Machine Learning in Acute Stroke Care: A Novel Model for Assessing the Need for Enteral Nutrition

Abstract

Aim: Early enteral nutrition is often recommended for patients with acute stroke who have difficulty with oral intake. This study aimed to develop a predictive model to assess the need for enteral nutrition in older patients with acute cerebrovascular disorders. The model employs a machine learning algorithm using observational parameters related to swallowing ability. Methods: Ninety patients experiencing a cerebrovascular accident for the first time were included in this study. Swallowing function was assessed using the Food Intake LEVEL Scale. Nine specific variables were used to create a model for determining the need for enteral nutrition. Initially, variable selection was conducted through correlation analysis. Subsequently, the data were randomly divided into training and test groups. Five machine learning methods were applied to identify the most effective algorithm: logistic regression, decision tree, random forest, support vector machine, and XG Boost. Results: Through correlation analysis, we identified the independent variables Functional Independence Measure, motor and cognitive scores and speech intelligibility. The logistic regression model demonstrated high performance (accuracy, 0.82; area under the curve, 0.82). Conclusion: We demonstrated that a predictive model, employing machine learning and integrating Functional Independence Measure motor and cognitive scores and speech intelligibility, exhibits superior predictive efficacy and ascertains the necessity for enteral nutrition. This model can be expediently appraised even by individuals not specialized in dysphagia. Additionally, it is applicable to patients who are incapable of adhering to conventional swallowing assessment protocols owing to compromised consciousness or cognitive impairments, or those with an exceptionally elevated risk of aspiration.