0

I have an observational, non-randomized longitudinal study with 3 time point + baseline (t0 ... t3). Analysing solely the post-values in such trials is meaningless. I want to analyze the change from baseline, as the typical activity in medical studies.

I'm wondering, whether it makes any difference to model the change scores (Time_x - baseline) and obtain the differences (T3-T0, T2-T0, T1-T0) vs. modelling the post values itself and then test the appropriate contrasts after that: T3 vs T0, T2 vs T0, T1 vs T0?

For me, the outcome must be the Dunnett contrast: all vs. baseline, that's all. So this suggests to model post-values and then use appropriate contrasts, rather than modelling change scores - because how to adjust using the Dunnett, right?

Scenario 1: Model: post_response ~ covariates Dunnett: post_t3 vs t0, post_t2 vs. t0, post_t1 vs. t0 + Dunnett adjustment based on the multivariate t distribution.

Scenario 2: Model: change_from_baseline ~ covariates Dunnett: have no idea how?

But how to use Dunnett when I analyze change score?! To employ Dunnett I need the post-analysis (adjusted for baseline), so the answer seems clear.

I only wonder if the two approaches make any difference for modelling itself?

Saronaya
  • 1
  • See https://stats.stackexchange.com/questions/3466/best-practice-when-analysing-pre-post-treatment-control-designs – kjetil b halvorsen May 20 '22 at 14:30
  • Unfortunately, it doesn't answer my question. I'm wondering if analysing post-values followed by analysis of contrasts vs. baseline is different from analysing change scores. And I'm wondering how to employ the Dunnett procedure on change scores. – Saronaya May 20 '22 at 14:44

0 Answers0