study in their June members’ bulletin. The European Space Agency advertised the study via its mailing list. To ensure responses were genuine, access to the study website was restricted to listed members of these groups and the study was not publicised on social media platforms. Data collection Data collection took place from 2 June to 23 July 2021. The study comprised a sequence of 12 tasks from a library available on the Cognitron server (www. cognitron.co.uk). The test took about 30 minutes to complete. The tasks were selected on the basis of previous data, which showed they can be used to measure distinct aspects of human cognition, spanning planning and reasoning, working memory, attention, and emotion processing abilities. Previous work has shown that the battery of tasks is robust to the type of device that a person uses to complete the test; sensitive to population variables of interest such as age, gender, and education level; and not so strongly correlated as to measure just one overarching ability. As a result, the raw scores on each task are not of interest; instead, the meaningful findings are obtained by comparing standardised scores between individuals or groups to showcase differences in the measures. Before doing the test, participants completed a questionnaire comprising six questions related to specialist area, gender, age, geographical location, handedness, and level of experience (years in specialty). Task designs and data preprocessing The 12 tasks were prospective word memory, digit span, spatial span, block rearrange test (two dimensional spatial problem solving), four towers test (three dimensional spatial problem solving), the Tower of London test (spatial planning), two dimensional manipulation, target detection, verbal analogies, rare word definitions, emotional discrimination, and delayed recall of words (see supplementary figure 1). Each task was scored, and, except for the rare word definitions task, was based on reaction time (ie, speed of response). Data were preprocessed in a similar fashion to previous studies using the Cognitron platform.9 Briefly, only those datasets in which all tasks had been completed were included for analysis. In addition, we excluded participants who we considered had lost task focus—that is, the window had been inactive (in the background) for more than two seconds. We also performed a manual check for inconsistencies in questionnaire responses and excluded these datasets. Scores for each task >8 standard deviations from the mean were winsorised to reduce the effect of spurious outliers. Statistical analysis Confounding variables (age, handedness, and gender) were regressed out of the raw task scores and reaction times using generalised linear modelling, leaving adjusted scores. Through factor analysis, the eigenvalues of the correlations between adjusted task scores and reaction times were used to split the scores into several domains, with each task contributing weights to the domain. Two factor analyses were conducted, one for the task scores (12 tasks) and one for the reaction times (11 tasks). The number of domains for each factor analysis was based on the Kaiser criteria on 16 April 2022 by guest. Protected by copyright. http://www.bmj.com/ BMJ: first published as 10.1136/bmj-2021-067883 on 13 December 2021. Downloaded from RESEARCH the bmj | BMJ 2021;375:e067883 | doi: 10.1136/bmj-2021-067883 3 (eigenvalue >1). Using generalised linear modelling we then compared the domain scores between groups. The comparator group comprised members of the UK general population who had completed the same set of tasks. This is a subset of the more than 250000 participants who completed the GBIT but incorporates all eligible participants who had completed the battery of tests undertaken by the aerospace engineers and neurosurgeons. Only 18257 were eligible because the initial battery of GBIT went through iterative amendments before settling on this final battery. The GBIT cohort was recruited through diverse sources, including the BBC Two’s Horizon programme, the BBC, and BBC News home pages and news meta-apps. Members of the cohort were predominantly white (226257/269264; 84.0%), had completed secondary school (84860/269264; 31.5%), and had a university degree (154656/269264; 51.4%). The task weightings derived from this study were applied to create domain scores for the general population. Z scores for each participant were then generated using the mean and standard deviation of domain scores from the general population. To assess if the z scores from each group were different from those of the general population, we used two tailed one sample t tests. All data processing, analysis, and visualisation were conducted on Matlab v2020b (Mathworks). P values This publication marks a time of significant growth for our departments: Patient volume has risen steadily, we have added stellar faculty members in key strategic areas, and our research and educational programs continue to expand. Today, all of our subspecialties are staffed by physician leaders, including many nationally and internationally recognized clinicians and researchers. Our divisions include one of the largest, most sophisticated epilepsy centers in the nation; the