Development and Validation of a Natural Language Processing Algorithm to Identify Social Isolation in Older Surgical...

Area of Science:

• Geriatric Medicine
• Health Informatics
• Surgical Care

Background:

• Social isolation is a growing concern for older adults, linked to negative health outcomes.
• Its prevalence and impact within surgical patient populations are not well understood.
• Limited systematic measurement in surgical settings hinders research on social isolation.

Purpose of the Study:

• To evaluate a natural language processing (NLP) algorithm for identifying social isolation in older surgical patients.
• To validate the NLP algorithm's accuracy against a standard Social Isolation Index.
• To assess the feasibility of using routine electronic health record (EHR) data for social isolation screening.

Main Methods:

• A prospective validation study involved 249 adults aged 65+ admitted to inpatient surgical services.
• A semi-automated, rule-based NLP algorithm analyzed clinical notes for social isolation indicators.
• Algorithm performance was compared to an in-person administered 6-item Social Isolation Index.

Main Results:

• The NLP algorithm identified social isolation in 18.9% of patients, compared to 22.1% by the reference standard.
• Key performance metrics included high sensitivity (0.84), specificity (0.99), and accuracy (0.96).
• The algorithm demonstrated strong positive predictive value (0.98) and negative predictive value (0.96).

Conclusions:

• A semi-automated NLP algorithm effectively identifies social isolation in older surgical patients via EHR documentation.
• This EHR-based approach shows promise for complementing existing screening tools in perioperative care.
• Further automation and external validation could enhance its utility in clinical practice.