Department of Epidemiology and Preventive Medicine (DEPM)

Logistic Regression when the Outcome is Measured with Uncertainty

Implements the method described in the paper, "Logistic Regression when the Outcome is Measured with Uncertainty," by Larry Magder. Published in the American Journal of Epidemiology 1997, 146:195-203.

This macro can be used to perform logistic regression when the binary outcome of interest is measured with an imperfect test, and the sensitivity and specificity of the test are known. It allows for the possibility that the sensitivity and specificity differ for different observations in the dataset.

A more general macro will be written soon which will make it possible to use all available information regarding the presence of a particular outcome (e.g. the results of a continuous diagnostic test, or serial diagnostic tests) in the fitting of logistic regression models.

View more documentation about the macro code.

View Macro code (in plain text format).

Comments:

Notice that you have to create your own intercept variable.

If you wanted to allow for differential misclassification you could allow the value of sensitivity and specificity to be different for different groups. For example, you could add a statement such as "If sex="M" then sensvar=.9;

One way to test the program is to run it with both sensitivity and specificity set to 1.00 for all observations. The results should be the same as those produced by PROC LOGISTIC.

Please direct any questions or comments to lmagder@epi.umaryland.edu.

Printer friendly    Email this page