0

I am investigating the prophylactically effect of drug A regarding onset of epileptic seizures. Drug A is given to the test group and not the control group. The hypothesis is that drug A does not prevent seizures.

I am comparing 17 studies in a meta-analysis regarding brain tumors. They differ greatly; the median age, gender ratio, location of tumor, tumor histology (benign vs. malignant tumor), follow up, publish year, prescribed medicine etc.

I would intuitively choose the Random Effect Model as all beforementioned factors are associated with the risk of developing seizures.

The RE model reveals the following about heterogeniety: I^2=0.0%, p=0.27, Q=8.5.

This indicates homogeniety between the studies - should I use til Fixed Model? And why/why not?

Thank you, C.

amoeba
  • 93,463
  • 28
  • 275
  • 317
cmirian
  • 239
  • 3
  • 11
  • How can you use either? Perhaps I am missing something obvious, but it does not seem that you have panel/longitudinal data? – Repmat Jan 11 '18 at 19:52
  • It is longitudinal data. It is 17 randomized controlled trials investigating the effect of drug A vs placebo in preventing seizures as a consequence to the tumor. The "longitudinal" aspect is that the trials are investigating the prophylactically effect of development of seizures regardless of what type of brain tumor/patient/age etc. – cmirian Jan 11 '18 at 20:00
  • Well okay, the key difference between random and fixed effects methods is whether the individual effect is correlated with $X_i$ or not. I you think there is no such relationship use random effects, otherwise use fixed effects. – Repmat Jan 12 '18 at 07:02
  • Sorry - I am really new at this. How do I test if it's correlated with Xi? Thanks for answering. – cmirian Jan 12 '18 at 08:04
  • 1
    My answer to a previous questions might be helpful.https://stats.stackexchange.com/questions/97182/what-makes-the-results-differ-for-fixed-effects-models-vis-%c3%a0-vis-random-effects/97248#97248 https://stats.stackexchange.com/questions/160681/heterogeneity-statistics-with-fixed-effects-model-in-revman/160850 – abousetta Jan 12 '18 at 10:23
  • 1
    Just to emphasise the points made in @abousetta comment and his.her answers linked to there you should make the decision on scientific grounds not observed heterogeneity. Fixed effects estimates the single underlying effect, random effects estimates parameters of the distribution of true effects. – mdewey Jan 12 '18 at 11:36

0 Answers0