Risk-based versus benefit-based targeting of early physiotherapy in older patients hospitalized for heart failure: a target trial emulation using causal machine learning in the J-Proof HF registry

Physiotherapy for older patients hospitalized for heart failure is usually prioritized by prognostic risk, yet the patients at highest risk may not be those who benefit most. This preregistration describes a secondary analysis of the J-Proof HF registry of the Japanese Society of Cardiovascular Physical Therapy (96 institutions; about 10,000 patients aged 65 years or older). Among patients who were independent in activities of daily living before admission (Barthel Index ≥ 85), the study emulates a target trial comparing physiotherapy started within 2 days of admission with later initiation, using a clone-censor-weight approach with a 2-day grace period. The primary outcome is functional decline or death at 1 year. The average effect is estimated with augmented inverse probability weighting (H1); heterogeneity of the effect is tested with a causal forest and the rank-weighted average treatment effect (H2); and prioritization by estimated benefit is compared with prioritization by predicted risk using Qini curves (H3). Outpatient cardiac rehabilitation started within 30 days of discharge is analysed as an exploratory second decision point. The frozen analysis code, outcome-free feasibility checks, and a plasmode simulation of the planned procedures are deposited with this registration. Outcome variables had not been loaded for this study at the time of registration.