High four-pillar guideline-directed medical therapy adoption in heart failure with reduced ejection fraction: A tertiary-centre registry from eastern India, 2019–2024

BACKGROUND: Whether dedicated tertiary heart-failure programmes in low- and middle-income countries can approach guideline-directed medical therapy (GDMT) targets, and whether temporal trends reflect prescribing optimisation or case-mix shift, remains unresolved. We characterise four-pillar GDMT adoption in an eastern Indian tertiary centre with systematic robustness testing. METHODS: Repeated cross-sectional analysis in a prospective HFrEF registry (2022-2024 primary cohort, n = 319 patients). Half-yearly adoption assessed by Cochran-Armitage trend tests (Benjamini-Hochberg FDR), MI-adjusted logistic regression (m = 20), and Bayesian regression. Robustness: delta-adjusted MI, tipping-point, FMI, sacubitril-valsartan target-dose, E-value, time-by-diabetes interaction, quarterly ITS at ≥ 8 bins. RESULTS: Four-pillar composite rose from 48.6% (H12022) to 76.3% (H2 2024); MI-adjusted OR per half-year 1.24 (95% CI 1.05-1.47; p = 0.013). SGLT2 inhibitor drove the composite (OR 1.51; 1.22-1.86; p < 0.001); RAAS-i 1.24, MRA 1.11 and beta-blocker 0.98 not significant. Bayesian regression concurred (SGLT2i posterior probability 0.972; four-pillar 0.896). SGLT2i trend was robust to delta-adjusted MI, tipping-point beyond +/- 50 mL/min/1.73 m^2, and unmeasured confounding (E-value point 1.76; CI lower 1.44). Time-by-diabetes interaction was null; sacubitril-valsartan target-dose attainment 20.0%. Quarterly ITS at ≥ 8 bins returned non-significant level and slope changes at every feasible policy event. CONCLUSIONS: Tertiary specialist care approached but did not reach the four-pillar prescription ceiling (76.3%; residual 23.7-percentage-point gap). The SGLT2i trend was robust to MNAR, unmeasured confounding, and diabetes case-mix shift; quarterly ITS at ≥ 8 bins detected no policy-event association. Sacubitril-valsartan target-dose attainment (20.0%), single-centre design, and fragile non-SGLT2i E-values constrain generalisability.