cvd-glassbox-screening: code for target leakage in machine learning models for myocardial infarction screening from BRFSS survey data (v2.1.0)
Year: 2026
Type: software
Abstract: Complete analysis pipeline for a leakage-tiered, fairness-audited benchmark of ten model classes (logistic regression through EBM, gradient boosting, and the TabPFN v2 / TabICL tabular foundation models) on CDC BRFSS 2022-2023 self-reported myocardial infarction. Version 2.0.0 adds the major-revision analyses for npj Digital Medicine: seed-0 paired-DeLong non-inferiority and TOST equivalence engine (pre-specified delta = 0.005), the T1-ns stroke-exclusion tier, fairness v2 (common selection objective, equalise-TPR control arm, effective per-group thresholds, 1000-draw stratified paired bootstrap), conformal v2 (four-decimal coverage, empty-set accounting, split-validation optimism check, 18-39 carve-out hybrid), survey-weighted evaluation and weight-informed training, the FM-edge-by-weight-quintile test, LIME-on-EBM faithfulness, and a full timing/hardware manifest. Frozen experiment configurations under configs/ document the pre-specified protocol.
Institutions: Abu Dhabi University, Zayed University
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7202394737