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

Background

Social isolation is increasingly prevalent among older adults and is associated with adverse health outcomes, yet its impact in surgical populations remains poorly defined. Lack of systematic measurement during surgical care has limited large-scale investigation of social isolation and its associations with postoperative outcomes.

Methods

We conducted a prospective validation study of adults ≥ 65 years old admitted to inpatient surgical services at an academic medical center to evaluate the accuracy of a semi-automated, rule-based natural language processing (NLP) algorithm for identifying social isolation from routine electronic health record (EHR) clinical notes from index hospitalizations. NLP performance was validated against a 6-item Social Isolation Index administered in-person during hospitalization. The primary outcome was sensitivity of the NLP algorithm for identifying social isolation.

Results

Among 249 enrolled patients (median age, 74 years; 55% female), 55 (22.1%) were classified as socially isolated by the patient-reported reference standard, and the NLP algorithm identified social isolation in 47 patients (18.9%). Sensitivity was 0.84 (95% CI, 0.71-0.92); specificity was 0.99 (95% CI, 0.97-1.00); positive predictive value was 0.98 (95% CI, 0.89-1.00); negative predictive value was 0.96 (95% CI, 0.92-0.98); accuracy was 0.96 (95% CI, 0.93-0.98); F1 score was 0.90 (95% CI, 0.84-0.96).

Conclusions

In this prospective validation study of older surgical patients, a semi-automated, rule-based NLP algorithm demonstrated strong performance in identifying social isolation using routine EHR documentation. With further automation and external validation, EHR-based identification of social isolation could complement embedded screening tools in perioperative care strategies.