Can I use ordinal logistic regression to develop a propensity score?

[jemima spencer]

Hi,

Could someone kindly help me to develop a propensity score for an ordinal treatment variable?

(Apologies, I have had to rewrite my STATA code as I'm working on a secure database)

I am looking at whether degree of kidney impairment (ckdstage - my ordinal variable) affects odds of receipt of coronary angiography (binary) after a heart attack (MI). I want to compare multivariable and propensity score-adjusted models.


My multivariable model is:

logistic angiography i.ckdstage i.age_grp gender obese smoking ethnic.....


To develop a propensity score, I initially tried the following:

logistic ckdstage i.age_grp gender obese smoking ethnic (several more variables)

predict propensity


I think this is wrong, and did some reading, then tried using ordinal logistic regression as follows:

ologit ckdstage i.age_grp gender obese smoking ethnic (several more variables)

predict propensity

graph tw kdensity propensity if ckdstage==0 || kdensity propensity if ckdstage==1 || kdensity propensity if ckdstage==2 || kdensity propensity if ckdstage==3


The ologit regression appeared to work, but I can't get a graph and can't find a resource on the internet to show me where I might be going wrong. Could someone possibly suggest where I am going wrong?

Thanks so much

Jemima

[Rich Goldstein]

I think that the problem is that you don't have a variable called "propensity" after your -ologit- result; note that "propensity" is a stub for variable names after -ologit-, not the variable name itself (of course, unless you dropped the variable called propensity after your -logistic- model, you will have such a variable but it won't be what you expect); note that questions like this are much easier to answer if you provide a -dataex- data example and use CODE blocks (please read the FAQ);

also note that because you will have several "propensity*" variables, you need to be careful when graphing and when interpreting the graph

[jemima spencer]

Thanks so much for replying.

I can't unfortunately use -dataex- as I'm using a secure intranet site for my STATA work and cannot export directly from there. Apologies for the code not being too clear.

I've tried the following, but had to quit STATA on iteration no 54 so I assume I must still be doing something wrong! Could you possibly point out my error(s)?

ologit propensity_NST i.ckdstage i.age_grp gender obese smoking ethnic (several more variables)


Following a bit more reading, I wondered if I could just use "pscore" but I can't now find the command in STATA. I wonder if it is a historic one?

Thanks again

Jemima