 |
 |
Wait a minute You loose info if you don't calculate LR(-) For A about 0.6, for B about 0.2 and for C 0 The intuition is to choose C as a screening test. If it is negative (LR-) you rule out disease. To rule in disease all would be equally useful (LR+) as demonstrated. You do not loose info if you are aware of all LRs for the test. I definitely like LRs more when one deals with non-dichotomous results. That is when they really shine! Victor -----Original Message----- From: Sachin Dave [mailto:[log in to unmask]] Sent: Monday, September 20, 1999 7:33 AM To: Atle Klovning Cc: [log in to unmask] Subject: Re: Likelihood ratios (and a CAUTION) Do we really loose any information when it comes to applicability of LR's to a clinical problem ? Let us say for example there are 3 tests A, B, C with sens. and spec. as outlined by you for a given disease. My bottom line would be what is the post-test probability of that disease given a positive result of test A or B or C. Does post-test probability change ? Would clinician care for the sens. or specificity of the test A or B or C ? What he wants to know is how sure he can be about a disease given a positive test result. Sachin Dave, MD. Atle Klovning wrote: > The holy LR is a vast step forwars in understanding tests- BUT when > combining sens and spec into ONE measure- we lose information. E.g. at > times having a test with high specificity is very important, whereas at > other times high sensitivity is the thing you want. > > Also, and quite disheartening: > > A sens 40% and spec 98% => LR+ =20 and > B sens 80% and spec 96% => LR+ =20 and > C sens 100% and spec 95% => LR+ = 20 > > A, B and C should generate very different test interpretations > > ******************************************************************* > Atle Klovning, MD, Research Fellow, > Specialist in General Practice/Family Medicine > Div. for General Practice, > University of Bergen, Norway > (+47) 55 58 67 09 > (+47) 55 58 67 10 (FAX) > > http://www.uib.no/isf/people/atle/ > > ******************************************************************* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%