Deep Natural Language Processing Identifies Variation in Care Preference Documentation.

Area of Science:

• Geriatric Medicine
• Critical Care Medicine
• Health Informatics

Background:

• Documentation of care preferences within 48 hours of intensive care unit (ICU) admission is a National Quality Forum-endorsed metric for older adults.
• Administrative data poorly captures patient care preferences, necessitating alternative methods for documentation.

Purpose of the Study:

• To determine the rate of care preference documentation in free-text clinical notes using deep natural language processing (NLP).
• To assess patient-specific factors associated with care preference documentation in the ICU.

Main Methods:

• A retrospective review of 11,575 clinician notes from 1350 ICU admissions for adults aged 75 years and older.
• Deep NLP was employed to identify documentation of care preferences, including goals-of-care and treatment limitations, within 48 hours of ICU admission across five ICU types.
• Covariates included patient demographics, ICU type, Sequential Organ Failure Assessment (SOFA) score, and mechanical ventilation status.

Main Results:

• Overall, 64.7% of ICU admissions had documented care preferences.
• Patients with documentation were significantly older (median 85 vs. 83 years) and more often female (53.8% vs. 43.4%).
• Adjusted analysis revealed higher documentation rates in older patients, females, nonelective admissions, and medical ICUs compared to cardiac or surgical ICUs.

Conclusions:

• More than one-third of ICU admissions for patients aged 75 years and older lacked timely care preference documentation.
• Documentation rates varied significantly by ICU type, with medical ICUs showing higher rates than cardiac or surgical ICUs.