Materials and Methods
The HPFS is an ongoing prospective cohort study including 51 529 male health professionals who were between 40 and 75 years old at enrolment in 1986. At baseline, the men completed a mailed questionnaire on lifestyle characteristics, medical history and a food frequency questionnaire. Since then, participants have been followed every 2 years by mailed questionnaires to ascertain the occurrence of new medical diagnoses as well as to update exposure status. Information on dietary intake has been updated every 4 years. Between 1993 and 1995, 18 225 participants provided a blood sample, collected in blood tubes containing liquid EDTA and shipped to the laboratory via overnight courier while chilled on ice. At the laboratory, the blood was centrifuged, divided into plasma, erythrocytes, and buffy coat and stored in liquid nitrogen freezers. The study was approved by the Institutional Review Board of Harvard School of Public Health and Brigham and Women’s Hospital in Boston, Massachusetts, and written informed consent was obtained from all participants.
In the present study, we included participants who were selected for a nested case–control study on prostate cancer. The nested case–control study included prostate cancer cases verified by medical records and control subjects selected among participants who were alive and free of diagnosed cancer at the date of the case’s diagnosis, and who had a prostate-specific antigen (PSA) test after the date of blood draw. For each prostate cancer case, one control participant was matched on year of birth (±1 year), PSA test prior to blood draw (yes/no) and timing of blood draw – time of day (midnight to before 9 am, 9 am to before noon, noon to before 4 pm and 4 pm to before midnight), season (winter, spring, summer and fall) and year (exact).
Measurement of Vitamin D and Testosterone
The cases and corresponding controls were identified in three waves of follow-up resulting in three assay batches: blood draw to 1996, 1996–1998 and 1998–2000. Case–control pairs were analysed together with the laboratory personnel being blinded to case–control status.
Plasma concentrations of 25(OH)D were determined by radioimmunosorbent assay (RIA; DiaSorin Inc., Stillwater, MN, USA) in the laboratory of Dr Bruce Hollis as described previously.[7] The mean intrapair coefficient of variation calculated from blinded quality control samples was 5·4%. Missing data occurred because of too low plasma volume (two samples).
Plasma concentrations of sex steroid hormones and sex-hormone-binding globuline (SHBG) were measured in the laboratory of Dr Nader Rifai at the Children’s Hospital, Boston, Massachusetts. Total testosterone was measured by means of a chemiluminescent immunoassay (Elecsys autoanalyser; Roche Diagnostics, Indianapolis, IN, USA), free testosterone by means of enzyme immunoassay (Diagnostic System Laboratories, Webster, TX, USA), oestradiol by means of a third-generation RIA (Diagnostic Systems Laboratory) and SHBG by means of coated tube noncompetitive immunoradiometric assay (Diagnostic Systems Laboratory). The mean intrapair correlation coefficients of variation were 4·9% for total testosterone, 8·4% for free testosterone, 5·2% for oestradiol and 10·7% for SHBG. Missing data occurred because samples were too lipaemic for free testosterone assay (13 samples) or because of too low plasma volume (three samples).
Plasma total cholesterol concentration was measured by means of Infinity Total Cholesterol Enzymatic Assay kit (Sigma Diagnostics, St Louis, MO, USA) for the first and second analysis batches and an enzymatic assay using reagents from Equal Diagnostics (Exton, PA, USA) for the third analysis batch. The mean intrapair coefficient of variation for cholesterol calculated from blinded quality control samples was 10·9% for the Infinity assay and 9·1% for the Equal Diagnostics assay.
As reported previously,[8,9] to assess the intraperson consistency of 25(OH)D or sex steroid hormones over time, 25(OH)D, total testosterone, free testosterone, oestradiol and SHBG were measured for 144 HPFS participants who were free of a cancer diagnosis and who provided a blood sample in 1993/1994 and again in 1997 (mean of 3·03 ± 0·46 years apart). Adjusting for age, race and season of the year, the Pearson’s correlation coefficient between the two time points was 0·70 for 25(OH)D and Spearman’s correlation coefficients were 0·68 for total testosterone, 0·39 for free testosterone, 0·55 for oestradiol and 0·74 for SHBG (all P < 0·0001).
Statistics
To avoid confounding by race/ethnicity, we excluded nine African-American and five Asian participants. All remaining participants from the nested case–control study for whom both 25(OH)D and plasma sex steroid hormones were available (n = 1362) were included in the present analysis. Neither 25(OH)D nor sex steroid hormones have been related to total prostate cancer in HPFS.[8,9] Thus, to increase power, we included both cases (who were not diagnosed with cancer at time of blood draw) and controls in our final analysis, but all analyses were repeated including control participants (n = 678) only. In sensitivity analyses, we restricted the study population to controls and additionally excluded participants who reported major morbidities such as diabetes, high cholesterol and high blood pressure, which could be an indicator for lower 25(OH)D levels.
We used multivariate linear regression with robust variance (PROC MIXED with empirical statement, version 9.2; SAS Institute, Cary, NC, USA)[10] to investigate the association between 25(OH)D and total testosterone, free testosterone or oestradiol. This method allows for valid inference without assumption of normal distribution in the dependent variable. To account for differences in absolute concentrations between analysis batches, free testosterone levels were batch standardized.[11] Using regression coefficients for batch derived from a linear regression model with free testosterone as dependent variable and batch indicators as independent variable, free testosterone levels were standardized to the levels that would be obtained if all measurements were from batch 1. Batch standardization of total testosterone and oestradiol was not necessary because mean values were similar across batches. As an alternative approach, we calculated free testosterone from total testosterone and SHBG under the assumption of a constant albumin concentration of 43 g/l according to Vermeulen et al. [12] Multivariate-adjusted mean sex steroid hormone concentrations and corresponding 95% confidence intervals (95% CI) were calculated by batch-specific quintiles of 25(OH)D. To test for linear trend, participants were assigned the median values of the batch-specific 25(OH)D quintiles and the resulting variable was entered continuously to the model, deriving the P-values for linear trend from this variable’s Wald test. In addition, continuous parameter estimates representing the change in testosterone per 25 nmol/l increment in 25(OH)D (based on trend variable) are presented. The fully multivariate-adjusted model included age, body mass index (BMI), analysis batch, time of blood collection (four categories: before 9 am, 9 am–12 pm, 12 pm–4 pm and after 4 pm), season, geographical region (West, Midwest, South and Northeast), smoking status and physical activity [metabolic equivalent-hours/week (MET-h/wk)]. Total testosterone was additionally adjusted for SHBG in a separate model, which can be interpreted similarly to the multivariate model for free testosterone. The presented age- and batch-adjusted mean values were centred to age 60 years and batch 1; age-, batch- and BMI-adjusted models were centred to age 60 years, batch 1 and BMI 25 kg/m2; multivariate-adjusted models’ means were centred to age 60 years, batch 1, BMI 25 kg/m2 and mean physical activity (40 MET-h/wk). In a sensitivity analysis, we tested whether additional adjustment for batch-specific deciles of plasma cholesterol, which is a precursor for both testosterone and vitamin D, changes associations. We examined possible nonlinear associations using restricted cubic splines[13] with four knots to divide continuous 25(OH)D concentration into five intervals. To make the graph more stable, highest and lowest percentile of 25(OH)D levels were excluded.
To investigate whether the association between 25(OH) vitamin D and testosterone varies by age, BMI, vasectomy or season analyses stratified by these factors were performed. Tests for statistical interaction were performed by creating a cross-product variable of the medians of 25(OH)D quintiles and the respective stratification variable and testing the difference of the model with and without this variable by means of the likelihood ratio test. Because testosterone levels measured in morning samples are considered more accurate,[14] we repeated all analyses excluding blood samples drawn in the afternoon (n = 236).
In an additional analysis, we defined hypogonadism based on low total testosterone levels (<11 nmol/l)[15] and estimated the relative risk (approximated by the odds ratio) of hypogonadism by means of logistic regression. The multivariate-adjusted model included all covariates described earlier. We determined the seasonal variation of 25(OH)D, total testosterone, free testosterone and oestradiol by calculating age- and batch-adjusted mean concentrations by month of blood collection for both biomarkers using linear regression with robust variance as described earlier.
