VQSR bias against rare variants

Question

[drabiza1]

I read that VQSR is inherently biased toward filtering out variants that are very rare (ie, only encountered once in the analyzed data set).

Can someone explain why that is the case.