Annals of Oncology Advance Access published online on January 22, 2009
Annals of Oncology, doi:10.1093/annonc/mdn686
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
A comparison of models used to predict MLH1, MSH2 and MSH6 mutation carriers
1 Department of Medical Genetics, SMBD-Jewish General Hospital, Montreal, Canada
2 Genetic Counseling Program, Brandeis University, Waltham, USA
3 Department of Human Genetics, McGill University, Montreal, Canada
4 Molecular Pathology, SMBD-Jewish General Hospital, Montreal, Canada
5 Department of Psychology, Brandeis University, Waltham, USA
6 Departments of Medicine, Human Genetics and Oncology, McGill University
7 Department of Medical Genetics, McGill University Health Centre
8 Departments of Medicine, Human Genetics and Oncology, McGill University
9 Department of Medical Genetics, Oncology and Medicine, SMBD-Jewish General Hospital, Montreal, Canada
* Correspondence to: Ms. C. J. Pouchet MS, Cancer Prevention Centre, SMBD-Jewish General Hospital, E-758, 3755 ch. de la Côte Ste-Catherine, Montreal, Quebec H3T 1E2, Canada. Tel: +1-514-340-8222 ext 4315; Fax: +1-514-340-8600; E-mail: cpouchet{at}jgh.mcgill.ca
Background: MMRpro, prediction of mutations in MLH1 and MLH2 (PREMM1,2) and MMRpredict are models which were developed to predict the probability that an individual carries a Lynch syndrome-causing mutation. Each model utilizes data from personal and family histories of cancer. To date, no studies have compared these models in a cancer genetics clinic. The purpose of this study was to determine each model's ability to predict the probability of carrying a Lynch syndrome-causing mutation in individuals with a family history of colorectal cancer and to determine their clinical applicability.
Methods: We obtained family pedigrees from 81 individuals who presented for Lynch syndrome testing due to a personal and/or family history of cancer. Data from each pedigree were entered into the models and analyzed using SPSS.
Results: We found that MMRpredict, PREMM1,2 and MMRpro showed similar performances with areas under the receiver-operating characteristic curve of 0.731, 0.765 and 0.732, respectively. MMRpro showed the least dispersion of mutation probability estimates with a P value of 0.205, compared with 0.034 for PREMM1,2 and 0.001 for MMRpredict.
Conclusion: We found all three carried out well in a cancer genetics setting, with PREMM1,2 giving slightly better estimates. There were some significant discrepancies between the models in cases where the proband had endometrial cancer.
colorectal cancer, genetic testing, Lynch syndrome, mismatch repair, prediction models
Received for publication August 18, 2008. Revision received September 25, 2008. Accepted for publication September 26, 2008.