estimating cancer penetrances from genetic test results

ESTIMATING CANCER PENETRANCES
FROM GENETIC TEST RESULTS:
Analysis of the Families from the Italian
Registry of Hereditary Breast/ovarian Cancer
Fabio Marroni, Paolo Aretini, Emma D'Andrea, Maria
Adelaide Caligo, Alessandra Viel, Laura Cortesi, Enrico
Ricevuto, Simona Agata, Roberta Bisegna, Mauro Boiocchi,
Luigi Chieco-Bianchi, Giovanna Cipollini, Massimo
Federico, Chiara Ghimenti, Clelia De Giacomi, Arcangela
De Nicolo, Lara Della Puppa, Sergio Ferrari, Corrado
Ficorella, Davide Iandolo, Paolo Marchetti, Chiara Menin,
Marco Montagna, Manuela Santarosa, Vittorio Silingardi,
Daniela Turchetti, Generoso Bevilacqua, Joan E. BaileyWilson, Giovanni Parmigiani, Silvano Presciuttini
Participating Institutions
Department of Oncology, Transplants and New Technologies in Medicine, Section of
Pathology; Department of Biomedicine
University of Pisa, Italy
Department of Oncology and Surgical Sciences, Section of Oncology; IST, Section
of Viral and Molecular Oncology; Azienda Ospedaliera University of Padua, Italy
Experimental Oncology ; Medical Oncology 3; Oncology Referral Centre, IRCCS
Aviano (PN), Italy
Department of Experimental Medicine
University of L'Aquila, Italy
Department of Biomedical Science, Section of Biological Chemistry; Department of
Oncology and Hematology
University of Modena and Reggio Emilia, Italy
Inherited Disease Research Branch, NHGRI
Baltimore, MD, USA
Departments of Oncology and Biostatistics, JHU
Baltimore, MD, USA
Introduction
 A relevant proportion of breast and ovarian
cancers is attributable to germline mutations in
BRCA1 and BRCA2
 Accurate evaluation of the probability that an
individual carries a germline pathogenic
mutation at BRCA1 or BRCA2 is therefore
essential to help counselors and counselands
decide whether testing is appropriate.
Computing carrier probability
 BRCAPRO is a widely used software that compute the
carrier probability for BRCA1/2 genes in probands, based
on their family history of breast and ovarian cancer in firstand second-degree relatives.
 In a previous validation study, we applied BRCAPRO to a
series of 568 Italian families tested for BRCA1/2 (80
mutations found in BRCA1, and 53 in BRCA2).
 The overall performance of BRCAPRO was better than that
of other models; however, our results revealed promising
prospects for substantial improvement.
A case of inaccurate predictions
• A larger-than-predicted number of BRCA1/2 mutations
was identified in families at relatively low risk
Number
of families
Expected
number of
mutations
Observed
number of
mutations
<10%
305
7.96
32
10-30%
98
18.11
25
>30%
165
89.96
76
Total
568
116.03
133
Carrier
probability
A second example of inaccurate
predictions
• The ability of discriminating between genes was limited in
families stratified by profile
BRCA1
Expected
number of
mutations
HBC
HBOC
HOC
MBC
Total
47.4
35.8
3.7
3.5
90.5
BRCA2
Observed
number of
mutations
27
45
7
1
80
Both genes
Expected Observed
number of number of
mutations mutations
11.8
5.4
0.3
8.0
25.6
34
12
1
7
54
Expected
number of
mutations
59.3
41.2
4.0
11.6
116.0
Observed
number of
mutations
61
57
8
8
134
Aim of this work
 In the present study we obtained improved estimates of the
parameters (allele frequencies and cancer penetrances in
carriers and non-carriers) that compose the genetic model on
which prediction is formulated
 We maximized the retrospective likelihood of the model,
given the observed test results, using a MCMC approach;
BRCAPRO was used as a probability calculation tool.
 A total of 13 parameters were estimated:


three for each of four penetrance functions (breast and ovarian
cancer in BRCA1 and BRCA2 carriers)
one for the ratio of BRCA1/BRCA2 mutation frequencies
Study design
GENETIC
MODEL
Stop when convergence
criteria are met
Accept/Reject
modified
model
Individual
prediction for
all families
Calculation
of total
log-likelihood
Explore new
Parameter values
Compare
log-likelihoods
Re-calculation
of total
log-likelihood
MODIFIED
MODEL
Results
 Likelihood reached a stable value after about 5,000
iterations; chains were continued for about 10,000
additional iterations. The final acceptance rate was
about 17%.
 The final log-likelihood was about –326, vs. an
initial value of about –396, with a difference of 70
log units; this difference is substantial,
corresponding to a ratio of likelihood of over 1030.
Our new genetic model is well supported.
1) new penetrance function of
breast cancer in BRCA1 carriers
Risk of Breast Cancer
Age
Original
values
MCMC
estimates
30
3%
2%
50
46%
22%
70
69%
46%
New estimate
2) new penetrance function of
ovarian cancer in BRCA1 carriers
Risk of Ovarian Cancer
Age Original MCMC
30
0.02%
1%
50
11%
12%
70
30%
52%
New estimate
3) new penetrance function of
breast cancer in BRCA2 carriers
Risk of Breast Cancer
Age Original MCMC
30
1%
1%
50
28%
23%
70
67%
49%
New estimate
4) new penetrance function of
ovarian cancer in BRCA2 carriers
Risk of Ovarian Cancer
Age Original MCMC
30
0.03%
0.1%
50
3%
2%
70
19%
22%
New estimate
Examining the consequences of
the new model (1)
Expected/ observed number of mutations in families
stratified by risk
Number
of families
Expected
number of
mutations
Observed
number of
mutations
<10%
230
8.8
14
10-30%
132
25.1
28
>30%
206
109.2
91
Total
568
143.1
133
Carrier
probability
Examining the consequences of
the new model (2)
Discriminating between the two genes
BRCA1
Expected
number of
mutations
HBC
HBOC
HOC
MBC
Total
23.2
43.9
10.0
1.8
78.9
BRCA2
Observed
number of
mutations
27
45
7
1
80
Both genes
Expected Observed
number of number of
mutations mutations
36.9
14.6
0.6
13.0
65.0
34
12
1
7
54
Expected
number of
mutations
59.8
58.1
10.6
14.6
143.2
Observed
number of
mutations
61
57
8
8
134
Discussion



We estimated the penetrance functions by maximizing the
likelihood of the genotypes given the observed phenotypes;
this is therefore what is called the “retrospective likelihood”
(Kraft and Thomas, AJHG 2000).
The parameter estimates based on the retrospective
likelihood remain unbiased even when the ascertainment
scheme cannot be modeled.
The retrospective likelihood is not efficient in estimating
absolute penetrance; we therefore estimated penetrance odds
ratios and obtained absolute penetrances for carriers and noncarriers based on incidence of cancer in the general
population
Conclusions
 Our data refer to a substantial proportion of the
families that currently require genetic counseling in
Italy; therefore our new penetrance estimates
provide the most accurate genetic model so far
available for this population segment, and may lead
to a mutation-predicting model specifically adapted
to this country and to development of an Italian
customized version of BRCAPRO.