Design-Induced Circularity in Alzheimer Biomarker Trials and Trial-Ready Cohorts: Enrichment, Data-Driven Selection, and Estimand Mismatch

BackgroundBiomarkers increasingly gate Alzheimer disease trials and clinical access, raising risk of design-induced bias when biomarker data are reused for both selection and inference. ObjectiveTo quantify circularity patterns in Alzheimer disease biomarker trials and trial-ready cohorts and, using mechanistic experiments, estimate effects on discrimination, calibration/transportability, and false-positive risk. MethodsWe audited PubMed-indexed biomarker interventional trials and trial-ready cohorts (2020-2024) using dual independent full-text review and mechanism classification. Two simulation experiments examined (1) enrichment-induced range restriction with transport to an unselected population and (2) selection-inference reuse ("double dipping") versus split-sample confirmation. ResultsAmong 72 studies, 10 (13.9%) raised possible/definite non-independence; enrichment/range restriction predominated (7/10). In simulation experiment 1, enrichment preserved nominal type I error under the null but shifted the estimand under signal, attenuating discrimination and degrading calibration when enriched-fit models were applied to unselected populations. In simulation experiment 2, double dipping produced P(p < 0.05) = 0.994 under the null (n = 200, p = 100) versus 0.047 with split-sample confirmation, with winners-curse effect inflation. ConclusionsCircularity was uncommon but nontrivial and most often reflected enrichment with implicit generalization beyond the enriched population. Safeguards include explicit target populations/estimands and separation of discovery from confirmation (or prespecified multiplicity control), with clear validation and leakage prevention when biomarker claims are intended to guide decisions.