Investigation into the applicability and optimization of the Dutch matrix sentence test for use with cochlear implant users

Abstract

Objective: Matrix sentence tests use words from a fixed word matrix to compose syntactically equivalent, but semantically unpredictable sentences. These tests are suitable for monitoring performance of cochlear implant (CI) users by repeated speech intelligibility testing. This study evaluates the Dutch matrix sentence test in CI users in quiet and in noise. It then investigates the possibility to improve the test-retest reliability for CI users by selecting subsets of sentences. Design: Repeated speech intelligibility testing was performed in quiet and in noise. The effect of sentence selection on the test-retest reliability was predicted by computer simulations and experimentally evaluated using a cross-over design. Study sample: Fifteen post-lingually deafened CI users, of which eleven participated in the cross-over study. Results: The test-retest reliability equaled 2.3 dB in quiet and 1.3 dB in noise. The simulations predicted an improvement in test-retest reliability, especially in quiet. The cross-over study did not confirm the predictions. Conclusions: The results of the study suggest that the homogeneity of the sentences is not the prime component underlying the test-retest reliability. The Dutch matrix speech material and the selected subsets of sentences were equally suitable for speech intelligibility testing in CI users.

Key Words::

• Speech intelligibility
• cochlear implant
• matrix test
• speech-in-noise test
• speech reception threshold

Notes

Acknowledgements

We would like to thank the subjects for their time and effort. Also we would like to thank Cochlear^TM for their technical support. Furthermore we would like to thank Dr. M. Boymans for her clinical support during the measurements.

Declaration of interest: Financial support was received from Cochlear^TM.

Notes

1. The large number of ten alternatives per word type and the unpredictable semantics of the sentences preserve the validity of testing in the closed set configuration, even with a higher predictability per word type than in open sets.

2. The Dutch matrix test was referred to by Houben et al, 2014 as the Dutch matrix speech-in-noise test. However, because we focus both on its applicability in quiet and in noise, we chose to omit ‘speech-in-noise’ in this paper.

3. The computer simulations were also used to perform the power calculations for the cross-over study.

4. Because all subjects were patients of the audiological center in the Academic Medical Center (AMC, Amsterdam), their T-levels and C-levels were set according to the same protocol. Also, all subjects used the default amplitude mapping function (i.e. the function that relates channel magnitudes to stimulation levels during processing) of their respective device as set by the manufacturers.

5. Because of this slope restriction, the median instead of the mean of the subject-specific slopes was required as a measure for the steepness of the intelligibility function. Using the median gives a better estimate than simply omitting the unrealistic fits, because that would introduce an unwanted bias towards the mean.

6. The presentation levels were relative to the subject-specific SRTs, which is a common method used to correct for inter- individual differences in performance (Smits & Houtgast, 2006). The reference logistic function had a SRT of 0 dB and a steep but realistic slope of 15%/dB according to measurement results for NH listeners (Houben et al, 2014). Sentences were considered for selection if the criteria for selection were met by the majority of the observed intelligibility scores.

7. The SD for the test-retest reliability used in the power analysis was 1.08 dB. This value was based on the variability in test-retest reliability for sets of 10 triplets of simulated SRTs. The common SD for the median slope used in the power analysis was 1.5 %/dB. This value was based on the variability in subject-specific slopes between NH listeners (Koopman et al, 2007). Using a test significance level of 5% and power of 80%, the minimum number of subjects to confirm the theoretical effects on the test-retest reliability and median slope was nine.

8. It should be noted that nine test-retest differences > 3 dB were excluded because the subject lacked the required concentration to perform the task.