We made a mistake in Fig. 4 due to an abnormality in the order of the statistical parameters in the software (Stata 12.0 with ado-file metan)
Our statistician used an incorrect calculation by mistake:
(N_treat/(N_treat+N_contr)) : (event_treat/(event_treat + event_contr)
instead of:
(event_treat/N_treat) : (event_contr/N_contr)
The error does not have any impact on the results, but
Fig. 4(a), (b) did not show the correct data. Here are the recalculated correct figures(Reference Doege, Grajecki and Zyriax1).