Temporal prediction of in-hospital falls using tensor factorisation.

Area of Science:

• Healthcare Informatics
• Predictive Analytics
• Medical Data Science

Background:

• In-hospital falls are a significant concern, impacting patient outcomes and healthcare costs.
• Effective prediction of fall incidents is crucial for proactive patient safety measures and resource management.
• Existing predictive models often lack the ability to identify high-risk areas proactively.

Purpose of the Study:

• To develop and evaluate a novel tensor factorization framework for predicting in-hospital fall incidents.
• To capture latent temporal features for enhanced fall prediction accuracy.
• To identify high-risk locations that may not have recent fall records.

Main Methods:

• A tensor factorization-based framework was proposed to model complex spatio-temporal patterns of fall incidents.
• The model was trained and validated using real-world in-hospital data from Hong Kong.
• Performance was compared against traditional time series models.

Main Results:

• The proposed tensor-based framework achieved a high predictive performance, indicated by an Area Under the Curve (AUC) score of approximately 0.9.
• The model successfully identified high-risk locations even in the absence of recent fall incidents.
• Tensor-based models outperformed baseline time series models in predicting fall incidents.

Conclusions:

• Tensor factorization offers a powerful approach for predicting in-hospital falls by uncovering hidden patterns.
• The developed framework can aid healthcare providers in optimizing resource allocation and enhancing patient safety.
• This method provides a valuable tool for proactive risk identification in healthcare settings.