Sample size determination for a prospective bridging study to harmonize retrospective data from multiple studies-Application to a biofluid biomarker study of Alzheimer's disease.

Abstract

Biomarkers measured from biofluid samples and imaging scans are important to aid diagnosis of diseases and track their progression over time, especially in neurodegenerative diseases such as Alzheimer's disease (AD). Combining retrospectively obtained biomarker data across multiple studies can increase statistical power, but existing biomarker data may be generated using different assay platforms, scanner types, or processing protocols by different studies, which may significantly affect the measurements of biomarkers and hence render it necessary to harmonize the data across the studies. An optimal way of biomarker data harmonization is to re-analyze all the biofluid samples or imaging scans together on a single platform in a central reference lab, but this is often not practical because of the substantial cost involved as well as the limited amount of biofluid samples available. A more practical solution is to prospectively design a bridging study by re-measuring a subset of biofluid samples or imaging scans from the studies in a reference lab to evaluate how biomarker values may be harmonized across studies. An important design question for such a bridging study is the size of the retrospectively collected biofluid samples or imaging scans that will be re-measured. We aim to address this question by conceptualizing a latent but true biomarker that underlies the observed versions of the biomarker measured across retrospective studies and proposing methods to determine the sample size of a bridging study for estimating the biological correlation of the true biomarker with a standard and validated clinical outcome. We also report bridging data of several analytes from cerebrospinal fluid (CSF) in a multi-center biomarker study of AD and demonstrate that a small proportion of the CSF samples may be sufficient to design a future bridging study for estimating the correlations of CSF biomarkers with a cognitive and functional outcome.