A machine learning model for prediction of risk of dementia superimposed on delirium in intensive care patients.

BACKGROUND

The objective of this study was to construct a predictive model using multiple machine learning algorithms to predict the risk of dementia superimposed on delirium (DSD) in dementia patients in the ICU.

METHODS

The data for this study were sourced from the Medical Information Mart for Intensive Care IV database. The dataset was divided into a development set (70%) and a test set (30%). Feature selection was conducted using the Boruta algorithm to identify clinically relevant predictors, followed by model development using logistic regression, decision trees, random forest, XGBoost, glmnet, k-nearest neighbors (KNN), support vector machine (SVM), and neural network (NNET). The performance of the model was evaluated using several performance metrics, including the area under the curve (AUC), accuracy, precision, sensitivity, specificity, and the F-beta score.

RESULTS

The findings demonstrated that the XGBoost model showed stable and acceptable discriminative performance, with an AUC of 0.7072 in the development set and 0.7258 in the test set, along with reasonable agreement between predicted and observed risks. The model achieved a balanced classification profile, with satisfactory specificity and sensitivity, and an F-beta score of 0.7594 in the test set. Decision curve analysis indicated that the XGBoost model provided net clinical benefit across a range of decision thresholds.

CONCLUSIONS

XGBoost-based machine learning models demonstrate moderate discriminative ability for predicting the occurrence of DSD in ICU patients with dementia and may support risk stratification in similar ICU settings, pending further external validation in independent cohorts.