Proefschrift_vd_Beek

is called random. These random effects often introduce correlations between cases and therefore should be taken into account to elucidate the fixed effects which impact the population. Using mixed linear models enables the investigation of the effects of each parameter separately as well as of the interaction between different parameters. Furthermore, mixed linear models can effectively use all data, even when one or more data points are missing [Fitzmaurice et al., 2004]. The predictive models for Tand M-levels were based on randomly selected subgroups of 70% of the subjects in order to be able to predict levels in the remaining 30% and correlate those predictive values with the measured values. To improve reliability, 10 different random selections per predictive model were performed. RESULTS Figure 1 shows the percentiles for T-levels (fig. 1a), Mlevels (fig. 1b) and DRs (fig. 1c) at the 1-year follow- up. Data are presented in clinical units to enable comparison of levels with different pulse widths. T-levels, M-levels as well as DRs showed an increase towards the basal end. The T-levels reflected real measurements of the individual levels at each individual electrode contact, whereas the M-levels were set for the subject using a profile fitting method with emphasis on the higher frequencies. The ratio of T-/M-level is shown in a box plot in figure 2a. The median T-/M-level ratio for all the electrode duos was between 20 and 35%, corresponding to a DR of 9–14 dB. The whiskers are located at 1.5 × interquartile range (IQR). A 10% or smaller ratio only occurred in a very limited number of cases (>1.5 × IQR). Assuming a normally distributed data set, this means that about 1 out of 50 (theoretically 2.15%) has a ratio of 10% or below. Furthermore, from figure 2a, it can be seen that the ratio was fairly stable along the array. Figure 2b shows that about one third of the variance of the T-/M-level ratio could be predicted by the T-level (r = 0.61, p < 0.01). On the other hand, the T-/M-level ratio did not show any correlation with the M-level (fig. 2c), while it had a significant negative correlation with the DR (r = –0.42, p < 0.01; fig. 2d). The overall T-level turned out to be very weakly correlated with duration of deafness (approx. 4 CU per decade; r = 0.22, p < 0.05) and not correlated at all with age at implantation (p = 0.63). In contrast, the overall M-level was not significantly correlated with duration of deafness (p = 0.57), but a small but significant negative correlation was found with age at implantation (approx. 15 CU per decade; r = 0.23, p < 0.01). The changes in T-level, M-level and DR during the first year are shown in figure 3. Figure 3a shows the levels along the electrode array during initial fitting and after 1 year of follow-up, expressed in clinical units. Figure 3b shows the same data, now plotted using a decibel scale. M-levels (fig. 3a) showed a larger increase (40.6 CU; SD 83.8) than T-levels (11.0 CU; SD 24.3), resulting in an increase in DR (29.8 CU; SD 73.0). T-and M-levels expressed in decibels (fig. 3b) showed an approximately equal increase (1.8–1.7 dB; SD 3.58–2.78), resulting in a stable DR in the first year (–0.2 dB; SD 3.2).

100 | Chapter 5