Genetic analysis suggests high misassignment rates in clinical Alzheimer's cases and controls

Genetic case-control association studies are often based on clinically ascertained cases and population or convenience controls. It is known that some of the controls will contain cases, as they are usually not screened for the disease of interest. However, even clinically assessed cases and controls can be misassigned. For Alzheimer's disease (AD), it is important to know the accuracy of the clinical assignment. The predictive accuracy of AD risk by polygenic risk score analysis has been reported in both clinical and pathologically confirmed cohorts. The genetic risk prediction can provide additional insights to inform classification of subjects to case and control sets at a preclinical stage. In this study, we take a mathematical approach and aim to assess the importance of a genetic component for the assignment of subjects to AD-positive and -negative groups, and provide an estimate of misassignment rates (MARs) in AD case/control cohorts accounting for genetic prediction modeling results. The derived formulae provide a tool to estimate MARs in any sample. This approach can also provide an estimate of the maximal and minimal MARs and therefore could be useful for statistical power estimation at the study design stage. We illustrate this approach in 2 independent clinical cohorts and estimate misdiagnosis rate up to 36% in controls unscreened for the APOE genotype, and up to 29% when E3 homozygous subjects are used as controls in clinical studies.