From 1996 to 2001, 1961 men were enrolled in the Western New York Health Cohort Study (WNYHCS). A detailed description of the WNYHCS study design, methods and participants' characteristics is available elsewhere [14].
In brief, all participants provided informed consent; the Human Subjects Review Board of the University at Buffalo, School of Medicine and Biomedical Science approved procedures for protection of human subjects in the study. At the time of recruitment, trained interviewers collected extensive data on demographics and life style during in-person interviews. The use of a standardized protocol allowed for the collection of anthropometric data. The study participants donated morning spot urine which was kept at -80°C until biochemical determinations.
From January 2003 through September 2004, we completed the Western New York Health Cohort (WNYHC) re-call and follow-up. For the purposes of the present case-control study (PROMEN II study), the re-call process included male participants who met the following inclusion criteria: age at recruitment between 50 and 85, baseline history negative for malignancies, cardiovascular diseases and clinically defined type-2 diabetes. On this basis, the re-call and follow-up process involved 1092 cohort participants. Among them, 52 were not eligible for medical reasons other than Pca, 46 had died from causes other than Pca, 22 had moved out of Erie and Niagara Counties, and 117 were not able to be contacted by mail or phone. Among the remaining 855 study participants, 232 refused to join the study, 40 were scheduled but cancelled the appointment and 8 were still in-course of assessment at the end of the follow-up period. In this group of non-participating subjects, all of the cohort members referred to being free from Pca in their telephone interviews. Thus, 575 participants joined the study, accounting for an overall participation rate of 67% (575/855).
Pca cases were men who had been diagnosed with incident, histologically confirmed Pca within the time-frame between their recruitment in the WNYHC and the end of the follow-up period. Identifying Pca cases was based on the participants' reports at the re-call, which was subsequently validated by clinical records provided by their urologists. We identified and validated a total number of 41 incident prostate cancer cases. The 534 control subjects were male members of the WNYHC who, based on their report, were free from clinically evident Pca at the time of diagnosis of the related case. The control status was validated with a serum PSA assessment on a blood sample donated at the time of recall. We used a PSA cut-off value of 4 ng/ml [15]. Among the study participants whose PSA level was higher than 4 ng/ml, we ultimately included in the control group only those who tested negative at the prostate biopsy. We requested and obtained the pertinent medical records from the urologists.
For each case, four control subjects were randomly chosen after matching for age (within a 3-year-range), race and date of recruitment. The independent variables of interest, namely 2-OHE1, 16α-OHE1 and the 2-OHE1 to16α-OHE1 ratio, were available for 110 controls and 26 cases, thus we conducted the present analysis on 136 subjects.
Hormonal Determinations
For standardization purposes, we collected morning spot urine between 7:00 a.m. and 9:00 a.m. from all participants. We then transferred the aliquoted urine samples to the Eppley Institute, University of Nebraska Medical Center (UNMC), and stored them at -80°C until analysis. Each sample was thawed only once prior to analysis. We handled urine samples identically and located them in the laboratory runs randomly. All laboratory personnel were blinded in regards to case-control status. All of the study samples were analyzed in duplicate. Two-milliliter aliquots of urine were partially purified throughout solid phase extraction (SPE) with a phenyl cartridge (Varian, Palo, Alto, CA) and ultra-performance liquid chromatography/tandem mass spectrometry (LC/MS-MS). Analytes were identified based on their retention time and tandem mass spectrometry. Standards of the catechol estrogens 2-OHE1(E2) and 16α-OHE1(E2) were purchased from Steraloids Inc. (Newport, RI).
To avoid the artifacts and errors introduced by maintaining the urine samples at 37°C for 8 hours, we carried out all the analyses without glucuronidase/sulfatase treatment. We adjusted urine samples to pH 7 with 1 M NaOH or 1 M HCl.
We performed the LC/MS analyses through a Waters Acquity ultra-performance liquid chromatography (UPLC) system connected with a high performance Quattro Micro triple quadruple mass spectrometer designed for LC/MS-MS operation. We performed the analytical separations on the UPLC system using an Acquity UPLC BEH C18 1.7 μm column (1 × 100 mm) at a flow rate of 0.15 ml/min. We then moved the elutions from the UPLC column to the Quattro Micro mass spectrometer.
The ionization method used for MS analysis was Electrospray ionization (ESI) in both the positive ion (PI) and negative ion (NI) mode with an ESI-MS capillary voltage of 3.0 kV, an extractor cone voltage of 3 V, and a detector voltage of 650 V. We performed the MS-MS in the multiple reaction monitoring (MRM) mode to produce structural information about the analytes by fragmenting the parent ions inside the mass spectrometer and identifying the resulting daughter/fragment ions. We processed the resulting data and quantified the estrogen metabolites using the QuanLynx software (Waters).
To calculate limits of detection, we injected various concentrations of the analytes to LC/MS-MS. The detection limit was considered to be the injected amount that resulted in a peak with a height at least two or three times higher than the baseline. The detection limits of 2-OHE1 and 16α-OHE1 were 18 fmol and 349 fmol, respectively. Intra-assay coefficients of variation for 2-OHE1 and 16α-OHE1 were 3.2% and 3.0%, respectively. Inter-assay coefficients of variation were 1.9% and 3.5%, respectively.
We had previously measured the intra- and inter-individual variability for 2-OHE1, 16α-OHE1 determinations and their ratio over a one year period [13]. The intra-class correlation coefficients (ICCs) and lower limit of 95% CI (in parentheses) were 0.70 (0.46), 0.63 (0.35) and 0.78 (0.62), respectively. We had previously provided a detailed description of the procedures related to the reliability assessment [13].
Systematic Review
We conducted a systematic search of the literature to identify additional studies published up to August 2009 which examined the association between estrogen metabolites and Pca risk using our standard methods [19–22]. We searched MEDLINE (January 1966 onwards) and EMBASE (January 1980 onwards). An expert librarian designed a search strategy combining terms for estrogens, estrogen metabolites and prostate specific antigen (PSA) with terms for Pca (available upon request). We screened titles and abstracts in duplicate using the following inclusion criteria: observational studies investigating prostate cancer risk in relation to estrogen metabolism. We included studies providing at least one measure of either urinary or circulating levels of 2-OHE1, 16α-OHE1 and the 2-OHE1 to 16α-OHE1 ratio.
Statistical analysis
We examined distributions for all variables of interest by determining the frequencies, mean, median and measures of variance. To evaluate the statistical significance of the unadjusted associations between case/control status and participants' characteristics, we used either Fisher's exact tests or Pearson's chi-square tests for categorical variables.
The 2-OHE1 and 16-αOHE1 urinary levels were standardized by total urinary creatinine. We used unconditional logistic regression to compute crude and adjusted odds ratios (OR) and 95% confident interval (CI) of Pca in relation to 2-OHE1, 16-αOHE1 and the ratio of 2-OHE1 to 16α-OHE1 by tertiles of urine concentrations. We used the same models to test for significance in trends of association for any of the independent variables. We computed the cut-off points of the previously mentioned tertiles based on the distributions of estrogen metabolites in control subjects. We analyzed each independent variable separately. Based on the published literature, we identified age, race, education level, BMI and waist-to-hip ratio as possible covariates and tested them using regression models. Although none of them was a confounder for the investigated associations, we included age in years in further analyses based on its biological relevance in prostate carcinogenesis [2].
We verified several sources of potential bias. Because the exclusion of participants with missing data for any of the two outcome variables could have introduced a source of bias in our final sample, we examined data by subsets. Each of the two datasets included men with no missing data for either urinary levels of 2-OHE1 or 16-αOHE1. We then examined by case-case and control-control comparing the characteristics of the 136 subjects (110 controls and 26 cases) with no data missing for any of the considered variables and those of the subjects (534 controls and 41 cases) who fulfilled our study eligibility criteria. Finally, we compared the subjects in the latter category [575] to the 517 original cohort members who did not join the study either because they did not fulfil the inclusion criteria, were lost to follow-up or were not willing to participate.
To date, no data exists related specifically to any of these three categories (i.e. co-morbidity data pertinent to the WNYCS). Thus, we considered these 517 male subjects as part of an overall, although heterogeneous, category. As expected, the 517 males from the original cohort who did not ultimately join our study showed statistically significant differences when compared to the 575 included study participants. We analyzed these data using SPSS version 14.0 (SPSS, Inc., Chicago, IL).
Meta-analysis
We planned to combine the results from the current study with those identified in the systematic review using the DerSimonian-Laird random effects method expressing the pooled estimates in terms of summary OR and 95% CI. We calculated I2 to assess heterogeneity across study results applying the following interpretation for I2 (J Higgins, personal communication): 0-50 = low; 50-80 = moderate and worthy of investigation; 80-100 = severe and worthy of understanding; 95-100 = aggregate with major caution [23]. We used Revman 5.0 for the meta-analysis (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008).