Multivariate regression versus multivariable regression

#1
I am doing research to predict if alcoholism is an independent risk factor for bleeding after surgery. I will perform a univariate analysis with a chi squared test to know if other factors, such as age, sex, comorbidities,... are associated with an increased risk for postoperative bleeding. Thereafter, I will perform a multivariate or multivariable logistic regression analysis to correct for the independent predictors (significant risk factors after univariate analysis) of postoperative bleeding, so I will know if alcoholism remains a significant risk factor for postoperative bleeding.

Is this a correct plan? And do I need to use multivariate or multivariable logistic regression analysis for this research?
 

ondansetron

TS Contributor
#2
I am doing research to predict if alcoholism is an independent risk factor for bleeding after surgery. I will perform a univariate analysis with a chi squared test to know if other factors, such as age, sex, comorbidities,... are associated with an increased risk for postoperative bleeding. Thereafter, I will perform a multivariate or multivariable logistic regression analysis to correct for the independent predictors (significant risk factors after univariate analysis) of postoperative bleeding, so I will know if alcoholism remains a significant risk factor for postoperative bleeding.

Is this a correct plan? And do I need to use multivariate or multivariable logistic regression analysis for this research?
If you mean a single outcome (i.e. does the person bleed after surgery (Y/N)) as a function of 2 or more predictor variables, then you want a multivariable regression. If you have multiple outcomes that are being modeled simultaneously then you would want a multivariate regression. The latter isn't well defined for binary outcomes so I'm guess you just mean a multivariable logistic regression. In general the "univariate" screening is pretty useless with poor performance on new data, despite its misplaced popularity within medicine. Pre-specifying a multivariable model predictor set is much better when using subject matter expertise.
 
#3
If you mean a single outcome (i.e. does the person bleed after surgery (Y/N)) as a function of 2 or more predictor variables, then you want a multivariable regression. If you have multiple outcomes that are being modeled simultaneously then you would want a multivariate regression. The latter isn't well defined for binary outcomes so I'm guess you just mean a multivariable logistic regression. In general the "univariate" screening is pretty useless with poor performance on new data, despite its misplaced popularity within medicine. Pre-specifying a multivariable model predictor set is much better when using subject matter expertise.
Thank you for the clarification. This helps a lot.
 

noetsi

Fortran must die
#4
I have never heard the term multivariable used in statistical analysis and I have spent a fair amount of time reading it :p To me a single response variable with two or more predictors is a multivariate analysis.