External validation of an AI-based preoperative frailty index using real-world data

Abstract

Background: Preoperative frailty assessment is crucial for surgical risk stratification in older adults. Traditional frailty measurements are often too time-consuming and resource-intensive in preoperative settings. This study aimed to externally validate an artificial intelligence (AI)-based frailty index developed using electronic health records (EHR).

Methods: We externally validated an AI-based frailty index, previously developed by our team, on a cohort of 1 52 364 surgical patients aged 65+ years from the OneFlorida+ Clinical Research Consortium. We examined the association between the predicted frailty and three postoperative outcomes: 30-day mortality, length of hospital stay, and discharge disposition. We also compared the predictive performance of general and service-specific frailty indices (the latter developed using data from patients undergoing specific surgeries) in predicting postoperative outcomes.

Results: The AI-based frailty index demonstrated a strong and stepwise association with adverse postoperative outcomes. Patients in the highest frailty level (top 20%) had significantly higher odds of 30-day mortality (OR 4.33, 95% CI 3.91-4.80), longer hospital stays (2.53 times longer, 95% CI 2.47-2.60), and a higher likelihood of unfavorable discharge dispositions compared to the lowest frailty level, after adjusting for demographics and comorbidities. The general frailty index performed comparably to or slightly better than service-specific indices across surgical specialties.

Conclusion: The developed preoperative frailty index effectively predicts postoperative outcomes in a large and diverse external cohort. The index's efficiency and predictive performance in stratifying surgical risk can potentially improve surgical care and outcomes.

Keywords: Artificial intelligence (AI); Frailty index; Preoperative frailty assessment; Surgical risk stratification.