Regret graphs, diagnostic uncertainty and youdens index

Jørgen Hilden*, Paul Glasziou

*Corresponding author for this work

Research output: Contribution to journalArticleResearchpeer-review

155 Citations (Scopus)

Abstract

Regret is the difference in outcome between the action we took and the best action we could, in retrospect, have taken. Tent graphs, representing decision problems under uncertainty in terms of expected regret, offer an instructive geometric supplement to standard formulae, allow instant calculations, and suggest, as we shall illustrate, novel algebraic properties. For instance, the problem of finding the pre-test probability range in which it is worthwhile doing a diagnostic test becomes trivial, even when test costs depend on the unknown disease status; we take the opportunity to explore the (non-trivial) algebraic properties of this problem. The graphs also provide a simple way to illustrate and read off the expected value of information and expected value of perfect information. This property is used to derive a clinical interpretation of Youdens Index (sensitivity + specificity - 1), namely, it is the maximum proportional reduction in expected regret achieved by the test. Finally, we examine the relationship between Youdens Index and the area under an ROC curve.

Original languageEnglish
Pages (from-to)969-986
Number of pages18
JournalStatistics in Medicine
Volume15
Issue number10
DOIs
Publication statusPublished - 1996
Externally publishedYes

Fingerprint

Dive into the research topics of 'Regret graphs, diagnostic uncertainty and youdens index'. Together they form a unique fingerprint.

Cite this