134 BIOLOGICAL EFFECTS OF LOW LEVEL EXPOSURES
zero, the lowest value possible, is not unusual since zero is the lower normal
limit of a 95% normal range (Table 7.5A).
Tables 7.6 and 7.7 contain sensitivity probabilities for detecting hormetic
effects in SR studies. These depend on the relative risk (RR), defined in the
present context by
RR = (true expected number)/(calculated expected number)
If the null hypothesis of no difference in risk is true, then the RR is unity. In
such case (the last columns in Tables 7.6 and 7.7), the actual effect is
neutral, and the corresponding tabular entries are the probabilities of Type
I errors. These cannot exceed 2.5%, and are often considerably below 2.5%
because of the discreteness of the Poisson distribution. For all the other
columns, where the RR is lower than one, there is a hypothesized beneficial
effect relative to the standard population, and the tabular entries are the
probabilities of TLs for the corresponding hypothetical RRs. These sensitiv
ity probabilities tend to increase as the expected cases increase (though in an
erratic manner due to discreteness) and do increase systematically as the RR
decreases.
Meta-Analytic Techniques
Sometimes there may be several studies available with low-level exposures
to a substance, yet each individual study has insufficient sensitivity to detect
a hormetic effect. However, when the studies are pooled or viewed collec
tively, a clear pattern may emerge. Meta-analysis is a collection of statistical
techniques for objectively synthesizing or pooling similar studies. Meta-
analytic results are not always definitive. An attempt to synthesize the
literature on effects of passive smoking via meta-analytic techniques yielded
inconclusive results.11
There are numerous philosophical and practical problems involved in
meta-analysis. Many of these stem from dissimilarities between the several
studies being synthesized. No two studies are exact duplicates of each other,
and it may be impossible to statistically adjust for differences between them
in a satisfactory way. Meta-analysis has been described by some as contro
versial and by others as the wave of the future.12 Like any statistical tech
nique, it can be misused. A list of problems and questions concerning meta
analysis has been compiled by Spitzer.13
Meta-analytic studies of hormetic effects can be expected to suffer from
“publication bias” —nondetrimental results tend not to be submitted for
publication by investigators, and such nondetrimental results that are sub
mitted tend to be rejected by editors. Publication bias could result in signifi
cant overestimates of risk in humans and be detrimental to rational health
policy decisions. A group of epidemiologists recently concluded that publi
cation bias was a definite problem in their field.114 Nevertheless meta-