Digital biomarker-based individualized prognosis for people at risk of dementia

Abstract

Background: Research investigating treatments and interventions for cognitive decline fail due to difficulties in accurately recognizing behavioral signatures in the presymptomatic stages of the disease. For this validation study, we took our previously constructed digital biomarker-based prognostic models and focused on generalizability and robustness of the models.

Method: We validated prognostic models characterizing subjects using digital biomarkers in a longitudinal, multi-site, 40-month prospective study collecting data in memory clinics, general practitioner offices, and home environments.

Results: Our models were able to accurately discriminate between healthy subjects and individuals at risk to progress to dementia within 3 years. The model was also able to differentiate between people with or without amyloid neuropathology and classify fast and slow cognitive decliners with a very good diagnostic performance.

Conclusion: Digital biomarker prognostic models can be a useful tool to assist large-scale population screening for the early detection of cognitive impairment and patient monitoring over time.

Keywords: Altoida Neuro Motor Index; Alzheimer's disease; artificial intelligence; augmented reality; cognitive aging; digital biomarker; machine learning; risk prediction.

Conflict of interest statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.The authors declare the following financial interests/personal relationships which may be considered potential competing interests:

© 2020 The Authors. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring published by Wiley Periodicals, Inc. on behalf of the Alzheimer's Association.

Figures

FIGURE 1

Visual representation of three motor features during augmented reality task. Visual representation of three features (out of 109 key functional motor behaviors) corresponding to the “distribution of the path complexity when finding the augmented reality objects” (see Table S1). LEFT PANEL, Typical digital phenotype of mild cognitive impairment (MCI) subjects that will remain stable over time (n = 113). RIGHT PANEL, Typical digital phenotype of MCI that would progress to dementia over time for a period of 3 years

FIGURE 2

Flowchart showing methodology for training and evaluating machine learning models

FIGURE 3

Receiver operating characteristic (ROC) curves for internal cross‐validated results of predicting mild cognitive impairment (MCI) to dementia. Internally five‐fold cross‐validated ROC‐area under the curve (AUC) curves AUC‐ROC analysis, in identifying MCI subjects that progress to dementia in a 3‐year period. LEFT PANEL, Assessment versus MCI subjects that will remain stable over time (n = 113). RIGHT PANEL, Assessment versus the combined population of MCI and healthy control subjects that will remain stable over time for a period of 3 years

FIGURE 4

Receiver operating characteristic (ROC) curves for internal cross‐validated results of predicting amnestic mild cognitive impairment (aMCI) to Alzheimer's disease (AD) dementia. Internally five‐fold cross‐validated ROC curves predicting the conversion of subjects with aMCI to dementia due to AD versus a stable pooled population of non‐converters (LEFT PANEL) and predicting aMCI subject undergoing rapid conversion (

FIGURE 5

Individual performance of cognitive domains. Individual performance in each cognitive domain engaged by the Neuro Motor Index neuro‐motor parameters most critically involved in the Altoida iADL task performance that predict the conversion into either dementia (LEFT PANEL) or Alzheimer's disease dementia (RIGHT PANEL). The standard deviation is computed over five cross‐validation folds