Patient-Level Fall Risk Prediction Using the Observational Medical Outcomes Partnership's Common Data Model: Pilot...

Area of Science:

• Healthcare Informatics
• Clinical Data Standardization
• Predictive Modeling in Healthcare

Background:

• Patient falls in acute care settings pose significant safety risks.
• Existing fall risk prediction models face limitations due to small sample sizes, restricted variables, and lack of standardized methods.
• Developing robust, replicable fall prevention strategies requires overcoming these data and methodological challenges.

Purpose of the Study:

• To standardize fall-related electronic health record (EHR) data into the Observational Medical Outcome Partnership (OMOP) common data model.
• To develop and evaluate predictive models for patient fall risk within two distinct hospital timeframes.

Main Methods:

• A pilot study converted EHR data (nursing notes, assessments) into the OMOP common data model using ETL processes.
• Two machine learning algorithms, LASSO logistic regression and random forest, were employed to build fall risk prediction models.
• Models were developed for predicting falls within 7 days of admission and across the entire hospital stay.

Main Results:

• Extensive EHR data, including nursing statements and patient acuity scores, were successfully transformed into the OMOP common data model format.
• The developed fall risk prediction models demonstrated superior performance compared to the Hendrich II Fall Risk Model (AUC 0.692-0.726).
• Key predictors identified included patient acuity score, fall history, advanced age (≥60 years), movement disorders, and central nervous system medications.

Conclusions:

• The study successfully demonstrated the feasibility of using standardized EHR data in the OMOP common data model for fall risk prediction.
• Ongoing efforts to incorporate all nursing records into the OMOP format are expected to further enhance model performance.
• Future work will focus on leveraging comprehensive nursing data and external validation to improve the accuracy and generalizability of fall risk prediction models.