Predictive Modeling for Pressure Ulcers from Intensive Care Unit Electronic Health Records.

Area of Science:

• Medical Informatics
• Clinical Research
• Patient Safety

Background:

• Pressure Ulcers (PUs) are a significant concern in Intensive Care Units (ICUs).
• The Braden scale, a common assessment tool, has limitations in ICU settings, including missing critical risk factors and generating false positives.
• Electronic Health Records (EHRs) offer a rich source of data for improving PU risk prediction.

Purpose of the Study:

• To identify novel risk factors for Pressure Ulcers (PUs) in Intensive Care Unit (ICU) patients.
• To develop and evaluate predictive models for PU incidence that outperform existing methods.
• To leverage patient EHR data, including medications and diagnoses, to enhance PU risk assessment.

Main Methods:

• Extracted medication and diagnosis features from patient EHRs.
• Evaluated six types of predictive models using Braden scale, medication, and diagnosis features.
• Assessed model performance using Area Under the Curve (AUC) and other relevant metrics.

Main Results:

• Diagnosis features significantly improved the predictive power of PU incidence models.
• The best predictive models combined Braden scale features with diagnosis features.
• Top diagnosis features enhanced model AUC by 10% compared to the Braden scale alone.

Conclusions:

• Diagnosis features extracted from EHRs are crucial for accurate Pressure Ulcer risk prediction in ICUs.
• Integrating diagnosis data with the Braden scale offers a superior approach to identifying patients at high risk for PUs.
• This research provides a foundation for developing more effective clinical decision support tools for PU prevention.