LSS Report 12: Supplementary Tables

This document describes supplementary tables summarizing site-
specific cancer mortality data for the Life Span Study (LSS)
during the period from October 1, 1950 through the end of
1990. The primary analyses and discussion of these data are
given in Part 1 of LSS Report 12. In addition to these
tables, RERF has released two datasets that provide detailed
stratification of these mortality data by city, sex, organ
(colon or marrow) dose, age at exposure, and time since
exposure. The results presented herein are based on organ-
dose-specific tables of cases and person years.

The tables are included in an Excel 7.0 worksheet file called
R12Supp.xls.

Three tables are presented for each of the cancers or groups
of cancers listed in Table 1. The last two tables in each set
summarize the distribution of the observed and expected
numbers of cases together with person-years by dose and time
period or age at exposure categories. These tables also
indicate the total number of people by dose category and sex.
The expected numbers of cases in each table were computed on
the basis of a stratified excess relative risk (ERR) model
that was linear in dose and allowed for age-at-exposure and,
where relevant, sex effects on the ERR. Expected numbers of
cases are computed from the stratified background model, with
stratification on city, sex, age at exposure, and attained
age. The ERR estimate is simply observed divided by expected
minus one.
            
       Table 1:  Cancers used in supplementary tables

Cancer /         ICD codes     Organ dose
Cancer Group     (9th Rev)
      
Solid cancers    140 - 199     colon
Leukemia         204 - 208     bone
                               marrow
Stomach             151        stomach
Lung                162        lung
Liver          155 (0, 1, 2)   liver
Colon               153        colon
Rectum              154        bladder
Pancreas            157        pancreas
Esophagus           150        bone
                               marrow
Gallbladder         156        liver
Bladder             188        bladder
Uterus           179 - 182     uterus
Female breast       174        breast
Ovary               183        ovary
Prostate            185        bladder
Other Solid      all other     colon
Cancers        codes between
                140 and 199
Malignant         200-202      bone
Lymphoma                       marrow
Multiple            203        bone
Myeloma                        marrow

The first table in each set presents parameter estimates and
the results of selected hypothesis tests for the site. With
the exception of the test for a non-linear dose response,
which is based only on those people with DS86 shielded kerma
estimates of less than 4 Gy, these estimates and tests are
based on analyses of all LSS for whom DS86 doses have been
computed.

Detailed results for both excess relative and excess absolute
risk models are presented for sites for which the number of
cases was large and there was evidence of a radiation-related
excess risk. For other sites parameter estimates and
hypothesis tests were carried out for a limited number of
factors only for ERR models.

Models used for analyses and hypothesis tests

The ERR analyses were based on stratified background models.
The stratification variables were city, sex, age at exposure,
and attained age. In the more detailed analyses the basic
model for the ERR was linear in dose with sex and age at
exposure as modifying factors. This ERR used in this model
can be written as:
                               
       ERR = b	d exp ( b (e - 30))
              1s         2               

where b  is a sex-dependent dose effect and b  represents the effect
       1s				     2
of age at exposure (e). Because of the way in which e appears
in the model, the dose effect parameters correspond to the
risk for a person who was 30 years of age at exposure. The
effect of attained age on the ERR was examined by adding term
that is linear in log attained age to the exponential term in
the above model. In looking at city effects, it was assumed
that the effect would be the same for men and women. In those
sites for which the number of excess cases was insufficient to
allow working with the full model, neither age-at-exposure nor
sex effects were routinely included in the models.

With the exception of breast cancer, excess absolute risks
were modeled relative to a fully parametric model for the
background risks. Under the model the logarithm of the
background risks was described as a sex-specific quadratic
spline in log attained age with a single knot at age 70. The
intercept in this model was allowed to depend on sex, city,
and year of birth. Background risks for breast cancer were
modeled in terms of a log-linear spline in attained age with a
single knot at age 50. The intercept was allowed to depend on
city and birth cohort. The basic model used for the excess
absolute risk is
                               
       EAR = b  d exp ( b  ln(age/50))
              1s         2         

where age is attained age and thus the sex-specific dose
coefficients describe the risk at age 50.

Technical notes on hypothesis tests and confidence intervals

In most cases hypothesis tests are based on likelihood ratio
tests and confidence intervals were computed by direct
evaluation of the profile likelihood. However, there are
situations in which it is not possible to compute the
likelihood ratio statistic or, somewhat more often, likelihood-
based confidence bounds. These situations arise when the dose
effect is negative for some subset of the population either at
the maximum likelihood estimate or at some point in the
confidence region for the parameter of interest. If the MLE
did not exist the score test was used. Cases in which
confidence bounds could not be computed are indicated as N.C.
in the tables. A related problem occurs in the computation of
confidence intervals for ratios of effects (i.e. sex and city
effect ratios). If the joint confidence interval for the
numerator and denominator includes 0 for both effects, all
values of the ratio are consistent with the data. This is
indicated by the phrase 'all values' in the tables. If the
joint confidence interval includes 0 for the denominator but
not the numerator, the confidence interval for the ratio
consists of all points outside of some interval. In such
cases we report only the upper interval indicating a positive
lower bound and an infinite upper bound (indicated as inf in
the tables.). In most cases the negative interval cannot be
computed for the models that we consider.




Copyright 2003.
Radiation Effects Research Foundation