Empirically derived composite cognitive test scores to predict preclinical and clinical stages of Alzheimer’s disease

• Published in: Alzheimer's & dementia Vol. 17; no. S5
• Main Authors: Shishegar, Rosita, Chai, Tze Young, Cox, Timothy, Lamb, Fiona, Robertson, Joanne S., Laws, Simon M., Porter, Tenielle, Fripp, Jurgen, Doecke, James D., Tosun‐Turgut, Duygy, Maruff, Paul T., Savage, Greg, Rowe, Christopher C., Masters, Colin L., Weiner, Mike W., Villemagne, Victor L.L., Burnham, Samantha C.
• Format: Journal Article
• Language: English
• Published: 01.12.2021
• ISSN: 1552-5260; 1552-5279
• DOI: 10.1002/alz.053040

Background Alzheimer’s disease (AD) clinical trials require cognitive test scores that assess change in cognitive function accurately. Here, we propose new composite cognitive test scores to detect earlier stages of AD accurately by using the full neuropsychological testing battery (in ADNI) and a manifold learning dimension reduction technique namely UMAP. Method Data for this study included N=1585 ADNI participants ([492 cognitively normal (CN), 804 mild cognitively impaired (MCI), 289 AD; aged 73.8±7.1; 708 females]; Table 1). Subjects with 3 or more follow‐up sessions were included. Cognitive test scores with more than 60% missing data were excluded. Missing data within included test scores were imputed using the MissForest algorithm. A linear mixed model using all follow‐up data was applied to calculate the random slope (rate of change) and random intercept for each cognitive score and for each subject. The scores and demographic measurements: age, gender, years of education and APOE‐ɛ4 status were used to inform the UMAP. Levels for the output variable were defined as: 1) stable CN, 2) CN who progressed to MCI or probable dementia due to AD, 3) stable MCI, 4) MCI who progressed to dementia AD and 5) dementia due to AD. The model calculated two composite scores. These cognitive stages were predicted using Support Vector Machine (SVM) analysis of both the new composite scores and the traditional clinical rating measures of Clinical Dementia Rating (CDR) and Mini‐Mental State Examination (MMSE). Result Predicting cognitive stages using the proposed composite scores show a highly significant improvement with a 0.981 accuracy and 0.976 reliability (evaluated by Cohen's kappa coefficient), compared to using the combination of CDR and MMSE scores covaried for demographics, which had 0.660 accuracy and 0.567 reliability. Individuals’ clinical and preclinical stages with regards to UMAP two‐dimensional embedding and the clinical rating measures, CDR and MMSE, are presented in Figure 1. Table 2 reports the importance of the test measures on the UMAP components used in AD staging predictions. Conclusion The results here suggest that the proposed empirically derived composite cognitive test scores provides a practical solution to differentiate cognitive stages with a high accuracy and reliability.