Low Free Testosterone and Prostate Cancer Risk: A Collaborative Analysis of 20 Prospective Studies

Background Experimental and clinical evidence implicates testosterone in the aetiology of prostate cancer. Variation across the normal range of circulating free testosterone concentrations may not lead to changes in prostate biology, unless circulating concentrations are low. This may also apply to prostate cancer risk, but this has not been investigated in an epidemiological setting. Objective To examine whether men with low concentrations of circulating free testosterone have a reduced risk of prostate cancer. Design, setting, and participants Analysis of individual participant data from 20 prospective studies including 6933 prostate cancer cases, diagnosed on average 6.8 yr after blood collection, and 12 088 controls in the Endogenous Hormones, Nutritional Biomarkers and Prostate Cancer Collaborative Group. Outcome measurements and statistical analysis Odds ratios (ORs) of incident overall prostate cancer and subtypes by stage and grade, using conditional logistic regression, based on study-specific tenths of calculated free testosterone concentration. Results and limitations Men in the lowest tenth of free testosterone concentration had a lower risk of overall prostate cancer (OR = 0.77, 95% confidence interval [CI] 0.69–0.86; p < 0.001) compared with men with higher concentrations (2nd–10th tenths of the distribution). Heterogeneity was present by tumour grade (phet = 0.01), with a lower risk of low-grade disease (OR = 0.76, 95% CI 0.67–0.88) and a nonsignificantly higher risk of high-grade disease (OR = 1.56, 95% CI 0.95–2.57). There was no evidence of heterogeneity by tumour stage. The observational design is a limitation. Conclusions Men with low circulating free testosterone may have a lower risk of overall prostate cancer; this may be due to a direct biological effect, or detection bias. Further research is needed to explore the apparent differential association by tumour grade. Patient summary In this study, we looked at circulating testosterone levels and risk of developing prostate cancer, finding that men with low testosterone had a lower risk of prostate cancer.


Introduction
Experimental and clinical evidence implicates testosterone in the aetiology of prostate cancer. Nearly all metastatic prostate tumours overexpress the androgen receptor, and androgen deprivation therapy is the mainstay treatment approach for many prostate tumours [1]. Two large randomised controlled trials of 5a-reductase inhibitors (which block the conversion of testosterone to the more biologically active dihydrotestosterone [DHT]) showed a reduction in prostate cancer risk [2][3][4]. Genome-wide association studies and animal models also support an association between androgens and risk [5][6][7][8].
Despite the strong biological evidence of an association between testosterone concentration and prostate cancer risk, previous epidemiological studies have not found evidence of an association [9]. This may be because the association is nonlinear; variations across the normal range of circulating testosterone may not lead to alterations in prostate growth because the stimulation of prostatic androgen receptors may remain relatively constant, due to relatively constant intraprostatic DHT concentrations and/or saturation of the androgen receptors [10,11]. However, when the supply of testosterone to the prostate is abnormally low, prostate growth may decrease [12,13]. Therefore, we hypothesised that men with very low circulating testosterone concentrations may have a reduced risk of prostate cancer but that, above these low concentrations, prostate cancer risk is not associated with further increases in circulating testosterone concentrations. Less than 2% of testosterone circulates unbound to carrier proteins or "free", and is able to pass out of the blood into the prostate tissue [14]; therefore, the focus of our analysis was on free testosterone.
The Endogenous Hormones, Nutritional Biomarkers and Prostate Cancer Collaborative Group (EHNBPCCG) is a pooled individual participant dataset of prospective studies and prostate cancer risk. A previous analysis by this group found no association between prediagnostic androgen concentrations and prostate cancer [9]. However, this dataset has since been expanded to include almost double the number of prostate cancer cases and now comprises 20 prospective studies with a total of 6933 cases and 12 088 matched controls with calculated free testosterone data. This large dataset now provides sufficient power to examine whether

Abstract
Background: Experimental and clinical evidence implicates testosterone in the aetiology of prostate cancer. Variation across the normal range of circulating free testosterone concentrations may not lead to changes in prostate biology, unless circulating concentrations are low. This may also apply to prostate cancer risk, but this has not been investigated in an epidemiological setting. Objective: To examine whether men with low concentrations of circulating free testosterone have a reduced risk of prostate cancer. Design, setting, and participants: Analysis of individual participant data from 20 prospective studies including 6933 prostate cancer cases, diagnosed on average 6.8 yr after blood collection, and 12 088 controls in the Endogenous Hormones, Nutritional Biomarkers and Prostate Cancer Collaborative Group. Outcome measurements and statistical analysis: Odds ratios (ORs) of incident overall prostate cancer and subtypes by stage and grade, using conditional logistic regression, based on study-specific tenths of calculated free testosterone concentration. Results and limitations: Men in the lowest tenth of free testosterone concentration had a lower risk of overall prostate cancer (OR = 0.77, 95% confidence interval [CI] 0.69-0.86; p < 0.001) compared with men with higher concentrations (2nd-10th tenths of the distribution). Heterogeneity was present by tumour grade (p het = 0.01), with a lower risk of low-grade disease (OR = 0.76, 95% CI 0.67-0.88) and a nonsignificantly higher risk of high-grade disease (OR = 1.56, 95% CI 0.95-2.57). There was no evidence of heterogeneity by tumour stage. The observational design is a limitation. Conclusions: Men with low circulating free testosterone may have a lower risk of overall prostate cancer; this may be due to a direct biological effect, or detection bias. Further research is needed to explore the apparent differential association by tumour grade.

Study design
The majority of the studies were matched case-control studies nested within either prospective cohort studies or randomised trials. Four studies were cohort or case-cohort analyses. To apply a consistent statistical approach across all studies, the cases from the case-cohort studies were matched to up to four participants who were free of prostate cancer at the age at diagnosis of the case on the basis of our minimal matching criteria (Supplementary Table 2

Data processing
Free testosterone concentrations were calculated from total testosterone and SHBG concentrations using the law of mass action [15,16]

Statistical analysis
Conditional logistic regression was used to calculate the odds of prostate cancer diagnosis by hormone concentration. The analyses were conditioned on the matching variables and adjusted for age at blood collection, body mass index (BMI), height, usual alcohol consumption, smoking status, marital status, and education status as categorical variables, with an additional category for missing data, except for age (continuous). As we were interested a priori in the risk for prostate cancer in men with very low free testosterone concentrations, we categorised free testosterone concentrations into study-specific tenths, with cut points defined by the distribution in control participants, to allow for any systematic differences between the studies in assay methods and blood sample types [17], using the highest tenth as the reference category. To explore the association with greater power, these tenths were also grouped (1, 2-4, 5-7, 8-10), with the 8th-10th tenths combined as the reference category. In all further analyses, the 2nd-10th tenths were combined and used as the reference category. Where more than two categories of exposure were compared, variances were used to calculate floating confidence intervals, which facilitate comparisons between any two exposure groups [18,19].
PSA, IGF-I, and C-peptide concentrations at blood collection were available for subsets of participants. The main analyses of the relationships between low free testosterone and prostate cancer risk were examined in these subsets before and after further adjustment for these variables (log transformed PSA [continuous] and study-specific fifths of IGF-I, and C-peptide concentrations).
Heterogeneity among studies was assessed by comparing the x 2 values for models with and without a study Â analyte interaction term.

Results
A total of 20 studies, comprising 6933 cases and 12 088 controls, were eligible for this analysis. Mean age at blood collection in each study ranged from 33.8 to 76.2 yr (overall mean = 59.8 yr, standard deviation [SD] = 11.5 yr), and the year of blood collection ranged from 1959 to 2004. Study participants were predominantly of white ethnic origin (82%). The average time from blood collection to diagnosis was 6.8 yr, the average age at diagnosis was 67.9 yr (SD = 7.2), and most cases were diagnosed between 1995 and 1999 (39%). Prostate cancers were mostly localised (55%) and/or low grade (68%; Table 1). The free testosterone concentration cut points used for each study are shown in Supplementary Table 4. Men in the lowest study-specific tenth of free testosterone were older, and had a higher mean BMI and lower PSA at blood collection than men with higher free testosterone concentrations (Table 2).
There was evidence of heterogeneity by tumour grade (x 2 1 = 7.35; p het = 0.01); a low concentration of circulating free testosterone was associated with a reduced risk of low-grade prostate cancer (OR = 0.76; 95% CI 0.67-0.88), while there was a nonsignificantly increased risk of high-grade prostate cancer (OR = 1.56; 95% CI 0.95-2.57; Fig. 4. There was no evidence of heterogeneity in the association by tumour stage, aggressiveness, PSA era, or other characteristics (Fig. 4).

Discussion
Our results indicate that men in the lowest study-specific tenth of calculated free testosterone concentration have a 23% reduced risk of prostate cancer compared with men with higher concentrations. Above this very low concentration, prostate cancer risk did not change with increasing free testosterone concentration. We also found evidence that this association varied by tumour grade. This is the largest collection of data on hormones and prostate cancer risk available, and is the first large-scale prospective evidence supporting an association between low free testosterone concentrations and prostate cancer risk. The observed association between low free testosterone and lower prostate cancer risk may be due to a direct biological effect. Across the normal range of circulating free testosterone concentrations, stimulation of prostatic androgen receptors may remain relatively constant, due to stable intraprostatic DHT concentrations and/or saturation of androgen receptors [10,11]. Therefore, variation across the normal range of circulating free testosterone concentrations may not be associated with a prostate cancer risk. However, when circulating concentrations are very low, reduced androgen receptor signalling may lead to a reduction in prostate cancer risk [10,11,20].
An alternative explanation for the main findings may be detection bias. Controls with low free testosterone concentrations had low PSA concentrations at blood collection, and adjustment for PSA concentration in a subset of our dataset attenuated the association of low free testosterone and prostate cancer risk towards the null. However, there was no evidence of heterogeneity in the associations between men diagnosed before and after 1990, before which there was relatively little PSA testing [21]. PSA is partly regulated by the androgen receptor [12,22]; therefore, it is difficult to disentangle the relationship between these variables in this observational analysis [12,21,22]. While there was no evidence of heterogeneity in the association of free testosterone with prostate cancer risk by tumour stage or aggressiveness, there was evidence of heterogeneity in this association by tumour histological grade; a low free testosterone concentration was associated with a lower risk of low-intermediate-grade prostate cancer, and there was a nonsignificantly increased risk of high-grade disease. Although it is possible that this heterogeneity is a chance finding due to the multiple tests [ ( F i g . _ 1 ) T D $ F I G ] Fig. 1 -Associations between risk of overall prostate cancer and study-specific tenths of hormone concentrations. Estimates are from logistic regression conditioned on the matching variables and adjusted for age, BMI, height, alcohol intake, smoking status, marital status, and education status. The position of each square indicates the magnitude of the relative risk, and the area of the square is proportional to the amount of statistical information available (inverse of the variance of the logarithm of the relative risk). The length of the horizontal line through the square indicates the 95% floated confidence interval. BMI = body mass index; FCI = floated confidence interval; OR = odds ratio; SHBG = sex hormone-binding globulin.
[ ( F i g . _ 2 ) T D $ F I G ] Fig. 2 -Odds ratio (95% FCIs) for overall prostate cancer associated with study-specific tenths of concentrations of free testosterone. Estimates are from logistic regression conditioned on the matching variables and adjusted for age, BMI, height, alcohol intake, smoking status, marital status, and education status. BMI = body mass index; FCI = floated confidence interval; OR = odds ratio.
conducted and the relatively small number of high-grade tumours, this pattern has been reported previously in the Health Professionals Follow-up Study [23], several clinical case studies [24][25][26], and the PCPT and Reduction by Dutasteride of Prostate Cancer Events (REDUCE) trials. These two trials investigated the effect of 5a-reductase inhibitors, which can reduce intraprostatic DHT concentration by approximately 80-90%, on prostate cancer risk [27]. Both trials reported a 23-25% reduction in overall prostate cancer [3,4]. However, the PCPT reported a 27% increase in high-grade (Gleason grade !7) tumours (n = 517) [4], and the REDUCE trial reported a 58% increased risk of high-grade (Gleason grade !8) tumours (n = 48) [3].
There are several possible explanations for the observed heterogeneity in the associations by tumour grade. Prostate tumour grade stays stable over several years [28], suggesting that high-grade tumours develop de novo rather than from the dedifferentiation of low-intermediate-grade tumours. Mechanistically, prostatic androgen-androgen receptor binding is an important modulator of cell differentiation [29]; thus, prostate cells with reduced androgen exposure may be less differentiated and more likely to develop into high-grade tumours [30]. Alternatively, this may be a differential growth response of early lowgrade cancer lesions to a low androgen environment. Another possibility is differential detection bias as discussed in relation to PCPT and REDUCE [31][32][33][34][35][36]. Owing to the clinical importance of high-grade tumours, this observed heterogeneity by grade, with a possible higher risk of high-grade tumours, requires further investigation.
Our study has a number of limitations. Free testosterone was calculated using the law of mass action [15,16], which is based on testosterone and SHBG concentrations and assumes a constant albumin concentration. Although this is a commonly used method of estimating free testosterone concentration, it has not been validated within each individual study via equilibrium dialysis [14]. The assay methods used to measure analytes varied, with the majority of studies using nonextraction assays to measure testosterone. While this may introduce some misclassification, this would be expected to be nondifferential and therefore tend to bias any association towards the null. Mass spectrometry is often considered the gold standard method to measure sex hormone concentrations [37], but high-quality immunoassays are able to measure reliably low adult male testosterone concentrations [38,39]. Although these assays may not be suitable for determining absolute clinical cut points, they are considered appropriate for the determination of relative concentrations within studies [39]. Our study relied on single measurements of testosterone and SHBG, with an average time from blood collection to diagnosis of 6.8 yr, to represent participants' hormone concentrations over medium to long term. While several studies show that a single measure of these analytes has moderately good reproducibility over periods of up to 1 yr [40], it is unknown whether these measures are reliable over the longer term.
[ ( F i g . _ 3 ) T D $ F I G ]   4 -ORs (95% CIs) for prostate cancer associated with free testosterone in the study-specific 1st tenth compared with the 2nd-10th tenths, according to characteristics of cases and controls. Estimates are from logistic regression conditioned on the matching variables and adjusted for age, BMI, height, alcohol intake, smoking status, marital status, and education status. BMI = body mass index; CI = confidence interval; IGF = insulin-like growth factor; OR = odds ratio; PSA = prostate-specific antigen. Tests for heterogeneity for case-defined factors were obtained by fitting separate models for each subgroup and assuming independence of the ORs using a method analogous to a meta-analysis. Tests for heterogeneity for non-casedefined factors were assessed with a x 2 test of interaction between subgroup and the binary variable. a 1st study-specific tenth of free testosterone. b 2nd-10th study-specific tenths of free testosterone. 5.

Conclusions
In summary, the findings from this pooled prospective analysis of 6933 prostate cancer cases and 12 088 controls support the hypothesis that very low concentrations of circulating free testosterone are associated with a reduced risk of prostate cancer. Further research is needed to elucidate whether the association is causal or due to detection bias, and explore the apparent differential association by tumour grade.
Author contributions: Eleanor L. Watts had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.