Related Experiment Videos
Can Unsupervised Machine Learning Support Malnutrition Risk Screening in Older Adults? A Preliminary Study Using
Karolina Kujawowicz1, Iwona Mirończuk-Chodakowska1, Monika Cyuńczyk1
1Department of Food Biotechnology, Medical University of Białystok, ul. Szpitalna 37, 15-285 Białystok, Poland.
Nutrients
|August 13, 2026
Summary
Unsupervised machine learning identified three distinct phenotypes of malnutrition risk in older adults. The most vulnerable group exhibited higher frailty, depressive symptoms, and poorer functional status, indicating a need for targeted interventions.
Area of Science:
- Geriatric Medicine
- Computational Biology
- Public Health
Background:
- Malnutrition in older adults is a significant public health issue requiring early identification.
- Multifactorial risks necessitate advanced analytical approaches for geriatric care.
Purpose of the Study:
- To explore the utility of unsupervised machine learning (ML), specifically hierarchical clustering on principal components (HCPC), in characterizing multidimensional phenotypes of malnutrition risk in older adults.
- To identify distinct geriatric and nutritional profiles associated with malnutrition risk.
Main Methods:
- An exploratory cross-sectional study involving 105 older adults (aged 60-95) from community and residential settings.
- Comprehensive data collection including anthropometrics, body composition, nutritional status, depressive symptoms, physical fitness, frailty, sarcopenia, appetite, and biological parameters.
- Statistical analysis using Multiple Factor Analysis (MFA) for dimensionality reduction followed by HCPC for cluster identification.
Main Results:
- HCPC identified three distinct phenotypes.
- Cluster 1: Preserved nutritional and geriatric status with lower frailty.
- Cluster 2: Male-dominant with preserved muscle mass and functional reserve.
- Cluster 3: Most vulnerable phenotype, characterized by higher malnutrition risk, frailty, polypharmacy, depressive symptoms, and poorer functional status, strongly linked to body composition, muscle mass, grip strength, and functional performance.
Conclusions:
- Unsupervised ML, particularly HCPC, shows promise in characterizing geriatric and nutritional profiles related to malnutrition risk in older adults.
- The most vulnerable phenotype, exhibiting multidimensional vulnerability, presented the highest malnutrition risk.
- Findings warrant validation in larger cohorts to refine targeted interventions for at-risk older adults.