1

 Radiation Carcinogenesis: Applying Basic Science to Epidemiological
Estimates of LowDose Risks

2

 Bayesian methods and CML
 LinearQuadraticExponential model
 Likelihood and prior data sets
 Baseline LQE estimate of CML risk
 Improved risk estimates based on BCRtoABL distances and CML target cell numbers
 Net lifetime CML risk: Can it have a
Ushaped low dose response?

3

 Priors+ likelihood estimates Þ
posteriors
 Posterior information equals
prior plus likelihood information
 Posterior means are informationweighted averages of prior and
likelihood means
 Posteriors are normal if
the prior and likelihood
estimates are normal
 Priors act as soft constraints on the parameters
 Priors and structures come from the same data

4

 CML is homogeneous, prevalent, radiationinduced, and caused by BCRABL
 The a2 intron of ABL is unusually large
 Leukemic endpoints have rapid kinetics
 White blood cells need fewer stages
 Linear CML risk is not
biologicallybased
 Linearquadraticexponential CML
risk does have a biological basis

5


6


7


8

 CML is practically absent in Nagasaki
 High dose HF waiting times are too long
 HM data is consistent with prior expectations

9


10

 C_{1} and k: SEER data
 k_{t} : Patients
irradiated for BGD
 b_{k}, a_{k }and a_{kn }: CAFC and MRA assays
 b/a and a_{n}/a: Lymphocyte dicentric yields
 C_{2} : Depends on a,
k_{t}, N, and P(baT)
 N: SEER and translocation age structure data
 P(baT): BCR and ABL intron sizes, the genome size

11


12


13


14

 Linear model
 R_{g} = 0.0075 Gy^{1 }and Q_{g} = 0.0158 Gy^{1}
 LQE posterior model
 R_{g} = 0.0022 Gy^{1 }and Q_{g} = 0.0042 Gy^{1}

15

 A comparison of age responses for CML and total translocations suggests
a CML target cell number of 2x10^{8}
 10^{12} nucleated marrow cells per adult and one LTCIC per 10^{5}
marrow cells suggests 10^{7}
CML target cells
 P(baT) = 2T_{abl}T_{bcr}/G^{2 }may
not hold

16


17


18


19


20

 Transplant doses of 10, 100, and 1000 CRU => CRU levels 120%
or 1560% normal Blood
(1996) 88: 28522858
 Broad variation in human HSC levels
Stem Cells (1995) 13:
512516
 Low levels of HSCs in BMT patients Blood (1998) 91: 19591965

21


22


23


24


25

 Bayesian methods provide a natural framework for biologically based risk estimation
 BCRtoABL distance data and
knowledge of CML target
cell numbers can be useful in a biologically based approach to CML
risk estimation
 Low dose hypersensitivity to killing might lead to a Ushaped low dose
response if there is a deadband in the control of target cell numbers

26

 Rainer Sachs
 David Hoel
 NIH and DOE
