Harmonizing neuropsychological test data across prospective studies

TL;DR

This study combines data from three Alzheimer's research studies to improve understanding of the disease by harmonizing cognitive test scores.

Contribution

A validated method using MissForest imputation to harmonize cognitive test data across Alzheimer's disease studies.

Findings

MissForest imputation showed high accuracy with mean absolute errors ≤ test–retest variability in most cases.

Composite cognitive scores reflected known disease patterns and showed significant stratification across clinical–pathological groups.

Digit Symbol Coding and Trail-Making Test Part B had slightly higher MAEs but still within individual variation.

Abstract

Alzheimer's disease (AD) research relies on large datasets and advanced statistical models. However, individual population studies often lack sufficient sample size for conclusive results. Harmonizing cognitive test data across studies can address this gap, despite differences in testing protocols. This study harmonizes cognitive data from three major AD cohorts to support robust clinical–pathological modelling.

Information from the Alzheimer's Disease Neuroimaging Initiative (N = 1446); Australian Imaging, Biomarkers and Lifestyle (N = 1764); and Open Access Series of Imaging Studies‐3 (N = 440) were integrated, including cognitive scores, demographics, genetics, and clinical and neuroimaging data. Neuropsychological tests relevant to AD were harmonized using MissForest, a machine learning–based imputation method. Validation involved assessing imputation accuracy and analyzing composite cognitive scores across clinical–pathological groups.

Imputation showed high accuracy (mean absolute error ≤ test–retest variability in cognitively unimpaired participants). Composite scores reflected known disease patterns with significant stratification across clinical–pathological groups.

The validated harmonization approach demonstrated reliable imputation, enabling more powerful AD models and supporting future diagnostic and therapeutic advances.

MissForest imputation reliably harmonized cognitive test scores across Alzheimer's disease (AD) studies.Harmonized dataset included 1446 Alzheimer's Disease Neuroimaging Initiative; 1764 Australian Imaging, Biomarkers and Lifestyle; and 440 Open Access Series of Imaging Studies‐3 subjects.Mean absolute errors (MAEs) ≤ test–retest variability in cognitively unimpaired participants for most tests.Digit Symbol Coding and Trail‐Making Test Part B MAEs slightly > test—retest, < individual variation.Composite scores had lower baseline and faster decline with increasing AD severity.

MissForest imputation reliably harmonized cognitive test scores across Alzheimer's disease (AD) studies.

Harmonized dataset included 1446 Alzheimer's Disease Neuroimaging Initiative; 1764 Australian Imaging, Biomarkers and Lifestyle; and 440 Open Access Series of Imaging Studies‐3 subjects.

Mean absolute errors (MAEs) ≤ test–retest variability in cognitively unimpaired participants for most tests.

Digit Symbol Coding and Trail‐Making Test Part B MAEs slightly > test—retest, < individual variation.

Composite scores had lower baseline and faster decline with increasing AD severity.

Figures