population.

310
In conclusion, in the largest retrospective longitudinal study of its kind, we have demonstrated 311 evidence of a sexually dimorphic role for androgens in mediating the risk of T2DM. Reduced SHBG 312 levels in both sexes, but particularly in women, significantly increase the risk of T2DM. These data 313 further define androgens as a novel metabolic risk factor in men and women, but potential mechanisms 314 underpinning these observations remain to be clarified. We suggest that women with androgen excess and 315 men with androgen deficiency should be systematically screened for T2DM. Future studies will be 316 required to show if reducing androgens in women, and increasing androgens in men, will improve overall 317 metabolic health and risk of progression to overt T2DM.