Developing an Inpatient Electronic Medical Record Phenotype for Hospital-Acquired Pressure Injuries: Case Study Using...

Area of Science:

• Medical informatics
• Clinical informatics
• Health informatics

Background:

• Hospital-acquired pressure injuries (HAPI) surveillance using administrative data (ICD codes) is often delayed and undercoded.
• Electronic medical records (EMRs) offer a potential solution for more accurate and timely HAPI identification.
• Natural language processing (NLP) can extract valuable information from free-text clinical notes within EMRs.

Purpose of the Study:

• To demonstrate that EMR-based phenotyping algorithms, utilizing NLP, outperform traditional ICD-10-CA codes for HAPI detection.
• To validate the accuracy of clinical logs recorded via NLP in nursing notes for HAPI identification.
• To enhance the precision and timeliness of HAPI surveillance in acute care settings.

Main Methods:

• HAPI cases were identified from nursing notes in EMRs during a 2015-2018 clinical trial in Calgary, Alberta.
• Text classification models (Random Forest, XGBoost, deep learning) were developed using sequential forward selection of clinical notes.
• Model performance was evaluated using sensitivity, specificity, positive predictive value, negative predictive value, and F1-score, with thresholds tuned for specificity.

Main Results:

• The study analyzed data from 280 patients, identifying 97 with HAPIs.
• The Random Forest model achieved a sensitivity of 0.464 and specificity of 0.984, with an F1-score of 0.612.
• Machine learning models demonstrated higher sensitivity than ICD-based algorithms without significant loss of specificity.

Conclusions:

• EMR-based NLP phenotyping algorithms significantly improve HAPI detection compared to ICD-10-CA codes alone.
• Daily nursing notes in EMRs are a rich data source for machine learning models to detect adverse events accurately.
• This approach enhances automated healthcare quality and safety surveillance systems.