Diversity of Model Approaches for Breast Cancer Screening

A Review of Model Assumptions by the Cancer Intervention Surveillance Network (CISNET) Breast Cancer Groups

Published in: Statistical Methods in Medical Research, v. 13, no. 6, 2004, p. 258-538

Posted on RAND.org on December 31, 2003

by Rob Boer, Sylvia K. Plevritis, Lauren D. Clarke

Read More

Access further information on this document at smm.sagepub.com

This article was published outside of RAND. The full text of the article can be found at the link above.

The National Cancer Institute-sponsored Cancer Intervention and Surveillance Network program on breast cancer is composed of seven research groups working largely independently to model the impact of screening and adjuvant therapy on breast cancer mortality trends in the US from 1975 to 2000. Each of the groups has chosen a different modeling methodology without purposeful attempt to be in contrast with each other. The seven groups have met biannually since November 2000 to discuss their methodology and results. This article investigates the differences in methodology. To facilitate this comparison, each of the groups submitted a description of their model into a uniformly structured web based 'model profiler'. Six of the seven models simulate a preclinical natural history that cannot be observed directly with parameters estimated from published evidence concerning screening and therapy effects. The remaining model regards published evidence on intervention effects as prior information and updates that with information from the US population in a Bayesian type analysis. In general, the differences between the models appear to be small, particularly among the models driven by natural history assumptions. However, we demonstrate that such apparently small differences can have a large impact on surveillance of population trends. The authors describe a systematic approach to evaluating differences in model assumptions and results, as well as differences in modeling culture underlying the differences in model structure and parameters.

This report is part of the RAND Corporation external publication series. Many RAND studies are published in peer-reviewed scholarly journals, as chapters in commercial books, or as documents published by other organizations.

The RAND Corporation is a nonprofit institution that helps improve policy and decisionmaking through research and analysis. RAND's publications do not necessarily reflect the opinions of its research clients and sponsors.