Predicting malnutrition-based anemia in geriatric patients using machine learning methods.

Area of Science:

• Geriatric Medicine
• Computational Medicine
• Nutritional Science

Background:

• Anemia in the elderly, often from nutritional deficiencies (iron, folate, vitamin B12), increases morbidity and mortality risks.
• Early anemia diagnosis and treatment are critical for improving geriatric patient outcomes.
• This study focuses on predicting anemia in outpatient geriatric populations.

Purpose of the Study:

• To predict anemia diagnosis in geriatric patients using machine learning (ML) methods.
• To evaluate ML model performance with and without hemogram data.
• To provide a valuable dataset for future geriatric anemia research.

Main Methods:

• Anemia classification using ML on hemogram, biochemistry, malnutrition, and physical/cognitive activity scores.
• Comparison of ML algorithms, including J48 and Random Forest.
• Analysis of predictive performance using only non-blood-test attributes (malnutrition, physical activity).

Main Results:

• The J48 algorithm achieved 97.77% accuracy when using all available data.
• Excluding hemogram data, the Random Forest algorithm achieved 85.39% accuracy using only malnutrition and physical activity scores.
• The dataset comprised 438 geriatric patient observations.

Conclusions:

• Anemia in geriatric patients can be accurately predicted without relying on hemogram data.
• The study highlights the potential of ML for non-invasive anemia prediction in the elderly.
• The shared dataset facilitates further research into ML methodologies and geriatric disease prediction.