Example 12 for PROC MIANALYZE: MNAR Sensitivity Analysis

Difficulty Level

Beginner

Published on : 16/12/2023

The script first generates a simulated dataset (Mono1) with arbitrary missing values. It then performs two distinct analyses to compare results: a first standard multiple imputation (MAR - Missing At Random) followed by linear regression and combined analysis, then a second approach using an MNAR adjustment to model Missing Not At Random data, also followed by regression and synthesis of results via PROC MIANALYZE.

Data Analysis

Type : CREATION_INTERNE

The 'Mono1' dataset is generated entirely in the first DATA step using loops and random functions (rannor, ranuni).

1 Code Block

DATA STEP Data

Explanation :
Creation of the 'Mono1' dataset simulating clinical trial data with two treatment groups (Trt). Missing values are artificially introduced into variable y1.

Copied!

1	DATA Mono1;
2	DO Trt=0 to 1;
3	DO j=1 to 5;
4	y0=10 + rannor(999);
5	y1= y0 + Trt + rannor(999);
6	IF (ranuni(999) < 0.3) THEN y1=.;
7	OUTPUT;
8	END; END;
9
10	DO Trt=0 to 1;
11	DO j=1 to 45;
12	y0=10 + rannor(999);
13	y1= y0 + Trt + rannor(999);
14	IF (ranuni(999) < 0.3) THEN y1=.;
15	OUTPUT;
16	END; END;
17	drop j;
18	RUN;

2 Code Block

PROC PRINT

Explanation :
Display of the first 10 observations to verify the generated data.

Copied!

1	PROC PRINT DATA=Mono1(obs=10);
2	var Trt Y0 Y1;
3	title 'First 10 Obs in the Trial Data';
4	RUN;

3 Code Block

PROC MI Data

Explanation :
First execution of multiple imputation (20 imputations) assuming a monotone pattern and a standard regression method. Imputed data is saved in 'outex12a'.

Copied!

1	PROC MI DATA=Mono1 seed=14823 nimpute=20 out=outex12a;
2	class Trt;
3	monotone reg;
4	var Trt y0 y1;
5	RUN;

4 Code Block

PROC REG Data

Explanation :
Execution of a linear regression model on each imputed dataset (via the BY _Imputation_ group). Parameter estimates are exported to the 'regparms' table.

Copied!

1	ods select none;
2	PROC REG DATA=outex12a;
3	model y1= Trt y0;
4	BY _Imputation_;
5	ods OUTPUT parameterestimates=regparms;
6	RUN;
7	ods select all;

5 Code Block

PROC MIANALYZE

Explanation :
Combination of the results of the 20 regressions to produce valid statistical inferences (Rubin's rules) for the first approach.

Copied!

1
2	PROC MIANALYZE parms=regparms;
3	modeleffects Trt;
4	RUN;
5

6 Code Block

PROC MI Data

Explanation :
Second execution of multiple imputation. This time, the MNAR statement is used to specify a model for Missing Not At Random data on variable y1, specifically for the Trt=0 group.

Copied!

1	PROC MI DATA=Mono1 seed=14823 nimpute=20 out=outex12b;
2	class Trt;
3	monotone reg;
4	mnar model( y1 /modelobs=(Trt='0'));
5	var y0 y1;
6	RUN;

7 Code Block

PROC REG Data

Explanation :
Linear regression on data from MNAR imputation, repeated by imputation.

Copied!

1	ods select none;
2	PROC REG DATA=outex12b;
3	model y1= Trt y0;
4	BY _Imputation_;
5	ods OUTPUT parameterestimates=regparms;
6	RUN;
7	ods select all;

8 Code Block

PROC MIANALYZE

Explanation :
Final combination of results for the MNAR approach.

Copied!

1
2	PROC MIANALYZE parms=regparms;
3	modeleffects Trt;
4	RUN;
5

This material is provided "as is" by We Are Cas. There are no warranties, expressed or implied, as to merchantability or fitness for a particular purpose regarding the materials or code contained herein. We Are Cas is not responsible for errors in this material as it now exists or will exist, nor does We Are Cas provide technical support for it.

Difficulty Level

Published on : 16/12/2023

Data Analysis

1 Code Block

2 Code Block

3 Code Block

4 Code Block

5 Code Block

6 Code Block

7 Code Block

8 Code Block

Related Documentation