I am confused with the two kinds of cohort analyses composed of discovery cohort and validation cohort shown in Figure2 e-h in the attached file.According to ROC curves, both analysis tells that exosome concentration is superior to exosome size to detect the malignancy. According to ROC curves. Why the authors have to demonstrate ROC analyses twice in the different way? What is the difference between discovery and validation cohort?
http://www.nature.com/nature/journal/v523/n7559/full/nature14581.html
Kindly follow the links below.
http://journals.plos.org/plosmedicine/article/file?type=supplementary&id=info:doi/10.1371/journal.pmed.1001916.s009
http://www.proofcentre.ca/resources/biomarker-solutions/what-is-cohort/
https://aasldpubs.onlinelibrary.wiley.com/doi/full/10.1002/hep4.1188
Hello Go J Yoshida, as far as I know, your problem is somewhat different from how you are looking at it. Unless things have changed a lot, you will find nice insights after reading the article I am suggesting to you:
Sackett DL, Haynes RB “The architecture of diagnostic research”. BMJ 2002; 324:539-41.
I hope this will help you.
Sincerely,
Suppose, if we identify different patient subgroups with distinct alterations of specific marker, that will be discovery. And for validation of these findings, the data from a sub group of patients will have to be taken as follow up study, this is validation.
Discovery cohort are essentially data collection, wet/dry experiments and building models while Validation cohorts are machine learning steps with the input from the discovery cohort to TEST and APPLY it to suspecting population (or sub-population) to check and see if the model or predictions hold true. So you build a strong model in discovery cohort and apply the suitability of this model in a different population/sub population to see your power of prediction in a Validation cohort
Using biomarkers as an example: Discovery cohort is the study population (cases of condition A and controls) from which you collect blood and screen for metabolite hits. You then identify a narrow number of metabolites from the hit that differ significantly between cases of condition A and controls. Now you want to know if a good candidate metabolite from your discovery could be used to diagnose or predict condition A: This is when you recruit a new set of condition A case population with controls (validation cohort) to confirm diagnostic utility of the biomarker. I hope this helps. If it is still not clear, I will be at TMDU on January 27th, 2017.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics