When can Multi-Site Datasets be Pooled for Regression? Hypothesis Tests, ℓ ₂-consistency and Neuroscience...

Area of Science:

• Biomedical data science
• Statistical genetics
• Health sciences research

Background:

• Biomedical studies often face small sample sizes due to logistical and financial limitations.
• Pooling data across multiple sites is a common strategy to increase statistical power for detecting weak associations.
• Existing methods for multi-site data analysis do not clearly indicate when pooling is beneficial, irrespective of the inference algorithm used.

Purpose of the Study:

• To develop a hypothesis test for determining the conditions under which pooling multi-site datasets is statistically advantageous.
• To identify specific regimes where data pooling enhances analytical power in both classical and high-dimensional linear regression.
• To provide practical guidelines for deciding on data pooling strategies before data transfer, focusing on applications in Alzheimer's disease research.

Main Methods:

• Development of a novel hypothesis test to assess the utility of pooling data from diverse research sites.
• Application of the test to classical and high-dimensional linear regression models.
• Empirical validation using simulated data and a real-world Alzheimer's disease study dataset.

Main Results:

• The study precisely identifies conditions (regimes) where pooling datasets across multiple sites is statistically sensible.
• A simple, site-executable check is proposed to guide data pooling decisions prior to data transfer.
• Empirical results demonstrate improved statistical power when pooling local data with international Alzheimer's disease study data under the identified regimes.

Conclusions:

• A statistically sound method is provided to determine the benefit of pooling multi-site data, independent of specific inference algorithms.
• The findings offer practical guidance for researchers, enabling informed decisions about data sharing and pooling to maximize analytical power.
• The approach has direct implications for enhancing the efficiency and effectiveness of collaborative research, particularly in complex areas like Alzheimer's disease research.