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
- /
- Health Care and Pharma
- /
- SAS in Health Care Related Fields
- /
- HELP! Matched case-control study analysis

Topic Options

- 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
- RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-14-2007 08:36 AM

Hi guys... so glad I found this board.

I need help with a matched case-control study. I had an outcome of interest 'TMC' and matched 1 case of 'TMC' to 2 controls by age. Now, I want to analyse with SAS to see whether variables of interest are different (2-sided) in cases vs. controls. The data is non-parametric, non-symmetrical for the most part.

'TMC' is the condition (case=1, control=0), 'daysadmit' is the continuous variable

For the continuous variables, I used the following code:

**PROC GLM DATA = descriptive ; **

CLASS match tmc ;

MODEL daysadmit = tmc match(tmc);

MEANS tmc;

RUN;

Is this correct? Do I need to look at the p-value for match(tmc), or do I need that nested variable in there at all? Or, is sufficient to have 'match' in the class statement and only have daysadmit=tmc as the model?

Also, is there any value to putting in this statment:

TEST H = tmc E = match(tmc);

For the categorical variables I used:

**data binomial;**

set descriptive;

if TMC=0 then status=2;

if TMC=1 then status=1;

if TMC=1 then censor=1;

if TMC=0 then censor=0;

run;

proc phreg data=binomial;

model status*censor(0)= fever /ties=discrete risklimits;

strata match;

run;

Where I made status=1 controls, status=2 cases (of toxic megacolon), censor=0 controls, censor=1 cases.

The problem here is that the numbers are strange. I want to see whether the frequency of a bunch of variables is different in the case group from the control group. The the predictor variable listed above ('fever'), 90% of the case group ('TMC') had fever while 25% of the control group had fever. Using the above program, I get a p of 0.026. With leukocytosis 'leuko' as the predictor, 80% of the case group had leukocytosis while 5 % of the control group had leukocytosis. It looks like it should be significant, but I get a p value of 0.996.

**HELP! Can someone help me with what to do?**

I need help with a matched case-control study. I had an outcome of interest 'TMC' and matched 1 case of 'TMC' to 2 controls by age. Now, I want to analyse with SAS to see whether variables of interest are different (2-sided) in cases vs. controls. The data is non-parametric, non-symmetrical for the most part.

'TMC' is the condition (case=1, control=0), 'daysadmit' is the continuous variable

For the continuous variables, I used the following code:

CLASS match tmc ;

MODEL daysadmit = tmc match(tmc);

MEANS tmc;

RUN;

Is this correct? Do I need to look at the p-value for match(tmc), or do I need that nested variable in there at all? Or, is sufficient to have 'match' in the class statement and only have daysadmit=tmc as the model?

Also, is there any value to putting in this statment:

TEST H = tmc E = match(tmc);

For the categorical variables I used:

set descriptive;

if TMC=0 then status=2;

if TMC=1 then status=1;

if TMC=1 then censor=1;

if TMC=0 then censor=0;

run;

proc phreg data=binomial;

model status*censor(0)= fever /ties=discrete risklimits;

strata match;

run;

Where I made status=1 controls, status=2 cases (of toxic megacolon), censor=0 controls, censor=1 cases.

The problem here is that the numbers are strange. I want to see whether the frequency of a bunch of variables is different in the case group from the control group. The the predictor variable listed above ('fever'), 90% of the case group ('TMC') had fever while 25% of the control group had fever. Using the above program, I get a p of 0.026. With leukocytosis 'leuko' as the predictor, 80% of the case group had leukocytosis while 5 % of the control group had leukocytosis. It looks like it should be significant, but I get a p value of 0.996.

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

Posted in reply to deleted_user

10-15-2008 12:13 AM

You have one case matched with two controls - from your description I understand the variable "match" shows who is matched with who, and the variable "TMC" shows who is the case and who is the control.

If your variable of interest "daysadmit" is approximately normal, then to compare cases and controls run the following code:

proc mixed data=descriptive;

class match TMC;

model daysadmit=TMC/solutions;

random match;

run;

If your variable of interest "daysadmit" is approximately normal, then to compare cases and controls run the following code:

proc mixed data=descriptive;

class match TMC;

model daysadmit=TMC/solutions;

random match;

run;