Odds ratio and relative risk from procedure causaltrt

[lbrion]

This is a follow-up question regarding the poster "odds ratio from procedure causaltrt".

I am using SAS version 9.4.

I am trying to obtain OR and RR from proc causaltrt.

Here are my codes for causaltrt:

ods graphics on;

proc causaltrt data=nec METHOD=AIPW  covdiffps poutcomemod ppsmodel  nthreads=10 ;

class  INVASIVE (ref=first) multbirth (ref=first)  GAgrp4 (ref=first)   BM34 (ref=first)

sepsisNEC (ref=first)  group  (ref=first)  support  (ref=first) invasive  (ref=first)  ;

psmodel EarlyfeeddBM (ref=FIRST) =  multbirth support  group  gagrp4/plots=(pscovden weightcloud psdist lps) WGTFLAG=10  ;

model NECIII (event=last) = BM34 sepsisNEC invasive /dist=bin ;

bootstrap bootci(all) nboot=5000 seed=1234 plots=hist(effect)  ;

output out=out ps=ps ipw=AIPW;

ods output CausalEffects=ce;

run;

Using StatDave's recommendation I obtained negative values for the lower limit of CI for RR and OR see below.

Nonlinear Function Estimate

Label	Estimate	Standard Error	Wald Chi-Square	Pr > ChiSq	Alpha	Lower	Upper
Rel Risk of POMs	1.1374	0.6287	3.27310	0.0704	0.05	-0.09480	2.3697

Name	Estimate
B_p1	0.01810
B_p2	0.01592

Nonlinear Function Estimate

Label	Estimate	Standard Error	Wald Chi-Square	Pr > ChiSq	Alpha	Lower	Upper
Odds Ratio of POMs	1.1400	0.6411	3.16200	0.0754	0.05	-0.1165	2.3965

Therefore I cannot use this method for my data.

I would like to follow Michael's  recommendation but am unable to do so:

"Instead what you can do is use the potential outcome mean estimates that are reported to compute the odds ratio, i.e. (POM1*(1-POM0) )/ (POM0 * (1-POM1) ). To compute confidence limits for this estimate you could apply the delta method. The Analysis of Causal Effects table reports estimates for the standard deviation of POM1 and POM2 and you can obtain the estimate for their covariance by using the identity VAR(ATE)=VAR(POM1 - POM0 ) = VAR(POM1) + VAR(POM0)- 2*COVAR(POM1, POM2)."

Could you provide me the codes I need to use to  accomplish this?

thanks

[StatDave · Accepted solution]

In your original post on this, I first showed how to estimate the POM difference using the NLMeans macro, but then I showed code that implements Michael's suggestion. The delta method does not guarantee that the lower confidence limit will be nonnegative. In these cases, a negative lower limit is typically treated as zero.

[lbrion]

Thanks a lot for your response!

Much appreciated.

Luc

[lbrion]

Thanks a lot!

Luc