In Reply.—Publication
bias and time lag bias are distinct terms
with complementary implications. One may exist in the absence of the other,
for example, when negative trials take longer to complete (eg, due to lower
event rates, smaller treatment effects, or both), even if subsequently published
as quickly as positive ones. Another example is when, looking retrospectively,
all trials on a subject have been published, but negative ones were delayed
in the process. As Drs Clarke and Stewart highlight, we could be misled if
meta-analyses (and medical decisions) are based on early appearing evidence
misrepresenting the true treatment effects. Unfortunately, considering only
evidence from trials that started before a certain date may be a suboptimal
solution because it means disregarding the latest evidence, which may be most
pertinent to current practice. In fields with rapid new drug development,
the marketing honeymoon of drugs is short before being ousted by newer agents.
Delays of 2 to 3 years (and occasionally longer) for negative trials in such
fields can largely mislead practice, while meta-analyses excluding recent
evidence to escape bias or waiting 20 years to be all-inclusive run the risk
of becoming archaeological exercises. I fully agree with Clarke and Stewart
that we need more research on the impact of time lag bias, and efforts should
be prospective.1- 2 Empirical
prospective evaluations in continuous updating initiatives, such as the Cochrane
Collaboration,3 may offer more solid evidence
on how reported treatment effects for various treatments may change over time
and may offer hints on how to anticipate and dissect such heterogeneity.1 The problem of time lag bias only heightens the
importance of careful meta-analysis efforts, such as those led by the Cochrane
Collaboration.