Objective Pressure injuries (PIs) are a common issue among patients, particularly in home care centers. PIs lead to increased complications and increased healthcare costs. This study aimed to predict the occurrence of PIs among patients in Taiwan’s home care centers using electronic health records (EHRs) from the proposed Home Care Management System (HCMS). Design A retrospective study was conducted. Setting and Participants Nurses completed the holistic healthcare assessment (HHCA) for 36,896 patients. The data collection duration covered the period from October 2021 to May 2023. The study analyzed 44,188 cases, with a PI prevalence of approximately 31%. Methods This study employed an automated machine learning (AutoML) approach. The AutoML systematically and efficiently selected modeling strategies and tuned hyperparameters using a partial factorial design. This approach reduced the number of trials needed for optimal model performance. Results On the independent hold‐out testing set, the final model achieved an accuracy of 77.56% and an AUC of 81.82%. Not only can top risk factors be identified but also the Shapley additive explanations (SHAP) are employed at an individual level over time. Conclusions and Implications This study represents one of the first large‐scale applications of AutoML for PI prediction in the home care setting, a context that differs substantially from both nursing homes and hospitals. Our findings emphasize Braden assessment, chronic disease burden, polypharmacy, and nutritional status as the dominant predictors in community‐dwelling patients receiving professional home care. By extending predictive analytics into this underexplored domain, the study demonstrates how risk models can be operationalized to support home care nurses. In particular, SHAP‐based waterfall plots guide nurses to detect whether specific risk features are improving or worsening and to adjust care plans accordingly.
Tsay et al. (2026) studied this question.