topic Re: small sample conditional logistic in Statistical Procedures

small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 00:02:42 GMT

Hi,

I am trying to apply exact conditional logistic regression to my samll sample matched data with th efollowing code but t does not give me odds ratio. It gives coefficients with their CI.

Can yo ucorrect my code? Thanks

proc logistic data=pe.matchedata_21 desc;
strata group;
class CMV_ca_2_1 (ref='Absent');
format CMV_ca_2_1 CMV_igg.;
model PEstatus (event='1')= CMV_ca_2_1;
exact CMV_ca_2_1 / estimate = both;
run;

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 00:09:05 GMT

Use an explicit ODDSRATIO statement or use the EXPB option in the model statement.

proc logistic data=pe.matchedata_21 desc;
strata group;
class CMV_ca_2_1 (ref='Absent');
format CMV_ca_2_1 CMV_igg.;
model PEstatus (event='1')= CMV_ca_2_1/ expb clodds=both;
exact CMV_ca_2_1 / estimate = both;
oddsratio cm_ca_2_1;
run;

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 00:20:09 GMT

Thanks Reeza. However, it does not work. With 'oddsratio cmv_ca_2_1' no change.

With 'expb clodds=both' it even does not work; log states '
ERROR: The CLODDS=PL option is not allowed when a STRATA statement is specified.

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 00:29:19 GMT

Change clodds to Wald or remove it.

Please post

1) log from code

2) parameter estimate table

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 01:12:11 GMT

proc logistic data=pe.matchedata_21 desc;
strata group;
class CMV_ca_2_1 (ref='Absent');
format CMV_ca_2_1 CMV_igg.;
model PEstatus (event='1')= CMV_ca_2_1;
exact CMV_ca_2_1 / estimate = both;
oddsratio cmv_ca_2_1;
run;

Below is the table which I get for Exact analysis after using the above code. PLease note that before this table or exact analysis, the result contains non-exact output with 95%CI (output 1 below).

SAS Output 1

Odds Ratio Estimates and Wald Confidence Intervals Odds Ratio Estimate 95% Confidence Limits CMV_ca_2_1 Present vs Absent

2.386

0.374

15.232

SAS Output 2

Exact Parameter Estimates Parameter Estimate Standard
Error 95% Confidence Limits Two-sided p-Value CMV_ca_2_1 Present

0.4348

0.4729

-0.6886

1.7146

0.6250

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 01:50:04 GMT

LOG?

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 01:53:21 GMT

My apologies!

81   proc logistic data=pe.matchedata_21 desc;
NOTE: Data file PE.MATCHEDATA_21.DATA is in a format that is native to another host, or
      the file encoding does not match the session encoding. Cross Environment Data
      Access will be used, which might require additional CPU resources and might reduce
      performance.
82   strata group;
83   class CMV_ca_2_1 (ref='Absent');
84   format CMV_ca_2_1 CMV_igg.;
85   model PEstatus (event='1')= CMV_ca_2_1;
86   exact CMV_ca_2_1 / estimate = both;
87   oddsratio cmv_ca_2_1;
88   run;

NOTE: Writing HTML Body file: sashtml1.htm
NOTE: PROC LOGISTIC is modeling the probability that PEstatus='1'.
NOTE: Convergence criterion (GCONV=1E-8) satisfied.
WARNING: Odds ratios for CMV_ca_2_1 in the EXACT statement are not computed unless the
         reference parameterization is used.
NOTE: There were 73 observations read from the data set PE.MATCHEDATA_21.
NOTE: PROCEDURE LOGISTIC used (Total process time):
      real time           2.08 seconds
      cpu time            1.37 seconds

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 01:53:42 GMT

So I got some sample data and replicated your code. My log says this:

70 proc logistic data = exlogit desc;

71 class female;

72 freq num;

73 model admit = female apcalc;

74 exact female apcalc / estimate = both;

75 run;

NOTE: PROC LOGISTIC is modeling the probability that admit=1.

NOTE: Convergence criterion (GCONV=1E-8) satisfied.

WARNING: Odds ratios for female in the EXACT statement are not computed unless the reference parameterization is used.

NOTE: There were 8 observations read from the data set WORK.EXLOGIT.

NOTE: PROCEDURE LOGISTIC used (Total process time):

real time 0.28 seconds

cpu time 0.21 seconds

Once I changed the code to use reference parameterization, it worked.

I would assume that your log has a similar warning?

class CMV_ca_2_1 (ref='Absent')/param=ref;

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 01:55:01 GMT

PLEASE read the log in the future and post it in your question if you have issues.

It tells you exactly what you need to do to solve your issue.

@wajmsu wrote:

WARNING: Odds ratios for CMV_ca_2_1 in the EXACT statement are not computed unless the
reference parameterization is used.

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 02:06:13 GMT

Thanks Reeza; it worked this time.

Both tables are below, exact and non-exact analysis.

My question is that why we use exact when we get wide 95%CI, usually we prefer narrow CI.

Exact Odds Ratios
Parameter		Estimate	95% Confidence Limits	Two-sided p-Value
CMV_ca_2_1	Present	2.386	0.252	30.852	0.6250

Odds Ratio Estimates
Effect	Point Estimate	95% Wald Confidence Limits
CMV_ca_2_1 Present vs Absent	2.386	0.374	15.232

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 02:12:56 GMT

I don't know the answer to that, someone else will have to chime in now.
Cheers!

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 06:45:50 GMT

Hi again Reeza,

As I start buidling multivariable model it again creates problem. Here are my codes and logs, once again thanks for your help.

proc logistic data=pe.matchedata_21 desc;
strata group;
class CMV_ca_2_1 (ref='Absent')/param=ref bmicat (ref='Less than 25')/param=ref;
format CMV_ca_2_1 CMV_igg. bmicat bmicat.;
model PEstatus (event='1')= CMV_ca_2_1 bmicat;
exact CMV_ca_2_1 bmicat / estimate = both;
oddsratio cmv_ca_2_1 bmicat;
run;

153 proc logistic data=pe.matchedata_21 desc;
NOTE: Data file PE.MATCHEDATA_21.DATA is in a format that is native to another host, or
      the file encoding does not match the session encoding. Cross Environment Data
      Access will be used, which might require additional CPU resources and might reduce
      performance.
NOTE: PROCEDURE LOGISTIC used (Total process time):
      real time           0.01 seconds
      cpu time            0.01 seconds

NOTE: The SAS System stopped processing this step because of errors.
154 strata group;
155 class CMV_ca_2_1 (ref='Absent')/param=ref bmicat (ref='Less than 25')/param=ref;
                                               ------
                                               22
                                               76
ERROR 22-322: Syntax error, expecting one of the following: ;, CODING, CPREFIX, DESC,
              DESCENDING, LPREFIX, MISSING, ORDER, PARAM, REF, REFERENCE, TRUNCATE.
ERROR 76-322: Syntax error, statement will be ignored.
156 format CMV_ca_2_1 CMV_igg. bmicat bmicat.;
157 model PEstatus (event='1')= CMV_ca_2_1 bmicat;
158 exact CMV_ca_2_1 bmicat / estimate = both;
159 oddsratio cmv_ca_2_1 bmicat;
160 run;

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 06:49:53 GMT

Param=ref is only specified only once at the end of your class statement.

If you really want multiple methods you can use multiple class statements.

In this case the error isn't very helpful 😉

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 06:51:34 GMT

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 06:52:45 GMT

It did not change putting Param=ref once.

Problem remians the same.

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 06:58:57 GMT

Post your code and log.

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 07:10:35 GMT

This is the same I posted in previous message. It is here again.

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 07:30:35 GMT

And you didn't change your code... Why would I want to see the same code/log?

Change your code as indicated in my first answer, if that doesn't work post the code and log.

Re: small sample conditional logistic

Reeza — Mon, 25 Jul 2016 07:40:05 GMT

class CMV_ca_2_1 (ref='Absent') bmicat (ref='Less than 25')/param=ref;

Re: small sample conditional logistic

wajmsu — Mon, 25 Jul 2016 08:13:46 GMT

As I mentioned earlier, it did not work. Here is the log:

161 proc logistic data=pe.matchedata_21 desc;
NOTE: Data file PE.MATCHEDATA_21.DATA is in a format that is native to another host, or
      the file encoding does not match the session encoding. Cross Environment Data
      Access will be used, which might require additional CPU resources and might reduce
      performance.
162 strata group;
NOTE: PROCEDURE LOGISTIC used (Total process time):
      real time           0.01 seconds
      cpu time            0.01 seconds

NOTE: The SAS System stopped processing this step because of errors.
163 class CMV_ca_2_1 (ref='Absent') bmicat (ref='Less than 25')/param=ref;
164 format CMV_ca_2_1 CMV_igg. bmicat bmicat.;
165 model PEstatus (event='1')= CMV_ca_2_1 bmicat;
166 exact CMV_ca_2_1 bmicat / estimate = both;
167 oddsratio cmv_ca_2_1 bmicat;
                          ------
                          22
                          202
ERROR 22-322: Syntax error, expecting one of the following: ;, /.
ERROR 202-322: The option or parameter is not recognized and will be ignored.
168 run;