*2.6. Statistical Analysis*

Statistical analyses were performed using the SigmaStat® software (Systat Software, San Jose, CA, USA) and R version 4.1.2 (R Core Team 2021, R Foundation for Statistical Computing, Vienna, Austria). As they followed a normal distribution (confirmed by a Kolmogorov–Smirnov test), SRT70 values measured at T1 were compared between groups using a *t*-test. In each group, SRT70 values measured at T2 were compared to T1 values using paired *t*-tests.

To control for potential differences in demographics (age at testing, deafness duration prior to implant, years of implant experience) between groups, the *t*-test and Wilcoxon test were used. A possible correlation between demographics and SRT70 improvement between T1 and T2 was also tested.

In contrast, the total number of games played and the total duration of play were not normally distributed. The correlations of these 2 variables with each other and with SRT70 changes as a result of training were assessed using Spearman's correlation tests.

#### **3. Results**

Patient characteristics are summarized in Table 1.

Among all participants, two from the study group did not complete the training and were excluded (one moved, the other gave up), leaving 13/15 subjects (87%) who completed training and post-training Fr-matrix assessments. The time spent playing varied between 4 h 24 min and 39 h (mean 13 h) for a total of 141 to 973 exercises performed (mean 368); the number of games played was significantly correlated with the duration of play (Spearman rho = 0.951; *p* < 0.001).


**Table 1.** Demographic data for trained and untrained participants (CI = cochlear implant; HA = hearing aid; RE = right ear; LE = left ear; SNHL = sensori-neural hearing loss).


**Table 1.** *Cont.*

Before the intervention, the initial results from the Fr-matrix assessments were not significantly different between the study and control groups (t = 0.688 with 26 degrees of freedom; *p* = 0.49). Mean age differed between the study and control groups (*t*-test, *p* = 0.039). Age at testing, however, was not correlated with SRT70 improvement between T1 and T2 (Pearson test, *p* = 0.525). Moreover, neither deafness duration prior to implant (*t*-test, *p* = 0.449) nor the number of years of implant experience (Wilcoxon test, *p* = 0.487) differed between groups. Further, SRT70 improvement between T1 and T2 did not correlate with deafness duration (Pearson test, *p* = 0.071) nor with CI experience (Spearman test, *p* = 0.360).

In the control group, the mean difference in SRT70 between T1 (12.66 dB) and T2 (11.60 dB) was not significant (t14df-test = 0.655; *p* = 0.523, Table 2).

In the study group, a significant difference in speech-in-noise intelligibility was found between pre- and post-test assessments. The mean SRT70 in the study group was 15.5 dB at T1, and 11.5 dB at T2 (t12df-test = 4.521; *p* < 0.001; Figure 3). The mean SNR gain at SRT70 was −3.98 dB, with 6 of the 13 subjects evaluated having gained at least −4 dB SNR (Median = −2.8 dB SNR). All trained subjects improved their hearing abilities in noise, with decreased SRT70 after training, except Patient 5 (a 70-year-old male with 1 year of CI experience) whose SRT70 remained stable post-training (Table 2; Figure 3). The largest reduction in SRT70 was −10.2 dB SNR (Patient 12). Changes in SRT70 between T1 and T2 were not correlated with the number of games played (Spearman rho = −0.130; *p* = 0.693) nor with the total duration of play (Spearman rho = 0.033; *p* = 0.915).


**Table 2.** Individual and mean signal-to-noise ratio (SNR) results from Fr-matrix for the study and control groups at enrollment (T1), at 5 weeks (T2), and, for the study group, 5 weeks postintervention (T3).

**Figure 3.** Changes over time in signal-to-noise ratio (dB) at a 70% speech reception threshold. Individual results are shown on the left and mean group results on the right in the study group (**top panel**) and control group (**bottom panel**); testing at enrollment (yellow) and at 5 weeks (blue). The difference is significant only in the study group (noted \*).

All 13 participants in the study group responded 'Yes' to the two questions in the exit interview, i.e., "Did you enjoy the training program?" and "Did the training improve your listening-in-noise skills?".

At T3, eight out of the nine re-tested subjects still presented a decrease in SRT70 compared to T1, and the mean difference between T1 and T3 was of −2.28 dB. The mean SRT70 difference between T2 and T3 was +1.13, ranging from −4.0 in Patient 9 to +10.2 in Patient 8. Only one patient (Patient 5) did not show an overall improvement between T1 and T3 (Figure 4).

**Figure 4.** Changes over time in signal-to-noise ratio (dB) at a 70% speech reception threshold for nine subjects of the study group, 5 weeks after serious game-based AT (green).

#### **4. Discussion**

This study provides evidence of the impact of serious gaming on speech-in-noise intelligibility in adult CI users.

The Fr-matrix SRT70 was used as a measure of speech-in-noise intelligibility for assessing the effectiveness of a 5-week AT and its persistence. To remain as close as possible to real-life listening situations, the training assessment was performed using sentences and informational masking noise. Our group previously reported that, among speech-innoise tests suitable for French-speaking populations, the Fr-matrix provides the lowest intra-subject variability (±0.6 dB for SRT50) [34,35].

Herein, the post-training improvement in SRT70 was measured at a mean of −3.98 dB, a result that cannot be attributed to either intra-individual variation or to procedural learning alone. The latter is, in fact, evaluated at 1.8 dB for the Fr-matrix test [33]. Moreover, the improvement in SRT70 was observed in 12 of the 13 trained subjects. In the patient who did not improve, the SRT70 degradation was minimal (+0.1 dB SNR). Conversely, the control group did not show an overall improvement. More precisely, eight subjects from the control group showed an improvement in SRT70 ranging from −0.9 to −13.3 dB SNR (mean −5.2 dB SNR), while seven showed a degradation ranging from +0.5 to +14.6 dB SNR (mean +3.6 dB SNR). Even when excluding the control patient with the highest SRT70 degradation after 5 weeks (+14.6 dB SNR), the mean SRT70 values after 5 weeks were still not significantly different from those measured initially (t13df = 1.733; *p* = 0.107). Among the participants' demographic characteristics, only mean age differed between the study and control groups. None of the demographic characteristics, including deafness duration and experience with the implant, were found to correlate with improvement in SRT70.

In the nine subjects of the study group re-tested 5 weeks after the end of the intervention, only one had a worse SRT70 than before training (difference T3-T1 = +6.2 dB SNR). For the other eight patients, the SRT70 remained better than before training: three subjects had a gain of between −0.5 and −2 dB SNR and five maintained a gain of greater than −3 dB. However, the mean difference in SRT70 (−2.28 dB SNR) measured between

inclusion and re-test at 10 weeks was not significant. To date, only one study has measured the persistence of the efficacy of computer-assisted AT on speech-in-noise intelligibility in CI users [16]. These authors showed that, in 10 adult CI subjects, the benefit of AT on SRT50 could be observed up to 4 weeks after the end of the training with a gain of 2dB SNR. Future studies should more systematically integrate follow-up evaluation sessions to assess the long-term benefits of AT [36].

The serious game we used was developed specifically for this study. The software and its content had not been subject to a previous validation study. During the procedure, participants performed one training session face-to-face in the laboratory each week to ensure that the game's instructions were understood and well-followed during training, and to collect the user's experience over the previous week. The rest of the training was carried out remotely via the online gaming platform. In order to preserve the playful nature of AT, the duration of the training, the choice of activities among the six available options, and the initial difficulty level were left to the participant's will. However, an adaptive training procedure was used, in order to minimize the potential effect of inter-individual differences in initial SNR values.

Each participant was instructed to do a minimum of four training sessions per week, which was the case for each of them. The number of games played per session, however, was left up to the players in order to encourage their adherence. The relationship between the magnitude of improvement and the cumulative duration, in hours, over the 5 weeks of training could be assessed, since training logs were collected. Although the duration of training was highly variable between subjects, it was not associated with SRT70 improvement. The patient who participated the most showed an improvement at T2 (−6.9 dB SNR) compared to T1, which was higher than the mean SRT70 improvement. However, other subjects with less total training time (Patients 1, 10, and 12) showed a higher improvement (−7.9, −7.4, and −10.2, respectively) even though they had completed fewer games than the mean number of games played (336, 162, and 141, respectively). Furthermore, the patient with the highest improvement was the one who played the least. This result indicates that, while training had an overall beneficial effect and was measurable in almost all participants, there were large inter-individual disparities in the magnitude of SRT70 improvement, which prevailed over the total training time. While a weekly training schedule was set in the present study, only one study, to our knowledge, has evaluated the impact of AT schedule on speech recognition performance in degraded listening situations [37]. By training NH adults to recognize modulated vowels via a CI simulator, the authors did not find any influence of the pace of the training sessions on recognition improvement.

All or part of the inter-individual variability observed in speech-in-noise intelligibility improvement could be due to differences in the supraliminal abilities of the participants. Meta-analyses conducted in adult CI users provided evidence that demographic factors such as deafness duration or age at onset were predictive of CI outcomes, although they only explained 20% of the variance [38–40]. Furthermore, the sole SRT70 as a supraliminal measure does not account entirely for the patient's ability to recognize speech in noise. A recent meta-analysis identified the involvement of particular cognitive domains associated with speech-in-noise intelligibility, namely, processing speed, inhibitory control, working and episodic memory, and crystallized intelligence [41]. However, taken together, these cognitive abilities explain less than 10% of the inter-individual variability. A more recent review underlined the relationship between profound deafness of genetic origin and the occurrence of central auditory processing disorders in mice [42]. This is in full agreement with the fact that for a given degree of hearing loss, supraliminal auditory performance may considerably vary from one subject to another. Further studies on serious gamebased AT are needed in order to better control cognitive biases potentially affecting speech comprehension in noise.

**Author Contributions:** Conceptualization, H.T.-V., V.A., E.V.; methodology, E.V., H.T.-V.; software, C.B., V.A.; validation, P.R., E.V., H.T.-V.; formal analysis, P.R., C.-A.J., V.A.; investigation, V.A., P.R.; resources, P.R.; data curation, P.R.; writing—original draft preparation, P.R.; writing—review and

editing, C.-A.J., S.I., H.T.-V.; visualization, P.R., R.H.; supervision, H.T.-V.; project administration, H.T.-V.; funding acquisition, H.T.-V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study is part of a larger project (Neurosyllabic) that has been funded by a grant to Hung Thai-Van from the French Fonds Unique Interministériel. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board (local ethics committee (CPP Sud-Est IV 14/034 ID RCB 2014-A00345-42). The study protocol was registered on ClinicalTrials.gov (NCT02323256).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the subjects to publish this paper.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to ethical, legal and privacy issues.

**Acknowledgments:** The authors acknowledge the support to the Paris Hearing Institute from Fondation pour l'Audition (FPA IDA09). They would like to thank all the subjects who accepted to participate in this study, and are grateful to Verena Landel for carefully proofreading the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.
