A Colorectal Cancer Risk Prediction Tool

Vortragende: 
Dr. Ruth Pfeiffer, PhD
Biostatistics Branch, Division of Cancer Epidemiology and Genetics,
National Cancer Institute, National Institutes of Health, Bethesda, USA 
Zeit: 
11.11.2010, 16:00 Uhr s.t. 
Ort: 
IMIBesprechungsraum (S05170), Auenbruggerplatz 2/5, 8036 Graz 
Einladende: 
Institut für Medizinische Informatik, Statistik und Dokumentation 
Abstract: 
Several modifiable risk factors have been identified for colorectal cancer (CRC), the second leading cause of cancer death in the USA. We developed models to estimate the probability of first incidence of proximal, distal or rectal cancer in White men and women, aged 50+ years. We 1) estimated relative risk parameters from populationbased casecontrol data separately for proximal, distal, and rectal cancer; 2) estimated baseline agespecific cancer hazard rates based on incidence rates from 13 populationbased cancer registries which are representative of the U.S. population, and attributable risks; and 3) combined competing risks from national mortality rates, relative risks and baseline hazards to estimate the probability of developing the first of proximal, distal or rectal cancer over a prespecified time interval (e.g., 5 or 10 years) given age and risk factors. We derived the variance of the risk estimates. A large cohort study was used to validate the model. We compared expected (E) number of CRC cases predicted by the model to the observed (O) number of CRC cases identified. The overall E/O ratio was 0.99 (95% CI: 0.951.04) in men and 1.05 (95% CI: 0.981.11) in women indicating that the absolute risk model was well calibrated.
To assess the usefulness of this model for screening and prevention we proposed two new criteria. The first criterion, the proportion of cases followed, PCF(q), is the proportion of individuals who will develop disease who are included in the proportion q of individuals in the population at highest risk as determined by the model. The second complementary criterion, is the proportion needed to followup, PNF(p), namely the proportion of the general population at highest risk as determined by the model that one needs to follow in order that a proportion p of those destined to become cases will be followed. We show the relationship of those two criteria to the Lorenz curve and its inverse, and present distribution theory for estimates of PCF and PNF. We develop new methods, based on influence functions, for inference for a single risk model, and also for comparing the PCFs and PNFs of two risk models, both of which were evaluated in the same validation data. We assess our methods using simulated data, and we compute PCF and PNF for data from a validation study for the CRC risk prediction model. 
