turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Large confidence interval

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-05-2017 07:18 AM

Hi I'm in the process of doing a multiple logistic regression.

I'm trying to model outcome of an emergency admission using various predictors.

I have a mix of categorical and continous explanatory variables. For each of the continous variables I carried out a simple logistic regression beforehand which included just age, gender & the contiuous variable.

When running the logistic regression for haematocrit, including age and gender, I get a OR (95% CI) of 0.008 (0.002 - 0.040).

```
proc logistic data=predictors;
class gender;
model emerg(event='1')=gender age haematocrit;
run;
```

However, when I run the full model, the confidence interval for this variable becomes very large OR 2.845 (0.084 - 95.841) for this variable. I wonder if anyone can explain this phenomenon? Is this a true effect?

```
proc logistic data=predictors;
class gender smoking alcohol ethnicity no_medication lives_alone bereaved ho_ptca ho_cabg average_glucose_gt200 / param=ref ref=first;
model emerg(event='1')=gender smoking alcohol ethnicity no_medication lives_alone bereaved ho_ptca ho_cabg average_glucose_gt200
age bmi num_deficits num_comorbidites systolic_bp blood_urea_nitrogen creatinine haematocrit;
run;
```

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-05-2017 07:39 AM - edited 05-07-2017 03:42 PM

You've changed the parameterization and the reference levels, so you should expect the estimates and odds ratios will change.

The first call to PROC LOGISTIC implicitly uses PARAM=EFFECT REF=LAST.

The second call uses PARAM=REF REF=FIRST.

Of these two changes, it is probably the REF= option that has the biggest effect on what you are seeing.

For example, suppose the only covariate is GENDER and the odds ratio for GENDER is 0.5 when REF=LAST. If you change REF=FIRST and rerun, then the new odds ratio is 2.0 (=1/0.5). These two numbers mean exactly the same thing: one gender has twice the odds than the other. The only thing that has changed is if you are comparing females to males or vice versa.

Try using the same parameterization and reference level for each call and report back.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-05-2017 09:02 AM

Hi Rick,

Sorry I should've thought about it a bit harder before I posted this question. I realised the problem was the unit of the variable. When I changed the units from ratio to percentages, it made the output make sense.

Thank you for your help as always!

Sorry I should've thought about it a bit harder before I posted this question. I realised the problem was the unit of the variable. When I changed the units from ratio to percentages, it made the output make sense.

Thank you for your help as always!