Comparing the predictive accuracy of frailty instruments applied to preoperative electronic health data for adults undergoing noncardiac surgery

Background — Preoperative frailty is associated with increased risk of postoperative mortality and complications. Routine preoperative frailty assessment is underperformed. Automation of preoperative frailty assessment using electronic health data could improve adherence to guideline-based care if an accurate instrument is identified.

Methods — We conducted a retrospective cohort study of adults >65 yr undergoing elective noncardiac surgery between 2012 and 2018. Four frailty instruments were compared: Frailty Index, Hospital Frailty Risk Score, Risk Analysis Index-Administrative, and Adjusted Clinical Groups frailty-defining diagnoses indicator. We compared the predictive performance of each instrument added to a baseline model (age, sex, ASA physical status, and procedural risk) using discrimination, calibration, explained variance, net reclassification, and Brier score (binary outcomes); and explained variance, root mean squared error, and mean absolute prediction error (continuous outcomes). Primary outcome was 30-day mortality. Secondary outcomes included 365-day mortality, length of stay, non-home discharge, days alive at home, and 365-day costs.

Results — For this study, 171 576 patients met the inclusion criteria; 1370 (0.8%) died within 30 days. Compared with the baseline model predicting 30-day mortality (area under the curve [AUC] 0.85; R² 0.08), the addition of Hospital Frailty Risk Score led to the greatest improvement in discrimination (AUC 0.87), explained variance (R² 0.09), and net reclassification (Net Reclassification Index 0.65). Brier and calibration scores were comparable.

Conclusions — All four frailty instruments significantly improved discrimination and risk reclassification when added to typically assessed preoperative risk factors. Accurate identification of the presence or absence of preoperative frailty using electronic frailty instruments may improve perioperative risk stratification. Future research should evaluate the impact of automated frailty assessment in guiding surgical planning and patient-centred optimisation amongst older surgical patients.