Machine Learning-Based Routine Laboratory Tests Predict One-Year Cognitive and Functional Decline in a Population...

Area of Science:

• Gerontology and Artificial Intelligence
• Biomarker Discovery
• Predictive Analytics in Healthcare

Background:

• Cognitive and functional decline are prevalent in adults aged 75+
• No specific plasma biomarker currently predicts decline in healthy older adults
• Machine learning (ML) offers potential for outcome prediction

Purpose of the Study:

• Evaluate routine laboratory variables for predicting cognitive and functional impairment
• Utilize ML algorithms in a cohort aged 75+ years
• One-year follow-up study to assess predictive capabilities

Main Methods:

• 132 adults aged 75+ evaluated at baseline and one year
• Functional and cognitive assessments included questionnaires and standardized tests
• Machine learning models (Random Forest, SVM, XGBoost) applied to routine lab tests

Main Results:

• Random Forest model demonstrated high accuracy (0.79 for cognitive, 0.92 for functional decline)
• Key predictors for cognitive decline included triglycerides, glucose, hematocrit, RDW, albumin, hemoglobin, TSH, and inflammatory markers
• Key predictors for functional decline included platelets, hemoglobin, cortisol, glucose, B12 vitamin, creatinine, and various blood cell counts and ratios

Conclusions:

• Routine laboratory variables are effective predictors of cognitive and functional decline
• ML algorithms can successfully identify predictive biomarkers in oldest-old populations
• This approach may facilitate early intervention and personalized care strategies