PetoLopez method for estimating tobaccoattributable mortality from mortality data alone

1
Peto/Lopez method for estimating
tobacco-attributable mortality from mortality
data alone
2
Death certification rates for lung cancer, age
45-9, selected populations, 1955-1995
3
Step 1 Use lung cancer mortality to index
population tobacco exposure by proportion of
smokers in a hypothetical mixture of current and
never smokers
In this example, the observed rate is inferred to
be that expected in an hypothetical mixture of
50 never smokers and 50 current smokers ie the
SIR is 0.5
Taken from American Cancer Society CPS II,
1984-88
4
Step 1 Use lung cancer mortality to index
population tobacco exposure by proportion of
smokers in a hypothetical mixture of current and
never smokers
Arithmetically Propn exposed (observed-ns
rate)/ (sm rate ns rate)
Taken from American Cancer Society CPS II,
1984-88
5
Step 2 For disease of interest, calculate
relative risk in the hypothetical mixture

• RR for mixture
• Propn not exposed x 1
• Propn exposed x RR (for disease of
  interest)

6
Step 3 Calculate excess risk (in mixture) and
halve it

• Excess RR halved
• (RR 1 ) / 2

7
Step 4 Calculate proportion of disease
attributable (PAF) as

• Excess RR halved /
• Total risk in mixture
• Excess RR halved /
• 1 Excess RR halved

8
Step 5 Calculate rate of disease of interest
attributable to smoking as

• PAF x observed rate

9
Step 6 Estimate disease attributable to smoking
in population of interest as

• Lung cancer attributable
• lung cancer observed lung cancer expected
  in never-smokers
• Sum of attributable rates of other smoking caused
  diseases

Taken from American Cancer Society CPS II,
1984-88
10

• Illustrate in Excel
• Peto Lopez for exposition.xls

11
Petos method

• Doesnt require comparable data on (lifetime)
  tobacco exposure
• Gives (surprisingly?) plausible results eg US

12
Utility of Peto/Lopez method

• Widely used in policy debates and not effectively
  challenged by tobacco industry
• See site http//www.ctsu.ox.ac.uk/tobacco/
• Allows estimation of effects of smoking on other
  disease rates
• Eg Contribution of smoking to
• Secular trends in mortality
• East/West differences in mortality
• Social class differences in mortality in the UK

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
Estimated vascular mortality not attributable to
smoking, ages 35-69, selected former socialist
and non-socialist countries, 1965 and 1989 (mf/2)
17
By how much would mortality from vascular causes
vary in the absence of smoking?
18
(No Transcript)
19
Step 3 Calculate attributable proportion in
hypothetical population assuming true excess is
half the theoretical excess
C
A
B
for conservatism ie to allow for confounding