topic Tests for comparing hospital lengths of stay in Statistical Procedures

Tests for comparing hospital lengths of stay

julicny — Fri, 04 Jan 2013 20:20:05 GMT

Hello all.

What are the most commonly used statistical tests for comparing the average length of a hospital stay? Should I be thinking in terms of comparing the medians instead of the averages? The data I have come from two separate groups of patients undergoing the same procedure. Thanks in advance.

Re: Tests for comparing hospital lengths of stay

MohammadFayaz — Fri, 04 Jan 2013 20:27:37 GMT

Truncated means and trimmed means are another approach for testing your problem.

Re: Tests for comparing hospital lengths of stay

Reeza — Fri, 04 Jan 2013 20:28:12 GMT

Check the distribution of the data. If normally distributed then use averages, times are rarely normally distributed though, so median tends to be used instead.

Another option is to use log(time) instead, but again you'll need to verify the distribution of the data.

Re: Tests for comparing hospital lengths of stay

MohammadFayaz — Fri, 04 Jan 2013 20:47:13 GMT

You can compare it with logistic regression ( for example if a patient stay more than 7 days 1 else 0) for each hospital.

(check proc logistic and proc genmod in SAS)

Re: Tests for comparing hospital lengths of stay

PGStats — Sun, 06 Jan 2013 03:52:01 GMT

Look at NPAR1WAY example :

SAS/STAT(R) 9.3 User's Guide

If you can show a significant location difference with the Wilcoxon test then you needn't look any further.

Re: Tests for comparing hospital lengths of stay

1zmm — Sun, 06 Jan 2013 18:38:23 GMT

You might also look at the differences in the distributions of the length of stay for the two groups of patients using PROC LIFEREG (multiple parametric distributions) or PROC LIFETEST (nonparametric distributions).

Re: Tests for comparing hospital lengths of stay

julicny — Tue, 08 Jan 2013 16:13:51 GMT

Thanks for all the suggestions. I did the analysis using NPAR1WAY and did find significance with the Wilcoxon test. The LIFETEST and LOGISTIC approaches seem interesting as well, but I am not sure I will have the time to pursue this analysis further.