turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- How do you get diagnostic test evaluation (sens, s...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

05-24-2016 01:13 AM

Hi,

Can anyone help?

I have a binary dependent variable (outcome of interest: severe dengue) with independent continuous variables (admission Hct, admission AST, admission ALT and AST^2/ALT). 365 observations.

1. How do you get diagnostic test evaluation (sens, spec, NPV, PPV, LR+, LR-, AUC of ROC) with their 95% CI?

2. Is there a way to do it straight from the raw data table without creating separate dummy tables for each independent continuous variable? - how do you partition a continuous variable eg AST^2/ALT with interval of 100.

Help much appreciated!

Accepted Solutions

Solution

05-29-2016
10:19 AM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Dcicantab5

05-24-2016 07:41 AM

Hi @Dcicantab5,

Given that your independent variables are continuous, I would start with PROC LOGISTIC and compute ROC curves for each continuous independent variable. (Later you can continue with multiple logistic regression models in order to define new predictors involving two or more independent variables.) I did this with SAS 8.2 (and other statistical software) long ago, but relevant new features have been added to PROC LOGISTIC in SAS 9.2: please see this paper. This includes the calculation of the AUC together with a confidence interval.

The question of finding an optimal cut-point (to dichotomize a continuous independent variable) is discussed in this recent paper.

Once you have determined a cut-point, you have a 2x2 table and you can present sensitivity, specificity etc. for this cut-point.

All Replies

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Dcicantab5

05-24-2016 01:27 AM

Here's how you calculate it, based on a 2x2 table.

http://support.sas.com/kb/24/170.html

If you have continuous variables you'll need to categorize them somehow? You can use proc format for this so, yes, you can do it straight from the raw table.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Reeza

05-24-2016 07:21 AM

Ok, gotcha on the link you provided, thanks.

Now I am stuck again; the example in the link has binary outcome (same as mine) and BINARY predictor (vs. mine is continuous).

Do you know of similar guide on how to categorize the continuous independent variables into categories (ranges, eg 0-100, 101-200, 201-300,..) using PROC FORMAT? Now, let's say i have done the categorizing of a continuous variable, (?)INFMT, it MAY contain both options of the binary outcome e.g. range 201-300 may contain both outcome *severe dengue *and *non-severe*. How do i deal with that?

help very much needed!

Solution

05-29-2016
10:19 AM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Dcicantab5

05-24-2016 07:41 AM

Hi @Dcicantab5,

Given that your independent variables are continuous, I would start with PROC LOGISTIC and compute ROC curves for each continuous independent variable. (Later you can continue with multiple logistic regression models in order to define new predictors involving two or more independent variables.) I did this with SAS 8.2 (and other statistical software) long ago, but relevant new features have been added to PROC LOGISTIC in SAS 9.2: please see this paper. This includes the calculation of the AUC together with a confidence interval.

The question of finding an optimal cut-point (to dichotomize a continuous independent variable) is discussed in this recent paper.

Once you have determined a cut-point, you have a 2x2 table and you can present sensitivity, specificity etc. for this cut-point.