Programming the statistical procedures from SAS

Mixed model with 1:1 matched pairs and repeated measures

Reply
Contributor
Posts: 24

Mixed model with 1:1 matched pairs and repeated measures

Dear all,

 

I am trying to fit a mixed model for a multi-centre study with 1:1 matched pair design with unequidistant repeated measures. The corresponding data structure looks this way: Obs denotes the observation number, Match ID the number of the matched pair, where patients have been matched 1:1 to have the same gender, the same centre ID and similar values regarding age and some other baseline lab values. Visit no is the unique visitnumber per patient, where each visit was assigned to a Month (1-12), which is the time difference to baseline. Since this is an observational study, there are unequal assessments per month, and in several months no data is available at all. Y is the continuous dependent variable and Treatment (A or B) the predictor.

Obs

Centre ID

Patient ID

Match ID

Visit No

Month

Y

Treatment

1

1

1

1

1

1

 

 

2

1

1

1

2

1

 

 

3

1

1

1

3

2

 

 

4

1

1

1

4

3

 

 

5

1

2

1

1

2

 

 

6

1

2

1

2

3

 

 

7

1

2

1

3

3

 

 

8

1

2

1

4

6

 

 

9

1

2

1

5

6

 

 

10

1

3

2

1

1

 

 

11

1

3

2

2

3

 

 

12

1

3

2

3

6

 

 

13

1

4

2

1

1

 

 

14

1

4

2

2

1

 

 

15

1

4

2

3

2

 

 

16

1

4

2

4

5

 

 

17

1

4

2

5

6

 

 

How can I fit a mixed model which accounts for the correlation within patients and within match id that tests the treatment effect and in a second model the treatment, time and treatment x time effect?

I found a similar post (https://communities.sas.com/t5/SAS-Procedures/How-to-adjust-for-matching-pair-in-mixed-effect-model/...), but unfortunately, the matching strategy is not 1:1 and there are not unequal numbers of visits.

I thought of something like to test the treatmen effect, but I am not sure whether this accounts for the nested structure (patient within match)

PROC MIXED DATA = ds;

     CLASS centre_id patient_id matching_id ;

     MODEL Y = Treatment;

random matching_id patient_id/ subject= patient_id;

RUN;

 

And how would the model with the time effect look like? Is this sufficient?

PROC MIXED DATA = ds;

     CLASS centre_id patient_id matching_id month;

     MODEL Y = Treatment month treatment * month;

random matching_id patient_id/ subject= patient_id;

RUN;

 

 

Thanks in advance and regards!

Ask a Question
Discussion stats
  • 0 replies
  • 92 views
  • 0 likes
  • 1 in conversation