INS NYC 2024 Program

Poster	Poster Session 02 Program Schedule 02/15/2024 08:00 am - 09:15 am Room: Majestic Complex (Posters 61-120) Poster Session 02: Aging \| MCI \| Neurodegenerative Disease - PART 1 Final Abstract #106 The Cognitive, Age, Functioning, and APOE4 (CAFE) Scorecard to Predict the Development of Alzheimer’s Disease Yumiko Wiranto, University of Kansas, Lawrence, United States Devin Setiawan, University of Kansas, Lawrence, United States Arian Ashourvan, University of Kansas, Lawrence, United States Amber Watts, University of Kansas, Lawrence, United States Category: Dementia (Alzheimer's Disease) Keyword 1: dementia - Alzheimer's disease Keyword 2: cognitive functioning Objective: Timely intervention in Alzheimer's disease (AD) relies on early detection; Current tools for this purpose (brain imaging, lumbar puncture) pose challenges due to cost, access, and practical complexities. The present study aimed to create a scoring system to predict the likelihood of developing AD using accessible variables, including demographics, informant-reported daily functioning, APOE4 status, and cognitive performance, in older adults with normal cognition (NC) and mild cognitive impairment (MCI). Participants and Methods: We analyzed data from 731 participants (Mean age = 73.44, SD = 6.88) with either NC or MCI diagnosis at baseline from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Among these participants, 57.36% remained stable, and 42.64% converted to AD after an average of 7.2 (NC) and 2.5 (MCI) years. A single subtest was selected to assess each cognitive domain: Category Fluency for language, Trail-Making Test A for attention, Trail-Making Test B for executive function, Rey Auditory Verbal Learning (RAVLT) for working memory, and logical memory delayed recall for long-term memory. The Mini-Mental State Examination provided a measure of global cognitive function. Informant-reported daily functioning was measured with the Functional Activities Questionnaire (FAQ). Age, sex, educational level, and APOE4 status were also integrated into the prediction algorithm. The scorecards were generated using the FasterRisk algorithm (Liu et al., 2022) that leverages sparse linear models with integer coefficients, offering memorability akin to traditional risk scores and allowing for an efficient technique to address complex mathematical programming. A cross-validation procedure was employed to determine the optimal sparsity parameter. Results: A sparsity level of 5 (AUC = 0.872) was selected for the generation of the final scorecards. Ten scorecards were generated with a test AUC range of 0.867 to 0.893. This abstract will demonstrate the detailed results from one scorecard. This scorecard (AUC = 0.893) features 5 variables with the following point assignments: (1) age < 76 (-2 points); (2) no APOE4 alleles (-3 points); (3) RAVLT <= 36 items (4 points); (4) logical memory <= 3 items (5 points); and (5) FAQ <= 2 (-5 points). Positive points indicate an elevated risk of AD, while negative points suggest a reduced risk. The probable AD development risk was 4.3% for a score of -10, 12.5% for a score of -7, 31.5% for a score of -3, 50% for a score of -1, 76.3% for a score of 1, 87.5% for a score of 3, and greater than 95% for a score of 6, 7, or 9. In sum, younger age, absence of APOE4 alleles, higher cognitive performance, and greater daily functioning contributed to reduced AD risk. Conclusions: Our findings highlight the potential of these interpretable scorecards to predict the likelihood of AD using accessible information. They possess wide-ranging applicability across diverse healthcare environments, accommodating resource-constrained settings and allowing for expert customization. While our initial scope centers on AD, the foundation we have established paves the way for similar methodologies to be applied to the early detection of other types of dementia.

Poster

02/15/2024
08:00 am - 09:15 am
Room: Majestic Complex (Posters 61-120)

Poster Session 02: Aging | MCI | Neurodegenerative Disease - PART 1

Final Abstract #106

The Cognitive, Age, Functioning, and APOE4 (CAFE) Scorecard to Predict the Development of Alzheimer’s Disease

Yumiko Wiranto, University of Kansas, Lawrence, United States
Devin Setiawan, University of Kansas, Lawrence, United States
Arian Ashourvan, University of Kansas, Lawrence, United States
Amber Watts, University of Kansas, Lawrence, United States

Category: Dementia (Alzheimer's Disease)

Keyword 1: dementia - Alzheimer's disease
Keyword 2: cognitive functioning

Objective:

Timely intervention in Alzheimer's disease (AD) relies on early detection; Current tools for this purpose (brain imaging, lumbar puncture) pose challenges due to cost, access, and practical complexities. The present study aimed to create a scoring system to predict the likelihood of developing AD using accessible variables, including demographics, informant-reported daily functioning, APOE4 status, and cognitive performance, in older adults with normal cognition (NC) and mild cognitive impairment (MCI).

Participants and Methods:

We analyzed data from 731 participants (Mean age = 73.44, SD = 6.88) with either NC or MCI diagnosis at baseline from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Among these participants, 57.36% remained stable, and 42.64% converted to AD after an average of 7.2 (NC) and 2.5 (MCI) years.

A single subtest was selected to assess each cognitive domain: Category Fluency for language, Trail-Making Test A for attention, Trail-Making Test B for executive function, Rey Auditory Verbal Learning (RAVLT) for working memory, and logical memory delayed recall for long-term memory. The Mini-Mental State Examination provided a measure of global cognitive function. Informant-reported daily functioning was measured with the Functional Activities Questionnaire (FAQ). Age, sex, educational level, and APOE4 status were also integrated into the prediction algorithm. The scorecards were generated using the FasterRisk algorithm (Liu et al., 2022) that leverages sparse linear models with integer coefficients, offering memorability akin to traditional risk scores and allowing for an efficient technique to address complex mathematical programming. A cross-validation procedure was employed to determine the optimal sparsity parameter.

Results:

A sparsity level of 5 (AUC = 0.872) was selected for the generation of the final scorecards. Ten scorecards were generated with a test AUC range of 0.867 to 0.893. This abstract will demonstrate the detailed results from one scorecard. This scorecard (AUC = 0.893) features 5 variables with the following point assignments: (1) age < 76 (-2 points); (2) no APOE4 alleles (-3 points); (3) RAVLT <= 36 items (4 points); (4) logical memory <= 3 items (5 points); and (5) FAQ <= 2 (-5 points). Positive points indicate an elevated risk of AD, while negative points suggest a reduced risk. The probable AD development risk was 4.3% for a score of -10, 12.5% for a score of -7, 31.5% for a score of -3, 50% for a score of -1, 76.3% for a score of 1, 87.5% for a score of 3, and greater than 95% for a score of 6, 7, or 9. In sum, younger age, absence of APOE4 alleles, higher cognitive performance, and greater daily functioning contributed to reduced AD risk.

Conclusions:

Our findings highlight the potential of these interpretable scorecards to predict the likelihood of AD using accessible information. They possess wide-ranging applicability across diverse healthcare environments, accommodating resource-constrained settings and allowing for expert customization. While our initial scope centers on AD, the foundation we have established paves the way for similar methodologies to be applied to the early detection of other types of dementia.