ITT and per protocol analyses. How to do both and how to compare baseline measures.

#1
Hello,

I have read about but cannot find the answer to my queries.

I am doing a study looking at the effects of positive thinking (PT) and exercise on depression. I'll be measuring at LEAST 2 outcomes. So far, depression levels and quality of life score levels using standardise inventories, like the beck depression inventory. Possibly a few more. The hypothesis is that exercise and PT will lower levels of depression and increase quality of life but exercise will be more effective.

So, I have one IV (group type) with 3 levels (exercise, PT and a no-treatment control group) and a minimum of two DVs (depression level and quality of life levels).

So, I am thinking that I should perform a MANOVA when I have my results. However, I have a few queries.

First of all, I would obviously want to compare baseline measures to ensure between group homogeneity. What is the best way/are the best ways to do this? (Do I even need to perform an analysis?)

Secondly, I want to do both intention to treat (ITT; all participants randomised to each condition) and per protocol (PP; only participants who did a certain amount of exercise etc) analysis. They both have separate advantages and I would like to include both.

What is the best way/are the best ways to do this? Would I perform a separate MANOVA for each and then do some kind of adjustment like bonferroni? Or include them in the MANOVA as separate groups?

I think I have read somewhere that the same participants should not be used in a MANOVA unless it is a repeated measures but not sure about this.

Lastly, if I wanted to do the above study, with ITT and PP analyses AND wanted to take group measures at 5 points in time and compare these between groups, how would I best go about doing that? I'm thinking some kind of multilevel model, but none of my books cover this, and I don't know if that would be getting out of my depth. Alternatively, could I do a repeated measures MANOVA?

I am working with SPSS.

A lot here I know. As I said, I have read about a fair bit but can't seem to find instructions/recommendations of what kind of test to do in any of my books.

Any help would be very much appreciated.

Thank you muchly guys

:)
 
Last edited:
#2
Re: ITT and per protocol analyses. How to do both and how to compare baseline measure

So, another connected question.

The more stats test you run on a set of data the more you increase the chance of a type 1 error, right? If you're running a test (say a MANOVA) on baseline measures (pre treatment) to ensure between group homogeneity are you increasing the risk of a type 1 error if afterwards you're running another test (MANOVA) on the post treatment data? I'm guessing not as these are 2 data sets but I'm just wondering whether I have missed something and the more incidences of running a stats test in general increases type 1 errors.

Also, would this be the standard way to go about this (testing between group homogeneity and then post treatment scores with 2 separate MANOVAs)?
 
#3
Re: ITT and per protocol analyses. How to do both and how to compare baseline measure

Can anyone help? Do I need to give more information for this?
 
#4
Re: ITT and per protocol analyses. How to do both and how to compare baseline measure

Anyone? Anything? I think I've been quite clear but if I need to clarify further then please say.
 

Karabiner

TS Contributor
#5
Re: ITT and per protocol analyses. How to do both and how to compare baseline measure

So, I am thinking that I should perform a MANOVA
There is no clear reason given for this. MANOVA is appropriate if the DVs
jointly operationalize one hypothetical construct. I suppose that
depression and quality of life, although usually highly correlated,
do not operationalize one construct -- at least you didn't say so.
First of all, I would obviously want to compare baseline measures to ensure between group homogeneity.
Why should you do this? Somewhere you mention that this is a randomized
study. Do you mistrust your randomization procedure? BTW, sample size
is nowhere described.
Secondly, I want to do both intention to treat (ITT; all participants randomised to each condition) and per protocol (PP; only participants who did a certain amount of exercise etc) analysis. They both have separate advantages and I would like to include both.
What is the best way/are the best ways to do this? Would I perform a separate MANOVA for each and then do some kind of adjustment like bonferroni? Or include them in the MANOVA as separate groups?
I must admit that I do not quite understand what you are after.
What has Bonferroni to do with handling of missing data?
And what do you mean by separate groups? You either perform
ITT or PP, or both (but separately).

Lastly, if I wanted to do the above study, with ITT and PP analyses AND wanted to take group measures at 5 points in time and compare these between groups, how would I best go about doing that? I'm thinking some kind of multilevel model, but none of my books cover this, and I don't know if that would be getting out of my depth. Alternatively, could I do a repeated measures MANOVA?
It would be easiest for you if you only used 1 dependent variable in each analysis,
and possibly you then could use repeated measures ANOVA. A multilevel model
might (perhaps) even better apply to your data, but since you are not accustomed
with that, it would probably be justified if you sticked with ANOVA and/or repeated
measures ANOVA.

Just my 2pence

K.
 
#6
Re: ITT and per protocol analyses. How to do both and how to compare baseline measure

Sorry for the extremely delayed reply.

Just posting to say a sincere delayed thank you for contributing your time :)