Clinical profile impacts the replicability of multivariate brain-behavioural associations

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Biostatistics

Background:

• Multivariate brain-behaviour analyses often require large sample sizes for replicability.
• The impact of sample size and cohort composition on Canonical Correlation Analysis (CCA) replicability is not fully understood.

Purpose of the Study:

• To investigate how sample size and participant cohort composition affect the replicability of CCA results.
• To determine the minimum sample size needed for reliable CCA findings in brain-behaviour studies.

Main Methods:

• Applied CCA to diffusion-weighted magnetic resonance imaging (dMRI) phenotypes and cognitive assessment scores from the UK Biobank (N=40,514).
• Defined four participant cohorts based on clinical profiles.
• Evaluated replicability of canonical correlations and variable loadings across varying sample sizes and cohorts.

Main Results:

• Sample sizes of approximately 500 individuals were generally sufficient for replicable canonical correlations and variable loadings across all cohorts.
• A highly targeted cohort required significantly fewer samples to achieve comparable or superior correlations.
• Variable loadings remained consistent from sample sizes of around 500 up to thousands.

Conclusions:

• Sample sizes in the hundreds may be sufficient for reliable CCA results in brain-behaviour research.
• Targeting specific subpopulations can substantially reduce the required sample size for CCA.
• These findings have implications for optimizing study design in neuroimaging and cognitive research.