CI calculation added for odds ratio

[runjin326]

Purpose/implementation Section

What scientific question is your analysis addressing?

Add confidence interval calculation for odds ratio in interaction-plots module

What was your approach?

Three columns - standard_error_or, or_ci_lower_bound and or_ci_upper_bound were added to the results column with the following calculation -

      standard_error_or = sqrt(1/mut11 + 1/mut00 + 1/mut10 + 1/mut01),
      or_ci_lower_bound = exp(log(odds_ratio) - 1.96 * standard_error_or),
      or_ci_upper_bound = exp(log(odds_ratio) + 1.96 * standard_error_or)

What GitHub issue does your pull request address?

#1203

Directions for reviewers. Tell potential reviewers what kind of feedback you are soliciting.

Which areas should receive a particularly close look?

Please check and see the addition of CI related columns are as wanted.

Is there anything that you want to discuss further?

No.

Is the analysis in a mature enough form that the resulting figure(s) and/or table(s) are ready for review?

Yes.

Results

What types of results are included (e.g., table, figure)?

All the *.tsv files in the results folder are updated with addition of CI related columns.

What is your summary of the results?

Spot checked a couple entries and the numbers make sense.

Reproducibility Checklist

• The dependencies required to run the code in this pull request have been added to the project Dockerfile.
• This analysis has been added to continuous integration.

Documentation Checklist

• This analysis module has a README and it is up to date.
• This analysis is recorded in the table in analyses/README.md and the entry is up to date.
• The analytical code is documented and contains comments.

[jashapiro]

Hi @runjin326-
Thank you for implementing this!

Unfortunately, the simple OR CI calculation that you implemented here is probably not quite what we will want, as it results in Inf/NA any time we have a 0 count. There are two potential solutions that I can think of:

One is to implement Haldane's Correction when there are zero counts: This would mean adding 0.5 to all counts in the table if any counts are zero, then performing the same standard error and CI calculation as before. This is probably the easiest to implement, but as it starts to get a bit messier, I would pull it out as a separate function, similar to the row_fisher() function. One note here is that the CI calculated this way starts from the "corrected" OR, which means that we often end up with an OR that does not include 0, so I would probably set the lower bound value to min(or_uncorrected, or_lowerbound)

Another option is to use the PropCIs::orscoreci() function, which is probably more accurate (it uses a method from Agretsi), but would involve a more backend work, as it requires adding a new package to the docker image.

[runjin326]

@jashapiro, thanks so much for your prompt review - I would probably take the first approach and ping you for another review when completed.

[runjin326]

@jashapiro, I pulled out function for CI and now the results only have one ci column with the calculation that you specified. Could you please check to see whether this is what you expect? Thanks!

[jashapiro]

Hi @runjin326 I saw that some of the intervals were still not making sense, which I think has to do with when the function is being applied. I also modified the code to more accurately reflect the full adjustment as it has to be applied (if odd_ratio is zero, you can't take the log, which is part of the reason for this procedure).

I have not tested the code I wrote, but hopefully it works, or does so mostly!

[runjin326]

@jashapiro , thanks so much for the detailed review and instructions. I fixed the scripts based on your suggestion - please check to see whether this now looks good.

[jashapiro]

LGTM! I spot checked some of the confidence intervals, and they seem correct to me.