Many empiric and computer-based risk/probability estimation models have been developed, particularly after the discovery of BRCA1/2 genes, for estimating counselee’s probability of being carrier or to predict her risk of developing breast cancer. In this study, the performances of 6 such models have been compared. These are the Claus and Ford programs from Cyrillic3, BOADICEA, Manchester Scoring System, Tyrer-Cuzick, and COS which is a new model and being validated for the first time in this study. Model pedigrees and 172 Grampian families have been used to ascertain models’ performance. In this study the Claus and Ford models have the highest sensitivity but their extremely low specificities make them useless for any clinical or epidemiological use. In contrast COS and MSS have the second highest sensitivity (94% and 90% respectively) at reasonable specificities (53% and 41% respectively) and PPVs (56% and 49% respectively) showing that they are the most useful models for reducing the likelihood of mutations in BRCA1/2 (where the result is negative) and having the lowest false negative rate. From the ROC plots, COS and MSS also have the highest accuracy (within a range of all possible cut-off points) indicating their ability to discriminate between carriers and non-carriers. BOADICEA and T-C generated the most accurate overall predicted prevalence of mutations for all types of family histories and also increased the likelihood of carrying a mutation. BOADICEA and T-C have a sensitivity of 67%, specificity of 76% and 74% respectively and PPV of 64% and 62% respectively. COS and especially MSS can discriminate between BRCA1- and BRCA2-mutation carriers better than other models. These models identified a larger proportion of BRCA1- and BRCA2-mutation carriers correctly. COS and MSS have the higher sensitivity (73% and 64% respectively) at reasonable specificities (76% and 67% respectively) for the families with 3 or fewer cases of breast cancers in comparison with other models, while BOADICEA and COS have the most reasonable combination of sensitivity (80% and 100% respectively) and specificity (56% and 44% respectively) for the families with 6 or more cases of breast and/or ovarian cancer. Interestingly this study has shown that combined use of COS and MSS would significantly increase the specificity to 66% at the expense of few present loss of sensitivity. The single most effective model for clinical use is COS. However mutation prediction could be further improved if different models for different clinical circumstances (e.g. different family histories) were used. However it is practically cumbersome to have all models available in a busy clinic.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:590986 |
| Date | January 2007 |
| Creators | Roudgari, H. |
| Publisher | University of Aberdeen |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.002 seconds