How can I perform covariate adjustment in SPSS (or AROC)? Seems that there are no guidelines, the only thing I can find are theoretical explanations which are not quite clear. Thank you in advance.
In simple terms, when you include a covariate into your model what you will get from your model will be adjusted for this covariate. You can search for "ANCOVA" for some explanations.
Hello Danica. As Mehmet said, it should be a simple matter of including the covariate as another explanatory variable in your model. What kind of model are you trying to estimate? It would make things clearer if you posted your command syntax. HTH.
Dear Bruce, thank you for your answer. I have a ROC curve which describes specificity and sensitivivty of biomarkers for mortality prediction. Since there are no controls included in the study I am supposed to use comorbidity indices and adjust the model for comorbidities in order to prove that biomarkers really are highly specific. Thank you in advance.
Hello Danica. If I follow, you have been using the ROC procedure in SPSS, and it takes only two variables, test result and gold standard result. In order to adjust for covariates, you'll have to use the LOGISTIC REGRESSION procedure and save the predicted probabilities to the working data file (SAVE sub-command, if I remember correctly). Then use that predicted probability variable as your Test variable for ROC.
If you do a Google search on , you'll find some examples including some YouTube videos. HTH.
I already did the additional value of multiple biomarkers in roc curve using binary logistic. But I was asked by a reviewer to adjust this by comorbidity scales. As far as I understood he/she wanted me to do the biomarkers minus comorbidity and see the level of biomarker prediction if all the patients are without comorbidities. As far as I went you can do that with logistic regression but it doesnt give an option to represent the results in roc curve.
Yes, SPSS does not give it automatically (AFAIK). You need to save predicted probabilities, as Bruce said, and then use ROC procedure to draw the curve after including your covariates into logistic model.
Hello! I did exactly what you said and got roc curves. The results showed that for one biomarker the roc curve is lower when comorbidities are added. I guess that this can also be taken as a valid result.
It came accross my mind that maybe I should use hierarchical binary logistics in order to measure the difference between biomarker results and comorbidities. And I still get the predicted values for roc. Should I consider this or not?