Length of stay (LOS) data are very skewed and often have genuine outliers, making it a hard situation to model adequately. A common objective is to compare the LOS between different health care providers, after adjusting for some patient characteristics. I have used multilevel negative binomial regression in the past and then got BLUPs for the random intercept as predictions of each hospital’s log discharge rate ratio compared to the overall average. So I was interested to see this new paper just out in BMC Med Res Meth. They compare various approaches on New Zealand ICU data (and I was relieved to see my one included). Worth reading or bookmarking should you ever encounter LOS data.
Published by Robert