Related Experiment Video
Updated: May 17, 2025
A Computer-Based Platform for Aiding Clinicians in Eating Disorder Analysis and Diagnosis
Published on: May 10, 2022
An artificial intelligence malnutrition screening tool based on electronic medical records
Xue Wang1, Kuanda Yao2, Zhicheng Huang3
1Department of Clinical Nutrition, Department of Health Medicine, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100730, China; Department of Neurology, Zhongshan Hospital, Fudan University, Shanghai, 200032, China.
An artificial intelligence (AI) tool effectively screens for malnutrition risk using electronic health records (EHR), improving early detection in inpatients. This AI approach enhances efficiency and diagnosis rates compared to traditional methods.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Healthcare
- Nutritional Science
Background:
- Nutrition screening is crucial for timely intervention in patients with malnutrition.
- Current methods can be inefficient, potentially delaying diagnosis and treatment.
- Developing automated tools using electronic health records (EHR) can improve efficiency and malnutrition diagnosis rates.
Purpose of the Study:
- To develop an artificial intelligence (AI) tool for malnutrition risk screening using electronic medical records.
- To compare the performance of the AI tool against traditional patient interview-based screening methods.
- To evaluate the efficiency and diagnostic accuracy of AI-driven nutrition screening.
Main Methods:
- A cross-sectional study was conducted at a tertiary hospital in China.
- Electronic health records (EHR) data were extracted to train and test AI models for malnutrition risk screening.
- Six machine learning algorithms were compared, with the GLIM framework used as a reference standard; feature selection and cross-validation were performed.
Main Results:
- The extreme gradient boosting (XGBoost) algorithm demonstrated the highest performance (AUC).
- Key features identified for malnutrition risk included weight loss, decreased food intake, prealbumin, white blood cell count, BMI group, and neutrophil percentage.
- Simplified models showed high accuracy, with Random Forest achieving an AUC of 0.97 using interview data and 0.87 using EHR data.
Conclusions:
- Artificial intelligence can effectively integrate inpatient EHR data for malnutrition risk detection.
- The developed AI-enabled tool shows significant promise for timely and efficient nutrition screening of newly admitted inpatients.
- This technology has the potential to enhance clinical workflows and improve patient outcomes related to malnutrition.
