Advancing dementia identification using machine learning and real-world sequential health data.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Gerontology

Background:

• Dementia identification is a global challenge, with many cases often unrecognized.
• Early detection is crucial for timely intervention and management.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in identifying dementia cases using routinely collected health data.
• To assess the impact of longitudinal data features on model performance for dementia case finding.

Main Methods:

• A nested case-control study involving 8195 individuals with dementia and 8195 matched controls.
• Development and testing of four machine learning models incorporating cross-sectional and longitudinal health data features.
• Utilized de-identified health records from a real-world setting in New Zealand.

Main Results:

• The best-performing model achieved an area under the curve (AUC) of 0.86, with 73.3% sensitivity and 87.5% specificity.
• Key predictors for dementia identification included ICD-10 codes, healthcare utilization, aged residential care referrals, and delirium assessments.
• Inclusion of longitudinal "timestamp" data uniquely enhanced model performance.

Conclusions:

• Machine learning models demonstrate good accuracy for dementia identification in real-world healthcare settings.
• Longitudinal data analysis is a valuable feature for improving machine learning model performance in dementia case finding.
• Routine health data, when analyzed with advanced ML techniques, holds significant potential for enhancing early dementia detection.