Graph trend line from meta-analysis (multiple pooled studies)

#1
Hi! I have searched the web and have found no convincing answer, although this should be pretty easy. My question separates into 2:
1- Is it reasonable to trace a trend line of the evolution of a parameter over time from the pooling of different studies, each of which do not contain all time points? See the pictures for exemple: the forest plot picture shows all the different studies' contribution to each time point. Then I compute a graph from the mean +- SD of each time point, with a trend line. Is this heresy?
2- If this is acceptable, can we (and how do we) compare these time point changes among each other?
Thanks!
MA forest plot.jpg Graph from forest plot.jpg
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Yeah maybe heresy. Is study A at all time points alway the same study, but its evolving outcomes across time? So your graph is pooling studies but not always the same studies?
 
#3
Well, study A only reports data at 3 and 6 months. But C reports 3, 6 and 12 months data. And so on and so forth. So I'm pooling different studies at the 3 different time points
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
How are you drawing that line as well?

Think about it as you surveyed some people about satisfaction then had an intervention then surveyed more people, some people in the post-group were in the pre-group and others were not. Can you make conclusions about the effectiveness of the intervention? This scenario is kind on similar to your's, you have partial indepenence and dependance between group s.
 
#5
Thanks for the insight!
The way I see it, is that all study groups had people in the pre-intervention, they all start at the same place. But during the follow-up, outcome data is collected at different times for different groups. As they started all together in the pre-intervention group and because studies had similar inclusion/exclusion criteria, I thought we could pool them together and draw a trend line between the different time points. No?
 
#6
When we look at studies describing the evolution of prevalence of a disease over time, they don't always follow the same group for the whole time period, the prevalence at each time point comes from different study groups no? And yet they still do regression analyses... The picture below comes from a meta-analysis which time points come from different groups, not a single cohort. Screen Shot 2020-08-17 at 11.38.28 AM.png
 

hlsmith

Less is more. Stay pure. Stay poor.
#7
I dont think i would draw the trend line given not all groups are represented at each cross-section. It is highly unlikely but there could be nonlinear trends, so study A has effect 2, 4, -6, 7 for every two month period, if we were missing the 3rd value,we could make erroneous conclusions, etc.

if desired the reader could stare at the pooled diamonds across studies and make the trend inference themself if they are comfortable with it.