To the Editor Winkelbeiner and colleagues¹ questioned the presence of treatment-by-patient interaction in antipsychotic drugs as their meta-analysis yielded a variance ratio (VR) below 1. The authors correctly stated that their study cannot rule out treatment-by-subgroup interactions. Actually, the authors cannot rule out treatment-by-patient either because treatment-by-subgroup interaction may entirely cancel out the association of treatment-by-patient interaction with VR.

The authors have closely followed the work of Senn,² which discussed that outcome variance, the main ingredient of VR, is driven by not only treatment-by-patient interaction but also other factors. Unlike treatment-by-patient interaction that only increases VR, other factors may decrease VR. Treatment-by-subgroup interaction is a key example. Under certain conditions (eg, when the treatment effect is greater in subgroups with more severe symptoms), treatment-by-subgroup interaction decreases VR. Additionally, incomplete follow-up and measurement error is often associated with the outcome as well as the treatment assignment, reducing the observed VR. Lastly, as mentioned by the authors, VR will be reduced by the 鈥渟tabilizing quality鈥� of treatment. All these factors could have pushed the VR downwards, cancelling out any upwards effect of treatment-by-patient interaction. Therefore, a VR less than 1 does not prove the absence of treatment-by-patient interaction; rather, it demonstrates the presence of downwards effects.