1. ## Change over time

Hi all,

I am looking at patient activity for a disease and have patient numbers for 5 years (2009-2013). I have numbers for 4 different types:

Alcohol-induced unique patients 153, 151 171 176 162
Alcohol induced discharges 215, 202, 226, 236, 222
Non-alcohol induced unique patients 394, 424, 459, 429, 460
Non-alcohol induced discharges 538, 548, 583, 552, 777

I've been advised to just state the % difference between 2009 and 2013, but I don't think this is the correct way to look at the data. Is there any way, with these limited data, to say whether or not there is a trend (increase) over time? I don't believe there is... but it would be nice to be able to say definitively.

I am currently using SPSS.

2. ## Re: Change over time

Well if you have proportions you are examining over time you may be able to use the armitage and Cochrane trend test (sp?).

I couldn't tell if you had more than two groups, but if you do there is a different test that I forget the name of, named after a person with a j-name, jaensky or something.

Also, it may be possible to use logistic regression with year as a continuous prediction variable, so 1,2,3,4 and see if the its coefficient is significant.

3. ## Re: Change over time

One simple way to go about it would be to create a 4x5 table and conduct a Pearson's chi-square test. The null hypothesis would be that the proportions of each type are the same over all 5 years. A rejection of the null would not inform you of whether there's an increasing or decreasing trend, but you can probably tell just by eyeballing it.

Just as a word of caution, though, since you have rather large sample sizes, even minute differences may appear as significant. So judgement should be exercised in interpreting whether the percentage change over the years really "matter" or not.

 Tweet