Peto’s et al method
Hybrid method by Lam
Variation
Prevalence of smoking is a
RR
respect
poor proxy for cumulative
different than studied is inconsistent.
different than studied is inconsistent.
proportional
hazards of smoking.
Designing a case-control study OR
A calculation procedure let us to
attribution
Defining
could be assessed.
calculate the RR.
method
impact ratio or Synthetic
SIR
(Smoking
extrapolation
to
Basic method
populations
RR
extrapolation
to
populations
prevalence) authors avoid
prevalence limitations.
Calculation
procedure
SIR 
C LC  N LC
*
*
S LC
 N LC
OR 
where CLC, NLC, S*LC, N*LC are
age-sex
specific
lung
a1b0
a (1  p1 )
RR  1
a0 b1
a0 p1
where p1 is the prevalence between the
smokers
and
never
smokers in the study and in
the reference population (*).
RRs=1+ ac((t-5))-t0)
Multistage-function in smokers
4.5
RRs=1+[(t-5) +ac(1+2ac)((t-5)4.5
cases and:
cancer mortality rates for
Packs-function in smokers
4.5
4.5
4.5
t0) +2ac((t-5) -t0 )]/(t-5)
Cases
Controls
Exposed
a1
b1
Where a is a constant, c is the
Non exposed
a0
b0
number
of
packs
of
cigarettes
smoked per year, t is the current age
and t0 is age at start of smoking.
In former smokers t1 replaces (t-5)
and t1 is age at stop smoking.
Table 2.- Methodologies’ modifications taking into account proportional
attribution method as base method.
1
Method
Data employed
Data source
Proportional
attribution
/
SAMMEC
Prevalence
Relative Risks: Smokers,
non-smokers and former
smokers
(n=52)
Method
proposed by
Peto
and
colleagues
(n=6)
Basic method
(n=1)
Prevent
method
(n=2)
Weaknesses
Strengths
Estimations calculated
National Statistics
Cohort study
Method applied to estimate
mortality due to:
Tobacco
consumption,
exposure to environmental
tobacco
smoke
(ETS),
obesity, alcohol intake,…
- Does not take latency into account.
- Generalizes the RR obtained in the
Cancer Prevention Study (CPS).
- Worldwide use.
- Application of risks other than
CPS.
Attributable mortality for all
causes.
Relative Risks: Smokers
and non-smokers
Lung cancer death rates:
Global (non smokers +
smokers
+
former
smokers),
non-smokers
and smokers.
Lung cancer death rates
CPS II
Tobacco consumption.
- Assumes constant worldwide lung cancer
mortality rates among never smokers.
- Assumes the same latency for all death
causes related to tobacco.
- Does not take into account former
smokers.
- Worldwide use.
- Mortality estimation in absence
of smoking prevalence.
- Takes latency into account for
lung cancer.
Attributable mortality for all
causes.
Tobacco consumption.
Prevalence
National statistics /
Estimated
Calculated
National statistics /
Estimated
National statistics /
Estimated
Previous studies
National Statistics
- Partial view of attributable mortality (only
used to estimate mortality by lung cancer).
- Use of constants.
- High need of information.
- Rate ratios for former smokers.
- Assumes constant worldwide lung cancer
mortality rates among never smokers.
- Takes into account induction
time.
- Estimates smoking-adjusted RR
in different time periods.
Lung cancer death rate
attributable
and
not
attributable to active smoking.
Tobacco consumption and
general
scenarios
of
effective health promotion.
- High need of information.
-Takes
into
account
the
multiplicity of cause or effect.
- Proportional decrease in risk
reduction related to time.
- To measure the results of
intervention policies.
Attributable mortality for all
causes.
Tobacco consumption and
exposure to ETS.
- Case-control study design.
- Recall bias.
- Specific risk dates.
Attributable mortality for all
causes.
Tobacco consumption and
alcohol intake.
- Partial view of the attributable mortality
(only used to estimate mortality by cancer).
- Assumes constant worldwide cancer
mortality rates among never smokers.
- Necessary dates are few.
- Does not use risks or
prevalence.
Cancer deaths attributable to
smoking.
Tobacco consumption.
- Availability of mortality registers.
- Has a population representative survey
about health-risks.
- Assumption: smoking status remains
steady since the survey about health-risks.
- Risks calculated ad hoc.
- The population division is more
reliable.
Attributable mortality for all
causes.
Lung cancer relative risk
Packs
of
cigarettes
smoked
Age of starting/giving up
tobacco consumption
Constants
Composition
of
the
population
Mortality (population) and
birth (women) rates
Latency and delay
National Statistics /
CPS II
National Statistics
National Statistics
Previous studies
Time-Tendency of tobacco
consumption.
Relative risks
Mortality observed
Exposure prevalence: case
or controls
Odds Ratios
Mortality observed
Personal interviews
Cancer mortality rates in
non smokers.
American
Society
Rogers´
method
Mortality observed (all
causes)
Prevalence (7 categories)
National Statistics
(n=1)
Odds Ratios
Discrete-time
hazard models
Hybrid method
by Lam
(n=4)
Garfinkel´s
method
(n=2)
CPS II
National Statistics
Case-control study
Case-control study
National Statistics
Cancer
Surveys
Table 1.- Methods used to estimate tobacco attributable mortality.
2