ITT vs Per Protocol in RCT

[ashar ata]

if ~2% of my data is missing on the outcome (continuous scale), out of a total of 200, two in control and three in intervention group, do I need to impute?
Or can I make a case that with such small missing data PP approximates ITT.
If I do impute, am I better off doing MI or can I do LOCF.

The study basically compares the improvement from baseline at time point 5 and there are 3 other time points between baseline and time point 5, where scores were taken. So I can carry the observation at 4th time point forward instead of MI for 5 missing.

I think I know the answer that the order of preference would be:
1. Multiple Imputation
2. LOCF
3. Make a case PP was as good as ITT because the missing data was minimal and there was nothing extra ordinary about the 3 patients that had missing data at time point 5.

Thanks for your input.

[Rich Goldstein]

1. I think there is no need to impute with this small an amount of missing data
2. I don't think that LOCF is ever acceptable

[ashar ata]

Does that mean # 3 would be correct?

Would my analysis be considered ITT?

Would the following statement be valid "The analysis was limited to patients with non-missing data on primary outcome. Missing data on primary outcome was due to loss to follow-up (patient not available). There were no patients that switched treatment and percentage of missing data was 2%. We therefore consider our analysis to be intention to treat."

[Joseph Coveney]

None of your three options for handling missing data will make the analysis ITT if it isn't otherwise already so. ITT is defined in terms of which treatment group the patient's outcome is analyzed in: according to the treatment group as randomized (experimental or control) versus the treatment, if any, actually received.

In a conventional ("superiority") study, what is sometimes done in order to assess the influence of missing values is to perform a "sensitivity analysis", which is to assign therapeutic success as the outcome to all missing-outcome patients in the control treatment group and therapeutic failure as the outcome to all missing-outcome patients in the experimental treatment group. If the worst-case analysis doesn't materially affect the conclusions drawn from the study's results, then you are in good stead. Unless your results are weak, with three patients' outcomes missing out of 200, I'm guessing that it would be unlikely that assigning worst-case outcomes for them would materially affect the conclusion, but you might want to go through the motions in order to be ready to respond to a referee's demand.

"The analysis was limited to patients with non-missing data on primary outcome." That's okay for the primary analysis of the primary outcome, but I would recommend using all available data for the secondary analyses, for example, if you're going to do exploratory analysis on the other three time points, then don't throw out those data for the three patients who have missing values for the fifth time point.

"Missing data on primary outcome was due to loss to follow-up (patient not available)." You'll need a little more explanation than that, for example:

* in which treatment group was each of the three patients?

* why was each patient "not available"?
- moved away?
- missed visit due to scheduling difficulties?
- incapacitated or died as a side effect of the experimental (or control) treatment?
- died or became incapacitated or withdrew as a result of lack of efficacy of the experimental (or control) treatment?

"There were no patients that switched treatment . . ." ITT allows for patients to switch treatments.