Predicting sequential bilateral cochlear implantation performance in postlingually deafened adults; A retrospective cohort study

To identify which preoperative patient characteristics influence sequential bilateral cochlear implantation performance and to create a statistical model that predicts benefit.


Conclusion:
Advanced age or a long interval between implantations does not necessarily lead to poor CI2 results. Patients who are successful HA users before CI2, who have a low PTA before CI2, a high CNC phoneme score with CI1 and a limited length of hearing loss before CI2, are likely to be successful CI2 recipients.

| INTRODUCTION
Bilateral cochlear implantation offers advantages over unilateral cochlear implantation in patients with bilateral profound hearing loss.
Bilateral implantation helps to restore sound localisation and improves hearing in noise and quality of life. [1][2][3][4][5][6][7][8] Cochlear implant (CI) teams need to decide which patients are likely to benefit from a second CI and which patients are not. In general, they will consider a patient's age, duration of deafness, cause of hearing loss, hearing aid (HA) use, the length of the interval between implantations, hearing results before CI2 and the performance level with the first CI when counselling patients whether a second CI would be successful. [9][10][11][12][13][14][15][16][17][18][19] The majority of literature on factors affecting CI outcomes is about unilateral implantation. In 2009, Roditi et al 20 presented a prediction model for unilateral CI performance in postlingually deafened adults based on duration of any hearing loss in the CI ear, preoperative speech understanding in quiet and the length of severe to profound hearing loss in either ear. With their model, they could predict 60% of the variance in postoperative consonant-nucleus-consonant (CNC) scores. Our research group recently performed a systematic review to determine whether similar factors play a role in the success of sequential bilateral implantation as in unilateral implantation. 21 We included ten papers on the effect of age, duration of hearing loss, time between implantations, preoperative hearing, aetiology of hearing loss, hearing aid use and duration of follow-up on sequential CI performance. 4,9,10,12,13,17,[22][23][24][25] Based on the best evidence available to date, advanced age, a long duration of deafness or a long interval between implantations does not necessarily lead to poorer sequential cochlear implantation outcome. The performance level with the first CI may be an important predictor for sequential implantation performance, but, to our knowledge, has only been examined in two studies. 10,26 Unfortunately, the included studies were heterogeneous, had relatively low sample sizes, and the influence of a certain prognostic factor was often a secondary outcome of the study. 21 It was therefore rather difficult to draw straightforward conclusions. 21 The aim of this study was to contribute to filling this gap in the existing literature. We retrospectively utilised a database with a large number of sequentially implanted adult CI recipients to determine which preoperative factors are related to sequential cochlear implantation outcome. This led to the development of a prediction model based on the factors that were significantly correlated to auditory performance with a second CI. Knowing which factors are related to sequential cochlear implantation outcome will help CI teams to more accurately counsel patients who are considering sequential implantation.

| Ethical consideration
The study was performed according to the principles expressed in the Declaration of Helsinki. The study was recognised as negligible risk and was granted an exemption from the Human Ethics Committee of the University of Western Australia (Reference number RA/4/1/8931).

| Study design and participants
This retrospective chart review was conducted within the CI audiology service managed by the Ear Science Institute Australia (ESIA). All March 2016 were included in this study. The study outcome measure was the 12-month CNC phoneme score (speech intelligibility in quiet at 65 dB SPL). Patients were only excluded if this measure was Keypoints • Advanced age or a long interval between implantations does not necessarily lead to poor CI2 results.
• Patients who are successful HA users before CI2, who have a low PTA before CI2, a high CNC phoneme score with CI1 and a limited length of hearing loss before CI2, are likely to be successful CI2 recipients.

SMULDERS ET AL.
| 1501 missing from the database and patient file. Two authors, YS and TH, verified whether the data in the database corresponded to the data in the patient records in the hospitals and adjusted the database if necessary. The data gathered were patients' gender, age of onset of any degree of hearing loss (age at which patients could remember their hearing loss started or when they started to use hearing aids), age at implantation of the first CI (CI1) and the second CI (CI2), side of first implantation, duration of deafness before CI1 and CI2 in each ear, interimplantation interval duration, origin of hearing loss for both ears, HA use, comorbidity expressed as the Charlson score (0 = no comorbidity, 24 = maximum comorbidity score) 27 and preoperative hearing details (pure tone average (0.5, 1, 2 kHz) in each individual ear and maximum speech intelligibility (CNC phoneme score) in each ear with and without HAs and with wearing CI1 only).

| Study outcome
The study outcome measure was the CNC phoneme score (%) with CI2. A full list of 25 words was presented in quiet at a fixed level of 65 dB SPL, from a speaker in front of the patient at 1 m distance. The outcome measure is the percentage phonemes repeated correctly.
The test was performed 12 months after the second implantation.

| Data analysis
The statistical analyses were performed with IBM SPSS Statistics for Windows version 24.0 (IBM Corp. Armonk, NY: IBM Corp.). The data were, overall, normally distributed, and means, standard deviation and ranges are displayed in the tables. We used a multiple imputation technique to account for the missing values in our database. Only 2.8% of the data were missing, including all patient characteristics (Table 1) and hearing test outcomes. Ten imputations were used. 28 To analyse which variables were correlated to the study outcome, we performed univariate linear regression analyses. A correlation R is considered very weak when R < 0.3, weak when R = 0.3-0.5, moderate when R = 0.5-0.7, strong when R = 0.7-0.85, very strong when R = 0.85-0.95 and extremely strong when R > 0.95. 29,30 Subsequently, we identified the variables that were significantly correlated to the outcome and entered these variables into a backward multiple linear regression analysis. This latter method analyses which factors are actual predictors for sequential cochlear implantation outcome and can be used to create a predictive model. The accuracy of the model is presented as the explained variance R 2 (<10% = very weak, 10%-25% = weak, 25%-50% = moderate, 50%-75% = strong, 75%-90% = very strong, >90% = extremely strong). 29,30 We will present the accuracy of the model based on the imputed data and based on the original data. The cause of hearing loss was extracted from all patient files. In many patients, a cause could be identified. However, when the cause was not clear, we described the progression of hearing loss, if known (eg, "sudden deafness," or "progressive hearing loss"). We    Table 2 shows the correlations between 18 preoperative factors and the study outcome. Only four factors correlated significantly with the postoperative CI2 CNC phoneme score. These factors were as follows: HA use before CI2 in this ear, length of hearing loss before CI2, preoperative pure tone average (PTA) before CI2 and the CNC phoneme score measured 12 months after CI1. We excluded "cause of hearing loss" from the analysis, because of the heterogeneity of this factor. Figure 2 shows the correlation between two predictive factors and CNC phoneme score for CI2.

| Predicting sequential CI outcome
Backward stepwise multiple regression analysis showed that three factors were significant contributors to predict the outcome of a sequential CI: Hearing aid use before CI2 in the second ear, the length of hearing loss before CI2 in the second ear and the CNC phoneme score with CI1 at 65 dB SPL after 12 months of unilateral CI experience.
We applied this model to the study population for internal validation. Figure 3 displays the predicted and the actual CNC phoneme scores with CI2. For the actual CNC phoneme score, the mean was 68% ± 22% (SD). For the predicted CNC phoneme score, the mean was 68% ± 12% (SD). The model based on the original data has a moderate accuracy of R = 59%, R 2 = 35%. The model based on the imputed data is R = 55%, R 2 = 30%.
The factor HA use appeared to play an important role; however, as it was based on only five patients, we repeated the analysis above after exclusion of this factor. Subsequently, the following equa- For the predicted CNC phoneme score, the mean was 67% ± 8% (SD). The model based on the original data has a weak accuracy of R = 38%, R 2 = 15%. The accuracy of the model based on the imputed data is R = 35%, R 2 = 12%.

| Synopsis
The aim of this study was to determine which preoperative factors affect performance with a second CI after sequential cochlear implantation and to create a mathematical model to predict speech intelligibility in postlingually deafened adult patients undergoing sequential cochlear implantation. This model was based on patient characteristics (Table 1) identified through retrospective chart review of included patients.
One of the key factors that appeared to determine the success of sequential bilateral cochlear implantation was wearing a HA before CI2, although only five of the 92 patients did not wear a HA before CI2. All five patients did not benefit from a HA, due to the severity of the hearing loss. This finding can be explained as follows: when a patient is a successful HA recipient before CI2, it is likely that he/she will perform well with an implant in that ear. This does not imply that every candidate for a second implant should wear a hearing aid; in some cases, it will have no benefit. As this factor appeared to play such an important role based on the results of only a small portion of the group, we created a second model after exclusion of the factor HA use before CI2. However, the accuracy of the second model was considerably lower than that of the original model.
Prolonged duration of hearing loss before CI2 was a negative predictor and a high CNC phoneme score with CI1 a positive predictor for sequential cochlear implantation performance. When we removed the HA factor from the regression analysis, preoperative PTA before CI2 also became an independent predictor for CI2 performance.
Our data also showed that several factors were not related to CI2 outcome, including patient's age, the length of the interval between implantations, the length of hearing loss before CI1 and a patient's comorbidity. This information is counterintuitive and is as valuable as knowing which factors are related to good or poor outcome.

| Strengths and weaknesses
A strength of this study is the high number of participants. A large study population is essential to perform a stepwise linear regression analysis and increases the internal validity of a study. Furthermore, the study has a low number of missing data and contains a large amount of information on each patient. We used a universally applied study outcome, which makes it possible to generalise our findings to other countries and studies. Literature has shown that bilateral cochlear implantation helps to restore sound localisation and improves hearing in noise. [1][2][3][4][5][6][7][8] Unfortunately, our patients did not undergo any specific binaural hearing tests. One may assume that better speech understanding in quiet in both ears will lead to better spatial hearing capabilities, but we could not prove this with the data available to us. Other weaknesses of the study are the retrospective design and that fact that the study was subject to selection bias.

| Comparison with the literature
A few other retrospective studies reviewed the influence of preoperative patient factors on sequential cochlear implantation outcome. In 2016, Boisvert et al 10 performed a study with 67 patients. They analysed the effect of six preoperative factors. As in our study, they found that the phoneme score with CI1 was an important, and in their study, the only significant predictor for performance with a second CI. In contrast to our findings, they did report a negative correlation between age and sequential cochlear implantation outcome, but all patients included were above the age of 50 years.
Other studies all had small sample sizes of 10-29 patients and reviewed a maximum of five different factors per study. 4,9,12,13,[23][24][25] There were several similar outcomes as in the current study.