Q: How do I reply to a reviewer's comment about my logistic regression analysis?

Detailed Question -

The sample size for my logistic regression analysis was 44896 people, and there are another 1146 people with missing value. Hence I removed 1146 people. Now the reviewer wants me to prove that removing 1146 people won't affect logistic regression analysis result. Now I have to add 1146 people to the data and run regression analysis again with original 44896 people. Then get the main results the same as previous. Do you think this will work?

3 Answers to this question
Answer: Hi, You need to carry out the analysis again with all the data from 44896 people. It is obvious that a slight variation in the results will be expected due to outliers / missing data. But, if your overall results remain near about the results obtained from previously, it should be easy to convince the reviewers. If your results with complete data from 44896 people drastically differs from the previous results (data for 43750 people) try to support the results with suitable arguments to convince the reviewers. I wish you good luck for this research publication.
Answer: The result should not differ, because multivariable analysis excludes anyway the patients with missing data for any variable introduced into the model.You should perform multiple imputation, which is considered the most correct way of dealing with missing data.

If you have at least one last value of the outcome (interim result if available) for the 1146 people with missing value, you can use that value for those people (as last value carried forward [LOCF] method) and repeat the analyses. If your results are similar to your main analysis, you can present these as sensitivity analyses/supportive analyses and write the response to the reviewer.

Since we don’t have a clear idea about your study and its results, it would be difficult to predict if the results would be similar or not. Even if we knew all the details, it would not be easy to predict the answer. The best way will be to try to re-run the analyses and check. If you need help with this, you might find some benefit in availing to professional publication support services, for example, Editage’s Statistical Review Service.

Alternatively, you can write to the reviewer saying the proportion of patients whose data were not used is <3% of the study population. Hence this should not affect the results of this study. If possible, you may enlist the reasons for dropout and explain whether or not it was related to the study conduct. Also you may present a baseline characteristics comparison of the persons included versus not included in the logistic regression analyses and see if there were any significant differences.

Related reading: