In Reply We appreciate the comments from Mühlau et al regarding our study¹ examining the effect of HLA genetic risk burden on magnetic resonance imaging (MRI) disease phenotypes in multiple sclerosis. They provide an opportunity to raise some known but seldom discussed issues relevant to genetic associations with clinical and MRI phenotypes.

We agree that there is a benefit of multicenter studies to limit potential bias and the value of replication in different centers. This is sometimes accomplished within a single study (preferable) or within a sequence of studies. In either case, in comparing results between centers, there are several factors that can lead to variance in the observed effect sizes and statistical significance leading to type 2 replication errors. These include (1) intrinsic bias in cohorts at the genetic or other fundamental level (time independent); (2) time-dependent bias when looking at phenotypes whose magnitude and variance can change over time (eg, clinical disability scores or MRI volume metrics); (3) differences in data acquisition instrument and protocol leading to differences in precision; (4) differences in data processing procedures; (5) differences in statistical model and variables included in the model; and (6) poor statistical power leading to fluctuations in significance about a marginally significant result.