Statistical analysis

Statistical analysis was performed using the program SPSS Statistics Version 22, IBM, Chicago, IL, USA. With respect to the thermal grill experiments, Kolmogorov–Smirnov test revealed that for NRS ratings of thermal sensation and for thermal threshold, data values were not normally distributed. Therefore, the median as well as the first and third quartiles (boxes) and range (error bars) were used for data description and Friedman’s ANOVA with post hoc testing (Wilcoxon’s signed-rank test) was employed for statistical analysis of the thermal grill-evoked sensations. To assess sex differences in the qualities of the thermal grill-evoked sensations, the Mann-Whitney U test was performed. Differences in sensations evoked by the uniform cold (20 °C) or warm (40 °C) stimulus and the thermal grill (20°/40 °C interleaved) stimulus are presented as means ± standard error of the mean (SEM; Δ values) for male and female subjects separately. For analysing sex differences in the intensities of grill-evoked sensations, Student’s t test for unpaired samples was performed. Linear regression analyses were carried out to evaluate the effects of thermal sensation and thermal pain sensation on the effects of the thermal grill-evoked warmth/heat and coldness. To assess dependency between two variables of non-normally distributed data, Spearman’s correlation coefficient (ρ) was calculated. A Bonferroni-type adjustment was made for multiple correlation analyses. P value < 0.05 was considered to indicate a statistically significant difference and is indicated by an asterisk (*) in the tables and figures.

To assure the quality of the retrospectively analysed QST data, the variances of all 12.874 records were analysed. The variances were distributed in the range 0.00–260.26. The distribution of variances is characterized as follows: median 0.280, interquartile range (IQR) 1.430, mean 2.577, and standard deviation (SD) 9.565. Skewness was reduced calculating an upper criterion (Eq. 1), following McGill and colleagues [27]

Any record with a variance above a crit value of 2.425 was excluded from further analysis. In addition, records with incomplete or implausible data were excluded, i.e. data lacking the information about sex or age or data obtained using incorrect stimulation sites. Some 9940 records remained for further analysis. Following the suggestions by Rolke and colleagues [2], all records were logarithmized (base 10) before being subjected to any statistical test. Sex differences were analysed post hoc using Tukey’s honest significant difference test [28]. Analyses were performed using the language for statistical computing R [29].

For analysis of multimodal distribution of CPTs, the CPT data were fitted by a Gaussian mixed model using the R-program AdaptGauss [30] (Additional file 1). For calculating the probability density function (PDF), the median of five repetitions of the CPT test of each subject was estimated and rescaled for stimulus intensity and a log transformation was subsequently carried out. The PDF was calculated for all subjects participating in the CPT measurements (N = 296).