## Tuesday, September 01, 2009

### That Cause-And-Effect Thingy Again...

I've written before about the importance of not assuming causality in studies which only find correlation, and I plan to write more about the oh-so-common misuse of simple correlation between two variables as somehow the Final Word On Something when the situation might be much more complex and full of sneaky omitted variables. Today's topic is related to that one but also serves as an example of some other difficulties that one stumbles upon in interpreting empirical research.

Sounds like fun, eh? I can't make it simpler because I'm fatigued (says she while reclining on her recamier). Here's the news that provoked all this:

No vacationer plans on getting sick, but many do fall ill, and seriously. All too often they land at hospitals that are anything but temples of healing.

In the popular sitcom Royal Pains, ritzy folks in the Hamptons hire a concierge doctor to tend their ills rather than an inept local hospital.

In reality, it's no comedy. A USA TODAY analysis finds two dozen hospitals near popular travel destinations, as compiled by the National Travel Monitor, have death rates among the worst in the USA. A separate analysis shows that one of every four hospitals with high death rates for heart attack, heart failure or pneumonia — 94 of 402 — are near state parks.

The quoted article recommends that travelers do some checking before picking a particular travel destination. But here's the problem with this interpretation: A hospital could have higher death rates for the very reason that it's close to a large tourist attraction. Tourists, by definition, are strangers to the place, far from their own doctors and their medical records, and that combination is unlikely to improve the outcomes of any illness attacks they may have.

This isn't necessarily the case, of course. It could be that the discussed hospitals just have worse outcomes, even when they treat local people. But in general outcomes are only meaningful if we control for the types of patients which enter the hospital. If those patients are, on average, high-risk cases then even an excellent hospital can look bad in the outcome statistics.