Center to Advance Palliative Care

There are a couple of possible solutions to dealing with LOS outliers. I'm not a big fan of eliminating cases when producing results for internal consumption in a practice, hospital, or system... and even less of a fan for using an arbitrary cut-off point of 14 or 30. In general, you should use medians rather than means to describe average LOS, to reduce the impact of a few outliers on an accurate description of the larger population or group. This does not require you to change the "n" -- the number of cases in your report -- from one analysis to another. If you do decide, however, to pursue the tactic of eliminating outlier csaes for some or all analyses, then yes, have someone first produce the mean & standard deviation for the relevant metrics (LOS, cost, etc.) and then eliminate cases whose metric value is greater than 2 (or 3) standard deviations above that mean. This will affect the "n", the mean, and the standard deviation of the remaining sample. This is more frequently done when publishing research studies or when comparing one institution to others or to industry-wide benchmarks. I don't recommend using propensity score-based matching for internal analyses. Let me know what you think. - Brian Cassel, VCU Palliative Care Leadership Center.