Dear all,
I'm using ROC analysis to identify the capacity of obesity indicators (BMIzscore, WCzscore, Waist to height ratio) to identify the caridovascular risk factors in children. I want to examine whether the AUC are different by sex, or by the level of obesity or not.
The first method, I'm using following command I saw there are different in AUC by graph (AUC in obesity is higher than in oveweight). However, I can't see exactly the AUC in each subgroup. For example, AUC in children with obesity or children with overweight (command and graph below)
rocreg E_BP z_wth [pweight = weight], probit ml roccov(who) cluster(school id) ctrlcov(age sex) noomitted
rocregplot, at1(who =0) at2 (who =2) at3 (who=3) legend(order(4 "reference" 1 "normal" 2 "overweight" 3 "obesity"))
The second method, I subset the dataset in level of obesity such as creating a group just only children with obesity and using the command
roctab E_BP wth
ROC -Asymptotic Normal--
Obs Area Std. Err. [95% Conf. Interval]
------------------------------------------------------------
2,076 0.5595 0.0129 0.53413 0.58485
The result showed AUC but it is quite low compared with the first method. So I'm confused that which method I should use.
Could you show me how to conduct the analysis to answer the question whether capacity to identify cardiovascular risk factors different in different level of obesity and whether Wais to height ratio add any value to to BMIz to identify CV or not?
It would be appreciated if you can support me.
I'm using ROC analysis to identify the capacity of obesity indicators (BMIzscore, WCzscore, Waist to height ratio) to identify the caridovascular risk factors in children. I want to examine whether the AUC are different by sex, or by the level of obesity or not.
The first method, I'm using following command I saw there are different in AUC by graph (AUC in obesity is higher than in oveweight). However, I can't see exactly the AUC in each subgroup. For example, AUC in children with obesity or children with overweight (command and graph below)
rocreg E_BP z_wth [pweight = weight], probit ml roccov(who) cluster(school id) ctrlcov(age sex) noomitted
rocregplot, at1(who =0) at2 (who =2) at3 (who=3) legend(order(4 "reference" 1 "normal" 2 "overweight" 3 "obesity"))
The second method, I subset the dataset in level of obesity such as creating a group just only children with obesity and using the command
roctab E_BP wth
ROC -Asymptotic Normal--
Obs Area Std. Err. [95% Conf. Interval]
------------------------------------------------------------
2,076 0.5595 0.0129 0.53413 0.58485
The result showed AUC but it is quite low compared with the first method. So I'm confused that which method I should use.
Could you show me how to conduct the analysis to answer the question whether capacity to identify cardiovascular risk factors different in different level of obesity and whether Wais to height ratio add any value to to BMIz to identify CV or not?
It would be appreciated if you can support me.