Statistical tests and identifiability conditions for pooling and analyzing multisite datasets.

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Biomarker Discovery

Background:

• Pooling smaller datasets can enhance the detection of weak associations between predictors and responses.
• Variations in data acquisition and distributional shifts across datasets can obscure real effects when combining data.

Purpose of the Study:

• To develop a rigorous statistical framework for correcting distributional shifts in pooled datasets.
• To provide an algorithm for identifiable shift corrections.
• To enable reliable analysis of combined datasets, particularly for small sample sizes.

Main Methods:

• Developed a statistical treatment to identify and correct distributional shifts.
• Proposed an algorithm for identifiable correction scenarios.
• Analyzed framework properties including model fit testing, confidence intervals, and consistency.

Main Results:

• Identified conditions for correcting distributional shifts between datasets.
• Demonstrated the framework's ability to harmonize protein biomarkers across different assay sites.
• Empirical results validate the proposed statistical approach.

Conclusions:

• The developed framework mitigates bottlenecks in pooling smaller clinical research datasets.
• Enables effective harmonization of data from diverse sources, crucial for rare disease research or resource-limited studies.
• Facilitates robust analysis of combined datasets, improving scientific discovery.