Response: Epileptic discharges in acutely ill patients investigated for SARS‐CoV‐2/COVID‐19 and the absence of evidence

Abstract

Epileptic discharges in acutely ill patients investigated for SARS-CoV-2/COVID-19 and the absence of evidence We appreciate the interest of Drs Rai, Gogia, and Tremont-Lukats in our preliminary report and their attempt to re-evaluate our findings using the Bayesian binomial statistics. 1 Their conclusion that more observations are needed is in close agreement with our manuscript's discussion and conclusions. One reason we published this work as a "preliminary report" was the low sample size of our case series and particularly of the COVID-19-negative group (n = 6), given that our study was done during the peak of COVID-19 pandemic in our region. 2 Along with this limitation, we were also cautious in our manuscript to highlight a number of other possible confounders that should be considered in future studies on the subject, among them false-negative rates of SARS-CoV-2/ COVID-19 testing and associated pre-existing and clinical data, as outlined also in the subsequent paragraphs. Whether one chooses the Bayesian or frequentist statistics, confidence upon their statistical outputs is strongly dependent on the sample sizes. A simple thought experiment is shown in Figure 1, using the same dataset that the authors used from our manuscript, that is, the rate of epileptiform discharges (EDs) in the COVID-19-negative (1/6, Group 1 or prior) and COVID-19-positive (9/22, Group 2 or posterior) cohorts. By merely increasing the sample size of the prior tenfold, while maintaining the same proportion of subjects with EDs over the total size (ie, from 1/6 to 10/60), and leaving the posterior (COVID-19-positive) dataset unchanged, both the simple sequential (SS) Bayesian A/B test and Fisher's exact test provide some level of statistical significance. However, extrapolating findings from small-sample exploratory studies of new patient populations, like our study, to larger populations without collecting real data is hard to recommend. Careful selection of the prior distributions needs to be done to incorporate in the hypothesis factors that may be important in positively or negatively controlling the likelihood of occurrence of a tested outcome. As shown in our cohort, acutely ill patients investigated for COVID-19suspected presentations have multiple clinical confounders that can either increase or decrease the likelihood of appearance of EDs on their EEG, as shown in table 1 of our report. 2