Meta-Analysis—Correlation between Spiral Ganglion Cell Counts and Speech Perception with a Cochlear Implant
Round 1
Reviewer 1 Report
Dear authors,
Greetings for the article.
The article is well defined, structured and easily readable.
Methodology and search strategy are well set and explained, according to PRISMA protocol.
I have only one question: Why did you choose 1946 as start date?
Discussion and post-hoc analysis are also well argued.
Author Response
Thank you.
The earliest scientific journal articles from Medline database dates back to 1946. We have now clarified that our searches included articles from the beginning of each database used.
More specifically, the search “was executed starting at the earliest journal articles in each database up to January of 2021 on PubMed/Medline, Biosis, Embase, and Web of Science.” Page 3 line 88
Reviewer 2 Report
Since the spiral ganglion cell count of 2800 is considered relevant in the included individuals I would ask that "2800" be marked with a thin line in Figure 2, otherwise the 9(10) individuals mentioned are not to be identified.
You mention 9 individuals with a cell count under 2800 (page 5 line 130) and later 10 individuals with a cell count under 2800 (page 5 line 147). Please clarify.
Author Response
Since the spiral ganglion cell count of 2800 is considered relevant in the included individuals I would ask that "2800" be marked with a thin line in Figure 2, otherwise the 9(10) individuals mentioned are not to be identified. Thank you. The new figure now has 2800 marked.
You mention 9 individuals with a cell count under 2800 (page 5 line 130) and later 10 individuals with a cell count under 2800 (page 5 line 147). Please clarify. Thank you for picking up this error, which has now been corrected. The correct number is 9 (Page 5 line 170).
Reviewer 3 Report
The authors performed a meta-analysis of published studies on the relation of SGC counts and speech perception scores after CI implantation in adults. A huge amount of published papers were screened and finally, no correlation was found.
The paper addresses an interesting topic but has some shortfalls (minor and major) that have to be addressed before publication. Unfortunately data analysis stopped already after the first step.
Line 58: SGC is used throughout the manuscript except here.
Line 63-64: Please provide the reference for the PRISMA-checklist
Line 71-75: It is very hard to follow. I guess, “scores were estimated” in line 71. You state, that CID and HINT were not directly correlated with NU-6 scores, but you use a correlation factor. A simple correlation factor would imply a direct correlation. Can you provide more information on this? It is the methods section and readers should be able to follow.
Line 83 (PRISMA flow chart): There is a mismatch between the flow diagram (164 out of 363 + 4) and the supplementary material (176 out of 363). Can you provide what other sources you used?
When you exclude 111 from 164, 53 should remain, not 51. What happened to the other two records?
Why qualitative and quantitative ‘synthesis’? Why the difference in numbers (14 vs 9)? It would also be helpful to provide an overview (table?) on the 14 or 9 studies being included (author, year, number of subjects, may be what type of additional data is available, etc…).
Line 103: What are the reasons for choosing 2800 and 3000 SGC for the calculations? Were there any scientific reasons? Currently, I do not see a scientific value.
Line 129: probably “above” instead of “about”
Line 147: Here, you state that 10 patients had SGC counts below 2800, in the results section (line 130) you mention 9 subjects. What is correct?
Line 158-164: You provide possible explanations for not having found a correlation and refer to explanations provided in the original studies such as small sample size. But one of the reasons to do a meta-analysis is to overcome some of these limitations. Therefore it should be discussed for your analysis why there was no correlation found.
Additionally, you stated in lines 89-92 that you also collected more information such as demographic information, type of implant and etiology. Why didn’t you work with it in your analysis? Especially with the earlier speech coding strategies, speech perception improved over years. Therefore also the duration of CI use (time from implantation to death) might have an influence on your results. This is only an example. You have the data, please work with it! With your larger dataset you might also be able to confirm or falsify some of the explanation provided in the different studies.
Line 331: This additional figure 1 is a low quality (old?) version of the figure that is included in the text and should be removed.
Author Response
Reviewer #3
The paper addresses an interesting topic but has some shortfalls (minor and major) that have to be addressed before publication. Unfortunately, data analysis stopped already after the first step.
Line 58: SGC is used throughout the manuscript except here. This has been corrected. Now SGC is used throughout the manuscript, without exceptions.
Line 63-64: Please provide the reference for the PRISMA-checklist. This has been done (see reference [26]).
Line 71-75: It is very hard to follow. I guess, “scores were estimated” in line 71. You state, that CID and HINT were not directly correlated with NU-6 scores, but you use a correlation factor. A simple correlation factor would imply a direct correlation. Can you provide more information on this? It is the methods section and readers should be able to follow.
Thank you. The score estimation was first described in Nadol et al. 2001, Khan et al 2005, Li et al 2007 and Kamakura et al 2016. We verified and applied the same estimation to 4 further ears from non-MEEI studies and 2 MEEI ears.
We have attempted to expand and clarify this section as follows:
In 5 subjects where CNC or NU-6 was not available, Nadol et al., Li et al., Khan et al. and Kamakura et al [6,7,10,28] reported estimated NU-6 or CNC scores. This was done based on the work of Rabinowitz et al. [29], who found that NU-6 word identification scores are highly predictable based on CUNY sentence identification scores (the Pearson correlation between predicted and measured scores was +0.93). Li et al., Khan et al. and Kamakura et al. based their predictions on scores from the HINT or CID tests rather than the CUNY sentence test, but these sentence tests are similar enough to make the predictions quite reasonable. In fact, a sensitivity analysis showed that NU-6 scores could be predicted within a 15 percentage point window based on HINT or CID scores. Thus, following the lead of these authors we applied the same estimation from CID sentence scores to four subjects from Xu et al. 2012 [15] and Ishiyama et al. 2019[4]. Subjects from Ishiyama et al 2019 who only had speech perception data from tests other than CNC, NU-6, or CID sentences were excluded from the analysis.
Line 83 (PRISMA flow chart): There is a mismatch between the flow diagram (164 out of 363 + 4) and the supplementary material (176 out of 363). Can you provide what other sources you used?
When you exclude 111 from 164, 53 should remain, not 51. What happened to the other two records?
Why qualitative and quantitative ‘synthesis’? Why the difference in numbers (14 vs 9)? It would also be helpful to provide an overview (table?) on the 14 or 9 studies being included (author, year, number of subjects, may be what type of additional data is available, etc…).
Thank you for identifying this error. The submitted diagram was an older version that was mistakenly included in the manuscript. This has been rectified.
Line 103: What are the reasons for choosing 2800 and 3000 SGC for the calculations? Were there any scientific reasons? Currently, I do not see a scientific value.
This was done to illustrate the point that a post-hoc selection of analysis methods (e.g., adding or removing analysis variables) or analysis methods (e.g., changing the threshold for this particular analysis from 2800 to 3000) can easily determine whether a result is deemed “significant” or “nonsignificant”.
To clarify this point in the manuscript itself we added the following sentence:
“This result illustrates the point that one should be wary of post-hoc analyses that were conducted after having seen the data, instead of being determined in advance and pre-registered.”
Line 129: probably “above” instead of “about” Thank you!
Line 147: Here, you state that 10 patients had SGC counts below 2800, in the results section (line 130) you mention 9 subjects. What is correct?
Thank you. This has been corrected.
Line 158-164: You provide possible explanations for not having found a correlation and refer to explanations provided in the original studies such as small sample size. But one of the reasons to do a meta-analysis is to overcome some of these limitations. Therefore it should be discussed for your analysis why there was no correlation found.
In our discussion section, we explored the possibility that the hypothesized correlation between WRS and SGC counts do not exist. That being highly unlikely, the alternative explanation (second paragraph of discussion) is that the relationship is non-linear and that a low threshold of SGC count may be sufficient for hearing with a CI. Possible reasons for failure to find a correlation are discussed in the Discussion section, where this passage has been expanded to clarify the point:
“One possible explanation for the absence of correlation in our analysis is that a small number of SGCs may be sufficient to deliver the information conveyed via electrical stimulation by a cochlear implant, and higher numbers of SGCs may not provide a substantial amount of additional speech information.”
Also in the Discussion section:
“Other possible explanations for the lack of correlation between SGCs and speech perception scores have been discussed in the studies included in our meta-analysis. These include varied etiology of hearing loss within the study population, and widely ranging intervals between time of last audiology test and death. Cognitive function at the time of testing is also not known, and it could have an impact on speech perception scores. Other known factors such as duration of bilateral deafness and compliance with using the CI every day are not taken into account either.”
Finally, the Conclusion section states:
“After examining all available published data, the present meta-analysis did not find a significant correlation between SGC counts and word recognition scores in CI users. This result is consistent with the possibility that a relatively small number of SGCs is sufficient to understand speech with a cochlear implant, and therefore the correlation in question does not exist for patients with SGC counts between 1000 and 25000 (the range found in this study). The absence of a correlation is also consistent with the possibility that any influence of SGC count on speech perception by CI users is obscured by other factors that are more strongly associated with speech perception outcomes.”
Additionally, you stated in lines 89-92 that you also collected more information such as demographic information, type of implant and etiology. Why didn’t you work with it in your analysis? Especially with the earlier speech coding strategies, speech perception improved over years. Therefore also the duration of CI use (time from implantation to death) might have an influence on your results. This is only an example. You have the data, please work with it! With your larger dataset you might also be able to confirm or falsify some of the explanation provided in the different studies.
We have since conducted several additional multivariate analyses and included it in the revised Methods section:
“In addition to the univariate analyses listed above, several multivariate analyses were conducted to see if a relation between speech perception scores and number of spiral ganglion cells was present when including other independent variables to the analysis. In total, six multivariate linear regressions were conducted with the corresponding independent variables being: number of spiral ganglion cells (#SGC) and age at implantation; #SGC and age at death; #SGC and duration of deafness; #SGC and duration of implantation; #SGC, age at implantation, and age at death; #SGC and age at implantation, age at death, and duration of deafness.”
Results are summarized in a new paragraph of the Results section:
“Lastly, none of the six multivariate linear regressions modified the findings observed in the univariate analysis. The only regression parameter whose point estimate was significantly different from zero, for any of these regressions, was the intercept. In particular, no significant relation between SGC count and speech perception scores was observed.”
Line 331: This additional figure 1 is a low quality (old?) version of the figure that is included in the text and should be removed.
An updated, higher resolution version of the PRISMA 2009 flow diagram (figure 1) has been included.
Round 2
Reviewer 2 Report
The proposed changes were implemented
Author Response
Thank you
Reviewer 3 Report
The paper has strongly improved. There are still two minor remarks:
- According to the flow chart, evaluation was based on refs 3-15. In the first sentence of the results section, you mention ref 28 instead of ref 5. Why? Please be consistent. Furthermore, if your intention was to reference these 13 studies, please do it earlier in the sentence. At the current place I would have expected the 9 refs for MEEI studies.
- Please provide the year of publication for ref 5.
Author Response
Thank you for drawing my attention to these issues.
The inconsistencies with references have now been corrected. References have been reviewed and updated now.
