JMIR Publications

ABSTRACT

Background:

In this study, we aim to develop an artificial intelligence-driven prediction model capable of detecting hypovigilance events using routinely collected physiological data in the ICU.

Objective:

In this study, we aim to develop a machine-learning algorithm capable of detecting hypovigilance events using routinely collected physiological data in the ICU.

Methods:

This derivation study used prospective observational data collected from eligible ICU patients in Lévis, Québec. We included patients admitted to the ICU between October 2021 and June 2022 who were aged ≥18 years and had an anticipated ICU stay of ≥48 hours. ICU nurses identified hypovigilant states every hour using the Richmond Agitation and Sedation Scale (RASS) or the Ramsay Sedation Scale (RSS). Routine vital signs (heart rate, respiratory rate, blood pressure, and oxygen saturation), as well as other physiological and clinical variables (premature ventricular contractions, intubation, use of sedative medication, and temperature), were automatically collected and stored using a GE CARESCAPE Gateway (for vital signs) or manually collected (for sociodemographic characteristics and medication) through chart review. Time series were generated around hypovigilance episodes for analysis. Random Forest, XGBoost, and LightGBM classifiers were then used to detect hypovigilant episodes on the basis of analyzing time series. Hyperparameter optimization was performed using a random search in a 10-fold group cross-validation setup. We report the results of this study using the TRIPOD+AI guidelines and potential biases were assessed using PROBAST.

Results:

Out of 136 potentially eligible participants, data from 30 patients (mean age: 69 years; 63% male) were collected for analysis. Among all participants, 30% were admitted to the ICU for surgical reasons. Following data preprocessing, the study included 1,493 hypovigilance episodes and 764 non-hypovigilant episodes. Among the three sets of models evaluated, LightGBM demonstrated the best performance. It achieved an average accuracy of 68% to detect hypovigilant episodes, with a precision of 76%, a recall of 74%, an area under the curve (AUC) of 60%, and an F1 score of 69%. The application of Shapley additive explanations (SHAP) values to elucidate the determinants of the model’s prediction revealed that the model was influenced by intubation status, respiratory rate, and noninvasive systolic blood pressure.

Conclusions:

All of the classifiers produced precision and recall values that show potential for further development, with slightly different yet comparable performances in classifying hypovigilant episodes. Machine learning algorithms designed to detect hypovigilance have the potential to support early detection of hypoactive delirium in ICU patients