Development of Alzheimer’s Disease Risk Score for Future Primary Care: A White-Box Approach
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Importance
Interpretable scoring system can contribute to bridge the gap between the timeliness and complexity of diagnosing Alzheimer’s disease (AD) and promote early intervention at non-specialist settings.
Objective
To develop a risk score to predict the likelihood of AD with interpretable machine learning using variables that are obtainable at integrated primary care settings.
Design
A secondary data analysis including cohort studies from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the National Alzheimer’s Coordinating Center (NACC) extracted in August 2023 and March 2024.
Setting
The ADNI and NACC are multi-site cohort studies in North America.
Participants
Participants with normal cognition or mild cognitive impairment at baseline visit were identified. Participants with the same diagnosis overtime were assigned to the stable group, and those converted to AD were placed in the progressive group.
Main Outcome(s) and Measure(s)
Cognitive tests and daily functioning measured with Functional Assessment Questionnaire (FAQ) at baseline visit.
Results
A total of 676 participants from ADNI and 4592 participants from NACC were identified. After removing incomplete data, 665 ADNI (mean age [SD]: 73.44 [6.90]; 293 [44.1%] female; 374 stable and 291 progressive) and 3657 NACC participants (mean age [SD]: 70.96 [10.03]; 2405 [65.8%] female; 2445 stable and 1212 progressive) remained. Combinations of 4 measures were selected to generate 10 scorecards using FasterRisk algorithm, showing strong performance (area under the curve [AUC] = 0.868-0.892) in ADNI and remaining robust when validated in NACC (AUC = 0.795). The features were Category Animal ≤ 20 (2 points), Trail Making Test B ≤ 143 (−3 points), Logical Memory Delayed ≤ 3 (4 points), Logical Memory Delayed ≤ 8 (3 points), and FAQ ≤ 2 (−5 points). The probable AD risk corresponded to total points: 7.4% (–8), 25.3% (–4), 50% (–1), 74.7% (2), and > 90% (≥ 6). We refer to this model as the (F)unctioning, (LA)nguage, (M)emory, and (E)xecutive functioning or FLAME scorecard.
Conclusions and Relevance
Our findings highlight the potential to predict AD development using obtainable information, allowing for applicability at integrated primary care. While our scope centers on AD, this foundation paves the way for other dementia types
Key Points
Question
Can accessible information, such as demographics, cognitive tests, and functioning questionnaire, yield in reliable results for predicting Alzheimer’s disease development using interpretable machine learning?
Findings
The results of 665 participants from the Alzheimer’s Disease Neuroimaging Initiative demonstrated robust performance of determining Alzheimer’s disease development using four separate measures of (F)unctioning, (LA)nguage, (M)emory, and (E)xecutive functioning or the FLAME scorecard. It remains reliable when externally validated with a separate dataset of 3657 participants from the National Alzheimer’s Coordinating Center.
Meaning
The FLAME scorecard shows potential to be implemented in integrated primary care settings to promote early detection and intervention of cognitive decline due to Alzheimer’s disease.
