Risk prediction methods

点击翻译按钮获取中文摘要

We continued to develop, refine and evaluate the National Cancer Institutes Breast Cancer Risk Assessment Tool (BCRAT). Dr. Mateo Banegas, a NCI Cancer Prevention Fellow, developed a new model for absolute invasive breast cancer risk for Latina women. There is interest in determining whether adding information from single nucleotide polymorphisms (SNPs) can increase the discriminatory accuracy and usefulness for screening of risk models. We demonstrated improved risk stratification of breast cancer by adding SNPs to a variety of more standard risk factors. Using data from 1.4 million women undergoing HPV testing and Pap smears in Kaiser Permanente Northern California (KPNC), we published on: (1) estimating absolute risks based on HPV genotyping tests; (2) calculating and comparing risk versus those in the New Mexico HPV/Pap Registry, and (3) calculating risks among those with equivocal Pap smears. We developed and validated models for risk of lung cancer incidence and death and used them to project the impact of risk-based selection of smokers for CT lung-cancer screening in the US. We developed absolute risk models appropriate under left/interval/right-censoring data occurring for screen-detected disease, which occurs for electronic health record data, called the "logistic-Weibull" and the "logistic-Cox" models. We explained why a method sometimes used to assess absolute risk models in case-control data does not assess calibration, but only fits of relative risk.