Revising the ABIDE MCI to dementia prediction model for automated cerebrospinal fluid assays

Abstract

Automated cerebrospinal fluid (CSF) biomarker assays have largely replaced manual immunoassays for measuring amyloid pathology in CSF. We refitted and validated the ABIDE model, predicting progression from mild cognitive impairment (MCI) to dementia, with CSF measurements from the automated Elecsys platform. We included 2413 MCI participants (998 [41%] amyloid-positive) from seven observational cohorts. Elecsys was used in 958 (40%) participants. The parameters of the previous ABIDE Cox model were re-estimated. Model discrimination and calibration were evaluated with leave-one-cohort-out cross-validation. During follow-up, 1034 (42%; 585 [58%] amyloid-positive) participants developed dementia. Discrimination was good with Harrell's C of 0.70 (95% confidence interval [CI]: 0.66-0.73). Calibration was good in the total population and amyloid-positive subgroup, with substantial predicted progression risks for all amyloid-positive participants. We refitted the ABIDE model, predicting MCI to dementia progression, with automated CSF measurements. The model was well calibrated in amyloid-positive patients and may support clinical discussions regarding ATTs.