Statistical pitfalls in pharmacological MRI

fMRI is increasingly used to monitor the modulating effects of pharmacological agents on human brain function [1]. Often a generalized linear model approach [2] (e.g. SPM, http://www.fil.ion.ucl.ac.uk/spm/) is applied to analyze fMRI data. Within this toolbox, a canonical form of the hemodynamic response function (HRF) is used as a standard regressor to analyze the signal course of the measured data. Despite potential effects of the examined drugs on the mechanisms of neurovascular coupling, most pharmacological MRI (phMRI) studies are analyzed under the premise of an unchanged hemodynamic response function (HRF). This assumption may lead to false negative activation changes when the HRF underlies considerable alterations induced by the employed drugs. Here we demonstrate this statistical pitfall on an exemplary phMRI study and show how this assumption may lead to incorrect conclusions if HRF changes are neglected in statistical analysis methods.