|
Abstract:
|
Diagnostic tests are used to detect if a condition of interest is present or absent. Often, test results are missing for some patients. Reasons include a patient sample is not of sufficient quality and the test score falls within a zone of indifference. If missing results are ignored, estimates of test accuracy can be biased. E.g., patients with observed test results can be easier to call correctly. Missing results also complicate the comparison of two or more diagnostic tests when the set of patients with missing results is different for each test. In this talk, I will discuss methods imputing for missing test data, including worst-case scenario, Buck's method for multiple correlated tests, and imputations that are not correlated with presence or absence of the condition. The methods will be discussed in the context of how a missing test result could affect patient management.
|
