Odds Ratios anyone?

[echod]

---

Members do not see this ad.

In table 3, having a PhD degree gives an odds ratio of 2.88 in matching into a competitive specialty. If you increase your odds by 2.88 times by having a PhD degree why does the paper rank it only 5th in terms of how well the factor predicts match success. Other criteria (except step I) that rank above it all have lower odds ratios.

http://www.nrmp.org/data/chartingoutcomes2007.pdf

 

[JHopRevisit]

Very interesting, thanks for posting this. Unfortunately, I can't help, and now I feel dumb, I need to take probability again.

How is the odds ratio for "USMLE < 200" just 1? I imagine we're comparing the group that got into their choice of competitive specialty with one that didn't. The odds that someone in the competitve specialty group has a USMLE < 200 is the same as the odds for someone in the denied group? That can't be right, can it? Obviously I'm confused about what we're comparing and/or don't understand odds ratios, so any help would be appreciated.

I'm sure we'll get an explanation here soon - this board has often argued about these sorts of issues. The success in matching into a preferred competitive specialty should provide good info on how valuable the PhD is in the match overall.

 

[mrsanator]

Check out footnote 5 on page 15 of that report &#8211; it says that they used log likelihood testing to judge the value of the different parameters. I'm not very familiar with this test, but it seems to involve putting each variable in and out of the regression and comparing the quality of the resulting model. I think the point is that because the vast majority of applications don't have PhDs, the addition of an MD/PhD variable adds less to the overall model than things where the population varies a lot (such as number of programs to match or step scores).

The odds ratio for step 1 < 200 is 1 because it was set that way. Remember, an odds ratio is a ratio &#8211; it has to be relative to a baseline. They set <200 as their baseline. If they had set 240 as their baseline, that would have equaled 1 and the <200s would have been < 1. The OR in this case is the odds of matching given your step 1 score relative to those students who scored <200.

A residency prediction model without 3rd year grades is pretty sketchy - I think it's fair to say that MD-PhD seems to help some, and that help is independent of board scores.

 

[JHopRevisit]

Thanks mrsanator, as I expected I just didn't understand what comparison we were making, which I could have if I had bothered to read the paper rather than just glance at the table. The odds ratios are between groups within each category, but in the case of binary results (PhD or no PhD) there was no reason to put the standard group in the table. That's what led me to think we were calculating odds ratios for a different set of groups.

Interesting. I agree with you that the failure to include clinical grades is a major weakness. Now, do we have any data on whether MD-PhDs are known to have higher clinical grades than their peers? If not, then the PhD seems to be a big help in matching into competitive specialties.