mid-p for p-value and confidence interval

[Joanna ericsson]

Hi,

Im using Stata/SE 17.0.
I've calculated incidence ratios w/confidence intervals using the "iri" command, but noticed that the default for calculating CI is fishers exact and the default for p-values is mid-p. Because of this the result in Stata is a significant p but the CI crosses 1, thus I would like to do both calculations with the same method. There is an option for changing the p-calculation to fishers exact, but I believe mid-p would be more correct for calculating both the p-value and CI in this case.

Anyone know if its possible to change the method for calculating CI with the iri command? Se sample of the data below.

Thanks in advance,

iri 42 18 1187 894

Incidence-rate comparison

| Exposed Unexposed | Total
-----------------+------------------------+-----------
Cases | 42 18 | 60
Person-time | 1187 894 | 2081
-----------------+------------------------+-----------
| |
Incidence rate | .0353833 .0201342 | .0288323
| |
| Point estimate | [95% conf. interval]
|------------------------+------------------------
Inc. rate diff. | .0152491 | .0010707 .0294274
Inc. rate ratio | 1.757372 | .9899295 3.24341 (exact)
Attr. frac. ex. | .4309684 | -.010173 .6916825 (exact)
Attr. frac. pop | .3016779 |
+-------------------------------------------------
Mid p-values for tests of incidence-rate difference:
Adj Pr(Exposed cases <= 42) = 0.9793 (lower one-sided)
Adj Pr(Exposed cases >= 42) = 0.0207 (upper one-sided)
Two-sided p-value = 0.0414

[Knut Stavem]

Hi,

We had the same problem, but could not find a solution in Stata SE 17.0.

You can get the mid-p CI using a calculator in Openepi here, comparing two person-rates:
https://www.openepi.com/PersonTime2/PersonTime2.htm