Hello Ahmed!

I would question whether a meta analysis on the number of total events is appropriate at all. The reason is that in every single study different outcomes were assessed. So it is possible that some studies report outcomes that are less relevant and others only report the primary outcomes.

Imaging you would make an meta analysis of chemotherapeutics. The studies report the following outcomes: nausea, vertigo, hospitalisation, deaths. It would make no sense to sum up these ouctomes and further make conclusions which is the best treatment.

My advise would be, that you choose the most patient relevant outcomes and compute a meta analysis for these. But don't combine different outcomes (except you have good reasons to do so).

As Robert has said you should do a meta-analysis for each individual outcome as a dichotomous outcome using odds ratios. Do not sum them into a continuous variable, it is exceedingly difficult to know what that kind of a continuous variable even means. If you really want a composite outcome you can create some but the end point needs to be either yes or no for that as well. 

You could generate a dichotomous outcome like "any complication", but you should also report an analysis for all individual outcomes as described.

I agree with what has been said above, I might also add that in making a composite outcome of "any complications" which I'm assuming you have done it means that you may have counted an individual twice in your analysis. For example (using the scenario above) a patient might have had nausea and vertigo which means they would contribute two events, ie. counting patient twice. This would violate a fundamental tenet of biostatistics in that data must be independant. As always I refer meta-analyses questions to the Cochrane Handbook on Systematic Reviews. All your answers will be there. Good luck!