Association rule learning in neuropsychological data analysis for Alzheimer's disease.

Abstract

Efficient methods of analysis readily available for clinicians continue to be limited within neuropsychological assessment at the raw data level. Here, a novel approach for generating predictive response patterns and analysing neuropsychological raw data is offered. In order to assess the usefulness of association rule learning as an analysis tool for neuropsychological raw data, Frequent Pattern Growth (FP-Growth) was used to mine patterns from the Consortium to Establish a Registry for Alzheimer's Disease Neuropsychological Battery (CERAD-NB) database. Complete assessment data for 84 post-mortem confirmed Alzheimer's disease (AD) cases and 294 age, race, and education matched controls were analysed across baseline and one-year follow-up using FP-Growth, for the purpose of assessing the clinical utility of a finer analysis at the raw data level and the feasibility of predicting response patterns for clinical/control groups. Output from FP-Growth, in terms of the number of frequent itemsets retained across both evaluation timepoints, was discernable between controls, mild, and moderate to severe Alzheimer's disease cases (p < .001, and η2 = .488). Patterns within raw data scores, both in terms of frequent itemsets and predictive association rules, appear to be differentiable across groups within neuropsychological analysis, which indicates that FP-Growth is applicable as a supplementary analysis tool for neuropsychological assessment by means of offering an additional level of data analysis, predictive item response capabilities, and aiding in clinical decision making.