Personalized Harmonic Baseline Encoding (PHBE): Reference Implementation and Simulation Pipeline

Reference implementation and complete reproducible simulation and analysis pipeline for Personalized Harmonic Baseline Encoding (PHBE), a learner-agnostic feature-engineering method that transforms regularly sampled exogenous time series into individualized deviation-based features anchored to a patient-specific harmonic-mean baseline. The repository contains the PHBE implementation, environmental data-processing workflow, synthetic cohort generation, multimodal prediction models, sample-size analyses, learning-curve analyses, permutation testing, attribution analyses, automated tests, figure-generation scripts, and the simulated data and numerical results used in the accompanying methodological study. The clinical development cohort is entirely synthetic. No electronic health records, identifiable patient information, or individual-level real patient data are included. Environmental exposures are based on data from the NASA Prediction of Worldwide Energy Resources (POWER) project. The repository is intended to enable full reproducibility and independent evaluation of the PHBE methodology. Accompanying article: “Personalized Harmonic Baseline Encoding: a multimodal machine-learning framework integrating satellite-derived environmental dynamics with electronic health records. A methodological development and simulation-based proof-of-concept study.”