Take-home: Combining data increases power, but a larger N won’t always improve inferences if data features differ across cohorts. When combining cohorts/studies (e.g., multiple lifespan segments of the HCP), data harmonization and reporting of how this was achieved is critical.

Harmonization can mitigate these HCP cohort differences, but implementation matters! Example here shows harmonization correcting for ‘protocol’ was more optimal than correcting for ‘site.’

These young-adult dips and blips are not typical lifespan patterns, and we show this using two independent reference lifespan datasets: NKI-Rockland Sample (NKI) and Dallas Lifespan Brain Study (DLBS). Neither showed the marked young-adult deviation seen in the HCP data.

The cohort-level signal differences in HCP-YA distort measurements of brain network organization, resulting in surprising decreases and increases in multiple measure of brain organization during young-adulthood.

The human connectome project (HCP) is an amazing open resource that spans from development to old adulthood. But the scanners/protocol weren’t identical across cohorts, resulting in cohort-differences in signal-to-noise ratio in the functional scans (e.g., resting-state).

Take-home: Combining data increases power, but a larger N won’t always improve inferences if data features differ across cohorts. When combining cohorts/studies (e.g., multiple lifespan segments of the HCP), data harmonization and reporting of how this was achieved is critical.

Harmonization can mitigate these HCP cohort differences, but implementation matters! Example here shows harmonization correcting for ‘protocol’ was more optimal than correcting for ‘site.’

These young-adult dips and blips are not typical lifespan patterns, and we show this using two independent reference lifespan datasets: NKI-Rockland Sample (NKI) and Dallas Lifespan Brain Study (DLBS). Neither showed the marked young-adult deviation seen in the HCP data.

The cohort-level signal differences in HCP-YA distort measurements of brain network organization, resulting in surprising decreases and increases in multiple measure of brain organization during young-adulthood.

The human connectome project (HCP) is an amazing open resource that spans from development to old adulthood. But the scanners/protocol weren’t identical across cohorts, resulting in cohort-differences in signal-to-noise ratio in the functional scans (e.g., resting-state).

Take-home: Combining data increases power, but a larger N won’t always improve inferences if data features differ across cohorts. When combining cohorts/studies (e.g., multiple lifespan segments of the HCP), data harmonization and reporting of how this was achieved is critical.

Harmonization can mitigate these HCP cohort differences, but implementation matters! Example here shows harmonization correcting for ‘protocol’ was more optimal than correcting for ‘site.’

These young-adult dips and blips are not typical lifespan patterns, and we show this using two independent reference lifespan datasets: NKI-Rockland Sample (NKI) and Dallas Lifespan Brain Study (DLBS). Neither showed the marked young-adult deviation seen in the HCP data.

The cohort-level signal differences in HCP-YA distort measurements of brain network organization, resulting in surprising decreases and increases in multiple measure of brain organization during young-adulthood.

The human connectome project (HCP) is an amazing open resource that spans from development to old adulthood. But the scanners/protocol weren’t identical across cohorts, resulting in cohort-differences in signal-to-noise ratio in the functional scans (e.g., resting-state).