Features predicting data exclusion in imaging studies of Alzheimer's disease
Shaney Flores, Jalen Scott, Ruijin Lu, Yi Su, Sarah J. Keefe, T. Hunter Smith, Russ C. Hornbeck, Nicole S. McKay, Jeremy F. Strain, Ashlee Simmons, Kaitlyn Dombrowski, Jacqueline Rizzo, Hope Shimony, Randall J. Bateman, Ganesh M. Babulal, Susan M. Landau, William Jagust, David A. Wolk, James Lah, Carey Gleason, Sterling Johnson, Erik D. Roberson, Reisa Sperling, Keith Johnson, John C. Morris, Chengjie Xiong, Brian A. Gordon, Tammie L. S. BenzingerAbstract
INTRODUCTION
Positron emission tomography (PET) without usable or accompanying magnetic resonance imaging (MRI) is typically excluded in quantitative analyses of Alzheimer's disease, potentially limiting study generalizability. We investigated participant features predicting data exclusion in magnetic resonance (MR)‐dependent analyses and evaluated an existing MR‐free PET pipeline to quantify these missing data.
METHODS
Imaging, clinical, cognitive, and sociodemographic data were analyzed for 2119 individuals in a multi‐site cohort. Agreement between MR‐dependent and MR‐free Centiloids (CL) assessed using intra‐class correlations and features predicting data exclusion were examined using logistic regressions.
RESULTS
MR‐free and MR‐dependent CLs generally agreed, but MR‐free CLs underestimated MR‐dependent cross‐sectionally and longitudinally. Approximately 19.5% ( n = 405) of our cohort would have been excluded in MR‐dependent analyses. Age and cerebrovascular comorbidities were consistent exclusion features across multiple sites.
DISCUSSION
Data exclusion in imaging studies is not entirely random. Flexible quantification methods like MR‐free PET could supplement traditional methods to improve generalizability in large, multi‐site studies.