Learning Effects of the Dutch/Flemish Matrix Test for Bimodal Cochlear Implant Users.

Abstract

Introduction: Cochlear implantation (CI) is the standard treatment for severe-to-profound sensorineural hearing loss, but CI users often struggle with speech understanding in noisy environments. The Dutch/Flemish Matrix test is frequently used to evaluate speech-in-noise performance due to its assumed immunity to learning effects. However, studies challenge this assumption, revealing significant learning effects that can confound research outcomes. In this study, we modeled the learning curves of the Dutch/Flemish Matrix test to assess the influence of both between-session and between-test effects. We hypothesized that a exponential model would describe the learning effects more accurately than a linear model.

Methods: The perceptual learning effects associated with the Dutch/Flemish Matrix test were assessed in 17 bimodal CI users. All participants performed the Matrix speech-in-noise tests across four sessions, with 13 randomized tests per session. The tests were conducted in a soundproof booth with an eight-speaker babble noise. The outcome parameter was the speech recognition threshold and was analyzed with a linear mixed model to account for confounders.

Results: The results showed a statistically significant learning effect between sessions that added up to a speech intelligibility increase of 1.3 dB signal-to-noise ratio (SNR) (equivalent to ∼10% word score) between the first and second sessions, 0.86 dB SNR (∼7%) between the second and third sessions and 0.67 dB SNR (∼5%) between the third and fourth sessions. In addition, a statistically significant within-session learning effect (i.e., between tests) was observed with a linear slope of -0.11 dB SNR/test (∼0.9% word score/test), which accumulates to a total of 1.7 dB SNR (13%) between session start and end. The between-session learning curve was described more accurately with an exponential fit than with a linear fit. The between-test learning curve can be described equally well with a linear and an exponential fit.

Conclusion: A robust between-test learning effect was observed, which could be accurately modeled using either a linear or exponential learning curve. Additionally, a between-session learning effect was evident and was best described by an exponential learning curve. This study provides an important handle for correcting these learning effects in future studies.