Prediction of Lynch Syndrome in incident cases of colorectal cancer'

1
Prediction of Lynch Syndrome in incident cases of
colorectal cancer.
Roger Green Patrick Parfrey Michael Woods H.
Banfield Younghusband Chris Hatcher
2
Definitions

• Lynch syndrome
• inherited mutation of a mismatch repair (MMR)
  gene.
• Familial colorectal cancer, type X
• FCC-X
• Meets Amsterdam criteria.
• Tumour is MMR proficient.
• HNPCC
• both of the above.

3
Importance of identifying cases of Lynch syndrome

• Clinical management is different.
• Clinical screening.
• Extended family will require genetic counselling.
• DNA testing to identify pre-symptomatic mutation
  carriers.

4
How are Lynch syndrome cases currently identified?

• Someone happens to notice a strong family history
  of CRC.
• A referral is made to a cancer genetics clinic.
• Follow-up is variable.
• No national standards.
• Molecular testing not a routine clinical
  procedure.

5
Objective criteria

• Amsterdam
• High specificity
• Low sensitivity.
• Will miss LS in small families.
• Will miss lower-penetrance mutations. (MSH6,
  PMS2.)
• Bethesda
• Low specificity.
• High sensitivity.

6
Need a better way to identify potential Lynch
syndrome cases.

• Four different groups developed predictive
  models.
• All used high-risk populations.
• Three models used a logistic regression approach.
• One used a Bayesian-Mendelian analysis.

7
Clinical Findings with Implications for Genetic
Testing in Families with Clustering of Colorectal
Cancer
Juul T. Wijnen, B.S., Hans F.A. Vasen, M.D.,
Ph.D., P. Meera Khan, M.D., Ph.D., et al
Leiden model

• Mutation prevalence 26.
• Mutations in only MSH2 and MLH1 were considered.
• Factors incorporated
• meeting the Amsterdam criteria (AC1).
• mean age at diagnosis of CRC of all family
  members.
• family history of endometrial cancer.

8
June 29, 2006
Identification and Survival of Carriers of
Mutations in DNA Mismatch-Repair Genes in Colon
Cancer
Edinburgh
Rebecca A. Barnetson,et al . and Malcolm G.
Dunlop.
MMRpredict model

• Mutation prevalence 4.4. (Age Dxlt50years.)
• Mutations in MSH2, MSH6 and MLH1 were considered.
  
• Variables age, sex, tumour location, multiple
  CRCs, CRC and endometrial cancer in first degree
  relatives.

9
20062961479-1487.
Prediction of MLH1 and MSH2 Mutations in Lynch
Syndrome
Judith Balmaña, MD David H. Stockwell, MD, MPH
Ewout W. Steyerberg, PhD Elena M. Stoffel, MD,
MPH Amie M. Deffenbaugh, BS Julia E. Reid,
MStat Brian Ward, PhD Thomas Scholl, PhD Brant
Hendrickson, MS John Tazelaar, MD Lynn Anne
Burbidge, BS Sapna Syngal, MD, MPH
Spain
PREMM1,2 model

• Mutation prevalence 14.5.
• Mutations in only MLH1 and MSH2 were considered.
• Proband-specific variables age at diagnosis of
  CRC, adenomas, endometrial cancer, and other
  LS-associated cancers.
• The number of relatives with CRC, endometrial
  cancer, and other LS-associated cancers
• relationship to the proband (first- versus
  second-degree)
• minimum age at diagnosis for each cancer in the
  family
• a relative with more than one LS-associated
  cancer.

10
2006 296 1469-1478
Prediction of Germline Mutations and Cancer Risk
in the Lynch Syndrome
Sining Chen, PhD Wenyi Wang, MA Shing Lee, ScM
Khedoudja Nafa, PharmD, PhD Johanna Lee, MPH
Kathy Romans, MS Patrice Watson, PhD Stephen B.
Gruber, MD, PhD, MPH David Euhus, MD Kenneth W.
Kinzler, PhD Jeremy Jass, MD(Lond), DSc(Lond)
Steven Gallinger, MD, MSc Noralane M. Lindor,
MD Graham Casey, PhD Nathan Ellis, PhD Francis
M. Giardiello, MD Kenneth Offit, MD, MPH
Giovanni Parmigiani, PhD for the Colon Cancer
Family Registry
MMRpro model

• Mutation prevalence 43.
• Mutations in MSH2, MLH1 and MSH6 were considered.
• Depends on the application of Bayes theorem and
  the Mendelian laws of inheritance.
• Developed a priori based on published values for
  mutation prevalence and penetrance.

11
Our study
Rationale

• All models were developed using high-risk cases
  (strong family history of CRC or early age at
  diagnosis).
• It is not known if the models can predict the
  risk of Lynch syndrome in CRC patients from the
  general population.

Objectives

• To test the diagnostic utility of four different
  predictive models in identifying cases of Lynch
  syndrome from cases of CRC in the general
  population.
• To compared their performance to that of the
  Amsterdam and Bethesda criteria.
• To determine whether variations in family size
  lead to biases in risk prediction.

12
A series of 725 Newfoundland colorectal cancer
patients diagnosed before the age of 75 years.
1999-2003
Mean age (yr)
60.4
Characteristic
n

69
9.5
MSI-High tumor
Meets Bethesda guidelines
364
50.2
CRC in 1 first-degree relative
241
33.2
Meets AM1 criteria
28
3.9
Meets AM2 criteria
31
4.3
Meets clinical criteria for FAP
7
1
13
A series of 725 colorectal cancer patients.
Mutations found in 25 cases (3.4)
14
600
600
Leiden
MMRpredict
500
500
Figure 1. Distribution of risk scores.
400
400
Number of cases
300
300
200
200
100
100
0
0
0
10
20
80
90
100
0
10
20
80
90
100
600
600
500
500
PREMM1,2
MMRpro
400
400
Number of cases
300
300
200
200
100
100
0
0
0
10
20
80
90
100
0
10
20
80
90
100
Risk score
Risk score
15
How accurate are the risk estimates?
16
Observed and expected numbers of MMR-mutation
carriers, for different categories of predicted
risk
,
17
(No Transcript)
18
Can the models be used to determine which cases
should be tested for MS!?

• How do they compare to the Bethesda guidelines?
• The ideal test should have
• high sensitivity (few false negatives).
• high specificity (few false positives).
• Compare sensitivity and specificity of models in
  predicting the presence of a MMR gene mutation.
• 18 of 725 cases (2.5) had a MMR mutation.

19
Receiver Operator Characteristics
AUC area under the curve.
20
Comparison of discriminatory power at a
sensitivity of 94.
21
Can we improve on these scores?

• Three models ignore unaffected members of the
  family. (Leiden, MMRpredict, PREMM1,2)
• The MMRpro model incorporates the ages of all 1st
  -and 2nd-degree relatives.
• Can we incorporate this information into the
  other three models?
• Will it make any difference?

22

• The SISE coefficient quantifies the
  informativeness of any family.
• Based on
• the number of relatives
• their relationship to the proband,
• their ages
• the age-dependent penetrance of HNPCC mutations.
• plateaus at age 70
• A large SISE coefficient is indicative of a
  large family containing many close relatives who
  survived into old age.

23
Discriminatory power when risk score is corrected
for family SISE.
24
Discriminatory power when risk score is corrected
for family SISE.
Sensitivity fixed at 94
25
Conclusions
With our population-bases data set

• All models over-estimated risk.
• 1.2-fold to 4.3-fold
• All models could discriminate between MMR
  carriers and non-carriers.
• MMRpredict performed best.
• Correcting for family size improved the
  specificity.

26
Distribution of individual risk scores.
Each point represents the risk score of an
individual patient.
27
725 incident CRC cases 100
Figure 4. Flow chart for identifying Lynch
syndrome mutations
Exceed MMRpredict cutoff
No
Yes
82 cases 11.3
643 cases 89
Tumour microsatellite testing
Not MSI-high
MSI-high
No further action
28 cases 3.9
54 cases 7.4
MSH2, MSH6, or PMS2-deficient
MLH1-deficient
Methylated or BRAF mut
11 cases 1.5
17 cases 2.3
7 cases
Unmethylated No BRAF mut
4 cases
Cutoff MMRpredict score / SISE coefficient
gt1.66. In tumor, MLH1 promoter is methylated
or BRAF has the V600 mutation.
Mutation search in appropriate gene