Tuesday, October 5, 2021; 8:45 AM - 10:15 AM
Section outline
-
LGD: Continuous Tests
ROC Curves, multilevel likelihood ratios and the relationship between them; how dichotmozing tests wastes information; the Walking Man approach to ROC curves; choosing cutoffsFaculty: Michael Kohn
-
This version has one sheet for demo and one with the formulas
-
Lecture 3 Segment 1 -- Multilevel Tests -- Why not make the test dichotomous? (13:34) Media Resource
-
Lecture 3 Segment 2 -- the ROC Curve (17:45) Media Resource
-
Lecture 3 Segment 3 -- Interval Likelihood Ratios (22:34) Media Resource
-
Lecture 3 Segment 4 -- What test result should prompt treatment? (11:29) Media Resource
-
Lecture 3 Segment 5 -- computer-drawn ROC curves (33:30) Media Resource
-
2020 (!) Zoom Recording of Large Group Discussion 10/6/2020 (1:31:00) Media Resource
3:00 – Michael: Review of material covered in the first 4 segments
10:45 – Michael: “Walking Man” approach to understanding computer-drawn ROC curves
46:30 – Student questions
1:03:30 – Tom: problem on using BNP to diagnose peripartum congestive heart failure (CHF)
Includes a detailed discussion of a published ROC table and why NOT to report LR(+) and LR(-) at each of multiple cutoffs.
-
I tried downloading the Zoom recording that includes only the shared screen and audio, not the gallery of faces. It's a bit odd because, when I'm not sharing my screen, there is no video.
-
M.A. Kohn, Key concepts in clinical epidemiology: reporting on the accuracy of continuous tests, Journal of Clinical Epidemiology, https:// doi.org/ 10.1016/ j.jclinepi.2021.07.012 File
-
PSet #3 Answer Key File