My study is a prospective observational study. My dependent variable (outcome) is development of surgical site infection (SSI) after surgery and my independent variables (predictors) are many factors containing socio-demographics, pre-operative, intra-operative and post-operative factors.

Outcome is dichotomous: SSI: 0=No , 1=Yes

Predictors are dichotomous as well as polychotomous( 3 or more categories), e.g. ASA score: 0=Class_1, 1=Class_2, 2=Class_3, 3=Class_4

I have already done the cross-tabulation (Chi square test) and i have also done univariate analysis using Enter method of binary logistics for every single variable.

Now i want to perform a multivariate analysis using all the predictors who came out to be significant in the univariate analysis (P= <0.25 as significant). I am now a bit confused which method i have to use in order to get more authentic results. I have seen literature similar to my study using simple logistic regression or forward step-wise regression as well. The references are as below:

Reference 1: http://www.ncbi.nlm.nih.gov/pubmed/23392976

Reference 2: http://www.ncbi.nlm.nih.gov/pubmed/11198018

My questions:

1) For polychotomous variables, i transformed them into dichotomous variables for one single category. e.g. I made 4 seperate columns for 4 classes of ASA score. and put them all individually in Univariate? and those who come out to be significant will be put in multivariate with 0=No as the reference category? Is this method acceptable?

2) Which method regarding binary logistics is the best as per my study?

i want to find out independent risk factors of SSI with Odds ratio?


Specify one model based on your knowledge prior to seeing your data. Fit that model. Perform your hypothesis tests.

Do not do any kind of stepwise variable selection, whether based on $p$ values, information criteria or anything else. Stepwise procedures invalidate subsequent inference. See here for a terse summary, and look through the references as needed. (Note that one author, Frank Harrell, knows what he is talking about.)


Not the answer you're looking for? Browse other questions tagged or ask your own question.