Study
PLCO
(Learn more about this study)
Project ID
2008-0005
Initial CDAS Request Approval
Mar 27, 2008
Title
The association between Longitudinal measurements on PSA and on DRE and incidence of Prostate Cancer
Summary
Attempts to understand the relationship between longitudinal measurements of prostate specific antigen concentration (PSA) and the risk of incident prostate cancer (CaP) have been complicated by the fact that the decision to confirm the presence or absence of CaP is influenced by the observation of these measurements. The screened arm of the Prostate component of the PLCO trial offers the unique opportunity to study the association between longitudinal measurements of PSA and DRE with the risk of incident CaP in a population of healthy screened men. In the intervention arm 28,673 subjects had complete information for analysis, 2,805 underwent biopsy and 910 prostate cancers were diagnosed within 18 months of last PLCO follow-up. We propose to explore the association between longitudinally measured PSA and risk of incident CaP along several channels. The first component of the project focuses on the effect of PSA velocity (PSAV), calculated as the slope in PSA between consecutive visits. The Risk of incident cancer will be modeled using Cox regression analysis treating screening variables as time-dependent covariates. The analysis of biopsied men will incorporate subject-level weights derived from an ancillary model for probability of biopsy so that inferences may be drawn within the entire cohort of screened men. We report upon the ability of PSAV to significantly enhance the model fit as compared with a model incorporating PSA alone. The second component of the project will investigate the relationship between longitudinal screening variables and incident cancer via a latent class model. According to this model, membership in the latent class will determine both prostate cancer risk and longitudinal screening variable measurements in the form of class specific main and subject-level random effects. As before, inverse weighting of biopsied men will be used to map the inferences back to the entire cohort. We compare the estimated risk of prostate cancer within various cofactor classes, obtained from the latent class approach with those obtained from the first approach. Finally, we assess the predictive accuracy of both approaches via the concordance indices obtained within a 5-fold cross-validation.
Aims
SPECIFIC AIMS 1-2: Cox regression with time dependent covariates: SPECIFIC AIM #1.a: Determine if the relative risk of PSA velocity controlled for current PSA, DRE and other patient and prostate specific cofactors, is significantly different than 1 when inference is done relative to the cohort of biopsied men. SPECIFIC AIM #1.b: Determine if the relative risk of PSA velocity controlled for current PSA, DRE and other patient and prostate specific cofactors, is significantly different than 1 when inference is done relative to the cohort of all screened men via the incorporation of weights inversely proportional to predicted probabilities of biopsy. SPECIFIC AIM #2.a-b: Determine if the concordance index of the model including PSA velocity, obtained in a 5- fold cross-validation, is significantly different than that of the model without PSA velocity. Inferences will be made (a) relative to the cohort of biopsied men and (b) relative to the cohort of all screened men. SPECIFIC AIMS 3-4: Latent class model: SPECIFIC AIM #3.a-b: Determine if parameters related to degree of separation of the latent classes are significantly different than 0, with inferences being drawn (a) within the cohort of biopsied men and (b) within the cohort of all screened men via incorporation of weights as above. SPECIFIC AIM #4.a-b: Determine if the concordance index of the latent class model obtained in a 5-fold crossvalidation, is significantly different from the best obtained in specific aim 2 above. Inferences will be made (a) relative to the cohort of biopsied men and (b) relative to the cohort of all screened men.
Collaborators
Robert Grubb (Washington University School of Medicine)
Grant Izmirlian (Division of Cancer Prevention)
Adam Kibel (Washington University School of Medicine)
