# Clustered data in ophthalmology, which statistical test to use?

I am doing a retrospective chart review comparing old treatment regimen (120 shots of laser) to a new treatment regimen (160 shots of laser). The main outcome measured is intra-ocular pressure (i.e. dependent variable).

Therefore my independent variable is dichotomous, and my dependent variable is continuous.

The problem is that in the study, some patients have submitted 2 eyes into the study, where as some patients have only submitted 1 eye into the study. Therefore to do an independent sample t-test would not account for the fact that some patients have submitted 2 eyes into the study, and these 2 eyes are likely to be similar in nature as they are from the same person.

What statistical test could I use for this? And would I be able to account for other variables, e.g. age, gender etc.?

I am currently using STATA.

You could use a linear mixed model to analyse this. You can include whatever covariates you are interested in (age, gender etc).

I specify a mixed model because you have both eyes for some patients; you can account for this by including 'patient id' as a random intercept. This will implicitly estimate a mean effect per patient (as suggested by @Łukasz Deryło) without any manual averaging step.

I do not know how to use STATA, but this is a reasonably simple model and should be easy to execute.

• Thanks for the help. What if I was comparing baseline characteristics e.g. 'number of pre-treatment medications' or 'age' between the two treatment groups. Again I would like to take into account that some patients have only submitted one eye and some two eyes. Would mixed effects linear regression still be possible for this, as I understand that 'number of pre-treatment medications' and 'age' are technically not dependent variables Jul 6, 2017 at 11:53
• I think a linear mixed effects model would be quite reasonable in the cases you describe here.
– mkt
Jul 6, 2017 at 19:46

I'd replace two results from people who submitted both eyes with their mean. After checking, of course, if results from left and right eye are in fact correlated.

• Thanks for the help. What if I was comparing baseline characteristics e.g. 'number of pre-treatment medications' or 'age' between the two treatment groups. Again I would like to take into account that some patients have only submitted one eye and some two eyes. Would mixed effects linear regression still be possible for this, as I understand that 'number of pre-treatment medications' and 'age' are technically not dependent variables Jul 6, 2017 at 11:49