Enhancing multimodal inpatient fall prediction via nursing statement integration within the OMOP common data model.

Area of Science:

• Clinical Informatics
• Machine Learning in Healthcare
• Patient Safety Research

Background:

• Accurate inpatient fall prediction is crucial for patient safety.
• Traditional models often neglect valuable clinical context from nursing records.
• Nursing documentation contains rich patient status information.

Purpose of the Study:

• To develop and validate machine learning (ML) models integrating structured clinical data and nursing statements.
• To assess the incremental predictive value of nursing documentation for fall risk.
• To enhance fall risk stratification using combined data sources.

Main Methods:

• Retrospective cohort study using adult inpatient data mapped to OMOP Common Data Model (CDM).
• Linked CDM data with nursing statements on functional status, mobility, mental status, and care dependency.
• Trained and evaluated Logistic Regression, Random Forest, Gradient Boosting, and XGBoost models on CDM-only, nursing-only, and combined feature sets.
• Assessed performance using AUROC, sensitivity, specificity, and SHAP analysis for interpretability.

Main Results:

• Integrated models (CDM + nursing statements) consistently outperformed single-source models in fall risk prediction.
• Nursing-only models achieved performance comparable to CDM-only models.
• XGBoost demonstrated the highest predictive discrimination.
• SHAP analysis identified nursing indicators like 'Reports weakness' as highly influential predictors.

Conclusions:

• Integrating nursing documentation with structured clinical data significantly enhances inpatient fall risk prediction accuracy and interpretability.
• Nursing notes capture critical safety information often missed in structured EHR data.
• Leveraging nursing documentation is vital for developing transparent, AI-driven clinical decision support systems for fall prevention.