Predicting future falls in older people using natural language processing of general practitioners' clinical notes

Area of Science:

• Gerontology
• Medical Informatics
• Machine Learning

Background:

• Falls are a significant health concern for older adults, leading to morbidity.
• Electronic health records (EHR) offer potential for automated fall risk prediction.
• Current prediction models often overlook valuable information within unstructured clinical notes.

Purpose of the Study:

• To evaluate the predictive performance of unstructured clinical notes for fall risk.
• To assess the added value of unstructured data over structured EHR data in fall prediction models.

Main Methods:

• Utilized primary care EHR data from individuals aged 65 and over.
• Developed three logistic regression models: structured data only, unstructured data topics only, and a combination.
• Employed least absolute shrinkage and selection operator (LASSO) and 10-fold cross-validation.

Main Results:

• Analyzed data from 35,357 individuals, with 4,734 experiencing falls.
• Natural language processing (NLP) identified 151 topics from unstructured notes.
• The combined model achieved the highest predictive performance (AUC: 0.718), outperforming models using only structured (AUC: 0.709) or unstructured data (AUC: 0.685).

Conclusions:

• Unstructured clinical notes represent a valuable data source for enhancing fall prediction models.
• Integrating unstructured data with structured EHR data improves prediction accuracy.
• Further research is needed to fully realize the clinical utility of these enhanced models.