Development of an explainable machine learning model for predicting device-related pressure injuries in clinical settings
Abstract Background Device-related pressure injury (DRPI) is a prevalent and severe problem for patients using medical devices. Timely identification of patients at high risk of DRPI is crucial for healthcare providers to make informed decisions and prevent DRPI quickly. Given the rapid advancements...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-07-01
|
| Series: | BMC Medical Informatics and Decision Making |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12911-025-03090-9 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Background Device-related pressure injury (DRPI) is a prevalent and severe problem for patients using medical devices. Timely identification of patients at high risk of DRPI is crucial for healthcare providers to make informed decisions and prevent DRPI quickly. Given the rapid advancements in computer technology, we aimed to develop an interpretable artificial intelligence (AI) model for predicting DRPI, utilizing SHAP (SHapley Additive exPlanations) to enhance the model’s transparency and provide insights into feature importance. Methods We enrolled 675 study subjects (225 in the DRPI group and 450 in the non-DRPI group) from a single medical center between January 2019 and December 2020. Python was used to perform classification models, including extreme gradient boosting (XGBoost), random forest (RF), decision tree (DT), Logistic Regression (LR), support vector machine (SVM), and K-Nearest neighbors (KNN). We evaluated the performance of the six models using area under the ROC curve (AUC), specificity, accuracy, and sensitivity, with the dataset split into a 80% training set and a 20% testing set. We utilized several analyses, such as SHAP and Uniform Manifold Approximation and Projection (UMAP), to explore the potential contribution of different characteristics in our risk prediction models. Results In the test set, XGBoost model outperformed the other models (AUC = 0.964). The interpretation of the model using SHAPscores revealed that the length of stay, instrument type, emergency admissions, instrument material, and instrument duration of use are the top five most important features in predicting DRPI. Conclusion Our study demonstrated that the development of DRPI in patients can be accurately predicted using the machine learning (ML) model. The findings not only provide clinical caregivers with a valuable framework to identify patients at high risk of DRPI, but also lay the groundwork for developing targeted preventive strategies and personalized interventions. |
|---|---|
| ISSN: | 1472-6947 |