CGEMS Prostate Cancer GWAS - Stage 1 - PLCO

The Cancer Genetic Markers of Susceptibility (CGEMS) prostate cancer genome-wide association study (GWAS) included genotyping approximately 550,000 SNPs (Phase 1A with HumanHap300 and Phase 1B HumanHap240, both from Illumina, San Diego, CA) in 1,172 prostate cancer patients and 1,157 controls of European ancestry from the Prostate, Lung, Colon and Ovarian (PLCO, http://www.cancer.gov/prevention/plco/) Cancer Screening Trial. The original analysis published in Nature Genetics [PMID: 17401363] included 2,282 subjects. After improvement and revisions of the original analysis, 2,252 subjects were submitted to dbGaP.]]>
We present results from two distinct analytic approaches. The first scheme reported here is more frequently used in case control studies. The second scheme takes full advantage of the prospective nature of the PLCO cohort and the power from incidence density sampling.

Cumulative density sampling
For this scheme, which will be more familiar to non-epidemiologists, does not account for the dynamic nature of the cohort. Genotypes of individuals that have been selected as a case in the relevant phenotype case group are counted once as a case and never as a control. Individuals who have been selected several times as controls but had not developed prostate cancer during follow-up are counted only once in the control group.

Incidence density sampling
Selection of controls from cases identified in a cohort that accounts for the dynamic nature of the cohort including development of disease during follow-up and timing of entry to and exit from follow-up may have more power to detect an association than the single-selection method. The main feature of incidence-density sampling, as used for control selection here, is that controls are selected independently for each case among those who are at risk at the time of the diagnosis of the case; i.e., among those who would become a case in the study had they developed disease at the same time. Inclusion as a control for a given case set is independent of future diagnosis as a case, of selection as a control for other case sets, and of entry and exit times. Thus, individuals may be included as a case and as a control. Genotypes of individuals who have been selected multiples times are taken into account each time he is selected; the man?s covariates that vary with time, such as age are defined differently each time, depending on the characteristics of the case set for which he was selected as a control.

The number of association model we fit increased from 4 in Build 1.0 to 32 in Build 2.0, including all combinations from the following four categories:

Sampling

Cumulative density
Whole genome association analysis of main effects for 554,291 SNPs on 1,151 cases diagnosed with tumors and 1,101 controls that were not diagnosed with prostate cancer at the start of the CGEMS project.

Incidence density
Whole genome association analysis of main effects for 554,291 SNPs on 1,151 cases diagnosed with tumors and 1,156 controls selected using an incidence density sampling strategy.

Dependent variable in model

Dichotomous
A dichotomous logistic model was constructed to contrast the risk of all prostate cancer cases (both non-aggressive and aggressive) against that of all controls (m=2).

Polytomous
A polytomous logistic model was constructed to separately contrast the risk of non-aggressive and aggressive prostate cancer cases against that of all controls (m=3).

Covariate adjustment

Unadjusted
A 3-by-m contingency table of genotypes by phenotypes was constructed.

Adjusted
The m phenotypes were regressed on indicator variables for genotype effects, age group at randomization (4 groups), region of recruitment (9 non-reference regions), and a single eigenvector to account for population stratification.

Genotype effects

Genotypic
The p-value was obtained from a score test of each estimated genotype effect with up to 2(m-1) degrees of freedom. (m is the number of phenotype categories)

Trend
The p-value was obtained from a score test for the estimated trend of the genotype effects with up to m-1 degrees of freedom.

Dominant
The p-value was obtained from a score test for the minor homozygote + heterozygote versus major homozygote effect with up to m-1 degrees of freedom.

Recessive
The p-value was obtained from a score test for the minor homozygote versus heterozygote + major homozygote effect with up to m-1 degrees of freedom.

]]> We present results from two distinct analytic approaches. The first scheme is more frequently used in case control studies. The second scheme reported here takes full advantage of the prospective nature of the PLCO cohort and the power from incidence density sampling.

Cumulative density sampling
For this scheme, which will be more familiar to non-epidemiologists, does not account for the dynamic nature of the cohort. Genotypes of individuals that have been selected as a case in the relevant phenotype case group are counted once as a case and never as a control. Individuals who have been selected several times as controls but had not developed prostate cancer during follow-up are counted only once in the control group.

Incidence density sampling
Selection of controls from cases identified in a cohort that accounts for the dynamic nature of the cohort including development of disease during follow-up and timing of entry to and exit from follow-up may have more power to detect an association than the single-selection method. The main feature of incidence-density sampling, as used for control selection here, is that controls are selected independently for each case among those who are at risk at the time of the diagnosis of the case; i.e., among those who would become a case in the study had they developed disease at the same time. Inclusion as a control for a given case set is independent of future diagnosis as a case, of selection as a control for other case sets, and of entry and exit times. Thus, individuals may be included as a case and as a control. Genotypes of individuals who have been selected multiples times are taken into account each time he is selected; the man?s covariates that vary with time, such as age are defined differently each time, depending on the characteristics of the case set for which he was selected as a control.

The number of association model we fit increased from 4 in Build 1.0 to 32 in Build 2.0, including all combinations from the following four categories:

Sampling

Cumulative density
Whole genome association analysis of main effects for 554,291 SNPs on 1,151 cases diagnosed with tumors and 1,101 controls that were not diagnosed with prostate cancer at the start of the CGEMS project.

Incidence density
Whole genome association analysis of main effects for 554,291 SNPs on 1,151 cases diagnosed with tumors and 1,156 controls selected using an incidence density sampling strategy.

Dependent variable in model

Dichotomous
A dichotomous logistic model was constructed to contrast the risk of all prostate cancer cases (both non-aggressive and aggressive) against that of all controls (m=2).

Polytomous
A polytomous logistic model was constructed to separately contrast the risk of non-aggressive and aggressive prostate cancer cases against that of all controls (m=3).

Covariate adjustment

Unadjusted
A 3-by-m contingency table of genotypes by phenotypes was constructed.

Adjusted
The m phenotypes were regressed on indicator variables for genotype effects, age group at randomization (4 groups), region of recruitment (9 non-reference regions), and a single eigenvector to account for population stratification.

Genotype effects

Genotypic
The p-value was obtained from a score test of each estimated genotype effect with up to 2(m-1) degrees of freedom. (m is the number of phenotype categories)

Trend
The p-value was obtained from a score test for the estimated trend of the genotype effects with up to m-1 degrees of freedom.

Dominant
The p-value was obtained from a score test for the minor homozygote + heterozygote versus major homozygote effect with up to m-1 degrees of freedom.

Recessive
The p-value was obtained from a score test for the minor homozygote versus heterozygote + major homozygote effect with up to m-1 degrees of freedom.

]]>A total of 1,361 subjects with prostate cancer met the eligibility criteria and were considered for the CGEMS project; 737 cancers were aggressive 624 cancers were nonaggressive. Of the eligible cases, all aggressive cases (n=737) were chosen to be cases in the CGEMS prostate cancer study. Of the 624 men found to have non-aggressive tumors, 493 men (70.4%) whose diagnosis was temporally closest to the first screening were included in this study. Controls were selected by incidence-density sampling. The first step was creation of non-overlapping sets of cases characterized by: Calendar year (FY) of entry into the cohort, Age at entry in five-year intervals (55-59, 60-64, 65-69, 70-74) Number of years under follow-up between enrollment and diagnosis of prostate cancer. Next, for each case set, we identified eligible men among all 28,251 men in the CGEMS cohort who met each of the following three criteria: Same year of entry into the cohort as the case set; Same five-year age-at-entry interval (55-59, 60-64, 65-69, 70-74) as the case set; and Observed through the year of follow-up in the case set with no prostate cancer diagnosis. ]]>
The PLCO Cancer Screening Trial is a large, randomized controlled trial of approximately 155,000 men and women. Participants are randomized to either a screening or control arm. Each year after enrollment, subjects are asked to notify the study of any cancers diagnosed in the past year using the Annual Study Update (ASU). The trial is designed to test the efficacy of cancer screening to prevent early death from prostate, lung, colorectal and ovarian cancer. The collection of questionnaire data and biospecimens (e.g., repeated blood samples and in some instances, buccal cell samples) allows investigation of early markers for cancer as well as etiology of common cancers. PLCO enrollment began in 1993 and ended in 2001. Recruitment included men and women, aged 55 to 74 with no reported history of prostate, lung, colon and ovarian cancer, although prior diagnoses of other cancers were acceptable. The CGEMS cohort consisted of men enrolled in the screening arm of the PLCO Trial who: were White and non-Hispanics; had no prior history of prostate of cancer before randomization; had at least one PLCO prostate cancer screen (PSA) before October 1, 2003; had completed a Baseline Questionnaire about risk factors for cancer; had signed informed consent; had provided a blood sample with at least 11 µg DNA at least 1 vial of buffy coat, or at least 7 vials of whole blood was available; and for controls, had returned at least one Annual Study Update (ASU). Based on these criteria, 28,521 men were included in the CGEMS sub-cohort. CGEMS distinguishes between non-aggressive and aggressive cases of prostate cancer at the time of diagnosis. The two subtypes are defined as follows: Non-aggressive: cases with a Gleason Score < 7 and Stage < III. Aggressive: cases with a Gleason Sore ≥ 7 or Stage ≥ III. Study enrollment began on October 1, 1993. Consequently, study years in the PLCO Trial are counted according to the Federal fiscal year, Oct 1 to the next September 30. All men diagnosed with prostate cancer between enrollment and the end of FY2001 were considered for inclusion in CGEMS. Because of our interest in the clinically more significant, but less common aggressive form of prostate cancer, we increased the fraction of aggressive cases in the CGEMS case series by extending eligibility for cases diagnosed with aggressive prostate cancer through the end of FY2003.]]>