Programming the statistical procedures from SAS

Modeling binary outcome with repeated measurements

Reply
Learner
Posts: 1

Modeling binary outcome with repeated measurements

Dear SAS community,

I have a data with 26 subjects in total and 70 observations with non-equally spaced time assessments and non-equal number of assessments.  I have tried GENMOD, GLIMMIX, NLMIXED and regular logistic regression regardless the correlation.

Each method come out with very different results. I understand each method use different algorism for parameter estimation,   but why the results come out so differently, how can I make judgement as which one should I report?  below please find the code and correspondent results. I can send the SAS data if necessary.

 

Thanks for your help.

SHa

 

proc nlmixed data=vocal;

*Where Time <6;

parms beta0 =-1.1938 beta1=0.0408;

eta = u+ beta0+beta1*CVR_Cube;

expeta = exp(eta);

p = expeta/(1 + expeta);

model Hydrops ~ binary(p);

random u ~normal(0, s2u) subject=subj_id;

run;

 

Parameter Estimates

Parameter

Estimate

Standard Error

DF

t Value

Pr > |t|

Alpha

Lower

Upper

Gradient

beta0

-5.1376

2.0194

25

-2.54

0.0175

0.05

-9.2967

-0.9786

1.10E-06

beta1

0.8079

0.3884

25

2.08

0.0479

0.05

0.00808

1.6077

5.48E-06

s2u

13.3115

10.7961

25

1.23

0.229

0.05

-8.9235

35.5465

-8.14E-

 

 

PROC GLIMMIX DATA= VOCAL;

  CLASS Subj_Id Time;

   MODEL hydrops = cvr_Cube / Dist=binomial Link=logit SOLUTION;

   RANDOM intercept / SUBJECT = Subj_Id;

RUN;

 

Solutions for Fixed Effects

Effect

Estimate

Standard Error

DF

t Value

Pr > |t|

Intercept

-2.8561

0.9429

25

-3.03

0.0056

CVR_CUBE

0.5005

0.2228

42

2.25

0.0300

 

PROC GENMOD DATA = VOCAL DESCENDING;

  CLASS Subj_Id ;

  MODEL hydrops = cvr_Cube / Dist= Binomial Link = Logit;

  REPEATED Subject = Subj_Id / TYPE = CS Corrw Covb;

RUN;

Quit;

 

Analysis Of GEE Parameter Estimates

Empirical Standard Error Estimates

Parameter

Estimate

Standard Error

95% Confidence Limits

Z

Pr > |Z|

Intercept

-1.1938

0.4486

-2.0730

-0.3147

-2.66

0.0078

CVR_CUBE

0.0408

0.0289

-0.0159

0.0975

1.41

0.1585

 

PROC LOGISTIC DATA = VOCAL DESCENDING;

 MODEL hydrops = cvr_cube;

RUN;

 

Analysis of Maximum Likelihood Estimates

Parameter

DF

Estimate

Standard Error

Wald Chi-Square

Pr > ChiSq

Intercept

1

-2.5059

0.5524

20.5783

<.0001

CVR_CUBE

1

0.4921

0.1297

14.4012

0.0001

 

 

 

 

 

 

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