Multivariate ''binary'' logistic regression...???


New Member
Hello all,
My name is Nick. I am a doctor and i would appreciate any of your help in the above problem. Recently, on purpose to delineate/propose a specific risc factor for a specific disease i proceeded in a binary logistic regression on SPSS. My model was characterized by one categorical (dichotomous) variable and 5 independent also categorical (and also dichotomous) variable. My model has been worked fine, i had a very big general population (2300 patients), a good one as regards my dependent variable (51), it seems that i managed to correlate my dependent variable with the one independent i wanted from the beggining, but in a conversation that i had with my supervisor proffessor we noticed one problem, and he proposed me something more that i really didn't understand so well. So regarding the problem, even we noticed that our data gave us a very good OR concerning the correlation between the 2 diseases that we thought from the beggining to be related, we noticed that 2 of our independent variables had negative OR. These obviously happened because we had few patients that combined each of the 2 independent diseases (2 and 3 respectively) with the dependent disease. So because our aim is to present only the positive relation with the disease that we wanted and not negative correlation of the dependent disease with the other 2 aforementioned (cause such an idea is wrong either way, or at least we don't want to show something like that) i decided to shorten these 2 categories to 1 (patients that have either the one or the other disease) with the hope in a new model i have a positive OR. My problem mainly has to do with another thing too. My professor told me that he would prefer a multivariate model and not a univariate like mine cause multivariate models have higher strengh. But i can not imagine how to do something like that. How can you proceed to multivariate logistic regression and not binary when you have a dichotomous dependent variable and 3 categorical (again dichotomous) variables to SPSS or generally eveywhere??? Please help...Thank you in advance


Omega Contributor
nok, I will guess that English is not your native tongue. Next post please break up the text into paragraphs. You giant block of text was hard to read.

Multiple logistic regression: multiple predictors and one dependent variable

Multivariate logistic regression: multiple predictors are an option, but it has multiple dependent variables.

Polynomial regression: Depend variable can have more than 2 groups.

What are you writing about in particular?

Also, your issue is that the IV is associated with your dependent variable, so they are both a result of another variable or is the IV a result of the DV? I am not following your statements. It also seem like you could have been writing about Berkson Bias?


New Member
Thank you a lot for your reply. I will put it in an example...

Let's say i am trying to examine if sex (male-female), hyperlipidemia (yes-no), clinical history of hypertension (yes-no) and clinical history of diabetes mellitus (yes-no) increase the risk of patients that visited a caldiology outpatient clinic cause of heart-attack (yes-no, my DV).

I suppose that the model i have to use to figure out what are the ORs for each IV is a binary logistic regression...Isn't that right???

The thing is that i have in a total of 2300 patients that visited the clinic 2 patients with both heart-attack and hypelipidemia and 1 patient with diabetes mellitus, so i have kind of curius ORs. I manage to proove a good OR for hypertention but the problem is with the other two...

Also as i said i suppose that the only model that is fixed to what i want to proove is binary logistic regression....

Thank you again...


New Member
Also i suppose i shouldn't proceed to that i was writing before, meaning creating a new IV like dibetes mellitus+hyperlipidaimia (yes-no) cause i suppose that it could include a lot of bias in the codext of ''the reason of your visit to the cardiology outpatient clinic''. The thing i am trying to say is that anyone could wonder why are you adding only these 2 diseases and not another 1 or 2 to your variable???