Abstract
BACKGROUND: Laboratory-based frailty indices (FI-Labs) offer potential adjuncts and alternatives to clinical assessments. Still, their optimal configuration and construct validity compared with nurse-assessed Clinical Frailty Scale (CFS) scores remain unclear. METHODS: In this retrospective cohort study, we evaluated five FI-Lab configurations against nurse-assessed CFS scores using data from 74 493 emergency department visits. We examined their association with clinical outcomes and assessed measurement reliability using mixed effects models. RESULTS: While nurse assessments demonstrated superior outcome discrimination (c-statistic 0.726 for 90-day mortality versus 0.718 for best FI-Lab), automated FI-Lab measures showed significantly greater between-visit reliability [intraclass correlation coefficient (ICC) = 0.51-0.76 versus 0.37 for nurse CFS]. The drug-adjusted FI-Lab demonstrated highest reliability (ICC = 0.76) but weaker age associations (β = 0.002, P = .08) compared to other configurations (β = 0.006-0.013, P < .001). In complex models adjusting for illness severity, nurse CFS scores showed stronger mortality associations (HR 1.55, 95% CI 1.45-1.66 per standard deviation) compared to FI-Lab configurations (HR range 1.19-1.29). Notably, all frailty measures showed effect sizes comparable to age (HR range 1.37-1.55 per SD). CONCLUSIONS: Automated FI-Lab measures offer a reliability advantage over nurse-assessed CFS scores despite slightly lower predictive validity for mortality. Their comparable effect sizes to age suggest these automated measures capture clinically meaningful patient characteristics. This trade-off between reliability and predictive validity suggests that integrated approaches combining automated screening with targeted clinical assessment may provide optimal frailty identification in emergency settings.