Functional assessment improves discrimination of clinical and biomarker-defined Alzheimers disease

Technical Summary: Functional Assessment Improves Discrimination of Clinical and Biomarker-Defined Alzheimer's Disease

Problem Statement
Accurately identifying the risk of Alzheimer's Dementia (AD) is critical for clinical screening and research recruitment. While numerous risk algorithms exist to quantify AD risk based on demographic, clinical, and lifestyle factors, their effectiveness in modern, diverse cohorts remains uncertain. These algorithms are historically grounded in specific populations and contexts; as societal conditions, educational opportunities, and lifestyle patterns evolve, the relevance of distal risk factors (e.g., BMI, specific dietary habits) may fluctuate. Furthermore, the operationalization of these complex algorithms in real-world datasets is often hindered by data granularity issues, where specific lifestyle proxies are unavailable or imperfectly aligned. Consequently, there is a need to evaluate whether established risk scores retain discriminatory power in contemporary, diverse populations and to identify which specific components drive their performance.

Methodology
This study conducted an exploratory secondary analysis using the Bio-Hermes-001 dataset, a diverse cross-sectional cohort of 1,001 community-dwelling adults aged 60–85. The cohort included significant representation from populations typically underrepresented in dementia research (19% of Healthy Cognitions, 24% of Mild Cognitive Impairment, and 33% of AD cases).

The authors evaluated seven established risk algorithms: CAIDE, LIBRA, BDSI (Brief Dementia Screening Indicator), ANU-ADRI, CogD, and CogD-AD. The study assessed the discrimination and calibration of these scores against six distinct outcomes:

1. Clinical Classifications: Healthy Cognition (HC) vs. Probable AD; HC vs. Mild Cognitive Impairment (MCI); HC vs. pooled MCI/AD; and MCI vs. AD.
2. Biomarker Status: Amyloid Beta (AB) PET scan positivity (binary) and plasma phosphorylated tau-217 (pTau-217) positivity (binary).

Performance was measured using Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC) for discrimination, alongside Hosmer-Lemeshow tests, Brier scores, and calibration plots for calibration. A post-hoc ablation analysis was performed on the best-performing algorithm (BDSI) to isolate the contribution of individual risk factors, specifically removing the "functional" item (difficulty managing money or medications) to assess its impact on discrimination and calibration.

Key Results

• Limited Performance of Established Scores: Most established algorithms (CAIDE, LIBRA, ANU-ADRI, CogD, CogD-AD) demonstrated limited discriminatory power, often performing at chance levels (AUC ≈ 0.50–0.60) for both clinical classifications and biomarker outcomes (AB PET and pTau-217). While calibration was generally acceptable for most scores (except the oldest algorithms), they failed to distinguish between groups effectively.
• Superiority of BDSI: The Brief Dementia Screening Indicator (BDSI) was the only algorithm to achieve clinically meaningful discrimination. It yielded an AUC of 0.86 for distinguishing HC from AD, 0.75 for HC from pooled MCI/AD, and 0.71 for MCI from AD. It also outperformed other scores in biomarker discrimination, achieving an AUC of 0.74 for AB PET and 0.67 for pTau-217.
• Critical Role of Functional Assessment: The ablation study revealed that the functional item ("needing help managing medications or money") was the primary driver of BDSI's performance. Removing this item resulted in a statistically significant reduction in AUC across all clinical and biomarker comparisons (e.g., dropping from 0.86 to 0.70 for HC vs. AD). DeLong tests confirmed these differences were statistically significant.
• Calibration vs. Discrimination: While removing the functional item reduced discrimination, the BDSI score (with or without the functional item) remained acceptably calibrated for most outcomes, suggesting the model structure remained sound but lacked a critical predictive variable.
• Depression Metric Discrepancy: The study noted that the BDSI's inclusion of antidepressant use (rather than just a depression symptom threshold) captured a significantly larger proportion of participants with depression-related risk factors compared to other algorithms, potentially contributing to its performance, though the functional item remained the dominant factor.

Significance and Claims
The paper claims that in a contemporary, diverse cohort, most traditional epidemiological risk algorithms have limited utility for discriminating current cognitive status or underlying AD pathology (amyloid and pTau positivity). The authors assert that the inclusion of a proximal functional measure markedly improves discrimination across both clinical and biomarker-defined outcomes.

The significance of these findings lies in the suggestion that functional assessment captures "integrated downstream consequences" of biological, cognitive, and environmental factors that distal lifestyle risk factors miss. The authors propose that functional independence represents a higher-order construct of cognitive vulnerability that may be less dependent on educational attainment or cultural heterogeneity. Consequently, functional assessment should be considered a critical component for scalable screening and referral pathways, potentially complementing or refining epidemiological risk algorithms. The authors conclude that while their findings are based on a cross-sectional design, they highlight the necessity of incorporating proximal indicators of daily life capacity to improve the identification of individuals with underlying AD-related pathology.