4
$\begingroup$

Repeated measurements of some continuous variable of interested is very common in clinical trials.

Usually patients are randomized between treatment arms. Hence it is reasonable to assume that all patients and especially all groups in the trial have equal value of this variable of interest.

Lets assume I run a clinical trial with three groups or arms and I make repeated measurements on variable of interest.

Against this reasoning I would say that only reasonable and justified linear mixed model would assume fixed intercept and random slopes. I would include group id as covariate and allow varying slopes for it. To me, it would be odd to include a random intercept since I assume that groups are balance regarding the baseline value of my measured variable.

So, is this model the only correct one or why should I use other models, like random intercept and fixed slope?

Improve this question
$\endgroup$
3
  • $\begingroup$ "I assume that groups are balance regarding the baseline value of my measured variable" ....which is very often a mistaken assumption (eg. selection bias) $\endgroup$
    – Robert Long
    Commented Oct 29, 2019 at 19:36
  • $\begingroup$ Selection bias in randomised clinical trial? $\endgroup$
    – arkiaamu
    Commented Oct 29, 2019 at 21:45
  • $\begingroup$ Sure there are "nuances" in balancing but I would say the baseline values across groups would be fairly same and within measurement error with large enough sample size. Indicating that fixed intercept should be used, right? $\endgroup$
    – arkiaamu
    Commented Oct 29, 2019 at 22:19

1 Answer 1

Reset to default
3
$\begingroup$

When you include a random intercept in the model you say that there is variability in the value of the outcome at baseline between individual patients, you do not say something about the average of the outcome in different groups of patients.

That is, it can well be that because of randomization the three arms have the same average at baseline, but still, individual patients vary around this average.

If you would have no variability at baseline, it would implicitly mean that all patients have exactly the same value at baseline.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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