Development of a Model for Predicting Prostate Cancer
Rationale: Effective prostate cancer screening is important for the alleviation of prostate cancer. Unfortunately, the current screening techniques lack specificity and sensitivity leading to many negative biopsies and missing a number of clinically relevant cancers. While there is new information regarding the environmental and genetic risk factors for prostate cancer, to date, there has been no effort to incorporate this information into a comprehensive model for prostate cancer prediction to develop a more accurate screen.
Goal: Develop a multivariate model that predicts clinically relevant prostate cancer by incorporating known risk factors such as PSA and age along with the recently identified genetic and environmental risk factors.
Specific Aim 1: We will develop a multivariate model for prostate cancer detection in a simulated population. A simulated population of at least 100,000 screened individuals will be created. We will then assign the following characteristics to each individual: age, ethnicity, family history of prostate cancer, PSA value, genotype, and dietary intake of low fat dairy products. Using this simulated population we will construct a number of models using multivariate logistic regression to determine the minimal number of parameters that can be used to discriminate between prostate cancer and control subjects. Each of the models will be ranked according to predictive value sensitivity and specificity.
Specific Aim 2: We will determine if any of the models developed in the simulated population can be used to predict prostate cancer in a screened population. Using the Personalized Medicine Research Project (PMRP) we will validate the multivariate logistic regression model developed above. Currently there are 3,931 individuals in the PMRP cohort that have a PSA value, 1,178 who have had a biopsy, and 453 who have been diagnosed with prostate cancer. We have obtained completed dietary history information on approximately 2/3 of these individuals. We will determine the genotype of this population for 12 polymorphisms that have been associated with prostate cancer and. We will test the model for accuracy in this population and test for potential interactions between the known risk factors.