kylehamilton / MAJOR

Meta-Analysis for JAMOVI

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Why to MAJOR risk ratio analysis results differ so much from RevMan?

apavlo89 opened this issue · comments

RevMan fever incidence

I am looking at fever risk ratio between two procedures. The results vary massively from RevMan as you can see from pic and the results from MAJOR here:

A total of k=8 studies were included in the analysis. The observed log risk ratios ranged from -1.5404 to 0.0000, with the majority of estimates being negative (62%). The estimated average log risk ratio based on the fixed-effects model was \hat{\theta} = -0.8472 (95% CI: -1.2507 to -0.4438). Therefore, the average outcome differed significantly from zero (z = -4.1162, p < 0.0001).

According to the Q-test, there was no significant amount of heterogeneity in the true outcomes (Q(7) = 4.3450, p = 0.7393, I² = 0.0000%).

An examination of the studentized residuals revealed that none of the studies had a value larger than ± 2.7344 and hence there was no indication of outliers in the context of this model. According to the Cook's distances, one study (Wen et al., 2018 [16]) could be considered to be overly influential.

Neither the rank correlation nor the regression test indicated any funnel plot asymmetry (p = 0.5484 and p = 0.3286, respectively).

If you dont get a reply from me in a couple days comment on this issue so Ill see it' first issue

I think I figured it out. It is because its log risk ratio while in Revman its not log risk ratio. I just don't get why jamovi uses log. There is only an option to remove log with odds ratio not risk ratio. And even if I chose odds ratio un-log option it only does this with the stats and not the forest plot.

Is there any chance someone could provide a code to remove log from the forest plot?

Also the numbers never really match up completely to cochranes Revman, whether is weighted mean difference, or risk/odds ratio. From what I can see it is because revman and MAJOR use different weighting statistical techniques. It would be nice to have more model options for MAJOR in the future....

I figured why the numbers don't fully match up. It is because Revman uses Dersmimonian & Laird (which you can select in MAJOR).

The annoying part is the logged forest plot. I wish MAJOR could implement this code:

https://www.metafor-project.org/doku.php/plots:forest_plot_revman

The code is basically there it would require much less work to implement in MAJOR.... Pleeaaaseee