turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- How to do subgroup analysis

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

10-02-2012 02:29 AM

Dear all,

I was asked to do subgroup analyses, which I have never done before.:smileyconfused: I was wondering if anyone can help?

Outcome of interest:carotid intima-media thickness (IMT) -continuous variable

Adjusted variables : age, sex, bmi, sbp, ldl, NO2, DM (we also did multivariate linear regression)

Question 1. subgroup analyses for 2 groups: 50yr (0=younger than 50, 1= older than 50)

Question 2. Can we include interaction term in subgroup analysis? ex. 50yr * NO2

Thanks,

Chi

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

10-02-2012 09:12 AM

Basic subgroup analyses are done by sorting the data by the grouping variable and then using the BY statement on the PROC. Another method, if only one subgroup is of interest, is to use the WHERE clause in the DATA= option of the PROC.

The subgrouping variable cannot be used in an interaction.

There are two general philosophies of analysis, lumping and splitting. Subgroup analysis is a form of splitting. Many statisticians (I'm one), prefer lumping. I prefer it because it increases power and I think that I can interpret one model with interactions more easily than interpreting two separate models. OTOH, I work with some epidemiologists who generally prefer to think about things as subgroups.

Doc Muhlbaier

Duke

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

10-03-2012 07:43 AM

And it seems that FDA reviewers are in LOVE with subgroups, given their fascination with homogeneity/heterogeneity of effect sizes in the subgroups. I suppose I understand it when it comes to writing labeling, but **YOU CAN ANSWER THE SAME QUESTION, WITH GREATER POWER TO DETECT DIFFERENCES IF YOU MODEL THE DATA WITH INTERACTIONS. **

Thanks, Doc, thought I was alone in the desert on this one.

Steve Denham