Across four major HFpEF randomized trials (TOPCAT, RELAX, NEAT-HFpEF, INDIE-HFpEF), an interaction-based individual treatment effect (ITE) framework identified four distinct drug-specific responder phenotypes: cardiorenal-inflammatory for spironolactone, NO-mediated anti-inflammatory for isosorbide mononitrate, afterload-reducing for inorganic nitrite, and anti-volume-overload for sildenafil. Critically, conventional responder analyses — those based on minimal clinically important differences — were shown to capture favorable prognosis common to both treated and placebo patients, not genuine drug benefit.
HFpEF has frustrated cardiologists for decades: it kills at rates comparable to heart failure with reduced ejection fraction, yet trial after trial returns neutral average results. This analysis reframes that failure not as drug inefficacy but as phenotypic dilution — the wrong patients averaged in with the right ones. The ITE approach, drawing on treatment-by-variable interaction modeling, offers a statistically grounded path to precision cardiology that standard subgroup analyses cannot. Each model's significance held only within its own trial, a meaningful cross-validation signal suggesting genuine mechanism specificity rather than overfitting.
Limitations are substantial: all four trials were originally powered for average effects, not subgroup detection; sample sizes within phenotype strata are small; and the framework remains exploratory. As a preprint posted on medRxiv and not yet peer-reviewed, these findings require independent replication before influencing clinical practice. Still, for a field starved of therapeutic wins, this methodology — if validated prospectively — could represent a genuine inflection point in how HFpEF trials are designed and patients are matched to therapy.