Dual-axis utility mapping (HAS-BLED x CHA2DS2-VA) after acute heart failure with atrial fibrillation: late bleeding hazard and a calibration-checked decision grid

Background Anticoagulation after acute heart failure (AHF) with atrial fibrillation (AF) is difficult because stroke, major bleeding and death compete over time. We tested whether a simple two-axis utility map combining HAS-BLED and CHA2DS2-VA can turn risk scores into actionable bedside decisions. Objectives (i) Compare discrimination of HAS-BLED, ORBIT and ATRIA for major bleeding; (ii) test time-heterogeneity of bleeding hazard; and (iii) build a HAS-BLED × CHA2DS2-VA grid with competing-risk net utility for oral anticoagulation (OAC). Methods We analysed a multicentre cohort of 716 AF with AHF admissions (2014–2020). First events were modelled under a competing-risk framework (stroke, bleeding, death). Fine–Gray models estimated hazard ratios per 1-SD in early (≤90 days) versus late windows with 90/180/365-day landmarks. Discrimination at 1 and 3 years used FGR-predicted risks with a binary ROC fallback; pairwise AUC differences used DeLong. Calibration used decile-level cumulative incidence (CIF), plotting observed versus predicted CIF with the 45° reference. For the decision grid, each score was split at the cohort 60th percentile (top 40%; ties counted as High), yielding cut-offs HAS-BLED ≥3 and CHA2DS2-VA ≥4. Net-benefit (NB) was 2×CIF_stroke − CIF_bleed − CIF_death (weights varied in sensitivity). Prespecified subgroups (renal impairment, anaemia, antiplatelet/NSAIDs) were examined for consistency. Results Median age 81 years [73–87]; 49% women. OAC at discharge 82%; among OAC, 64% DOAC. Median follow-up 512 days [IQR 82–969]. First-event totals: bleeding 85, stroke 38, death 140 (1y/3y: bleeding 59/78, stroke 23/30, death 78/122). Time-heterogeneity: HAS-BLED strengthened beyond 180 days (HR/SD early vs late ≈ 1.22 vs 1.77, p_hetero=0.036). Discrimination (1 year): AUC 0.587 (HAS-BLED), 0.655 (ORBIT), 0.659 (ATRIA); both were borderline-better than HAS-BLED (DeLong p≈0.06–0.07); differences attenuated at 3 years. Calibration: acceptable, monotonic alignment at 1 and 3 years. Utility map (Top-40%, 1 year, weights 2:1:1): quadrants ranked (higher=less unfavourable) Low-stroke/Low-bleed −0.079 (n=204) > Low-stroke/High-bleed −0.131 (n=18) > High-stroke/High-bleed −0.192 (n=285) > High-stroke/Low-bleed −0.193 (n=209); the ordering held across 30–50% thresholds and alternative weights. Subgroup checks preserved the same ordering. Conclusions In AF with AHF, bleeding risk strengthens after ~180 days, supporting intensified bleed-mitigation in the late phase. A HAS-BLED × CHA2DS2-VA utility map, calibration-checked and framed under competing risks, translates scores into decisions: flagging (i) Low-stroke/High-bleed candidates for OAC de-escalation and possible left atrial appendage occlusion, and (ii) High-stroke strata in whom stroke risk remains dominant so OAC is generally favoured unless bleeding risk is prohibitive, with targeted bleed-risk reduction mandatory.