ABSTRACT
PURPOSE: There is a considerable need to measure frailty retrospectively in clinical practice and research. We assessed agreement, accuracy, precision, and reliability of retrospectively compared with prospectively, and separately attained, retrospective Clinical Frailty Scale (CFS) scoring, respectively. METHODS: We studied 110 hospitalized patients aged ≥ 80 years, consisting of 70.9% females. Agreement was assessed by bias, accuracy by root mean square error (RMSE) and the percentages of one CFS scores within 20% and 30% of each other (P20 and P30), precision by interquartile ranges of the bias, and reliability by weighted Cohen's kappa (κ). RESULTS: Comparison of retrospective and prospective CFS scores demonstrated a modest bias of 0.26. Classification as robust, prefrail, or frail was generally correct in retrospectively compared to prospectively CFS scores. RMSE was low (0.28), while P20 and P30 values were high (90.0% and 96.6%, respectively). Precision of retrospective to predict prospective CFS scores was high with narrow interquartile ranges of 0 and 1. Reliability of retrospective with prospective CFS scores was high (κ = 0.89). Comparing two separately attained retrospective CFS scores demonstrated low bias (0.05) and RMSE (0.24), and a high P30 value (90.0%). Precision and interrater reliability of the comparison of retrospective CFS scores were high with narrow interquartile ranges and κ = 0.85. CONCLUSION: The retrospectively attained CFS score is a valid diagnostic instrument to measure frailty in older hospitalized patients with high agreement, accuracy, precision, and reliability compared to both prospective and separately attained CFS scores.
