Background Single-centre studies have shown that a low serum albumin at baseline forecasts enhanced mortality in chronic heart failure (CHF) possibly because it reflects aberrations (eg, inflammation, impaired nutrition, plasma volume expansion) that can exacerbate disease. We hypothesised that attenuations in serum albumin over time would be prognostically more ominous than baseline values, and would be so even in a multicentre setting.

Methods We analysed the survival implications of baseline albumin and ∆albumin in a derivation cohort of 246 CHF outpatients (mean [±SD] age 68±12 years, LVEF 29±8%, 48% NYHA class >2) from University College London Hospital and then in a validation cohort of 148 CHF outpatients (age 69±12 years, LVEF 28±10%, 41% NYHA class >2) from Imperial Healthcare (St Marys Hospital and Hammersmith Hospital, London).

Results In the derivation cohort, 51 (21%) patients died over 13 months. Baseline albumin independently predicted mortality (HR 0.89, 95% CI 0.84 to 0.94, χ²:18, p<0.0001). However, ∆albumin (unadjusted HR 0.89, 95% CI 0.84 to 0.92, χ²:53, p<0.0001) was even more predictive (Difference in ROC AUC for baseline vs ∆albumin 0.16, p<0.001) and did so independently of all covariates including baseline albumin. A reduction in albumin > 6 g/l optimally predicted death (ROC AUC 0.82, p<0.0001) and conferred a sixfold escalated risk of mortality (HR 6.42, 95% CI 3.67 to 11.22, p<0.0001). In incremental prognostic analyses, the addition of ∆albumin to the strongest four variable model (baseline albumin, NYHA class, ∆urea, ∆haemoglobin) dramatically augmented the χ² value (43 vs 84, p<0.0001). In the validation cohort, 43 (30%) patients died. ∆albumin (unadjusted HR 0.89, 95% CI 0.86 to 0.92, χ²: 44, p<0.0001) was again prognostically superior to baseline albumin with a fall >6 g/l predicting an ∼sixfold increased risk (HR 5.64, 95% CI 3.08 to 10.31, χ²: 35, p<0.0001). Addition of ∆albumin to the strongest three variable model (baseline red cell distribution width, ∆red cell distribution width, ∆urea) also augmented the χ² value (51 vs 65, p<0.001).

Conclusions A fall in serum albumin over time consistently predicts an amplified risk of death in systolic CHF and enables simple and cheap risk stratification.