Abstract: Objective: To develop and verify an intelligent early warning model of pressure injury risk in inpatients, and provide reference for the risk management of pressure injury in inpatients. Methods: The pressure injury cases and non-pressure injury cases were selected in a tertiary hospital in Changsha city by convenient sampling method. The intelligent feature extraction of early warning indicators was performed by structured query statements, and the consistency test was carried out. Random forest Gini index combined with the good parsimonious prediction model was used to rank the importance of features. The machine learning algorithms were applied to develop prediction model and evaluate the model performance. Results: The Kappa consistency coefficient between intelligent feature extraction and manual extraction of the indicators were 0.639-1.000. The feature importance of random forest Gini index ranged from 0.005 to 0.220. The good parsimonious prediction model with prediction accuracy screened 12 early warning indicators, including physical restraint, mechanical ventilation, use of vasoactive drugs, disturbance of consciousness, edema, and old age and so on. After comparing the model performance of each prediction model, the one established by random forest was the best. Conclusion: This study developed an intelligent prediction model for pressure injury in inpatients based on healthcare information systems with good model performance. This prediction model is suitable for clinical application.

Key words: pressure injury; healthcare information systems; machine learning; intelligent prediction model; wise information technology of nursing