Using electronic health record data to develop and validate a prediction model for adverse outcomes in the wards*

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Patient Safety

Background:

• In-hospital cardiac arrests are common and often preventable.
• Current risk scores for ward patients have limited accuracy, missing opportunities for early intervention.
• Inefficient resource utilization occurs due to inaccurate risk stratification.

Purpose of the Study:

• To derive and validate a prediction model for cardiac arrest in ward patients.
• To account for intensive care unit (ICU) transfer as a competing risk.
• To utilize electronic health record (EHR) data for improved prediction accuracy.

Main Methods:

• Retrospective cohort study at an academic medical center.
• Inclusion of adult patients with documented ward vital signs (November 2008 - August 2011).
• Development of a multinomial logistic regression model using EHR data (vital signs, demographics, lab results).

Main Results:

• The derived model demonstrated superior accuracy in predicting cardiac arrest (AUC 0.88 vs. 0.78) and ICU transfer (AUC 0.77 vs. 0.73) compared to the VitalPAC Early Warning Score.
• At 93% specificity, the model achieved higher sensitivity for cardiac arrest detection (65% vs. 41%).
• Model performance was validated using three-fold cross-validation.

Conclusions:

• A validated prediction tool was developed for ward patients, capable of simultaneously predicting cardiac arrest and ICU transfer risk.
• The new model significantly outperformed the VitalPAC Early Warning Score in accuracy.
• Implementation in EHR systems can provide real-time alerts for patient deterioration, enhancing caregiver awareness.