Programming the statistical procedures from SAS

Proc logistic- discrepancy in values in stratified and un-stratified results

Reply
Regular Contributor
Posts: 173

Proc logistic- discrepancy in values in stratified and un-stratified results

Hi,

I have performed proc logistic for un-stratified and age-stratified analyses, but strangle, my un-stratified Odds Ratio is way smaller than the individual stratified OR (ODDS ratio)  values. It should be close to an average of the other 2 stratified OR values.

Here are the ORs.

The red coloed value is no where close to the average of the 2 stratified ORs.

CKD Stage III/IV- Overall1.74 (1.67-1.82)
Age <652.02 (1.82 -2.25)
Age >=651.95 (1.86 - 2.03)


CKD Stage V/ESRD Overall 3.79 (3.62- 3.96)
Age <654.41 (4.04- 4.81)
Age >=653.52 (3.34- 3.70)


I checked and rechecked my coding and program but don't find anything wrong with it. I am not able to figure out why the overall OR is not close to the average of the other 2 ORs.

The program code are as below. The red colored variable is variable of interest. It has 3 levels, 0,1 and 2 and 0 is reference category, 1= CKD stage III/IV and 2- CKD stage 5. Each program gives 2 Odds ratio values for this variable; 1 compared to 0 and 2 compared to 0. Interestingly the OR values for CKD stage V look correct and the overall OR seems to be an average of the other 2 stratified ORs.

Any suggestions, advice would be greatly appreciated.

proc logistic data =

class   female  (ref= first) renalicd9_3 (ref=first) dm_all_comorbid htn_c  dyslipidemia aids alcohol ANEMDEF arth race1(ref=first)  

ZIPINC_QRTL(ref=first) hosp_location h_contrl(ref=first) hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph

lytes mets neuro obese para perivasc psych pulmcirc  tumor ulcer valve wghtloss cararrhythmia  sepsis pneumonia/param=ref;

model anyproc1_15_1=  age female los dm_all_comorbid TOTAL_DISC htn_c  dyslipidemia aids alcohol ANEMDEF arth race1  ZIPINC_QRTL hosp_location

h_contrl hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph lytes mets neuro obese para perivasc

psych pulmcirc tumor ulcer valve wghtloss cararrhythmia  sepsis pneumonia renalicd9_3;

where stemi=1;

weight discwt;

title 'Logi Reg anyproc vs renalicd9 in MI patients with RACE +income  for stemi';

run;

quit;

proc logistic data =library.Nismicathckdmiss_pcilos30_1 descending;

class  female  (ref= first) renalicd9_3 (ref=first) dm_all_comorbid htn_c  dyslipidemia aids alcohol ANEMDEF arth race1(ref=first)  

ZIPINC_QRTL(ref=first) hosp_location h_contrl(ref=first) hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph

lytes mets neuro obese para perivasc psych pulmcirc  tumor ulcer valve wghtloss cararrhythmia sepsis pneumonia/param=ref;

model anyproc1_15_1=   female los dm_all_comorbid TOTAL_DISC htn_c  dyslipidemia aids alcohol ANEMDEF arth race1  ZIPINC_QRTL hosp_location

h_contrl hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph lytes mets neuro obese para perivasc

psych pulmcirc tumor ulcer valve wghtloss cararrhythmia sepsis pneumonia renalicd9_3;

where stemi=1 and agecat3=1;

weight discwt;

title 'Logi Reg anyproc vs renalicd9 in MI patients with RACE +income  for stemi <65';

run;

quit;

proc logistic data =library.Nismicathckdmiss_pcilos30_1 descending;

class  female  (ref= first) renalicd9_3 (ref=first) dm_all_comorbid htn_c  dyslipidemia aids alcohol ANEMDEF arth race1(ref=first)  

ZIPINC_QRTL(ref=first) hosp_location h_contrl(ref=first) hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph

lytes mets neuro obese para perivasc psych pulmcirc  tumor ulcer valve wghtloss cararrhythmia sepsis pneumonia /param=ref;

model anyproc1_15_1=   female los dm_all_comorbid TOTAL_DISC htn_c  dyslipidemia aids alcohol ANEMDEF arth race1  ZIPINC_QRTL hosp_location

h_contrl hosp_teach bldloss chf chrnlung coag depress drug hypothy liver lymph lytes mets neuro obese para perivasc

psych pulmcirc tumor ulcer valve wghtloss cararrhythmia sepsis pneumonia renalicd9_3;

where stemi=1 and agecat3=2;

weight discwt;

title 'Logi Reg anyproc vs renalicd9 in MI patients with RACE +income  for stemi >=65';

run;

quit;

Regular Contributor
Posts: 152

Re: Proc logistic- discrepancy in values in stratified and un-stratified results

You state that the value of an unstratified odds ratio should be between the values of its component stratified odds ratios.  This is not true, as the following set of 2-by-2 tables shows:

                       10      10                                2          1                            12    11

                         1         2                             10        10                            11    12

  odds ratio         2.00                                      2.00                                  1.19

Regular Contributor
Posts: 173

Re: Proc logistic- discrepancy in values in stratified and un-stratified results

Thanks for your response. But this was pointed out by one of the reviewers from a very reputed Journal as a part of the manuscript review process. That is when I started wondering if it is really true/correct to have unstratified OR somewhere between the 2 startified ORs.

I tried with a couple of examples as you have shown above and I too didnot get the OR value in between its component stratified ORs .

Ask a Question
Discussion stats
  • 2 replies
  • 180 views
  • 0 likes
  • 2 in conversation