Malnutrition risk assessment using a machine learning-based screening tool: A multicentre retrospective cohort

Area of Science:

• Clinical Nutrition
• Health Informatics
• Machine Learning in Healthcare

Background:

• Malnutrition significantly increases patient morbidity, mortality, and healthcare expenses.
• Early identification of malnutrition is crucial for effective and timely patient intervention.
• This study evaluates a machine learning screening tool's impact on malnutrition detection in hospitals.

Purpose of the Study:

• To assess the effectiveness of the MUST-Plus machine learning tool in early malnutrition identification.
• To evaluate the tool's impact on improving malnutrition diagnosis and documentation rates.
• To determine the usability and acceptance of the MUST-Plus tool by registered dietitians (RDs).

Main Methods:

• A retrospective cohort study was conducted across six hospitals in a large urban health system.
• The study included adult patients (≥18 years) with a length of stay ≤30 days, excluding COVID-19 admissions.
• Data from 7736 hospitalizations were analyzed to compare malnutrition screening pre- and post-MUST-Plus implementation.

Main Results:

• MUST-Plus identified 25.2% (1947/7736) of hospitalizations as malnourished through RD evaluations.
• Implementation of MUST-Plus reduced the time lag between admission and malnutrition diagnosis.
• The tool demonstrated high usability (>90%) among RDs, with improved diagnosis and documentation rates observed.

Conclusions:

• The MUST-Plus machine learning tool shows significant promise for improving malnutrition screening in hospitalized patients.
• Effective implementation requires adequate RD staffing and training on the tool.
• Other health systems can leverage EHR data to develop similar ML-based processes for better malnutrition care.