Improving machine learning algorithm for risk of early pressure injury prediction in admission patients using probability feature aggregation.

Abstract

Objective: Pressure injuries (PIs) pose a significant concern in hospital care, necessitating early and accurate prediction to mitigate adverse outcomes.

Methods: The proposed approach receives multiple patients records, selects key features of discrete numerical based on their relevance to PIs, and trains a random forest (RF) machine learning (ML) algorithm to build a predictive model. Pairs of significant categorical features with high contributions to the prediction results are grouped, and the PI risk probability for each group is calculated. High-risk group probabilities are then added as new features to the original feature subset, generating a new feature subset to replace the original one, which is then used to retrain the RF model.

Results: The proposed method achieved an accuracy of 83.44%, sensitivity of 84.59%, specificity of 83.42%, and an area under the curve of 0.84.

Conclusion: The ML-based approach, coupled with feature aggregation, enhances predictive performance, aiding clinical teams in understanding crucial features and the model's decision-making process.