Data Analysis and Statistics

Because of skewed distribution, the Mann Whitney U test was used to compare sleep characteristics and number of daytime naps between patients and controls at baseline, and the related-samples Wilcoxon signed rank test was used to analyze sleep characteristics and daytime naps before and during SXB treatment in patients.

For subsequent analyses on the 24-h profiles, the mean temperature of each episode of 30 min was calculated. Group differences, group by time of day differences, and treatment by time of day effects on temperature were analysed with Generalized Linear Model for repeated measures with HuynhFeldt corrections (IBM SPSS 20, Inc., Chicago, IL, USA) with between factor narcolepsy, within factors SXB, and time of day and covariate geographical site (Leiden or Zürich). This analysis was separately performed with the real clock time data and with the data anchored for nocturnal bedtimes (with the actual bedtime coded as time-point 0). Analysis was run on the 24-h data, and separately for daytime and nighttime. Where narcolepsy or SXB related differences reached significance, post hoc t-tests were used to evaluate differences in the time of day.

Mixed effects logistic regression analysis (R version 3.1.1, R Foundation for Statistical Computing, Vienna, Austria) was performed to evaluate the effect of temperature on the onset of spontaneous daytime sleep attacks in patients at baseline. Since these temperatures were measured once per minute, and sleep scoring was performed in 30-s epochs, the temperatures were interpolated into 30-s values. Nocturnal temperatures were excluded. For all analysis the outcome variable was sleep onset, which was binomially coded for every 30-s epoch as wake = 0 and sleep onset = 1 (further sleep epochs were excluded from analysis. The different temperatures (proximal, distal, and DPG) and time of day were entered into the model as fixed effects. Intercepts for subjects were defined as random effects. To evaluate different epoch durations prior to sleep onset, 3 analyses were performed, each with a different regressor representing the temperature profile in an epoch prior to sleep onset. The first analysis evaluated the last temperature value during the epoch prior to sleep onset. The second and third regressor evaluated the predictive value of monotonic changes in temperature prior to sleep onset. To this end, the second regressor evaluated changes immediately preceding a sleep bout, quantified as the difference between the last temperature readout immediately prior to the 30-s epoch and the temperature 5 min before. The third regressor evaluated slower changes, which were quantified as the difference between the last temperature readout immediately prior to the 30-s epoch and the temperature 15 min before. These 3 analyses were repeated for each of the temperatures (proximal, distal and DPG). P values were obtained by likelihood ratio tests of the full model with the effect in question against the model without the effect in question.