Does Predictive Skill Survive a Target Swap? A Provenance-Aware Audit of Machine Learning on the 2025 CDC PLACES Diabetes Map
Year: 2026
Type: preprint
Abstract: This preregistration describes a fixed-prediction, provenance-aware target-swap audit. The study holds geography, leave-one-state-out folds, Ridge models, and predictions fixed while changing only the evaluation target: from the CDC PLACES 2025 modeled diabetes product, based on BRFSS 2023 observations, to same-year direct SMART BRFSS 2023 estimates. The primary estimand is the difference between the documented-generator-aligned feature block's squared-loss advantage on PLACES and its advantage on SMART. A 199-permutation structured-placebo severity diagnostic and a directional standard AskCHIS 2023 independent-survey falsifier are prespecified. LACHS 2023 is an access-failure replacement for CHIS, never an additional arm. All experiment computation is restricted to the author's Mac Studio.
Author: Chuhong Xu
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7204117415