Clean up Proportion

[statasaurus]

• CI for Single Proportions
• CI for Two Match Proportions
• CI for Two Independent Proportions
• Move CI risk difference from CMH into its own section
• CI for odds ratio and risk ratios

[petelaud]

Hi - I came to the CAMIS issues page to make a suggestion for the page about proportions, and thought it would be neater to add it to this existing issue. I'd like to make you aware of my package 'ratesci' (published on CRAN), which covers all of the above topics, as well as analysis of Poisson exposure-adjusted incidence rates, which is not well served by SAS or other packages.

In particular, it includes the pairbinci() function for paired proportions, which the CAMIS comparison table (at https://psiaims.github.io/CAMIS/Comp/r-sas_ci_for_prop.html) incorrectly says is not supported in R.

Also, more generally, ratesci provides one-sided p-values to accompany the confidence intervals, for use in non-inferiority hypothesis testing.

I realise it might be impractical to include all available packages in such an evaluation - it may be that the package has been considered for inclusion but rejected for some reason (if so, I'd be interested to know why). Happy to provide further input to the project if that would be helpful.

Thanks,
Pete

[statasaurus]

We would love to have your input. Thank you for catching this. If you want to put in a quick PR happy to have the update right away, you can do that. Or I am trying to get time to do this clean up and split out the sections a bit. But as you can see the issue is from September, I have been struggling with time. I should have it updated by the end of this month and I can included in then if you don't want to do a PR.

[DrLynTaylor]

Let's continue the discussion here. @petelaud , the table of contents now has rows for the below. I can create additional for EAIR if you prefer to split up further. If you are happy to create the additional files (splitting up the content) & update stat_method_tbl.csv to point to the new ones, then i'll pull things in for you. Christina is currently busy with project work, so I have more time :)
CI for Single Proportions
CI for Two Independent Proportions (incl risk difference, risk ratio, odds ratio)
CI for Two Paired Proportions (incl risk difference, risk ratio, odds ratio)
CI for Proportions in Stratified Designs (with cross-reference to CMH)
Could also add:
CI for Poisson exposure adjusted incidence rates

[petelaud]

Thanks Lyn. Just checking, as I'm new to collaborating on GH - for further edits, is it best practice to create an additional branch for each new pull request, or am I OK to continue working in my existing branch now that my PR has been accepted? (The 'Get started' page says 'You can then delete your branch in your forked repository and sync any other branches with the upstream CAMIS repository', but instructions on how to do so might be helpful.)

[DrLynTaylor]

Hi Pete, I'm pretty new to github myself (which I why I lean on @statasaurus a lot for github help!). However, I think it's best practice to create a new branch for each pull request. This ensures you branch is up to date with latest repository changes, and means you are less likely to have conflicts. Good feedback about 'how' to delete your branch, i'll open an issue so we remember to update the guidance. Thanks!

[DrLynTaylor]

@petelaud - I wonder if you can help me as I think I may have mis-understood something. I was just doing some testing using scoreci.sas and {ratesci::scoreci} to try to replicate the scenario where we have 2 treatment groups, Number of subjects in each group with 'Any AE' (n1=45, n0=50), time at risk of AE (exposure) in each treatment group (e1=13 patient years, e0=12 patient years), I want to derive the EAIR and 95% CI using Miettinen nurminen method for poisson risk difference [RD]. I can put this data in {ratesci::scoreci} and times results by 100, which results in: EAIR difference (95%CI) of -70.51(-229.65,83.44). CI looks wide, but EAIR logical for Any AE (high rate of events / subject exposed).
rcode:
eair <- scoreci(
x1 = 45, #events treatment 1
n1 = 13, #exposure treatment 1
x2 = 50, #events treatment2
n2 = 12, #exposure treatment 2
distrib = "poi", #poisson
contrast = "RD", #risk difference
level = 0.95, #alpha for CI (95%CI)
skew = FALSE, #FALSE indicates no skewness correction for miettinen-nurminen method
)

However, when I try to do the same using SAS, it gives me a warning?
data lyn;
n1=45; n0=50; e1=13; e0=12;
run;

V2.0 macro;
%SCORECI( DS=lyn,
CONTRAST = RD,
DISTRIB = poi,
delta=.,
LEVEL = 0.95,
SKEW = FALSE,
stratify=FALSE,
BCF=TRUE
);
Any ideas why and what i'm doing wrong? This was the error which repeats several times
NOTE: Invalid argument to function TAN(-1.570796327) at line 46 column 20.
NOTE: Invalid argument to function TAN(1.5707963268) at line 46 column 110.
N_1=45 N_0=50 e_1=13 e_0=12 row=1 okrow=1 delta=. level=0.95 alph=0.05 W_1=1 WR11=0.2888888889 WR01=0.24 MV1=0.0111111111 MU31=0.0000246914 DENS1=0.0111111111 wmu31=0.0000246914 DIFF1=0.0488888889 STHETA1=0.0488888889 WSTHETA1=0.0488888889 skew=FALSE z=0 pi=3.1415926536 ITER=4 COUNT=1 D1=-1 D2=1 incr=2 PERR=. ERR=-0.463800723 skipt=0 D=0 Zminus1=7071067.8119 ZOBS=0.4638007235 Zplus1=-6708203.932 Zzero=0.4638007235 D1t=. D2t=. incrt=. pt_est=. pt_back=-1 LL=. UL=. uc_est=. Qstat1=. i=2 HOMOGQ=. HOMOGP=. NT=95 CT=25 R1=0.2888888889 R0=0.24 lambda=1.0106382979 Dp=0 A=95 B=-25 C=0 MR0=0.2631578947 MR1=0.2631578947 Dt=0 SUMW=1 R1S=0.2888888889 R0S=0.24 SDIFF=0.0488888889 SSTHETA=0.0488888889 DEN=0.0111111111 score1=0.4638007235 scterm=0.0035136418 AA=. BB=. CC=. num=. dsct=. ERROR=1 N=1

PS: I am very much looking forward to getting a page into CAMIS for such analysis & showing vs SAS replication :)

[petelaud]

Thanks for flagging this up. You seem to have transposed the e's (# events) and n's (# subjects or exposure time - I accept this is a bit confusing), but that's not the main problem here.
SAS struggles with evaluating the TAN function at values too close to +/-pi/2, and it seems the solution I'd implemented might depend on local system settings (or possibly SAS version change) - it worked on my system at the time I developed the test code, but it isn't working now.

As a quick fix, at the lines:

%else %if "&distrib." = "poi" %then %do;
      D1t = round(tan(pi * max((-1 + 1E-15), min(D1, (1 - 1E-15))) / 2), &converge./100);
      D2t = round(tan(pi * max((-1 + 1E-15), min(D2, (1 - 1E-15))) / 2), &converge./100);
%end;

changing 1E-15 to 1E-9 seems to fix the issue and produce a result that matches the R function (let me know if that works for you).

I'll try to look into it in some more detail later - please raise an issue at https://github.com/petelaud/ratesci-sas/issues.

[DrLynTaylor]

Hi Pete, Thanks so much for the quick reply. This is super helpful. I changed the E-9 and all agrees now and fully aligned with R. I'll raise the issue on the ratesci-sas issues page.

[petelaud]

Hi again, just noticing that there are comments in the proportion CI page about an intention (presumably from @statasaurus ?) to add details about the cicalc package 'soon'. As I'm making extensive structural changes to this page, please could you hold off adding that information until I've made the next PR? (I'm not sure what happens with branches if 2 people are working on the same file at the same time, but I guess it could get messy because I'm splitting it into multiple different files.)

[statasaurus]

Totally! I am fighting with the GitHub actions and deployment right now, so that works for me. Thank you for doing this

[DrLynTaylor]

HI @petelaud just wondered how you were getting on? I love the content you added for CI's so far. I could add something basic for Poisson EAIR rows, and stratified Miettinen nurminen using your ratesci() package and scoreci() sas macro (+ cicalc() and others), but if you are already still working on this, then I know you'd write it better than me! So i'll wait until I hear from you. Thanks!

[petelaud]

Hi, I'm really sorry I've had to put this on the back-burner for a while. I've sent you an email to explain, hopefully I have the right address. I'm definitely intending to get back to adding more content soon.