[ RadSafe ] Comments on thyroid cancer paper by E. Cardis et al.

Scott, Bobby BScott at lrri.org
Sun Dec 11 15:42:13 CST 2005

Comments on Thyroid Cancer Paper by E. Cardis et al.  


Paper: E. Cardis et al., Risk of Cancer After Exposure to I-131 in

Journal: Journal of the National Cancer Institute Vol. 97, No. 10 pages
724-732, May 18, 2005.


The authors carried out a population-based, case-control study of
thyroid cancer in Belarus and the Russian Federation to evaluate the
risk of thyroid cancer after exposure to radioactive iodine (mainly
I-131) from the 1986 Chernobyl accident and to investigate environmental
and host factors that may modify the risk.  The study involved 276 case
patients with thyroid cancer through 1998 and 1300 matched control
subjects, all younger than 15 years at the time of the Chernobyl
accident.  Linear and linear-quadratic excess relative risk models were
fit to the data using conditional logistic regression.  A strong
dose-response relationship was found between the calculated absorbed
radiation dose to the thyroid received in childhood and thyroid cancer
(P<0.001). A linear dose-response relationship was stated to have been
observed up to 1.5 - 2 Gy.


This e-mail questions the claim of a linear dose-response relationship
in light of the large errors associated with the derived odd ratios (OR)
and also questions the methods used for employing nonlinear
dose-response models.  The data reported in Figure 1 of the paper that
were used to obtain odds ratio (an estimate of relative risk) are
provided below:


Dose interval (mGy), cases in the interval, controls in the interval:

Interval 0-15 mGy, cases=16, controls=138;

Interval 16-199 mGy, cases=76, controls=503;

Interval 200-399 mGy, cases=40, controls=200;

Interval 400-599 mGy, cases=31, controls=141;

Interval 600-799 mGy, cases=29; controls=92;

Interval 800-999 mGy, cases=26, controls=67;

Interval 1000-1249 mGy, case=14, controls=60;

Interval 1250-1499 mGy, cases=10, controls=22;

Interval 1500-1999 mGy, cases=10, controls=23;

Interval 2000-2999 mGy, cases=15, controls=25;

Interval 3000 mGy and larger, cases=9, controls=29;


OR was apparently evaluated relative to the interval 0-15 mGy. The crude
OR I get for the interval 16-199 mGy is: 1.303 (0.736, 2.307).  The two
values within the parenthesis defines the 95% confidence interval.  Note
that the result is not significant > 0 and based on the lower confidence
value is consistent with the possibility of either a threshold or a
hormetic response. Neither a threshold or hormetic model were considered
by the researchers.  The indicated results related to the threshold and
hormetic models are especially important in that there was no actual
unexposed group.  Not having an unexposed group and using a case-control
study design can greatly favor LNT over threshold and hormetic models.
Also, given that reconstructed doses (dose estimates) are likely at best
crude approximations, doses for some persons stated to be in the dose
interval 0-15 mGy could be much higher. The OR results I got appear to
differ from those reported by Cardis et al.  However, how they got their
OR values was not explained in their paper.  Possibly some unreported
adjustments were made.


The excess relative risk (RR) model used by Cardis et al. for OR (an
estimate of RR) had the structure:


OR = 1 + beta*D + gamma*D*D,


where D is the individual absorbed dose (usual definition in risk
assessment). Beta and gamma are fixed model parameters.  In the
regression analysis, the above equation was apparently regressed against
average dose for each dose group.  A fixed observed OR was assigned to
each dose interval.  Please note that this implies that OR averaged over
each dose interval was assumed.  For the linear model (gamma=0), there
is no problem because average OR (here indicated as Av{OR}) equals 1 +
beta*Av{D}, where Av{} again indicates average.  However, for gamma not
equal to zero, Av{OR} is not only a function of Av{D}, but also a
function of Av{D*D}.  Av{OR} is not a function of Av{D}*Av{D} as was
apparently used by Cardis et al. (systematic error in model
application?).  Thus, conclusions obtained by the authors related to
application of the linear-quadratic function may not be valid.


For some analyses, instead of use of the excess RR model the authors
used and exponential form of OR, where


OR = exp(beta*D + gamma*D*D + ...).


However, they apparently did not realize that in using this exponential
model that 

Av{OR} does not equal exp(beta*Av{D} + gamma*Av{D}*Av{D} + ...), but has
a more complicated structure that should have been used.  Thus, results
reported by the authors based on this modeling may not be valid.


Radsafe Digest readers, comments on the above would be welcomed.



Bobby R. Scott, Ph.D.

Senior Scientist

Lovelace Respiratory Research Institute

2425 Ridgecrest Drive SE

Albuquerque, NM 87108 USA



More information about the RadSafe mailing list