A detailed description of cohorts participating in VDPP and the general methods for the project are provided in a separate paper (18
). Of the 10 cohorts in VDPP, the following 8 contributed cases of kidney cancer: the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC); the Cancer Prevention Study II Nutrition Cohort (CPS-II); CLUE; the Multiethnic Cohort Study (MEC); the New York University Women's Health Study (NYU-WHS); the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO); the Shanghai Men's Health Study (SMHS); and the Shanghai Women's Health Study (SWHS).
From these 8 cohorts, plasma/serum samples from 783 kidney cancer cases and 783 controls, alive and not diagnosed with cancer at the time of case diagnosis, and matched on age at blood collection (±1 year), sex, race (white/black/Asian/other), and date of blood draw (±30 days), were sent to the laboratory for measurement of circulating 25(OH)D. Five cases were excluded after 25(OH)D measurement because they were found to have cancer of the ureter (International Classification of Diseases, Eighth Revision (ICD-8)/International Classification of Diseases, Ninth Revision (ICD-9) code 189.2 or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code C66 or International Classification of Diseases for Oncology (ICD-O) code C66.9) or cancer of a urinary organ, site unspecified (ICD-8/ICD-9 code 189.9 or ICD-10 code C68 or ICD-O code C68–C68.9). Furthermore, 2 cases were excluded after 25(OH)D measurement because their date of diagnosis was prior to their date of blood draw, and 1 control was excluded because no 25(OH)D data were returned from the laboratory. The remaining 776 cases and 782 controls were eligible to be included in the analytic data set. Because only matched pairs were analyzed, the final analytic data set consisted of 775 matched case-control pairs. Cases included 708 with renal cell carcinoma (ICD-8/ICD-9 code 189.0 or ICD-10 code C64 or ICD-O code 64.9) and 67 with carcinoma of the renal pelvis (ICD-8/ICD-9 code 189.1 or ICD-9/ICD-10 code C65 or ICD-O code 65.9).
All serum/plasma samples were measured for 25(OH)D at Heartland Assays, Inc., as described in Gallicchio et al. (18
). The quality control measures included use of samples of the vitamin D standard from the National Institute of Standards and Technology (NIST) at both a “normal” (~65 nmol/L) and a “low” (~35 nmol/L) level. As reported in Gallicchio et al. (18
) for the VDPP, interbatch and intrabatch coefficients of variation for low-level samples were 12.7% and 9.3%, respectively; interbatch and intrabatch coefficients of variation for normal-level samples were 13.6% and 11.0%, respectively. The median interbatch coefficient of variation for the cohort quality control samples was 13.2% (range: 4.8%–17.0%); the median intrabatch coefficient of variation for the cohort quality control samples was 9.9% (range: 3.8%–16.4%).
The main analyses described in this paper were conducted by using clinically defined cutpoints (4
): <25 nmol/L, 25–<37.5 nmol/L, 37.5–<50 nmol/L, 50–<75 nmol/L, 75–<100 nmol/L, and ≥100 nmol/L. The referent category chosen was 50–<75 nmol/L because this range includes the mean level of the US population (62.91 nmol/L, standard error of the mean, 0.81 for males; 61.54 nmol/L, standard error of the mean, 0.85 for females) according to 2000–2004 National Health and Nutrition Examination Survey data (21
In addition to the core VDPP variables, the following additional variables were requested for inclusion in the kidney cancer analyses: family history of renal cancer, history of high blood pressure at blood draw, and history of diabetes at blood draw. The Wald statistic, generated by using conditional logistic regression, was used to test the significance of differences between the kidney cancer cases and controls for selected variables, including median 25(OH)D concentrations. Variables in the VDPP data set that were 1) associated with both 25(OH)D and kidney cancer or 2) identified based on the literature as risk factors for kidney cancer, or both, were included in the fully adjusted models; these variables were education, body mass index, height, smoking status, current alcohol drinking, history of high blood pressure; and history of diabetes. All of these variables were treated as categorical in the model with the exception of height, which was treated as a continuous variable. Data on physical activity were also obtained because this variable has been shown to be associated with both kidney cancer and circulating 25(OH)D (22
); however, this variable may be in the causal pathway and not a confounder. For this reason, physical activity was not included in the fully adjusted model.
Conditional logistic regression models adjusting for the variables listed above were constructed to estimate the overall odds ratios and 95% confidence intervals for kidney cancer by 25(OH)D concentration, examined by using the 6-category clinically defined cutpoint variable. Trend tests were conducted by assigning values of 1–6 to the respective 25(OH)D categories and treating the new variable as continuous in the model. Unconditional logistic regression analyses using a 5-category 25(OH)D variable with the categories of 75–<100 nmol/L and ≥100 nmol/L combined adjusting for the matching factors and the other potential confounders listed above were also conducted to examine the associations between 25(OH)D and kidney cancer in strata based on predefined categories of sex, season (winter/summer), age at blood collection, race, latitude, length of follow-up, calcium intake/supplementation, body mass index, cigarette smoking, history of high blood pressure, and history of diabetes. No differences in the investigated association were observed in these stratified analyses; thus, with the exception of the sex- and season-specific analyses, the results of the stratified analyses are not presented here.
Additionally, pooled analyses using cohort-, sex-, and season-specific quartile cutpoints (using the 2-season variable), based on the distribution of circulating 25(OH)D among all controls from all VDPP cancer sites combined, were also conducted. Both conditional and unconditional models were run with the lowest fourth as the referent category; the results were similar. Analyses to adjust for season using the residual method were also conducted, as described in Gallicchio et al. (18
). Residual adjustment for season takes into account the gradual nature of changes in concentrations of 25(OH)D over the year, which may be better than adjusting for season. For the residual analyses, conditional logistic regression models were run with the lowest fourth as the referent category.
All analyses described above were conducted for all kidney cancer cases (and matched controls) and then for renal cell carcinoma cases (and matched controls) only. Analyses were also conducted by excluding cases (and matched controls) diagnosed within 2 years of blood draw. Because the resulting estimates from these analyses were similar, only the stratified results based on all cases (and matched controls) are presented.
Finally, meta-analyses were conducted to examine the overall and sex-specific associations of both low (<25 nmol/L) and high (≥75 nmol/L) concentrations of circulating 25(OH)D compared with the referent category of 50–<75 nmol/L with kidney cancer for all of the contributing study cohorts. Pooled multivariate log odds ratio estimates adjusted for the covariates in the models described above (e.g., education, body mass index, high blood pressure) were obtained by using inverse-variance weights in random-effects models. Statistical heterogeneity was assessed by using the DerSimonian and Laird Q
). Because of small numbers of cases and controls in the Shanghai Women's Health Study and the Shanghai Men's Health Study, data from these cohorts were combined for the overall meta-analyses. In addition, because of small numbers (≤1 case in the high or low 25(OH)D category), not all cohorts contributed to all of the meta-analyses. Specifically, the New York University Women's Health Study dropped out of the overall and female-specific high versus referent 25(OH)D concentration meta-analyses; CLUE, the Cancer Prevention Study II Nutrition Cohort, and the Multiethnic Cohort Study dropped out of the female-specific low versus referent 25(OH)D concentration meta-analyses; and the Cancer Prevention Study II Nutrition Cohort, the Multiethnic Cohort Study, and the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial dropped out of the female-specific high versus referent 25(OH)D concentration meta-analyses. To explore the influence of each cohort on the results, the meta-analyses described above were repeated by excluding one cohort at a time.
Statistical analyses were performed by using SAS software, versions 9.1.3 and 9.2 (SAS Institute, Inc., Cary, North Carolina). All meta-analyses were conducted by using the R function MiMa (25
) in R version 2.8.1 (The R Foundation for Statistical Computing, Vienna, Austria).