Skip to main content
NUTRITION 2023 Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Topical Area: Medical Nutrition/Case Study Vignettes

Medical Nutrition/Case Study Vignettes (Poster Session)

(P08-009-23) Machine Learning-Assisted Nutritional Risk Assessment in Critically Ill Patients

Saturday, July 22, 2023
12:00 PM – 1:00 PM ET
Location: Poster Board 91

    Presenting Author(s)

  • HH

    Hui Min Hsieh (she/her/hers)

    Director of Department of Food and Nutrition
    Taichung Veterans General Hospital
    Taichung, Taichung, Taiwan (Republic of China)

    • Co-Author(s)

  • CC

    Chao Hsiu Chen, Dietitian

    Taiwan
    Department of Food and Nutrition of Taichung Veterans General Hospital
    Taichung, Taichung, Taiwan (Republic of China)

  • YC

    Yi-Jui Chan, master

    The Department of Food and Nutrition of Taichung Veterans General Hospital
    Taichung, Taichung, Taiwan (Republic of China)

  • HH

    Hsiao Lin Hsu, Dietitian

    The Department of Food and Nutrition of Taichung Veterans General Hospital
    Taichung, Taichung, Taiwan (Republic of China)

  • CW

    Chen-Yu Wang

    Department of Critical Care Medicine, Taichung Veterans General Hospital

  • CW

    Chieh-Lieng Wu

    Department of Post-Baccalaureate Medicine, College of Medicine, National Chung Hsing University

Disclosure(s):
Hui Min Hsieh: No relevant financial relationship(s) with ineligible companies to disclose.


Objectives: Criticaliy ill patients are at high nutritional risk with higher mortality rates and require immediate nutritional support. The predominant method of nutritional risk assessment, mNUTRIC score, lacks information on the patients' body weight and feeding. This study established a predictive nutrition risk machine model using artificial intelligence and data integration to increase the accuracy in assessing nutritional risk of criticaliy ill patients. Enable real-time nutritional interventions for high-risk populations. This is a research study of the National Science and Technology Council of Taiwan. (MOST100-2321-B-075A-001)

Methods: This was a retrospective study that used the Taichung Veterans General Hospital's (TCVGH) intensive care unit database to select patients over 20 years old who were admitted to the TCVGH intensive care unit from January 2015 to December 2019. The data collection period was 24 hours before admission to the intensive care unit to 48 hours after admission. The data was annotated by dietitians after de-identification. Deep machine learning methods were employed for data exploration and modeling. Three algorithms (XGBOOST, CatBOOST, LightGBM) were used to analyze the important variables and establish a predictive nutrition risk model.

Results: This research was performed from January 2021 to December 2022. A total of 9,356 records were selected from the intensive care database of TCVGH and 1,994 of them were annotated for nutrition risk by seven participating dietitians. Out of the selected records, 701 were marked as high-risk and 1,293 were marked as low-risk. The model AUC of the three algorithms exceeds 0.92. For instance, with the LightGBM algorithm, the precision, specificity, sensitivity, and accuracy of the model were 0.86, 0.93, 0.80, and 0.88, respectively.

Conclusions: The predictive nutrition risk model can characterize the nutritional status and provide early warnings of patients, allowing high-risk patients to receive timely and accurate nutritional support. This model is still in its early stages of use in critical ill nutrition risk assessment. We plan to continue the effort of feedback collection and model refinement to make it more assistive to the daily work of our dietitians.

Funding Sources: This is a research project of the National Science and Technology Council of Taiwan. (MOST100-2321-B-075A-001)