Modeling video tic counts in a crossover trial of deep brain stimulation for Tourette syndrome.

A recent pilot crossover study of deep brain stimulation for Tourette syndrome involved the counting of motor and sonic tics from video recordings of patients. The evaluation of a five-minute video (divided into ten 30-second segments) in each of eight intervention states per patient was found to be very tedious and time-consuming. The present study sought to determine the statistical implications of reducing this data collection burden. To make maximal use of data from the small sample (n=5) pilot study, we fit linear mixed effects models to the tic count data. As suggested by an empirical examination of within-person correlations, a novel random effects covariance structure, which we refer to as a 'partitioned random effects model' was found to provide the best fit to the data. The best model for each tic type was then used to estimate relative efficiencies for specified data reductions. This analysis indicated that using a subset of five out of 10 segments would require only a 10% increase in sample size to maintain a specified power. Lastly, the bias of estimated treatment effects based on the reduced data collection was evaluated, and the particular five-segment subsets with the smallest estimated bias were determined.