Machine learning model predicted mortality among malnourished inpatients with IBD
Integrating such models into clinical practice could improve identification of inflammatory bowel disease (IBD) patients at risk for malnutrition and potentially reduce their mortality by prompting earlier intervention, according to the authors of a new study.
Researchers created and validated a model for identifying malnourished patients hospitalized with inflammatory bowel disease (IBD) who are most in need of medical and nutritional intervention.
To develop a machine-learning model that accurately predicts mortality in hospitalized IBD patients with protein-calorie malnutrition, the researchers used data from the 2016 to 2019 National Inpatient Sample in two machine learning models, Random Forest Classifier and Extreme Gradient Boosting. Seventy percent of the sample was used to train the model, and the remaining 30% was used to test it. For external validation, they used 2019 data that had been sequestered from model development. Results were published by the Journal of Clinical Gastroenterology on Jan. 20.
Among 879,730 malnourished patients hospitalized for IBD, 1,930 (0.2%) died. Although only 30.0% of hospitalizations involved patients age 60 years or older, this group accounted for 80.3% of inpatient mortality. White race, ulcerative colitis, higher comorbidity indices, Medicare insurance, nonelective admission, weekend admission, and rural hospitals were also associated with inpatient mortality.
The final model had excellent performance at predicting inpatient mortality in both testing and validation sets, the study found. The area under the receiver-operating characteristic curve (AUROC) was 0.93 in the internal validation set and 0.91 in the external validation set. Precision, sensitivity, specificity, and accuracy were 0.98, 0.99, 0.99, and 0.99, respectively.
Machine learning models can accurately predict mortality in malnourished patients hospitalized with IBD, solely relying on readily available clinical data, the authors concluded. Integrating such tools into clinical practice could improve risk stratification, for example, by using the electronic medical record to flag for patients at highest mortality risk who could then be prioritized for closer monitoring, medical and nutrition intervention, and escalation of care.
“Amid an acute care hospitalization with multiple teams focused on improving and/or stabilizing a myriad of conditions, the patient's nutritional needs can often be missed or unintentionally ignored,” the study authors wrote.