Evaluating Natural Language Processing Packages for Predicting Hospital-Acquired Pressure Injuries From Clinical...

Area of Science:

• Medical Informatics
• Clinical Nursing
• Natural Language Processing

Background:

• Hospital-acquired pressure injury (HAPI) is a significant indicator of nursing quality, linked to adverse outcomes, extended hospital stays, and increased healthcare costs.
• Current prediction models primarily rely on structured patient data, often overlooking valuable insights within unstructured clinical notes.
• Accurate prediction of HAPI is crucial for improving patient care and reducing the economic burden on healthcare systems.

Purpose of the Study:

• To investigate the efficacy of using unstructured nursing notes for predicting hospital-acquired pressure injury.
• To evaluate the performance of different natural language processing (NLP) techniques, specifically named entity recognition (NER), in extracting predictive features from clinical text.
• To compare the impact of using all clinical notes versus only nursing notes on prediction model accuracy.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care III (MIMIC-III) database, a publicly available ICU dataset.
• Employed Named Entity Recognition (NER) using scispaCy and Stanza NLP packages to identify salient keywords from unstructured clinical notes.
• Developed and compared HAPI prediction models based on extracted keywords, assessing the influence of vocabulary size by comparing all notes versus nursing notes.

Main Results:

• Named entity recognition extraction from nursing notes demonstrated the potential to create accurate HAPI prediction models.
• Extracted keywords from unstructured nursing notes were found to be significant predictors of hospital-acquired pressure injury.
• The study confirmed that nursing notes alone contain sufficient information for effective HAPI prediction.

Conclusions:

• Unstructured nursing notes are a valuable data source for predicting hospital-acquired pressure injury.
• NLP techniques, particularly NER, can effectively identify key predictive features within nursing notes.
• Leveraging nursing notes in prediction models offers a promising approach to enhance patient safety and reduce HAPI incidence.