RP25- Sigmoid methodology allows early prediction of  cognitive decline towards Alzheimer's disease across several cognitive domains

M. Cespedes; C. Gillis; P. Maruff; N. Maserejian; C. Fowler; S. Rainey-Smith; V. Villemagne; C. Rowe; R. Martins; C. Masters; J. Doecke

doi:10.14283/jpad.2021.58

Background: Clinical trials in Alzheimer’s disease (AD) depend on clinical endpoints that may lack sensitivity to the cognitive changes that characterize the disease prior to dementia diagnosis. Consequently, trials designed to determine whether new drugs can forestall dementia or prevent cognitive decline prior to onset of AD dementia will require outcomes measures that are sensitive to cognitive changes earlier in the disease. Objectives: The objective of this work was to identify individuals whose cognitive scores progress more rapidly (“accelerators”) from those whose cognition does not reach this same level over the same time period (“non-accelerators”). The specific aim was to assess the MMSE, CDR-SB and seven different cognitive composite scores to examine 1) their ability to cluster accelerators and non- accelerators, 2) which cognitive scores can be used as strong predictors of cognitive change, and 3) the proportion agreement of progression classification within specific subgroups of AD pathology. This process was entirely data driven in that it was performed without using the formal AIBL clinical classification to inform disease stage. Methods: Using longitudinal data from the Australian Imaging, Biomarkers and Lifestyle (AIBL) study of ageing, seven cognitive composite scores (the AIBL PACC score (CVLT-II, LMII, DSC & MMSE), attention and processing speed (DSC & total digit span), episodic memory (delayed recall, LMII & RCFT30), executive function (verbal fluency & category switching), language (BNT & verbal fluency [D-KEFS]), recognition (recognition CVLT-II & RCFT), visuospatial function (RCFT copy & clock along) along with MMSE and CDR-SB from 144 months of AIBL were selected from all participants with at least three time points (at baseline: total N=1269; 973 Cognitively Normal (CN), 143 with Mild Cognitive Impairment (MCI), 153 with AD). Using the mathematical properties of a proposed sigmoidal function, “cognitive turning points” are defined to allocate accelerator/non-accelerator groups for each of the nine cognitive scores. Proportions of AIBL participants groups are assessed in the complete cohort, and in groups stratified to include Ab-/Tau unknown (N=523), Ab+/Tau unknown (N=420), Ab-/Tau- (N=218), Ab+/Tau-(N=200), Ab+/Tau+ (N=75) and lastly in those without Ab or Tau information (N=325). Using the mathematical properties of the sigmoid function, we identified turning points (thresholds) for each score whereby a participant would be allocated to an “accelerator” or “non-accelerator” group. Group classifications were compared across all nine cognitive scores and then between stratified participant groups. Results: In the full cohort, visuospatial function identified accelerators with the highest proportion of accelerators among all of the measures explored, with a total of 68% of the cohort progressing past the derived “cognitive turning point” for accelerated decline. The cognitive test with the next highest classification of accelerators was 2) language (64%); 3) the AIBL PACC (61%); 4) attention and processing (59%); 5) executive function (42%); 6) recognition (38%); 7) episodic recall (26%); 8) MMSE (22%); and 9) CDR-SB (20%). In contrast to CDR-SB and MMSE, all of the cognitive composite scores, except episodic memory, classified a higher proportion of AIBL participants as accelerators as compared with non-accelerators. Of the top ranked cognitive measures, the highest proportions of agreement in identifying accelerators were observed for visuospatial function, language, the AIBL PACC, and attention and processing, with proportions of cross-classification ranging between 59 to 69%. Reducing the sample to those with known pathology provided the strongest classification rates. Amongst Ab+/Tau+ participants, the AIBL PACC had the highest proportion of classified accelerators (85%), followed by language (80%), visuospatial functioning (73%) and attention and processing speed (71%). For Ab+ participants where Tau levels were unknown, the AIBL PACC and language both identified 75% of participants as accelerators, while visuospatial function identified 73% as accelerators. Among Ab+/Tau- participants, visuospatial function (76%), language (66%), attention and processing speed (66%) and the AIBL PACC (56%) ranked the highest to classify accelerators. Among Ab- participants where Tau was unknown, proportions to classify accelerators were reduced (visuospatial function [62%], language [55%], attention and processing speed [55%] and the AIBL PACC [52%]). Lowest accelerator classification rates were seen for the Ab-/Tau- subgroup, with only visuospatial function (57%) and attention and processing speed (51%) reaching 50%. When considering the group where both Ab and Tau pathology were unknown, proportions were ranked similarly, however proportions were lower when compared to results in known pathological groups (visuospatial function [73%], language [64%], attention and processing speed [53%] and the AIBL PACC [58%]) when compared to results in known pathological groups. Conclusion: This is the first representation of data derived accelerator groups and group comparisons using a sigmoid function to determine “cognitive turning points” across multiple common cognitive measures. Our results suggest that the cognitive composite scores including the AIBL PACC, language, visuospatial function and attention and processing are well suited to detect change in cognitive function. Such scores continued their high classification proportions across all subgroups of baseline cognition and AD pathology. Disclosures: CG and NM are employees and stockholders of Biogen. This work was sponsored by Biogen Global Analytics and Data Sciences, Epidemiology in Collaboration with the Australian Imaging, Biomarkers and Lifestyle study (AIBL)

RP25- Sigmoid methodology allows early prediction of cognitive decline towards Alzheimer's disease across several cognitive domains

Files and links (1)

Abstract

Details

Metrics