topic Re: matched data analysis in SAS Programming

matched data analysis

wajmsu — Thu, 12 May 2016 20:44:22 GMT

Hi,

I have a matched data set (shown below). I want to carry out the following analysis:

1. Descriptive table

2. conditional logistic regression unadjusted (outcome is fev)

3. conditional logistic regression adjusted (outcome is fev)

4. conditional logistic regression with interaction terms (opwo*race) (outcome is fev). Please guide me.

Thanks

Group	Type	Control Id	id	age	edu	race	parity	wt	ht	opwo	fev
1	Case	.	5	32	9	1	0	90	5.2	1	1
1	Control	1	70	32	9	1	0	90	5.2	1	0
1	Control	2	99	35	10	1	0	90	5.2	0	0
2	Case	.	12	35	12	1	2	110	5	0	1
2	Control	1	43	35	12	1	1	110	5	0	0
2	Control	2	101	33	11	1	2	110	5	0	0
3	Case	.	22	25	14	3	0	110	5.4	0	1
3	Control	1	125	25	13	3	0	110	5.4	0	0
3	Control	2	51	28	12	3	0	110	5.4	0	0
4	Case	.	29	27	12	2	1	95	5	1	1
4	Control	1	32	29	12	2	2	125	6.2	1	0
4	Control	2	31	27	10	2	1	95	5.5	0	0
5	Case	.	40	28	10	3	2	130	6	1	1
5	Control	1	58	28	10	3	2	130	6	0	0
5	Control	2	93	27	11	3	2	130	6	1	0

Re: matched data analysis

ballardw — Thu, 12 May 2016 20:53:14 GMT

What should the descriptive "table" contain? Counts, percents, Means, standard deviations, max, min or other statistics of single variables? Combinations of variables? within groups? Correlations?

Re: matched data analysis

wajmsu — Thu, 12 May 2016 23:19:11 GMT

Thanks; I want to have counts with percents and mean with standard deviations and range.

Re: matched data analysis

Ksharp — Fri, 13 May 2016 01:10:09 GMT

Can you explain

2. conditional logistic regression unadjusted (outcome is fev)

3. conditional logistic regression adjusted (outcome is fev)

What is unadjusted and adusted ?

Did you check the example in documentation ,there are many examples about conditional logistic regression.

What is your STRATA(conditional) variable ? GROUP ?

About interaction terms opwo*race , Couldn't you write as

model ....=.... opwo*race ;

strata group;

or check EFFECT statement.

Re: matched data analysis

wajmsu — Fri, 13 May 2016 05:16:03 GMT

Sure.

Unadjusted means univatiate analysis (one variable assessed for the outcome, here fev)

Adjusted means multipvariate, when more than one independent variables are added to the model simultaneously where these variables adjust the effect/association of each other on the response variable.

My matching variables are: age, edu, race and parity. I think, not sure, these can be taken as 'strata'.

Thanks

Re: matched data analysis

Ksharp — Fri, 13 May 2016 07:43:30 GMT

You can find some examples relate to conditional logistic in documentation.

Something like :

Example 72.11: Conditional Logistic Regression for Matched Pairs Data



data have;
infile cards expandtabs truncover;
input  Group	Type $	ControlId	id	age	edu	race	parity	wt	ht	opwo	fev;
cards;
1	Case	.	5	32	9	1	0	90	5.2 1 1
1	Control	1	70	32	9	1	0	90	5.2 1 0
1	Control	2	99	35	10	1	0	90	5.2 0 0
2	Case	.	12	35	12	1	2	110	5   0 1
2	Control	1	43	35	12	1	1	110	5   0 0
2	Control	2	101	33	11	1	2	110	5   0 0
3	Case	.	22	25	14	3	0	110	5.4 0 1
3	Control	1	125	25	13	3	0	110	5.4 0 0
3	Control	2	51	28	12	3	0	110	5.4 0 0
4	Case	.	29	27	12	2	1	95	5   1 1
4	Control	1	32	29	12	2	2	125	6.2 1 0
4	Control	2	31	27	10	2	1	95	5.5 0 0
5	Case	.	40	28	10	3	2	130	6  1 1
5	Control	1	58	28	10	3	2	130	6  0 0
5	Control	2	93	27	11	3	2	130	6  1 0
;
run;
proc logistic data=have ;
strata group;
model fev(event='1')= age edu race parity;
run;

Re: matched data analysis

wajmsu — Fri, 13 May 2016 13:27:22 GMT

Thanks, it did work; will explore further. However, I can't take age edu race and parity in the model as they are matching variable and their effect has already been fixed.

Re: matched data analysis

wajmsu — Fri, 13 May 2016 16:31:49 GMT

Hi Ksharp,

What code should I use for count (with percents) for parity, race, opwo and for mean (with std deviation) for wt ht acros fev (1 and 0)?

Thanks.

Re: matched data analysis

Ksharp — Sat, 14 May 2016 07:05:30 GMT

Sorry. I don't know what you are talking about. Maybe you should start a new session to discuss this .