Proefschrift_vd_Beek

Analysis Signal processing was performed off-line using Matlab. eCAP amplitudes were automatically detected using Matlab software (as per Frijns et al, 2002) and plotted against the electrode positions along the array. Curves that did not show eCAP amplitudes above 0.1 mV were not included in the analysis. This criterion was not reached in 16% of the responses, mainly in the low current range. The average of the peak amplitude, for both selectivity and scanning, was 0.6 mV. The curves were normalized by taking the value at the electrode contact of interest and dividing all values along the array by this value. Next, both flanks of the selectivity and scanning curves were fitted by a 4th order polynomial. The width was defined as the number of electrode contacts (spaced 1.1 mm apart) from the stimulated contact to the point at which the normalized amplitude reduces to 0.6. For the middle contact both the width in the apical (EM-A) as well as the basal direction (EM-B) were calculated. In cases where the minimum value did not drop to 0.6 the width was set as the limit of the array in the apical or basal direction (as per Abbas et al, 2004). In previous studies both 50% and 75% of the peak amplitude have been used as a measure of the width of the region of excitation (Cohen et al, 2003; Hughes & Abbas, 2006a). For this study, 60% of the peak amplitude, determined on the basis of the fitted, normalized curves, was selected as a trade-off between obtaining as many curves as possible and being able to measure differences between distinctive profiles along the array. Figure 2 shows typical selectivity curves recorded in one subject. The figure shows the normalized eCAP amplitudes obtained at the three current levels. A horizontal line indicates 60% of the amplitude. The horizontal solid arrows then indicate the width of the curve (in basal or apical direction) as defined above for the highest current level. Statistics The design of this study is basically a within-patient analysis with three factors, which means that at the patient level the measurements are correlated. So-called linear mixed models take this correlation into account, by considering the responses from a subject to be the sum of so-called fixed effects, affecting the population mean, and random effects, associated with a sampling procedure (e.g. subject effects). The random effects often introduce correlations between cases and should be taken into account to elucidate the fixed, population affecting, effects. The SE (standard error of the mean) generated by the model is used in significance analysis. Using linear mixed models enables investigation of the effects of each parameter separately as well as the interaction between different parameters. Furthermore, linear mixed models can effectively use all data, even when one or more data points are missing (Fitzmaurice et al, 2004). In the present study SPSS 16.0 was used to construct mixed linear models to address the influence of the measuring technique, the electrode position, and the current level separately. For significance levels in this study t-tests are used, both in the context of descriptive statistics as well as with linear mixed models.

4

To create a comprehensive overview data are plotted in boxplots. However, it should be noted that boxplots

77