Routine flumazenil administration and the interpretation of a non‐inferiority comparison: a reply

Deng and Peng [1] question whether routine flumazenil administration confounded our comparison of remimazolam and propofol [2]. The authors' concern rests on the premise that flumazenil-induced central nervous system hyperexcitability could have influenced our delirium assessments. This is difficult to reconcile with either the pharmacokinetics of flumazenil or the timing of our measurements. Flumazenil has a short elimination half-life of approximately 40–80 min [3], and a single 0.2 mg dose given at the end of surgery is essentially eliminated within a few hours. In contrast, the confusion assessment method was first applied 24 h postoperatively and again at 48 h and 72 h. Any transient excitatory effect would therefore have resolved long before the first assessment. The concern appears to conflate acute emergence-related phenomena with postoperative delirium, which are distinct entities evaluated over different timescales. More fundamentally, the evidence cited does not support extrapolation to our setting. The meta-analysis by Penninga et al. [4] examined patients admitted to the emergency department with impaired consciousness due to known or suspected benzodiazepine overdose, and that review explicitly did not include patients undergoing surgery. The flumazenil doses were far higher than ours (mean 0.77–7.13 mg, up to 10 mg), and the reported adverse events occurred within approximately 10 min of administration. The serious adverse events were confined largely to patients with multi-drug intoxication involving pro-convulsant drugs, such as tricyclic antidepressants, and are attributed to reversal of the anticonvulsant action of benzodiazepines rather than a direct effect of flumazenil. The agitation reported in that population reflects, in part, benzodiazepine withdrawal in chronic users. None of these conditions applied to our patients, who had normal pre-operative cognition, received benzodiazepine exposure only intra-operatively, and were given a single low dose of flumazenil. In our institution, routine administration of flumazenil 0.2 mg to approximately 1500 patients has not been associated with notable adverse events [5]. We would also note that the specific analysis requested is not feasible with our data. Because every patient in the remimazolam group received flumazenil and no patient in the propofol group did, flumazenil administration is perfectly collinear with group allocation. It cannot be entered as an independent covariate alongside group in any regression model, as this produces complete separation, and there is no within-group variation from which its effect could be estimated. Isolating the contribution of flumazenil is therefore impossible by design in this trial and would indeed require the dedicated flumazenil-free arm that the authors propose. We agree this design would be worthwhile in future work. The two mechanisms offered, that flumazenil may have masked residual benzodiazepine-related impairment or that it may have increased delirium independently, predict effects in opposite directions. As both are speculative, they do not establish a directional bias in our findings. As stated in our methods, flumazenil was administered to all patients in the remimazolam group in accordance with our institutional standard practice. Our findings should be interpreted accordingly as applying to remimazolam anaesthesia as delivered in this real-world context, which we regard as a strength of the study's external validity rather than a hidden confounder.