Drs. Ruth Pfeiffer and Andrew Freedman were cochairs of an interdisciplinary workshop attended by over 100 experts who discussed the development and application of models to predict absolute disease risk. The workshop proceedings were published this year.? ? In preparation for the workshop and partly in response to the workshop discussions, Branch members wrote and published a paper on criteria for evaluating risk projection models. General criteria, such as the area under the receiver operating curve, which is a measure of disciminatory power, have advantages but may not be as useful as loss function-based criteria tailored to a specific application. In particular, the degree of disciminatory power needed for screening a population using a risk assessment model is much greater than that needed for certain medical management decisions that weigh the risks against the benefits of a certain intervention, with guidance from a risk model. ? ? Using case-control data from cohorts of women treated for Hodgkin Disease (HD) with radiation and chemotherapy, and coupling this information with data from NCI's Surveillance and End Results (SEER) Program, we developed models for projecting absolute breast cancer risk, depending on the nature of the treatments given. Some women have risks comparable to carrying a mutation in a BRCA gene following treatment, especially those with high dose radiation and no ovarian ablative chemotherapy.? ? We developed a relative risk model for projecting breast cancer risk that includes mammographic density, weight, family history, age at first live birth and number of previous breast biopsies. The model has modestly higher discriminatory power than the """"""""Gail model"""""""" that does not include mammographic density. Work is in progress to model absolute risk, based on the relative risk model.? ? A paper has been submitted that presents a model for predicting absolute melanoma risk, based on factors readily determined by a general practioner.? ? Relative risk models have been developed for proximal and distal colon cancer risk. These are being coupled with SEER data to project absolute colon cancer risk.? ? In collaboration with staff in DCCPS, we used data from the ATBC Trial to check the calibration of a model for projecting absolute lung cancer risk.?
Kovalchik, Stephanie A; Pfeiffer, Ruth M (2014) Population-based absolute risk estimation with survey data. Lifetime Data Anal 20:252-75 |
Pfeiffer, R M; Gail, M H (2011) Two criteria for evaluating risk prediction models. Biometrics 67:1057-65 |
Gail, Mitchell H; Graubard, Barry; Williamson, David F et al. (2009) Comments on 'Choice of time scale and its effect on significance of predictors in longitudinal studies' by Michael J. Pencina, Martin G. Larson and Ralph B. D'Agostino, Statistics in Medicine 2007; 26:1343-1359. Stat Med 28:1315-7 |
Gail, Mitchell H (2008) Discriminatory accuracy from single-nucleotide polymorphisms in models to predict breast cancer risk. J Natl Cancer Inst 100:1037-41 |