Shu Liu and colleagues, utilizing data from the world's largest neuroimaging study to date, discovered that reliable associations between brain structure and function and individual variations in behavior, cognition, and personality can only be replicated when data from thousands of participants are included. This finding challenges the typical practice of performing brain-phenotype association studies with tens or hundreds of participants, implying that all previously reported results may not be replicable.
Importance of large-scale neuroimaging data
Guido van Wingen, professor of Neuroimaging in Psychiatry, and his colleagues used the UK Biobank neuroimaging dataset, which contains information from over 37,000 participants. Their study focused on associations with six variables related to physical and mental health: age, body mass index, intelligence, memory, neuroticism, and alcohol consumption. By examining the improvement in replicability as sampling sizes increased, the researchers shed light on the importance of large-scale participant cohorts in brain-phenotype associations.
Generalizability of six representative variables
In order to evaluate whether the results for these six representative variables would generalize to other phenotypes from the same domains, the associations between the brain and 23 additional phenotypes derived from UK Biobank were investigated. Again, the replicability was estimated at varying sample sizes with a threshold of p < 0.05. In general, these results were comparable to those for the six representative variables.
Replicability in research
There is a growing concern about the replication of scientific results. In particular, inadequate statistical power in neuroimaging research is thought to lead to high levels of false positive results and assessing the replicability has always been a challenge because of the limited sample sizes. The heterogeneity of brain–phenotype associations in the population can lead to inconsistent findings, particularly when small sample sizes are used.
Fortunately, the field has witnessed the accumulation of large-scale, high-quality imaging datasets over time. Notably, the UK Biobank initiated the world's largest multimodal imaging study in 2014 and has made available data from approximately 37,000 individuals who underwent both T1 and resting-state fMRI scanning. These expansive datasets enable researchers to examine the replicability of brain-phenotype associations in a feasible manner and address the limitations posed by small sample sizes.