ABSTRACT
All the methods presented thus far for meta-analysis in this book are based on
large sample theory as well as the theory of large sample approximations. For
rare events, these methods usually break down. For example, when events are
zeros, the methods for risk-ratio and odds-ratio discussed in Section 4.2 cannot
be used and when the events are rare, but not all zeros, the variance estimates
for these methods are not robust which may lead to unreliable statistical
inferences. The typical remedies are to remove the studies with zero events
from the meta analysis, or add a small value, say 0.5, to the rare events which
could lead to biased statistical inferences as pointed out by Tian et al. (2009)
and Cai et al. (2010).