Extraction of Geriatric Syndromes From Electronic Health Record Clinical Notes: Assessment of Statistical Natural...

Area of Science:

• Gerontology and Health Informatics
• Computational Linguistics in Healthcare

Background:

• Geriatric syndromes are common in older adults and linked to adverse health outcomes.
• These syndromes are often under-documented in structured electronic health record (EHR) data.
• Clinical notes within EHRs contain valuable, yet unstructured, information about geriatric syndromes.

Purpose of the Study:

• To develop and assess statistical natural language processing (NLP) techniques for automatically identifying geriatric syndromes from free-text clinical notes.
• To determine the effectiveness of different NLP feature sets in detecting these syndromes.

Main Methods:

• Conditional random fields (CRFs), a machine learning algorithm, were applied to identify 10 geriatric syndrome constructs.
• Three feature sets were evaluated: base, enhanced token, and contextual features.
• The CRF model was trained on 3901 annotated notes, tuned on 50 patients, and evaluated on 50 held-out test patients from a Medicare Advantage population.

Main Results:

• The final CRF model achieved good patient-level performance (macroaverage F1=0.834, microaverage F1=0.851).
• Performance varied by syndrome construct, with high accuracy for absence of fecal control (F1=0.857) and vision impairment (F1=0.798).
• Lower accuracy was observed for malnutrition (F1=0.155), weight loss (F1=0.394), and severe urinary control issues (F1=0.532), often due to out-of-vocabulary words and lack of context.

Conclusions:

• Statistical NLP methods can successfully identify geriatric syndromes from unstructured EHR clinical notes.
• This automated approach offers new avenues for identifying at-risk patients and investigating their health outcomes.
• Further refinement of NLP models is needed to address challenges like context and novel terminology.