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
- /
- Absurd value for odds ratio in logisitc regression...

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
- Permalink
- Email to a Friend
- Report Inappropriate Content

02-20-2013 03:55 PM

Hi there,

I have run a multivariate logistic regression model and one of the odds ratio values I have got it is

<0.001 (<0.001- >999.999) |

I am not sure what this means and why I have got such a value. Does this mean there is something wrong with my model. Is this a sample size issue? The original sample size is 914 and after weighting it comes to be 4500

The model is as below. Greatly appreciate any help or advice.

With such absurd value, can I still use the results?

**proc** **logistic** data =library.nismipostcath2 descending ;

class died race1(ref=first) ZIPINC_QRTL(ref=first) dm_all (ref=first) cm_htn_c (ref=first) morbidobesity (ref=first) hyperlipidemia (ref=first) cm_perivasc (ref=first) chf (ref=first) afib (ref=first) liverdisease(ref=first) ckd(ref=first)/param=ref;

model died= age race1 ZIPINC_QRTL dm_all cm_htn_c morbidobesity hyperlipidemia cm_perivasc chf afib liverdisease ckd ;

where pci_postcath=**1** and postcathcompli1=**1** and female=**1**;

weight discwt;

title 'Logi Reg in-hosp mortality in POST-PCI- MI women with post cath complications ';

**run**;

**quit**;

Some of the odds ratio values are as below.

Diabetes Mellitus | 1.36 (1.05 - 1.77) | 0.02 |

Hypertension | 1.09 (0.83 - 1.42) | 0.5364 |

Morbid Obesity | <0.001 (<0.001- >999.9) | 0.9764 |

Hyperlipidemia | 0.43 (0.33 - 0.55) | <0.0001 |

Perivascular disorders | 1.81 (1.41- 2.34) | <0.0001 |

Chronic Heart Failure | 4.11 (1.43 - 11.85) | 0.0088 |

Atrial Fibrillation | 1.61 (1.21 - 2.15) | 0.001 |

Liver Disease | 1.57 (0.63 - 3.94) | 0.3335 |

Chronic Kidney Disease | 0.98 (0.72 - 1.35) | 0.9226 |

Thanks much!

Ashwini

Accepted Solutions

Solution

02-20-2013
03:59 PM

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

Posted in reply to Ashwini_uci

02-20-2013 03:59 PM

From what I saw yesterday it seems like if you're using weights for sampling then you need to use proc surveylogistic not proc logistic.

Additionally you do have estimates of negative infinity to infinity when you have a very small numbers in that particular variable (check your table of characteristics or table 1 or run a proc freq to verify).

Usually if it's another reason SAS will issue a warning in the log or the output.

All Replies

Solution

02-20-2013
03:59 PM

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

Posted in reply to Ashwini_uci

02-20-2013 03:59 PM

From what I saw yesterday it seems like if you're using weights for sampling then you need to use proc surveylogistic not proc logistic.

Additionally you do have estimates of negative infinity to infinity when you have a very small numbers in that particular variable (check your table of characteristics or table 1 or run a proc freq to verify).

Usually if it's another reason SAS will issue a warning in the log or the output.

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

Posted in reply to Reeza

02-25-2013 01:40 PM

Hi there

Thanks for our response!

It seems that I cannot increase the sample size or change the restrive criteria.

What I wonder is if it is still alright to use these results, that I posted in the original post. Morbid obesity is one of the independent variables in the model, my interest variable is diabetese and hypertension. I hope there is no harm using these results.

Appreciate your response.

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

Posted in reply to Ashwini_uci

02-25-2013 01:52 PM

Take out morbid obesity and see if the results change.

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

Posted in reply to Ashwini_uci

02-25-2013 01:55 PM

I think you are missing a critical point. When there are sampling weights attached to the design, PROC LOGISTIC does not give the correct results. You MUST switch to PROC SURVEYLOGISTIC to get meaningful results. See many of @Reeza's responses to your question.

Steve Denham

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

Posted in reply to Reeza

02-25-2013 05:47 PM

Thanks for your response Reeza and for suggesting me using Proc surveylogistic!. I did switch to proc surveylogistic; good thing- my results haven't changed at all.

**I am curious to know **why** surveylogistic is more meaningful than logistic where we are able to add the weight statement and get the **SAME **results as suyveylogistic? What and how does proc surveylogistic do differentlythat logisitc while analysing the weights?

** I read on the sasupport that PROC SURVEYLOGISTIC incorporates complex survey sample designs, including designs with stratification, clustering, and **unequal weighting. Would you please explain what unequal weighting refers to?**

** So does it finally mean that whenever we are analysing any unweighted database, proc surveylogisitc is a better choice over proc logistic?

Appreciate your response

Thanks

Ashwini

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

Posted in reply to Ashwini_uci

02-25-2013 07:10 PM

You shouldn't get the same results. Check the standard error and the confidence intervals of your odds ratio, though the parameter estimates are probably fine.

The SAS Docs specify, in the weight statement for proc logistic, that it does not calculate the variance properly, to use surveylogistic instead.

I don't know what unequal weighting means.

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

Posted in reply to Ashwini_uci

02-23-2013 10:24 AM

The odds ratio of zero for the variable, MORBIDOBESITY, implies that none of the study subjects meeting the WHERE condition of

PCI_POSTCATH=1 and POSTCATHCOMPLI1=1 and FEMALE=1.

died. So, this is a sample size issue.

You can include the MODEL statement option, FIRTH, following a forward slash ("/") after the independent variables to see one approach for dealing with this issue. Another approach is to increase your sample size. A third approach is to make your WHERE condition less restrictive.

I agree with Reeza that you should use PROC SURVEYLOGISTIC instead of PROC LOGISTIC when your sample respondents have weights.

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

Posted in reply to 1zmm

02-25-2013 01:40 PM

Hi there

Thanks for our response!

It seems that I cannot increase the sample size or change the restrive criteria.

What I wonder is if it is still alright to use these results, that I posted in the original post. Morbid obesity is one of the independent variables in the model, my interest variable is diabetese and hypertension. I hope there is no harm using these results.

Appreciate your response.