Reply: Replicability and impact of statistics in the detection of neural responses of consciousness.

Sir, We read with interest the letter by Gabriel and colleagues (2016) addressing the major issue of replicability when probing conscious processing in non-communicating patients. This question—as well as the choice of the optimal statistical methodology—concerns the whole field of functional brain imaging in cognitive neuroscience (Kriegeskorte et al. , 2009), but its importance obviously culminates in single-subject analyses of non-communicating patients (see for instance the recent debate in Cruse et al. , 2011, 2013; Goldfine et al. , 2012). Gabriel et al. reacted to a recent discussion (Naccache et al. , 2015; Tzovara et al. , 2015 a , b ) following a report by Tzovara et al. (2015 a ), who adapted our auditory ‘local-global’ bedside EEG test (Bekinschtein et al. , 2009) to test comatose patients. Briefly, in the local-global paradigm two levels of regularities are manipulated: local auditory irregularities correspond to a change of sound within a trial, whereas global irregularities correspond to a change of sound sequence across trials. When analysing data according to the local irregularities, one can typically extract a mismatch negativity response observable even in unconscious states. In sharp contrast, when analysing event-related potentials (ERPs) to violations of global irregularities, we previously showed that a late global effect was present only in conscious or minimally conscious patients (Bekinschtein et al. , 2009; Faugeras et al. , 2011, 2012). Two problems emerged from the study of Tzovara et al. (2015 a ), first, this ERP global effect was found positive in the vast majority of conscious controls we tested at two distinct sites using high-density EEG: 18/18 (100%) in Paris, France (with 256 electrodes), and 7 to 10/10 (70 to 100%) with the monaural and binaural versions of the task, respectively in Cambridge, UK …