3
$\begingroup$

I've got a study in which patients can have from 1 to 5 aneurysms (concurrently) and each may be treated differently (each aneurysm). We are interested to see whether one treatment is different than the other and what risk factors may contribute to adverse effects.

I've set the data up so that we have one observation per aneurysm and not per patient. That means that one patient may be recorded upwards to 5 observations with a variable aneurysm_id denoting which aneurysm the observation is referring to.

This may look like so:

Patient1 --- Aneurysm_id --- adverse effect? --- treatment

---------------1 ---------------no ----------------1

Patient1

---------------2---------------no------------------2

Patient1

---------------3---------------no------------------2

This patient has three aneurysms and therefore constitutes 3 observations. The patient had no adverse effect on any of the aneurysms and two of them were treated with treatment 2, and one was treated with treatment 1.

I'm running a mixed model grouping by aneurysm_id.

In the model I have alcohol consumption as one variable and I'm just confused on how the results can be significant for alcohol consumption = 3 as this is the table for alcohol consumption and the adverse effect. Alcohol consumption = 3 does not seem to be any different than the others, constituting about 94% without the adverse effect for all groups? Yet in the model it shows an increased risk with a coefficient of 1.06 (mixed logit) and a p value of 0.015.

Table (I did try to format into a picture, but it wouldn't work).

Does anyone know why this is?

$\endgroup$
3
  • 1
    $\begingroup$ You said you have "one observation per aneurysm and not per patient", but then "I'm running a mixed model grouping by aneurysm_id.". I would have thought the grouping variable should be patient ID $\endgroup$ Aug 24, 2020 at 20:21
  • $\begingroup$ Hi Robert, thanks for clarifying that point, it has been causing me great confusion as I thought this as well at times, but thought it was grouping per aneurysm for some reason as well. Grouping by record_id (patients) is causing a great deal of computation time, though...Perhaps I should try in R or python instead of Stata. $\endgroup$
    – Paze
    Aug 24, 2020 at 20:33
  • $\begingroup$ Well for some reason it's not achieving convergence when grouping by patients. Another day another headache. I'll post another question for that if I can't figure it out (shrug). $\endgroup$
    – Paze
    Aug 24, 2020 at 20:37

1 Answer 1

3
$\begingroup$

It appears that you are not using the correct grouping variable. Observations are clustered within patient, not aneurysm, so patient ID should be the grouping variable.

As for the results, presumably you are including other variables in the model so you can't just look at raw risk ratios. Also, 1.06 appears to be an odds ratio, and this is close to 1 (ie no effect). You appear to have a large sample size so even tiny effects can be statistically significant. Clinical significance is more important here.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.