A Simpler Model for the Impact of Atrial Fibrillation on Cognitive Trajectories in the Elderly

Abstract

In their 2013 paper “Atrial fibrillation and cognitive decline: A longitudinal cohort study”, Dr. Thacker and colleagues investigated whether, in the absence of clinical stroke, incident atrial fibrillation was associated with faster cognitive decline in the elderly. They performed regression analyses on longitudinally collected data. When developing the regression model they relied on both prior scientific knowledge and the results of data driven model selection techniques. As a result, the regression model used thirty-nine terms to model the impact of thirteen covariates on cognition. In this paper we first explain the statistical and scientific rational for the terms in their regression model. Then through two simulation studies we investigate the consequences of having chosen a simpler but possibly misspecified model. Ultimately when deciding to utilize a simpler, possibly misspecified, model the researcher must settle the question of whether a small bias is acceptable when the benefits will be greater precision and ease of communication with non-statistically minded readers.