Joint Effects of Radiation and Smoking on Lung Cancer Risk
among Atomic Bomb Survivors
Donald A. Pierce, RERFGerald B. Sharp, RERF &
NIAIDKiyohiko Mabuchi, NCI
2
Nature of this Talk
• The data results here are in the paper of the same title in Radiat. Res. (2003) 511-20
• After summarizing these I will turn to more general statistical issues
• These slides and the paper can be obtained at http://home.att.ne.jp/apple/pierce
3
Hypothetical RRs for Joint Effects
Radiation None Moderate High
Smoking
None 1 1.25 1.5
10.25 10.5
12.5 15
20.25 20.5
25 30
Moderate 10
Heavy 20
Upper values for additive and lower for multiplicative
4
Nutshell Background
• Previous LSS analyses could not distinguish between multiplicative and additive effects– Main reason was that apparent
smoking risks were quite small– Probably due to scarcity of cigarettes
during and soon after the war• In my view BEIR IV,VI results for
miners & radon were equivocal in this respect
5
Smoking Information & Usage
• From mail and clinical surveys: 52,000 persons presenting 600 lung cancers
• Used only as levels: 0,1-15,16-25,>25 cigarettes/day, averaging over multiple responses
• Could estimate pack-years measure, but we think smoking rate may be preferable
6
Dose Distributions
0
10000
20000
30000
0 1-15 16-25 >25
Cigarettes per Day
Peo
ple Men
Women
0
4000
8000
12000
16000
<0.005 0.005-0.5 0.5-1.0 >1.0
Lung Radiation Dose (Sv)
Peo
ple Men
Women
7
Smokers by Radiation DoseAnalysis does not assume independence of smokingand radiation dose, but this is of some interest
0
20
40
60
80
100
0 0 - 0.5 0.5 - 1 >1
Radiation Dose (Sv)
% S
mo
kers Hiro Men
Naga Men
Hiro Wom
Naga Wom
8
Simplest Analysis: Radiation Relative Risk within
Smoking Levels
0
0.4
0.8
1.2
1.6
2
0 1-15 16-25 >25
Smoking Level (cigarettes/day)
ER
R/S
v
If effects were multiplicative these ERRs would be equal: P-value = 0.02 for testing this
9
More Complete Analysis: Radiation Risk Relative to
Non-Smoker Baseline Rates
0
1
2
3
4
5
6
0 1-15 16-25 >25
Sm oking Level (cigarettes/day)
ER
R/S
v
Unconstrained w / Error Bars
Fitted Additive Model
Fitted Multiplicative Model
P-value for testing additive effects 0.20
10
Effects of Adjusting for Smoking
• Spuriously large female:male sex ratio in ERR is reduced to usual level
• Exposure-age effect contrary to usual direction is eliminated
• Entire pattern of ERR/Sv becomes similar to solid cancers in general
• For baseline rates female:male ratio is increased by factor 3-4 by adjustment
11
Statistical Modeling
• As for the basic LSS, and BEIR IV, developing suitable statistical models was challenging
• Is necessary to allow all RRs to depend on attained age, birth cohort, gender as well as radiation and smoking
12
Analysis Modeling
• For simplest analysis, though, need no explicit model for smoking and can use for lung cancer
, , , ,( , , , , ) {1 ( ) }s ga b g s b a
rate a b g s d d
a, b, s : categories of age, birth cohort, smoking levelg : genderd : dose continuous
13
Joint Effects Modeling
• Take model for smoking effect as
• Then, combining, the main model is
,, ,( , , , ) {1 }g s aa b g b
rate a b g s
,,, ,( )( , , , ) {1 }s g a bg s aa b g bdrate a b g s
14
Remainder of Talk
• For the rest of the talk I will discuss issues arising in this work that are of more general interest
• In part, aim to be mildly provocative
15
Something about Confounding
• It was seen that smoking level and radiation dose are not related
• There is however confounding for the lung cancer sex-by-radiation interaction
• The radiation ERR/Sv is not a number but a pattern depending on …
• Thus smoking level and radiation effect are confounded
16
Prospects for Other Joint Effects Studies
• Hopeless to distinguish between multiplicative and additive effects unless– Other risk factor has an RR of 5 or so– Or focus is where the radiation risk is
larger than usual
• More ambitious goals are even less attainable
17
Hypothetical RRs for Joint Effects
If Other Factor Has Modest Effect
Radiation None Moderate High
Risk Factor
None 1 1.25 1.5
1.5 1.75
1.56 1.88
1.75 2
1.88 2.25
Moderate 1.25
Heavy 1.5
Upper values for additive and lower for multiplicative
18
Use of Smoking Information
• Even if more than rough smoking rate were available, it would be difficult to use the information
• Must model the effect of smoking history for information to be useful
• Age of cessation is both unreliable and difficult to model
• Age of starting is similarly difficult to use
19
Use of Pack-Years
• Plausibly, the smoking RR for given rate is fairly constant in age, and then the RR for given pack-yrs will decrease with age
• Mutation modeling suggests the RR covariable pack-yrs/age , that is the lifetime average rate up to age-at-risk
20
Special Problem for Our Cohort
• Cigarettes were scarce during and soon after the war
• Causes a birth cohort effect in the smoking ERR, in terms of our smoking rates
• Again, pack-yrs/age might be a better covariable than either pack-yrs or smoking rate
21
Effect of Smoking on Radiation Exposure-Age
Effect• Baseline lung cancer rates increased
strongly over most of our follow-up --- due to smoking
• Since effects are additive, this causes the radiation ERR exposure-age effect to increase with exposure age
• This is opposite to most cancers, but “corrected” by adjusting for smoking
22