There’s a bit of excitement on Twitter today with a number of tweets and retweets about a paper just published in the Annals of Family Medicine which shows that uninsured patients are being released from hospitals significantly sooner than insured patients. The numbers don’t lie.
From the abstract: “Across all hospital types, the mean length of stay … was significantly shorter for individuals without insurance (2.77 days) than for those with either private insurance (2.89 days, P = .04) or Medicaid (3.19, P <.01).” These are statistically significant differences.
The authors conclude, “Future research should examine whether patients without insurance are being discharged prematurely.”
Let’s look a little closer at these numbers. The difference between the uninsured length of stay (2.77 days) and those with private insurance (2.89 days) is 0.12 days or to put it another way, 2.9 hours.
Do you really think that a difference in hospital length of stay of less than 3 hours is really clinically significant? I don’t.
Here’s another problem with the paper. Length of stay is what is called a “soft” endpoint. Having practiced surgery for 40 years, I can assure you that length of stay is very often not determined by the type of illness, treatment rendered, skill of the physician or any other parameter you can think of.
Here is what I mean. Just yesterday, a patient told me he could not go home on the day he had his laparoscopic cholecystectomy because his sister, whom he lives with, gets upset whenever he comes home from the hospital. He felt she needed another day to adjust. Patients have told me, “No one can come and pick me up today.” The care manager says, “The bed at the nursing home isn’t available today.” Three weeks ago we couldn’t send some patients home because there was a massive power outage in our area. This list of excuses goes on and on.
I have written before about the problem of things being statistically significant but not clinically significant.
The paper is another example of statistical significance not corresponding to clinical significance.