Amyloid biomarkers as predictors of conversion from mild cognitive impairment to Alzheimer's dementia: a comparison of methods

Background Amyloid-beta (A beta) PET is an established predictor of conversion from mild cognitive impairment (MCI) to Alzheimer's dementia (AD). We compared three PET (including an approach based on voxel-wise Cox regression) and one cerebrospinal fluid (CSF) outcome measures in their predictive power. Methods Datasets were retrieved from the ADNI database. In a training dataset (N = 159), voxel-wise Cox regression and principal component analyses were used to identify conversion-related regions (Cox-VOI and AD conversion-related pattern (ADCRP), respectively). In a test dataset (N = 129), the predictive value of mean normalized F-18-florbetapir uptake (SUVR) in AD-typical brain regions (composite SUVR) or the Cox-VOI and the pattern expression score (PES) of ADCRP and CSF A beta(42)/A beta(40) as predictors were compared by Cox models (corrected for age and sex). Results All four A beta measures were significant predictors (p < 0.001). Prediction accuracies (Harrell's c) showed step-wise significant increases from Cox-SUVR (c = 0.71; HR = 1.84 per Z-score increase), composite SUVR (c = 0.73; HR = 2.18), CSF A beta(42)/A beta(40) (c = 0.75; HR = 3.89) to PES (c = 0.77; HR = 2.71). Conclusion The PES of ADCRP is the most predictive A beta PET outcome measure, comparable to CSF A beta(42)/A beta(40), with a slight but statistically significant advantage.