For the video version of this post, click here. For the cost of a single night in a typical American intensive care unit, about $7500, you could stay for 10 days at the all-inclusive Overwater hotel in Bora Bora. For this reason, many health economists have looked at ICUs as something of a necessary evil. They are a requirement for the advanced care the US health system can deliver, and at the same time the embodiment of a system that directs way too many resources to care occurring at the end of life.

The problem when you study the effect of ICU admission is called confounding by indication. Sicker patients are more likely to get admitted to the ICU, and more likely to die, so observational studies tend to make ICU care look bad.

Thats what makes this study, appearing in JAMA, of pneumonia admissions by researchers from the University of Michigan so interesting.

They wanted to know if ICU admission improved outcomes for Medicare patients with pneumonia. Instead of relying on a traditional multivariable adjustment approach, they used a technique called instrumental variable analysis.

An instrumental variable is one that is associated with your exposure of interest, in this case, ICU admission, but not associated with the outcome of interest (in this case, death at 30 days) except via that exposure. The idea, then, is that the instrumental variable acts like a randomizer to one or another treatment, allowing us to fairly compare them. In this analysis, the researchers chose distance from a hospital that used their ICU a lot as the instrument.

To believe the results, you have to buy that the closer you live to a hospital with high ICU utilization, the more likely you are to be admitted to an ICU. Not exactly a stretch, and indeed, the data shows that people with pneumonia who live within about 3 miles of such a hospital get admitted to the ICU 36% of the time compared to 23% of the time among those living further away. You also have to buy that distance to such a hospital is not associated with death - this is quite a bit trickier to prove.

If you trust the instrumental variable, what you find is that ICU admission was associated with less mortality than ward admission, 15% versus 21%, and less medicare costs by about $1300 dollars.

Now, the results of an instrumental variable analysis should be interpreted as applying to the marginal patient. In other words, the benefit the researchers saw applies to those that could reasonably be admitted to the ward or the ICU.

While I have some concerns over the quality of the instrument, I tend to believe these results. The authors, prudently, call for a randomized trial to evaluate the effects of ICU admission on older patients with pneumonia, but for now, I hope their voice sticks in your head when youre seeing a patient with pneumonia in the ER and thinking should I send them to the unit? If youre asking the question, the answer seems to be yes.