Challenges in Age-Period-Cohort Modeling of Breast Cancer Incidence

• Activity Number: 145
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ronald Gangnon*+ and Brian Sprague and Natasha Stout and Oguzhan Alagoz and Amy Trentham-Dietz
• Companies: University of Wisconsin and University of Vermont and Harvard Medical School and Harvard Pilgrim Health Care and University of Wisconsin and University of Wisconsin
• Keywords: age-period-cohort models ; negative binomial regression ; splines ; identifiability ; prediction

Abstract

Breast cancer incidence in the US has recently declined after decades of steady increases. The age-period-cohort (APC) framework, in which temporal trends are decomposed into age, year of diagnosis (period) and year of birth (cohort), was applied to breast cancer incidence data from the Connecticut Tumor Registry (1935-1979) and the national Surveillance Epidemiology and End Results cancer registries (1973-2008). Numbers of incident breast cancer cases were modeled using a log-linear negative binomial regression model with indicator variables for SEER registry, spline terms for age, period and cohort and an offset term for (log) female population. Here, we discuss various challenges that we encountered, including (1) obtaining population estimates by single years of age from available estimates for 5-year age groups, (2) identifying separate cohort effects for pre-menopausal and post-menopausal breast cancer with a smooth transition, (3) separating the impact of mammography from the period effect, (4) predicting future breast cancer incidence rates, (5) generating coherent stage-specific incidence rate estimates and (6) estimating race-specific breast cancer incidence rates.