4940Predicting risk at the point of care: NT-proBNP improves performance of the LACE index among patients hospitalized for Heart Failure (HF)

Abstract

Background: The Length of stay, Acuity, Comorbidities, and Emergency department visits (LACE) index can predict 30‐d readmission and composite readmission/ death in patients hospitalized for HF, but only with modest risk discrimination. Purpose: To assess whether admission or discharge NT‐pro‐BNP can improve performance of the LACE risk prediction model at the point of care (POC) among patients hospitalized for HF. Methods: This is a sub‐study of the Patient‐Centered Care Transitions in HF (PACT‐HF) multi‐center stepped wedge cluster randomized trial, which offered transitional care services to patients hospitalized for HF. We measured LACE and admission + discharge NT‐proBNP at the POC. We obtained 30‐d outcomes using linkages to administrative databases. We used log‐binomial regression models with 30‐d all‐cause readmission or 30‐d composite all‐cause readmission/death as the outcome. We measured risk ratios (RR) with 95% confidence intervals (CI); model discrimination (C‐statistic); and model calibration (Hosmer‐Lemeshow test). We adjusted models for transitional care services, and performed internal validation using bootstrapping. Results: Of the 772 patients included in the analysis, 49.9% were female. Mean age was 77.0+12.4 years, and mean LVEF was 48.0+14.5%. Overall, 20.3% of patients were readmitted and 21.1% were either readmitted or died within 30 days. LACE predicted 30‐d readmission (RR 1.15 [95% CI 1.09‐1.22]/unit increase) and 30‐d readmission/death (RR 1.16 [95% CI 1.09‐1.22]/unit increase) with modest discrimination. Calibration of the LACE model for 30‐d readmission and 30‐d readmission/death improved with both admission and discharge NTproBNP. Risk discrimination of the LACE model for 30‐d readmission (C‐statistic 0.618 [95% CI 0.567‐0.670]) slightly improved with admission NT‐proBNP (Cstatistic 0.627 [95% CI 0.576‐0.679]) and discharge NT‐proBNP (C‐statistic 0.628 [95% CI 0.577‐0.678]). Discrimination of the LACE model for 30‐d readmission/ death (C‐statistic 0.617 [95% CI 0.566‐0.668]) improved with admission NTproBNP (C statistic 0.630 [95% CI 0.579‐0.681]) and discharge NT‐proBNP (Cstatistic 0.629 [95% CI 0.580‐0.679]). In absence of LACE, both admission and discharge NT‐proBNP had suboptimal risk discrimination for the 30‐d outcomes (C‐statistic <0.590). Adjusting for LACE: each 10‐fold increase in admission NTproBNP increased the risk of 30‐d readmission (RR 1.42 [95% CI 1.05‐1.92]) and 30‐d readmission/death (RR 1.47 [95% CI 1.09‐1.97]); and each 10‐fold increase in discharge NT‐proBNP increased the risk of 30‐d readmission (RR 1.52 [95% CI 1.05‐2.20]) and 30‐d readmission/death (RR 1.52 [95% CI 1.06‐2.19]). Conclusion: Among patients hospitalized with HF, admission and discharge NTpro‐ BNP improve calibration and discrimination of the LACE model in predicting 30‐d outcomes. NT‐proBNP cannot be recommended preferentially over LACE due to suboptimal risk discrimination for 30‐d outcomes.