External validation of the HALP score for predicting in-hospital and long-term outcomes in acute heart failure: a retrospective cohort study

Abstract

The hemoglobin–albumin–lymphocyte–platelet (HALP) score integrates nutritional, inflammatory, and hematologic status and has been proposed as a prognostic marker in acute heart failure (AHF), but external validation is limited. We retrospectively analyzed 197 consecutive patients hospitalized with AHF at a tertiary center. HALP was calculated at admission, and the pre-specified outcomes were in-hospital mortality, prolonged hospital stay (>7 days), all-cause mortality over follow-up, and heart-failure rehospitalization. Discrimination was assessed by the area under the receiver-operating-characteristic curve (AUC) with DeLong 95% confidence intervals (CIs), and associations were tested with crude and age/sex-adjusted logistic regression (in-hospital outcomes) and Cox regression (mortality over follow-up), using both data-derived (Youden) and pre-specified HALP thresholds. The cohort had a median age of 71 years (30.5% female) and a median follow-up of 15.5 months; in-hospital mortality was 12.7%, prolonged stay 19.4%, and all-cause mortality 31.0%. HALP did not discriminate any outcome (AUC 0.53 [95% CI 0.41–0.65] for in-hospital mortality, 0.47 [0.36–0.59] for prolonged stay, and 0.57 [0.48–0.66] for all-cause mortality; all CIs included 0.50). A Youden-optimized threshold appeared associated with mortality in logistic regression (adjusted odds ratio 2.02, 95% CI 1.02–4.00), but the signal vanished under pre-specified thresholds (median-split adjusted odds ratio 1.23, 0.67–2.28) and in time-to-event analysis (adjusted hazard ratio per 1-SD decrease 1.04, 0.78-1.38; Youden threshold 1.75, 0.97-3.16; log-rank across quartiles p = 0.15). In the same patients, the established EHMRG and MAGGIC scores discriminated mortality (AUC 0.71 and 0.65; CIs excluding 0.50), whereas HALP correlated only with severity markers (NYHA class rho = −0.31; NT-proBNP rho = −0.16; CRP rho = −0.17). In this independent cohort, HALP reflected disease severity but did not predict outcomes once cutoff-optimization bias was removed, and should not be used as a stand-alone risk-stratification tool in AHF.