I would like to find the optimal cut-point of HbA1c to predict prediabetes (using fasting plasma glucose and 2-h plasma glucose to define), by using ROC curve. I wonder if the subject with NDD should be kept in the model or excluded.
e.g. I have 100 subject: 50 normal, 30 prediabetes and 20 NDD (diabetes), to use HbA1c predict prediabetes, should I drop the 20 NDD and use remaining 80 subjects or use all 100 subjects?
I find if I exclude the NDD, the AUC of HbA1c predicting prediabetes is higher. Thank you.
I suggest do exclude in this first analysis and just measured the pattern of these subjects at least during 20-30 days to check the glucose values.
Hi,
Kindly go the following:
When you want to NDD (newly detected diabetes) be excluded when I want to find the optimal cut-point of HbA1c to predict prediabetes. The participants are not using any anti-diabetes drugs and find their HBA1C for some period. How its changes in pre-diabetes patients. Then you have to fix the cut off point of HBA1C.
Using patients HBA1C before anti-diabetes drugs are used we predicted the time interval from prediabetes progression to diabetes is found one study 2.49 years without any influence of age and gender.
Thanks.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2999978/
http://www.degruyter.com/dg/viewarticle/j$002fcclm.2011.49.issue-12$002fcclm.2011.728$002fcclm.2011.728.xml
http://guidelines.diabetes.ca/browse/Chapter3
You want to avoid Length Bias and Lead-Time Bias. Lead-time bias is when pre-diabetes is detected by screening for HbA1c, and it gives the impression that those who are screened have a longer pre-diabetic period than those who are not screened; but in fact both the screened and the non-screened develop diabetes have the same length of pre-diabetic time, for example. Length bias is where people with a long pre-clinical phase (i.e. a longer pre-diabetic phase) of diabetes are over-represented in screened populations simply because they're pre-diabetic for a longer period and more likely to be detected if screened.
If elevated HbA1c is in fact the early manifestation of diabetes itself, then you should not use the NDD to "predict" diabetes because they have elevated HbA1c. You want to determine who develops (and doesn't develop) diabetes after their HbA1c is measured, and measure it when your study population is still diabetes-free. I assume you will test every subject after a certain follow-up time to determine if they in fact have actual diabetes or not. With this information you can then determine a sensitivity and specificity of your HbA1c test, and adjust the cutoff values of HbA1c to optimize this sensitivity and specificity. If for some reason your subjects are being tested for Hb1Ac and diabetes at basically the same time, and this includes NDD patients, then you have a case-control study and cannot make it a screening study.
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology
- Clinical Pharmacology