Prediction of COVID-19 patients’ admission to the Intensive Care Unit based on the precision nursing framework

Objective: apply and compare machine learning algorithms to predict COVID-19 patients’ admission to Intensive Care Unit from the Precision Nursing theoretical framework. Methods: Retrospective study with 180 patients reported in the city of Florianopolis. The performance of the following algorithms was evaluated: multilayer perceptron - artificial neural network, AdaBoost, logistic regression, random forest, kNN, Naive Bayes, SVM and decision tree. Results: The predictor variables that most influenced the model were hospital admission, race and throat pain. The multilayer perceptron model achieved better prediction for AUC (0.917), sensitivity (0.861), and specificity (0.825). Conclusion: This application proved to be a viable method for predicting the admission of COVID-19 infected patients to ICU and the clinical biomarkers prove to be relevant for the clinical practice of Nursing because they are easily observable and can be quickly implemented.