Identifying optimal cut-point with continuous variables help

[Dana Knowles]

Hi folks!

I am trying to find optimal BMI for CVD risk using a continuous risk score. I'm familiar with ROC curves and Youden index, but do not want to reduce my continuous values to a binary variable. Any suggestions on a data-driven approach or package I should be using in STATA?

Thanks!

[Clyde Schechter]

This is actually much simpler than ROC curves. I suppose you have some data set with two variables: bmi and cvd_risk. To find the value(s) of bmi associated with the lowest cvd_risk:

sort cvd_risk bmi
list bmi if cvd_risk == cvd_risk[1]

[Dana Knowles]

Thanks Clyde! This certainly gives me the BMI values for all those who are low risk in my sample, but is it possible that I can generate some sort of "optimal" cut-point? I was thinking I would have to go with a spline method. My CVD risk has 4 levels and I would ideally want to find BMI cut-point (along with other anthropometrics) associated with each level of risk.

[Clyde Schechter]

What do you mean by "optimal" in this context, if it doesn't mean associated with the lowest CVD risk? It sounds like you want to fit some kind of parametric functional relationship between BMI and CVD risk and then find the value of BMI that minimizes that function. Well, this all depends on what kind of functional relationship you want to fit. So you need to make some decision about that. Splines are a good way to fit curves for variables in a regression that are included as covariates to adjust for their confounding effects, but for the use you are proposing here, I don't think they are very suitable.

So perhaps you can elaborate how you want to proceed. And, to be sure, no progress can be made without showing example data. Do use the -dataex- command for that to assure that it will be usable. If you are running version 18, 17, 16 or a fully updated version 15.1 or 14.2, -dataex- is already part of your official Stata installation. If not, run -ssc install dataex- to get it. Either way, run -help dataex- to read the simple instructions for using it. -dataex- will save you time; it is easier and quicker than typing out tables. It includes complete information about aspects of the data that are often critical to answering your question but cannot be seen from tabular displays or screenshots. It also makes it possible for those who want to help you to create a faithful representation of your example to try out their code, which in turn makes it more likely that their answer will actually work in your data.

Added: I also have reservations about the entire idea here. All CVD risk algorithms I am aware of make use of multiple variables. I'm not sure it is reasonable to assume that there is a single optimal BMI that applies to smokers and non-smokers alike, to diabetics and non-diabetics alike, etc.

[Dana Knowles]

Yes, I am looking to fit a parametric functional relationship! I've used ROC curves plenty in the past, but with 4 risk score categories, I do not want to end up with 30 curves by doing a one vs one approach (low vs high, low vs mod, low vs borderline) for each males and females. I figured using a data driven approach will maintain the continous risk score and easier interpretation than all those ROC curves. I will look into dataex to see if it will help answer my question.

Also, good point you raise about CVD risk algorithms and a single optimal BMI threshold. I will bring this up with my advisor. For reference, I'm using the latest recommended risk score calculator here: https://professional.heart.org/en/gu...ent-calculator

[Rich Goldstein]

I'm a bit confused about what you want, but I recommend you read Altman, DG, et al. (1994), "Dangers of using 'optimal' cutpoints in the evaluation of prognostic factors," Journal of the National Cancer Institute, 86(11): 829-835; the argument, also made later by other authors, is still important