Interpretation of odds ratio

Can anyone help me interpret the results on the attached figure (drivers of preterm birth in Missouri)? This is the result of a multivariate logistic regression analysis.

What can be inferred from these ORs? Is it fair to say that the greater the OR the greater the contribution of that factor will be? Would be great to be able to discuss it further.

• "Multivariate" refers to when there are multiple response (Y) variables, which I don't think is what you have. There is multinomial logistic regression where the is 1 variable, but w/ multiple categories. I suspect you are referring to standard logistic regression, but w/ multiple predictor variables. Please clarify your question. Regarding the interpretation of odds ratios, it may also help you to read my answer here: Interpretation of simple predictions to odds ratios in logistic regression. Commented Dec 16, 2014 at 2:50
• Sorry about that, here is the exact method: "Unadjusted and adjusted ORs for premature birth and recurrent pre-mature birth were calculated using logistic regression in SPSS, and risk ratios and CIs were calculated manually"
– dav
Commented Dec 16, 2014 at 3:10

If all these odds ratios come from the same model, then, for example, the first odds ratio could be interpreted as: when all the other variables in the model are held constant constant (e.g. 0), blacks have 3.2 times the odds of preterm birth compared with non-blacks (or whatever the reference group is).

• Thanks tvance. I am just a little confused about seeing so many odd ratios greated than 1 - is it fair to say that all these factors contribute to preterm birth but that some (like Placental abruption) contribute much more? Wondering if there is some sort of threshold for significance here.. (sorry I am really new to this)
– dav
Commented Dec 16, 2014 at 3:13
• That is probably a reasonable conclusion, but whether or not all of these odds ratios come from the same model will affect your interpretation. If all these come from the same model then it could be said that for any variable, the odds ratio is "adjusted" for the other variables in the model, whereas if these are all from simple logistic regression you cannot be certain whether certain variables explain above and beyond other variables and whether there is confounding. Commented Dec 16, 2014 at 3:29
• Thanks tvance, super helpful. Regarding the adjusted or non-adjusted ORs: in this particular example the ORs are 'unadjusted' (see column header). The authors also report the 'adjusted' ORs in a different figure. To clarify: if the ORs are adjusted, does that mean that by definition they come from the same model and thus can be compared to each other without worrying about confounding?
– dav
Commented Dec 16, 2014 at 3:44
• If all ORs come from the same model, and assuming there is no interaction, then an OR can be interpreted as controlling for the other variables in the model. If you assume that all the other variables in the model are important confounders, then the OR of interest is adjusted for potential confounders. However, I wouldn't say theres no need to worry about confounding, since there are several additional considerations for confounding and, especially for observational data, there is always the possibility of confounding due to unmeasured variables. Commented Dec 16, 2014 at 4:00
• Thanks a lot. 2 follow-up Qs. 1) How would you typically verify that there is 'no interaction' between the variables? 2) Can you clarify what you mean by 'If you assume that all the other variables in the model are important confounders, then the OR of interest is adjusted for potential confounders'? Thanks so much.
– dav
Commented Dec 16, 2014 at 4:20