EPI204 Forum

An MP4 of the 2018 version of this lecture is available to all registered students.

We will ask classroom support to look for this year's recording, but the link that is there now will take you to a useless 1.5-minute video that ends with me saying something like "What does this button do?"

During this year's course review, I did the following:
-- gave an example of calculating kappa that is summarized in a posted Excel file
-- also gave an example of going from an ROC table to an LR table that is also posted in an Excel file
-- threatened (facetiously) to fail anybody who calculates LR(+) and LR(-) for each row in an ROC table, i.e. at each of several possible cutoffs
-- re-iterated a point made previously in this forum about calibration errors: they matter only when predicted risk and observed proportion with the outcome are on opposite sides of the treatment threshold (Ptt).
* Predicted risk below Ptt and observed outcome proportion above Ptt leads to under-treatment.  If the miscalibration occurs in the lowest risk group, the decision curve will drop below the "Treat All" curve.
** Predicted risk above Ptt and observed outcome proportion below Ptt leads to over-treatment.  If  the miscalibration occurs in the highest risk group, the decision curve will drop below the horizontal axis, which is also the "Treat None" curve.