Cochlear implants (CI) have become the standard treatment for bilateral severe to profound hearing loss with over 30,000 recipients implanted per year worldwide. Cochlear implant (CI) processors must be appropriately programmed and customized for the recipient [
We’ve recently conducted a global survey to make an inventory of the current practice in CI fitting worldwide [
Hence, the authors believe that optimizing the process of CI fitting requires defining outcome measures and targets and adopting systematic approaches and algorithms to reach target. At present, there are no agreed standards or targets for both what should be adjusted, or the outcomes expected. Subjective loudness or other comfort measures are relevant, but it should be taken for granted that professionals in the field are aware of this and take care of this. Comfort as such can hardly suffice as target for such an intrusive and costly intervention as cochlear implantation. Placing an implant in the cochlea aims at taking over the function of this sensory organ and it seems obvious that any target should relate to a functional aspect of this organ. This function is the coding of sound and many features of this are well known. Psychoacoustic tests aim at testing the coding of these features in the clinic. Sound field audiograms provide a measure for the correct setting of MAP parameters and targets of 30 dB HL are used by many centers [
The Eargroup decided many years ago to use a fixed set of outcome measures to assess the state of the aided cochlea after implantation; this set of tests consists of tonal audiometry, speech audiometry, and two tests of the A§E psychoacoustical test battery (Otoconsult, Antwerp, Belgium), namely, the spectral discrimination and loudness scaling tests [
The software application Fitting to Outcomes eXpert (FOX) system, described in previous papers, introduced a systematic methodology to make adjustments to the MAP, based on the target outcomes from the A§E test battery [
The purpose of this study is to demonstrate the concept and feasibility of process optimisation by setting targets in a substantial set of psychoacoustic outcome measures and adopting a systematic methodology for reaching these preset targets in a large group of subjects, by more than one clinical centre.
A retrospective study was conducted to assess the results of computer assisted CI fitting in terms of a set of psychoacoustic outcome measures.
The data for 255 consecutive subjects fitted, almost all (
All the CI recipients were fitted by an experienced audiologist who was assisted by FOX according to the procedures outlined in Govaerts et al. [
Overview of the fitting procedure.
Session | Programming | Outcome measure |
---|---|---|
Switch-on | Auto MAPs loaded | None |
Session 2 (2 weeks) | Electrode deactivation (if required) | Impedance telemetry, free field audiometry |
Session 3 (4 weeks) | MAP optimization as recommended by FOX, but only if targets not reached | Free Field Audiometry, Phoneme Discrimination |
Session 4 (10–12 weeks) | MAP optimization as recommended by FOX, but only if targets not reached | Loudness scaling, speech audiometry |
Overview of outcome variables with value definitions for target and close to target.
Audiological test |
|
Outcome variable | Target | Almost on target | % on target at first | % on target at last | % almost on target at last |
---|---|---|---|---|---|---|---|
Audiometry | 255 | 250 Hz | ≤35 dB HL | ≤40 dB HL | 56 | 80 | 88 |
255 | 500 Hz | ≤30 dB HL | ≤40 dB HL | 71 | 84 | 92 | |
255 | 1000 Hz | ≤30 dB HL | ≤40 dB HL | 69 | 84 | 89 | |
255 | 2000 Hz | ≤30 dB HL | ≤40 dB HL | 64 | 85 | 90 | |
255 | 4000 Hz | ≤30 dB HL | ≤40 dB HL | 55 | 81 | 90 | |
255 | 8000 Hz | ≤30 dB HL | ≤40 dB HL | 55 | 77 | 89 | |
|
|||||||
Spectral discrimination |
102 | Set of 20 contrasts | ≥18/20 | ≥17/20 | 82 | 97 | 99 |
|
|||||||
Loudness scaling* | 177 | 250 Hz (30–40 dB SPL) | 1.1–2.8 | 0.8–3.1 | 47 | 71 | 76 |
178 | 250 Hz (45–55 dB SPL) | 1.9–3.6 | 1.6–3.9 | 62 | 82 | 88 | |
180 | 250 Hz (60–70 dB SPL) | 2.9–4.4 | 2.6–4.7 | 59 | 82 | 91 | |
182 | 250 Hz (75–85 dB SPL) | 4.1–5.8 | 3.7–6.1 | 42 | 70 | 90 | |
180 | 1000 Hz (30–40 dB SPL) | 1.2–2.3 | 0.9–2.6 | 58 | 76 | 81 | |
180 | 1000 Hz (45–55 dB SPL) | 1.9–2.9 | 1.6–3.2 | 49 | 73 | 87 | |
181 | 1000 Hz (60–70 dB SPL) | 2.7–3.7 | 2.4–4.0 | 45 | 67 | 83 | |
182 | 1000 Hz (75–85 dB SPL) | 3.4–5.1 | 3.1–5.4 | 75 | 88 | 90 | |
178 | 4000 Hz (30–40 dB SPL) | 0.6–2.1 | 0.3–2.4 | 71 | 90 | 94 | |
180 | 4000 Hz (45–55 dB SPL) | 1.3–2.7 | 1.0–2.4 | 41 | 67 | 80 | |
137 | 4000 Hz (60–70 dB SPL) | 1.9–3.4 | 1.6–3.7 | 37 | 56 | 68 | |
178 | 4000 Hz (75–85 dB SPL) | 2.6–4.2 | 2.3–4.5 | 46 | 60 | 80 | |
|
|||||||
Speech audiometry | 58 | Differential scores at 40 versus 55 dB SPL | −15–15% | −20–20% | 19 | 34 | 40 |
92 | Differential scores at 55 versus 70 dB SPL | −15–15% | −20–20% | 65 | 87 | 91 | |
89 | Differential scores at 70 versus 85 dB SPL | −15–15% | −20–20% | 81 | 94 | 96 |
The following outcome measures were used to assess the results.
This yielded 66 raw data points, some of which were grouped such that the final number was reduced to 22 outcome variables (listed in Table
For each of these 22 outcome variables, a target and near target for acceptable performance was defined as shown in Table
Descriptive statistics were used to present the results as histograms for THRs and box and whisker plots for SHRs. Nonparametric statistics were used to compare the initial and final THRs and SHRs (Wilcoxon paired rank tests) with a cut-off level of significance set at 0.05.
Sixty-six psychoacoustic points were measured to monitor the fitting in 255 consecutive CI recipients. Some results were grouped such that a total of 22 outcome variables were obtained to describe the “state” of the CI fitting process. For all variables a target was defined in a strict sense (on target) and a more tolerant sense (almost on target). Hence the state of the process was measured at two moments, marked as initial and final. The initial state refers to the first time that the outcome was measured, which is typically after the automated switch-on procedure. It therefore reflects the success rate of this start-up procedure. The final state is the last time the outcome was measured. Since all CI-recipients were fitted for target (by the audiologist assisted by FOX), this final state reflects the success rate of this fitting approach.
The THRs and tTHRs of all 22 outcome variables individually are shown in Figures
(a) shows the percentages of CI-users who performed on target (THR) at initial testing (black), final testing (gray), and almost on target (tTHR) at final testing (white) on Audiometry, A§E phoneme discrimination (A§E phoneme discrimin), and Speech Audiometry. Part (b) shows the interval between the initial and final measurement for those CI-users who did not reach target at the latest measurement.
The figure shows the results for Loudness Scaling at 250, 1000, and 4000 Hz. See Figure
The SHR results are shown in Figure
The figure shows the distribution of the success rates of 22 outcome variables (SHR) with the median value (central dot), quartile range (box) and range (whiskers).
Optimizing any process requires (1) a number of parameters to be adjusted within specific constraints, (2) quantitative or objective performance measures that need to reach predefined targets, and (3) a systematic approach with methods and algorithms rather than trial and error. When applying this to the process of CI fitting, the first requirement refers to the MAP parameters which can be modified by means of the CI programming software. The next 2 requirements are not obvious, as revealed by the global survey which has recently been conducted [
This report shows that the setting of well-defined outcome targets did allow a range of different centres to apply a systematic methodology to monitoring the quality of the programming provided. In an age where good clinical practice requires an evidence based approach, it is essential to have the ability to objectively monitor and audit the success of the treatment provided. The use of clear targets enables audiologists to define what is meant by an optimised MAP and provides consistency across different professionals and centres.
For outcomes to be effective they must be measureable in most clinical settings and reliably repeatable. They must also provide an accurate assessment of auditory performance and preferably be independent of the language spoken. The auditory system is complex and therefore requires a complex assessment system; one single measure is unlikely to be sufficient to provide all the information required. Like any other sensory organ, the cochlea is responsible for detecting its particular signal, sound, and for discriminating two sounds which differ in one of their components. In the absence of a global consensus on such targets, we have chosen the psychoacoustical targets as used in this study because we believe that, within the context of programming, they reflect well the state of the aided cochlea. They combine measures at the level of detection (audiometry), spectral discrimination (A§E phoneme discrimination) and identification (A§E loudness scaling and speech audiometry). They cover the coding of the sound features intensity and spectral content and most can be used for both adults and children, as they do not require a high cognitive or language level and are easy to implement across a wide range of centres. One can argue the choice of outcome variables and with this paper we do not intend to state that the variables chosen here compose the best selection. We do intend to open the debate and to make the point that a consensus would be very helpful in moving forward the discussion on the quality of CI fitting. The targets and “tolerant” targets set for each outcome were empirical though educated choices. The audiometrical targets were set at 30 db HL since this is close to the technological limit of the current CI devices, which is defined by a combination of the microphone sensitivity, the front end preprocessing and filter bank steps and the internal noise of the electronic circuitry [
Once these targets were set, we introduced a systematic approach to change the MAPs based on the outcome obtained. The FOX computer assisted programming system provides such a systematic approach across centres. All the centres included were able to use the FOX system effectively and to perform the tests required. In this study, the use of the FOX system significantly improved the audiologists’ ability to achieve the target outcomes set at the beginning of the study. The initial switch-on MAPs provided were solely based on the statistical derivation of T, M and Gain levels, with incremental increases in T and M levels applied as the MAP number used was increased. With these MAPs, more than half (57%) of the 22 targets were already achieved before any further optimisation took place. Once the FOX system was applied and optimisation began, there was a significant 24% increase in the number of target achieved, as measured at the last fitting session. Also, the spread of SHR across subjects decreased from 66% initially (range 16–82%) to 41% finally (range 56–97%) or even 31% if near to target scores are also tolerated (tSHR range 68–99%). This indicated that the approach under study is capable of delivering robust results across different CI recipients from different CI centres.
The approach of systematic fitting for target also allows looking at and interpreting the individual results for each outcome measure (THR). For instance, FOX was able to improve the THR for free field audiometry outcomes by a minimum of 13% and by as much as 26% at 4000 Hz and 24% at 250 Hz. Although this measure merely reflects the front-end technological capacity of the device, it is remarkable to see that it still requires customized programming to achieve good results in every individual. The results for phoneme discrimination were good without any optimisation, with 82% achieving target. This is in line with previous reports on the use of FOX [
For speech audiometry equivalent performance across presentation levels is considered to be an area where correct fitting of the device can directly impact performance. For the un-optimised Auto MAPs, at what could be considered to be the key intensity comparison of 55 and 70 dB levels, the THR with Auto MAPs was 65%. This was improved with FOX optimisation to 87%. However, the THR for the 40 and 55 dB comparison remained low at 40%, even after optimisation. This is in line with previous reports showing that speech intelligibility at these quiet levels is very challenging [
The loudness scaling targets were harder to achieve with the Auto MAPs, with the 4000 Hz frequencies being the most difficult. This was again also shown in the smaller sample reported by Vaerenberg et al. [
Results were based on the last available measurement and not the measurement when the fitting was considered to be optimal. Typically, a reasonable interval to achieve optimum would be around one month for audiometry and spectral discrimination and six months for loudness scaling and speech audiometry. Therefore, for some measures, the interval between the initial and final measurements was much less than that ideally required and if optimisation was continued, then further improvements in the percentage of subjects achieving target could be expected.
A final consideration is on the validity of the outcome variables chosen. Although the authors feel that the main justification for the current selection of outcome variables lies in the fundamentals of sound coding by the cochlea, as argued above, the ultimate proof of their validity will come from better speech understanding in quiet and in noise. It is beyond the scope of this paper to further address this issue, but two studies have been conducted in other centres than the Eargroup where speech understanding in quiet and in noise has been analyzed after conventional fitting compared to computer assisted and target driven fitting. The first study was conducted at MHH (Hannover, Germany), where 10 long term CI users who had always been fitted in the conventional way entered a single FOX iteration based on the 66 measured outcome points (Buechner et al., submitted). Speech audiometry with monosyllabic words in quiet improved instantaneously in 7 CI recipients and deteriorated in 3. Speech audiometry with sentences in noise improved instantaneously in 6 CI recipients and deteriorated in 4. Another Sentence Test with Adaptive Randomized Roving levels (STARR test) [
This study demonstrates that it is feasible to set targets and to report on the effectiveness of a fitting strategy in terms of these targets. This is demonstrated with the FOX-assisted strategy as example. The psychoacoustical measures chosen here were selected because they measure the behavioural response to acoustic stimulation, both in terms of loudness and frequency, and provide the building blocks for eventual speech perception and language development. This study also demonstrates that the application of the FOX system provides an effective tool for achieving a systematic approach to programming, allowing for better optimisation of the MAPs, when measured by the set targets. When recipients used the automated MAPs provided at switch-on, more than half (57%) of the 22 targets were already achieved before any further optimisation took place. Once the FOX system was applied there was a significant 24% increase in the number of targets achieved. The setting of well-defined outcome targets allowed a range of different centres to successfully apply a systematic methodology to monitoring the quality of the programming provided.
The last author Paul J. Govaerts received royalties from Advanced Bionics on the use of FOX in AB’s products. There are no other conflict of interests regarding the publication of this paper.
The authors gratefully acknowledge the following CI centres who contributed data to the FOX database: University La Sapienza (Rome, Italy), Medical University Hannover (Hannover, Germany), Yorkshire CI Service (Bradford, UK), UMS Radboud (Nijmegen, The Netherlands), Guys and St Thomas Hospital (London, UK), Hôpital Beaujon (Paris, France), RIFMED (Casablanca, Maroc), Hôpital Sacré Coeur (Beirut, Lebanon), Hôpital Avicenne (Paris, France), University Hospital Federico II (Naples, Italy), Midlands Adult CI Program (Birmingham, UK), Hörtherapiezentrum (Potsdam, Germany), MERF (Chennai, India), and Mesiarc (Calicut, India). Bart Vaerenberg received a Ph.D. Grant for this work from the IWT (Agency for Innovation by Science and Technology, Baekeland-mandaat IWT090287). Part of the work is supported by a 7th Framework for SME Grant (Opti-Fox) from the European Commission.