by Alyssa M. Voss, M.P.H.
“A model is a lie that helps you see the truth.” (Howard Skipper, quoted in the Emperor of All Maladies)
DCEG biostatisticians continue to innovate statistical models to uncover important etiologic truths hidden within large, complex data sets. During the last several years, Biostatistics Branch members Philip S. Rosenberg, Ph.D. and William F. Anderson, M.D., M.P.H., have developed sophisticated statistical models to analyze the complex, interrelated effects of age, time period, and birth cohort on cancer incidence trends for various cancer types in the U.S. The results of these age-period-cohort (APC) analyses have provided informative “big picture” clues about how cancer incidence and mortality trends change over time, what factors are driving them, and provide projections for future cancer burden.
But using the APC approach has challenges, particularly because of the statistical complexity, and because of the difficulty of implementing the model in common statistical software packages and interpreting the results. This has limited the use of the methodology within the broader scientific community. Over the last few years, Drs. Rosenberg and Anderson have shared their expertise in APC methods to help other investigators in DCEG and the extramural community describe incidence and mortality trends in the U.S. for lung cancer1, pancreatic cancer2, ovarian cancer3, and Burkitt lymphoma4, and global incidence rates of oral cavity cancers and oropharyngeal squamous cell carcinomas5 (see References).
Most recently, Drs. Rosenberg and Anderson used the APC model to forecast breast cancer incidence rates and burden in 20306(see References). “We were able to incorporate recent birth cohort data in this analysis,” stated Dr. Rosenberg, “as well as breast cancer sub-type specific incidence data, which had not previously been done. We were able to show that the incidence trends of the future will not resemble the patterns we see today.” Dr. Rosenberg presented the results at the American Association for Cancer Research Annual Meeting in April.
Through these collaborative efforts, Drs. Rosenberg and Anderson realized that it was important to make APC methods, and particularly the newer approaches, more accessible to researchers interested in evaluating disease trends. With the help of David Check, BB, and Sue Pan from NCI’s Center for Biomedical Informatics and Information Technology (CBIIT), Drs. Rosenberg and Anderson developed a free, open-access, web-based APC tool available through the DCEG website. A description of the APC tool was recently published in Cancer Epidemiology, Biomarkers, and Prevention.
The APC tool was designed to be intuitive and user friendly, and includes sample data for first-time users. Users input or “paste” their data into an open field on the site, and the tool does the rest. The available functions incorporate traditional and new approaches to APC analyses, which take into account various longitudinal and cross-sectional models and interpretations of the data. One of the most important features of the tool is its ability to easily calculate net drift, which is defined as the sum of the linear change in the period and cohort effects and is conceptually similar to the standard annual percentage change in the age-standardized rate. The tool also uniquely calculates local drifts, which show the estimated annual percentage change over time for individual age groups. This feature can often reveal striking age interaction over time, which is a signature consequence of underlying birth-cohort phenomena.
Another key feature of the tool is that its output summarizes the results in tables and graphs, which provide powerful illustrations of the analyses. The output files may be exported into a variety of different formats, such as Excel or the R statistical package. “What previously required years of expertise by a trained statistician can now be done by anyone from any discipline in very little time,” stated Dr. Anderson.
With funds awarded through the DCEG Informatics Tool Challenge, Drs. Rosenberg and Anderson and Mr. Check and CBIIT colleagues will develop a companion tool that will conduct comparative APC analysis of “two-hazard” problems. For example, the new tool will allow the user to conduct two independent APC analyses using data from different populations for the same disease outcome and compare them. “Developing this tool will be challenging,” stated Dr. Rosenberg “But we are excited to bring this additional tool to others so they can test their own unique hypotheses.”