Repeated measures for ordinal data :)

Discuss statistics related things

by dr_Primus » Sat May 04, 2019 6:04 am

Hello, this seems like a stupid question but I am not a full time statistician ;)

Anyway I have 5 measurements (baseline, 2, 6, 12 and 24h) of analgesic requirements which I ordered 0-none, 1-NSAID and 2-opioid. And would like to test for difference between groups of patients.

I know that Friedman test is intended for ordinal data RM but it is not possible to test between groups (at least in jamovi, jasp, past and blue sky statistics :)

I would like to know if I can treat that data as categorical variables (since I have them labeled that way) and just perform chi squared for each measurement.

Thank you
dr_Primus
 
Posts: 23
Joined: Wed Dec 20, 2017 12:07 pm

by jonathon » Sat May 04, 2019 7:14 am

hmm, this is a bit out of my depth.

i think what the right analysis here, will depend a lot on your specific research question, etc.

what's the pattern in the data? do people start out on opioids, move on to NSAID, before going onto 'none'? i'd almost be inclined to model this with (two) survival analysis (or life tables, i guess).

the nice thing about survival analysis (or life tables) is the results are very easily interpreted.

alternatively, if one of the groups is an intervention in which they hope to reduce the amount of pain-killers people require, you could create two sums for each participant. one sum representing how many times the person required an opiate (between 0 and 5), and one sum representing how many times a participant required an opiate *or* NSAID (between 0 and 5). then run two mann-whitney's.

but yeah, there's usually a way you can simplify your design that more neatly matches the question you're asking ... and if you've got enough data/power, then it may not be an issue.

jonathon
User avatar
jonathon
 
Posts: 896
Joined: Fri Jan 27, 2017 10:04 am

by dr_Primus » Sat May 04, 2019 9:49 am

Thanks for your reply.
Well everyone starts with none (baseline is before surgery), and two groups get different epidural drugs before incision and we tracked the patients' analgesic requirements during the early postoperative period.

Do you think I could do this with death watch module (Kaplan Meier curve for analgesic requirement)? How should I format the data then?
dr_Primus
 
Posts: 23
Joined: Wed Dec 20, 2017 12:07 pm

by jonathon » Sat May 04, 2019 12:21 pm

in order for survival analysis (or life tables) to work, the analgesic requirements would need to be always decreasing. so if any patient went from NSAID *up* to opiates, then that would rule out a survival analysis.

also, the fact that your time intervals aren't evenly spaced 0, 2, 6, 12, 24 *might* be a problem for a survival analysis too (i really am out of my depth).

is analgesic requirement always decreasing?

jonathon
User avatar
jonathon
 
Posts: 896
Joined: Fri Jan 27, 2017 10:04 am

by dr_Primus » Sat May 04, 2019 4:18 pm

Yes... Always decreasing.

Btw... I just wanted to say that one article I wrote and in which jamovi was used for statistical analysis (and JASP was used for graphing) has just been accepted for publication in journal of clinical monitoring and computing.

Thank you for everything
dr_Primus
 
Posts: 23
Joined: Wed Dec 20, 2017 12:07 pm

by jonathon » Sat May 04, 2019 11:18 pm

hey, 'always decreasing' simplifies this a lot. i'd recode the data so that you have two columns for each participant. column 1 with the time point they stopped needing opiates, column 2 with the time point they stopped needing analgesics altogether (they'd have 5 levels, 2h, 6h, 12h, 24h, more). ordinal dependent variable + categorical explanatory variable = mann-whitney U.

congrats on the paper! (you'll have to fill us in on why you couldn't use jamovi for your plots).

cheers

jonathon
User avatar
jonathon
 
Posts: 896
Joined: Fri Jan 27, 2017 10:04 am

by dr_Primus » Sun May 05, 2019 1:42 pm

Needless to say I quoted jamovi, but unfortunately not in a way that is present in latest versions.

I didn't use jamovi for the plots because at that moment I liked better JASP's black and white plots - found them more legible compared to shades of grey in jamovi - and some journals have a fee if there are colored pictures or plots.

However, after learning the Inkscape trick, I don't need JASP for plots anymore - I just change lines from solid to dashed.

Btw... here's a feature suggestion - it would be nice to have significant between group interactions (after post-hoc tests) labelled with an asterisk or something similar.

Anyway... I have attached a plot to demonstrate how I changed the plot in Inkscape (and have no need for JASP) and added an asterisk.
Attachments
Slika zaslona 2019-05-05 u 15.41.00.png
Slika zaslona 2019-05-05 u 15.41.00.png (42.83 KiB) Viewed 730 times
dr_Primus
 
Posts: 23
Joined: Wed Dec 20, 2017 12:07 pm


Return to Statistics