Phenomapping-derived tool to individualize the effect of sacubitril-valsartan in heart failure with preserved ejection fraction

Sacubitril/valsartan may exert heterogeneous treatment effect in heart failure with preserved ejection fraction (HFpEF). We aimed to develop a machine learning (ML)-based model to individualize the cardiovascular benefits of sacubitril/valsartan. Using data from 4161 patients in the PARAGON-HF trial, we constructed a phenomap based on 53 baseline characteristics, utilizing the Gower distance metric. A semiparametric proportional rates method was applied within each patient’s 20% phenotypic neighborhood to calculate individualized rate ratios (RRs) for treatment effects of sacubitril/valsartan versus valsartan on the composite outcome of cardiovascular death or total hospitalizations for heart failure. This metric, the phenotype-specific response score (PRS), quantifies treatment benefit, with negative PRS values indicating greater benefit. The median PRS was −0.62 (interquartile range: −1.21 to 0.05), without significant overall treatment benefit observed in the whole population (RR, 0.90; 95% CI, 0.77–1.06). However, 70.8% of patients had a PRS ≤ 0, showing an 18% risk reduction with sacubitril/valsartan (RR, 0.82; 95% CI, 0.69–0.98). Notably, 57.6% of men and 70.2% of patients with LVEF > 57% had a PRS ≤ 0, suggesting benefit in subgroups typically considered less responsive. To aid clinical use, we developed the PARAGUIDE precision tool using 16 readily available variables. The model was internally validated via repeated 4-fold cross-validation and externally validated in patients from the PARADIGM-HF trial. We developed an ML-driven model to predict the individualized treatment effects of sacubitril/valsartan compared with valsartan in patients with HFpEF. By identifying potential responders, the PARAGUIDE precision tool advances precision medicine beyond traditional subgroup analyses.