I am evaluating risk for HT and DM in different ethnic groups.
I wonder if anyone has used ROC to determine different cut off points for BMI or Waist circumference to determine levels of increased risk and how this was done
Since most my risk factors give about 65% AUC I also wonder how to combine these.
Can anybody help?
Hi,
For your question:
ROC is the most appropriate analysis for cross sectional study or research.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4097154/
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3959186/
http://apps.elsevier.es/watermark/ctl_servlet?_f=10&pident_articulo=90090723&pident_usuario=0&pcontactid=&pident_revista=326&ty=57&accion=L&origen=zonadelectura&web=www.elsevier.es&lan=en&fichero=326v27n05a90090723pdf001.pdf
http://www.dmsjournal.com/content/2/1/68
http://www.scielo.br/scielo.php?pid=S0066-782X2006001700011&script=sci_arttext&tlng=en
http://www.scielosp.org/pdf/rbepid/v17n1/1415-790X-rbepid-17-01-00243.pdf
An AUC of 0·65 doesn't sound very convincing. it means that, in round figures, there's a 65% chance that the hypertensive or diabetic person will be fatter than the normotensive or non diabetic and, so, a 35% chance that the normotensive or non diabetic will be fatter.
When you say you want to combine the risk factors, do you mean that you want to create a risk estimation system based on several risk factors? If so, my first reaction is that this has been done often, so you would need a strong reason to assume that risk factors operated in a different way in your population.
One limitation is the size of your study. Any multivariate model will "shrink" when applied to new data : it will predict less well. The smaller the sample and the greater the number of predictors, the greater the shrinkage.
- Badges
- Science topic