How to determine the best predictor of responses?
Exploiting the use of ROC to determine the best predictor for biomedical research
How do you know you have chosen the best biomarker or predictor? The easiest way to visualise is to plot a ROC curve, which also provides information on the sensitivity and specificity of a biomarker.
ROC curve values usually range between 0.5 to 1.0, where 0.5 (indicated by y = x line on the plot) indicates that the prediction on the biomarker is only as good as a coin toss. On the other hand, a value of 1.0 indicates a perfect biomarker or predictor, with 100% accuracy in prediction. Hence, a better biomarker is one with ROC values close to 1.0, as indicated by a ROC curve with steeper slope. You can plot a ROC curve with GraphPad Prism or with R, using the pROC R package.
Another frequent question: Is there a way to prove if a biomarker or predictor is statistically better than the other? This is where the DeLong test using the pROC R package will be useful.
As an illustration, we have used ROC curves to show that MCEMP1 and HLA-DRA expression levels were suitable candidates for early prognosis of severe COVID-19. The DeLong test further showed that combination of MCEMP1 and HLA-DRA (in red) outperforms HLA-DRA (in blue) for prediction of severe COVID-19 (See Figure above). You may check out our publication in eBioMedicine (Chan et al., eBioMedicine, 2023)for more details of the application!