100 / 162
98 | Chapter 5
this does not provide the DR in linear clinical units, the DR in decibels can easily be derived: DR (dB) =
20 log[100/(T-/M-level ratio)].
To assess intrasubject variation and to facilitate the comparison with previously published data [Pfingst
and Xu, 2004], the data were recalculated and expressed in decibels: I (dB) = 20 log[I (CU)/1,000 × 20.6
(CU)]. This, for instance, enables the data to be seen more in line with data presented in Cochlear’s current
levels, which are also on a logarithmic scale. In line with Pfingst et al. [2004], across-site mean (ASM) and
across-site variance (ASV) were calculated in order to be able to analyze fitting levels both across as well as
within subjects. Both Tand M-levels were determined during regular clinical fitting sessions, approximately
8 times during the first year. The Tand M-levels of the initial fitting (about 4 weeks after implantation) and
the levels obtained at 1 year of cochlear implant use were used for this study.
low-up. For 19 of the 151 subjects included in the study, speech
scores at the 1-year follow-up were not available for logistical rea-
sons. The standard Dutch speech test of the Dutch Society of Au-
diology, consisting of phonetically balanced monosyllabic (CVC)
word lists, was used [Bosman and Smoorenburg, 1995]. As de-
scribed previously [van der Beek et al., 2005], the speech material
was presented in free field in quiet at a level of 65 dB.
Statistical Analysis
All data analysis was performed using SPSS 19.0 (IBM, Ar-
monk, N.Y., USA). Mixed linear models were used to analyze the
data and to construct predictive models. These models aimed to
predict T- and M-level profiles using only one measured level at
one fixed individual electrode contact. In a mixed linear model,
responses from a subject are thought to be the sum of fixed and
random effects. The effects which affect the population mean are
called fixed. If an effect is associated with a sampling procedure
(e.g.,subjecteffect), it iscalledrandom.Theserandomeffectsoften
introduce correlations between cases and therefore should be tak-
en into account to elucidate the fixed effects which impact the pop-
ulation. 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 T-
and 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 val-
ues. To improve reliability, 10 different random selections per pre-
dictive model were performed.
15+16
13+14
11+12
9+10
7+8
5+6
3+4
1+2
100
0
200
Level (CU)
Electrode duo
300
400
a
90%
75%
50%
25%
10%
100
0
200
Level (CU)
Electrode duo
300
400
b
90%
75%
50%
25%
10%
15+16
13+14
11+12
9+10
7+8
5+6
3+4
1+2
100
0
200
Level (CU)
Electrode duo
300
400
c
90%
75%
50%
25%
10%
15+16
13+14
11+12
9+10
7+8
5+6
3+4
1+2
Fig. 1.
Percentiles for T-levels (
a
), M-levels (
b
) and DRs (
c
) in
clinical units. Data from two adjacent electrode contacts were
combined and plotted as an electrode duo to include subjects with
fewer than 16 active electrode contacts.
PM
Fig. 1.
Percentiles for T-levels (a), M-levels (b) and DRs (c) in clinical units. Data from two adjacent electrode contacts were combined
and plotted as an electr de duo to include subjects with few r than 16 active electrode ont cts.