Extending the Gail model for Breast Cancer Risk Prediction to

Download Report

Transcript Extending the Gail model for Breast Cancer Risk Prediction to

Extending the Gail
model for Breast
Cancer Risk
Prediction to
Account for
Modifiable Factors
in the Italian
Population.
Calza S, Ferraroni M, Decarli A
University of Brescia and University of Milan, Italy
Background
The wide international variation in breast cancer (BC) rates
suggests that there are potentially modifiable environmental
and lifestyle determinants of BC.
Two relevant components need to be considered, when the
individual estimated probability of cancer is of main concern:
- the first is related to the risk factors which are
“irreversible” (i.e. genetic factors, age at menarche,
age at first birth, number of full-time pregnancies, etc.)
-the second is the set of risk factors which are
“modifiable”, at least in principle, such as those
related to life-style, (i.e. diet, body mass index, alcohol
consumption, etc.).
Gail et al. (1989) proposed a statistical method (GM) to
estimate the individual probability of BC developing, as a
function of age and a set of irreversible factors.
This model, widely used to optimise intervention studies
on BC in USA, is one of the more promising in this field
of research.
Different authors have evaluated the model in terms of
reliability in predicting the number of BC cases expected
and validity of assumptions.
The model
The probability of developing BC between the ages a and  ,
for a women who is in risk group i is written as:
where the subscript 1 refers to the incidence of BC and
2 to all other causes of death.
where :
is the baseline incidence of developing BC at age t
in the reference group.
is the relative risk of developing BC at age t
compared to the baseline group.
is the mortality rate, at age t, from all causes of
death, except BC, in the population.
is the probability of surviving to all other causes of
death to age t.
The major difficulty is in the estimation of the baseline
breast cancer age-specific hazard
, for a subject
without known risk factors. It is given by
where:
is the overall age-specific invasive BC incidence
rate irrespective of risk group
is the proportion of cases in the i-th risk group as
estimated by the case-control studies
is the attributable risk of the specified factors
which, in turn, define the different risk groups.
This model calculates a woman’s absolute risk of
developing BC over various time intervals.
It includes information on:
age
age at menarche ( 14 years, 12-13 years, < 12 years)
age at first live birth (< 20 years, 20-24 years, 25-29 years or
 30 years)
n. of first-degree relatives with BC (none, 1 , > 1)
breast biopsies (none or unknown, yes)
nulliparous,
Decision about the use of
Mammography
The routine use of screening mammography in
women 50 years old or older reduces BC mortality
by approximately one third.
This reduction comes without substantial risks and
at an acceptable economic cost.
However the use of screening is more controversial
in women under the age of 50, for several reasons.
Targeting mammography to women at higher risk of
BC can improve the balance of risks and benefits.
The inclusion in the Gail model of other variables,
related to modifiable risk factors for BC would allow to
disentangle the relative contribution of the two
components “irreversible” or “modifiable”, in the
individualised probability of developing BC.
This approach has the potential to define guidelines with
relevant implications for cancer prevention and public
health. In particular, it allows to stratify women with high
risk of BC in subgroups which would benefit from
different policies of prevention.
The data-sets
EPIC-Florence cohort
Recruitment was carried out in the period January 1993March 1998 and 10,083 women volunteers, residing in the
Provinces of Florence and Prato covered by Cancer Registry,
were enrolled in the age interval 35-64 years, in the Florence
section of EPIC prospective study on diet and cancer.
Detailed information was collected about dietary and life-style
habits, reproductive history and family history for BC , for
each woman included in the study.
10,083 women volunteers residing in the provinces of Florence
and Prato in the age interval 35-64 years
10 subjects lost to follow-up
30 subjects prevalent BC cases
12 subjects incident BC cases but
diagnosed whitin 6 months from the
date of recruitment
10,031 subjects included in the analysis
A total of 142 incident invasive BC cases occurred during the period
of follow-up
Multicentre case-control study
The case-control study was conducted between June 1991
and February 1994 in six Italian areas.
Cases were 2569 women, aged 23-74 years ( median
age 55 years) admitted to the major teaching and general
hospitals of the study areas with histologically confirmed
BC diagnosed within the year before interview, and no
previous history of cancer.
Controls were 2588 women, aged 20-74 years ( median
age 56 years) and admitted to hospitals in the same
catchment areas of cases for acute conditions.
Both data-sets include information on
sociodemographic charateristics, such as age,
education, occupation end socio-economic indicators,
lifelong smoking habits, physical activity,
anthropometric measures, alcohol consumption, dietary
habits, personal medical history and selected questions
regarding family history of cancer.
A validated food frequency questionnaire was used to
assess the usual diet in order to estimate the mean
daily intake of calories and selected nutrients.
Individualized absolute risks for BC in the EPIC-Florence
cohort were estimated according different models, that
differ in the following aspects:
US-Gail model
age-specific invasive BC rates, ri(t) and AR(t) are those
derived from the Surveillance, Epidemiology, and End
Results (SEER) Program of the National Cancer Institute;
It-Gail model
ri(t) are derived by the multicentre case-control study on
diet and BC conducted in Italy, and h1(t), h2(t) are
obtained from the Florence Cancer Registry. The
estimated ri(t) were obtained by means of a logistic
regression model including the same variables as in the
US-Gail model.
Gail model extension
Different It-Gail models were fitted including dietary
variables. In particular we investigated the role of
alcohol consumption and monounsaturated fats
intake.
The aim is to quantify the proportional
relevance of the different risk factors in
defining BC risk.
Validation of the model
Gail model extension
Conclusions
 Gail model is reasonably valid in Italy: our projected
probabilities of developing BC have been validated on
women included in a screening programme.
 Gail model can be improved for use in populations other
than American one, by using BC incidence and RR
estimates for risk factors of interest which are more
appropriate to the target population.
 Gail model structure including potentially modifiable
factors can be an important tool in BC risk counselling.
The counsellor has the opportunity to look at the lifestyle
of the women and possibly suggest that this be modified
as an alternative or in conjunction with
chemioprevention.
Acknowledgements
This work was conducted with the contributions of the
Associazione Italiana per la Ricerca sul Cancro and the
Italian Ministry of Education (PRIN 2003)
Estimates of probabilities (% and 95% bootstrap confidence intervals) of
developing breast cancer for selected group of subjects
- Breast density is generally higher in younger women
and screening mammography is less likely to detect
early BC at a curable stage.
- More false positive tests.
- Women under 50 are less likely to have BC. Fewer
women in this age group will benefit from screening.
- Cost of mammography per year of life saved is
approximately 80.000 Euro.