Health Risks of Ionizing Radiation: - Clark University

Analysis of preconception exposure studies
Appendix C
This analysis is intended to supplement the review
of paternal preconception exposure studies in section
10. As indicated in that review several studies
have noted elevated risks of leukemia, non-Hodgkin’s
lymphoma, solid cancers, or adverse birth outcomes
following paternal and sometimes maternal
preconception exposure. This analysis deals with
leukemia following paternal exposure because this
subset of the evidence is the largest. Non-Hodgkin’s
lymphoma (NHL) was sometimes grouped with
leukemia (leukemia/non-Hodgkin’s lymphoma,
LNHL) so that the results were not separable. Only
case-control study designs were included and studies
were selected so that there was no overlap among
study populations. Table C-1 lists the studies that we
initially chose to consider with some methodological
notes. Included in this table is a study of atomic
bomb survivors that was not included in the analysis
(not a case-control study). This group of studies
is not always well-suited for meta-analysis because
there is very little consistency among exposure variables.
For example, diagnostic x-ray exposure studies
might use information about x-rays at any time
before conception or only for a fixed period of time
(one month, one year, etc.) before conception. Nevertheless,
a group of studies with generally positive
although not statistically significant results can be
combined for a result that will help establish the
presence of a significantly positive effect even if the
central estimate is not precise or even meaningful.
Study results were presented as odds ratios and
were combined following the methods presented by
Greenland (1987). Specifically, the natural log of
the pooled odds ratio (ln(ORp)) was estimated as
the weighted mean of individual ln(OR) estimates;
weights were the inverse variance of each estimate
(w = 1/SE2). The pooled standard error was calculated
as the inverse of the square root of the sum
187
of the weights (SEp = 1/√∑w). Approximate 95%
confidence limits were estimated as (exp(ln(ORp) ±
1.96 SEp)). The standard error for each individual
study was estimated as the difference in the natural
logs of the 95% confidence limits divided by 3.92.
In one case (Graham et al. 1966) the standard error
had to be estimated from the p-value (0.16). In this
case (SE= (ln(ORp)/1.4)) where 1.4 is the unit-normal
test statistic corresponding to a p-value of 0.16.
Pooled risk estimates were calculated for preconception
x-ray exposure, for any occupational exposure,
for total preconception dose greater than 50 mSv,
and for a dose in the 6 months leading up to conception
of greater than 5 mSv.
The pooled estimate for diagnostic x-ray exposure
is presented in Table C-2. The data from Shu
et al. (1988) for specific categories of exposure
were combined (using the same methods described
above) prior to pooling the studies. The final pooled
result, an odds ratio of 1.4 for preconception diagnostic
x-ray exposure, is significantly positive (95%
CI 1.2-1.7). Limitations of this comparison include
the different time periods considered for exposure
and the inclusion of infant leukemia in the comparison
with childhood leukemias. Interview-based
exposure information was not validated by Shu et
al. (1988), introducing possible recall bias, but there
was a significant dose-response trend in this study.
The odds ratio estimate for any x-ray with one year
(OR = 1.32) was higher than the estimate from that
study for an x-ray ever (1.08, 0.42-2.81) and lower
than that for any x-ray within one month (2.56, 0.67-
9.75). Because of the low weight attached to this
study, however, the pooled result would be virtually
the same using any of these estimates. It should also
be noted that this study found higher and significantly
positive estimates for x-rays of the abdominal
region; for any x-ray ever the odds ratio is 2.24
(1.44-3.47) and for any x-ray within one month the
odds ratio is 5.93 (1.52-23.1). Meinert et al. (1999)