Hello everyone.
I am investigating the possible significant predictors of an outcome in a retrospective observational study with 88 patients (outcome yes: 61 patients, outcome no: 27 patients). In a multivariable Cox proportional hazard regression with stepwise based on the Akaike information criterion, I obtained only one significant predictor of the outcome. Instead, in a multivariable logistic regression with stepwise based on the AIC, I obtained two significant predictors, different from the first one. I tested multicollinearity with variance inflation factors and variables with multicollinearity were sequentially removed from the model starting for the variable with the highest VIF. The outcome was verified at a specific time point (ICU discharge, which was different among patients), but not at a fixed time point (e.g., 28 days, 60 days). Can anyone explain to me the possible reasons for this and which model should I prefer?
Thanks in advance for your help.
A Cox model should only be applied if you have right-censored data. If you have complete data with no right-censoring, the logistic regression model should be preferred since it is a parametric model, whereas a Cox model is only semiparametric.
See the attached two papers for the difference between the two methods and what you do with each method. These are not interchangeable. David Booth
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression