ASA Connect

 View Only

  • 1.  Network Meta-Analysis - Consistency assumption

    Posted 04-17-2018 13:56
    Hi everyone,
    I'm hoping this is an appropriate place to post this question. I am wondering about how the consistency (or coherence) assumption in network meta-analysis impacts our interpretation of the results in a clinical healthcare setting. Specifically, I'm curious about cases where the direct estimate and network estimate would suggest different treatment recommendations (one estimate shows drug benefit, other estimate shows no benefit). If the consistency/coherence assumption is not met for a comparison, can we still trust that network estimate? Would we be better off trusting only the direct estimate? Or would that whole comparison be inconclusive?

    Thanks for your time and expertise!

    ------------------------------
    Megan Smith
    Data Specialist
    IBM Watson Health
    ------------------------------


  • 2.  RE: Network Meta-Analysis - Consistency assumption
    Best Answer

    Posted 04-18-2018 13:49
    ​I'd say inconclusive. But of course, they want conclusions.
    We need to bring back into science the excluded middle. 
    I'd say: "inconclusive at this time."

    ------------------------------
    Borko Jovanovic
    Associate Professor in Preventive Medicine
    Northwestern University, Feinberg School of Medicine
    ------------------------------

    Original Message


  • 3.  RE: Network Meta-Analysis - Consistency assumption

    Posted 04-20-2018 13:08
    Thank you!

    ------------------------------
    Megan Smith
    Data Specialist
    IBM Watson Health
    ------------------------------

    Original Message


  • 4.  RE: Network Meta-Analysis - Consistency assumption

    Posted 04-26-2018 20:05
    I think it depends on whether the cases and control sample sizes are equal we may obtain a weighted average using a geometric mean in a binomial distribution (0.5 equal +, 0.5 equal _ ) at alpha = 0.05. Therefore, p = proportion of + or _ studies from total studies. For a CI95, then critical value is 1.96for a 2- sides hypothesis. If study totals by comparison are radically unequal then even pooling will not give a significant effect size.

    ------------------------------
    Julie Tackett
    ------------------------------

    Original Message


Related Content