|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 24, No 4 (February 1), 2006: pp. 707-715 © 2006 American Society of Clinical Oncology. DOI: 10.1200/JCO.2005.01.9737 Optimal Selection of Individuals for BRCA Mutation Testing: A Comparison of Available MethodsFrom the Familial Cancer Centre and Centre for Biostatistics and Clinical Trials, Peter MacCallum Cancer Centre; and Genetic Health Services Victoria, Murdoch Children's Research Institute, Melbourne, Australia Address reprint requests to Paul A. James, MD, Department of Human Anatomy and Genetics, Oxford University, South Parks Rd, Oxford, OX1 3QX, United Kingdom; e-mail: paul.james{at}anat.ox.ac.uk
PURPOSE: Several methods have been described that estimate the likelihood that a family history of cancer is a result of a mutation in the BRCA1 or BRCA2 genes. We examined the performance of six different methods with the aim of identifying an optimal strategy for selecting individuals for mutation testing in clinical practice. PATIENTS AND METHODS: Two hundred fifty-seven families who had completed BRCA1 and BRCA2 mutation screening were assessed by six models representing the major methodologies used to assess the likelihood of a pathogenic mutation. The performance of each method as a selection criterion was compared with the result of mutation testing to produce sensitivity, specificity, and receiver operating curve data. The impact of incorporating breast cancer pathology data in the assessment was also analyzed. RESULTS: The highest accuracy was achieved by the Bayesian probabilistic model (BRCAPRO). The formal probabilistic methods were significantly more accurate than clinical scoring methods. The methods were further improved by the incorporation of information on breast cancer pathology (tumor grade and estrogen receptor/progesterone receptor status). The resulting combined probability figure was highly accurate when selecting individuals for BRCA1 testing. Some BRCA2 mutation carriers were missed by all of the models examined. CONCLUSION: Formal probabilistic models provide significantly greater accuracy in the selection of families for gene testing than the use of clinical criteria or scoring methods. The accuracy is further enhanced by incorporating information on the pathology of breast cancers occurring in the families.
Inherited mutations in two tumor-suppressor genes, BRCA1 and BRCA2, account for the high incidence of breast and ovarian cancer in approximately 30% of families with strong histories of these cancers.1-3 Screening for cancer-predisposing mutations in BRCA1 and BRCA2 is now a routine part of the investigation and management of familial breast and ovarian cancer. When a pathogenic mutation is detected, increased screening procedures and other forms of intervention are offered to those with a high cancer risk. An accurate predictive test can then also be offered to other at-risk family members to determine who has inherited the mutation. Mutation testing is expensive and is generally offered to families in whom family history and other clinical factors indicate that the expectation of finding a mutation in the BRCA1 or BRCA2 gene is increased. The American Society of Clinical Oncology has previously recommended that mutation testing should be offered in cases where the probability of identifying a BRCA1/2 mutation is 10% or greater.4 Internationally, other criteria for mutation testing are used.5-7 Families with the highest risk of carrying a mutation can generally be identified clinically by the large number of breast and ovarian cancers occurring in the family and the young age at which they are diagnosed. However, many families presenting for assessment at family cancer centers will have less prominent family histories, and deciding which of these families should proceed with mutation testing remains challenging. Several different approaches have been developed to assess the likelihood of a pathogenic mutation being present, including clinical criteria and scoring systems,5,6,8-10 empirical approaches that reflect the actual experience of large testing centers,11,12 and statistical approaches such as logistic regression13 or a Bayesian probabilistic method.14-16 In addition, specific pathologic features have been used to predict which breast cancers may be associated with underlying mutations, particularly in the BRCA1 gene.17 Each of these methods can be used as a basis for determining which families should be offered genetic testing. Some of the reported methods have not been independently validated. In addition, studies that have compared more than one model are limited,8,9,18 and most have not demonstrated a statistical difference between the methods used. In this study, we compare the performance of a representative group of previously described methods for selecting hereditary breast and ovarian cancer families for gene testing, or calculating the pretest probability of finding a BRCA1 or BRCA2 mutation. The aim was to identify an optimal method and a threshold level for initiating mutation testing of moderate- to high-risk families that could be applied in the setting of the standard clinical practice of a specialized familial cancer center.
Study Population All families completing a family history assessment and BRCA1 and BRCA2 gene mutation screening through the Familial Cancer Centre at the Peter MacCallum Cancer Centre (Melbourne, Australia) from 1997 to June 2003 were included in this retrospective study. During this period, the standard criterion for mutation testing was a validated history of at least two first- or second-degree relatives with breast or ovarian cancer and an additional high-risk feature. High-risk features included additional affected family members, an individual diagnosed with breast cancer before the age of 40 years or ovarian cancer before the age of 50 years, bilateral breast cancer or breast and ovarian cancer in the same individual, male breast cancer, and Ashkenazi Jewish decent. Testing was also offered to a small number of individuals (n = 15) with one or more high-risk features but without a significant family history.
Family History Assessment
Mutation Screening
Comparison of Methods
Pathology data were collected, where available, for all cancers. Any reported level for estrogen receptors and progesterone receptors in immunohistochemical assays was graded as positive. For breast cancers, the distribution of histologic grade and hormone receptor status was then used to create a set of conditional probabilities for each of these features with respect to gene mutation status that were used to adjust the probabilities from the clinical and family history by Bayesian methods, as follows:
The Frank and Couch models and BRCAPRO calculations were performed using the CancerGene3.3.2b software package from University of Texas Southwestern Medical Center (Dallas, TX). The accuracy of the BRCAPRO program relies on the inclusion of survival data for unaffected family members. If these individuals are omitted because, for example, the exact age of death is not known, then the mutation probability will be overestimated. To allow inclusion of all known family members, we used a population average figure when information such as date of birth or age at death were not known to the family and could not be established from the state registry of births, deaths, and marriages. An assumption of this sort was used for at least one individual in the majority of families, but in nearly all cases, the assumption involved second-degree or more distantly related relatives. The FHAT score and Manchester Score were calculated manually.
Sensitivity and specificity values for each model were calculated for all of the possible testing threshold values. To evaluate the overall performance of each model, receiver operating curves (ROCs) were constructed by plotting the sensitivity of the method for selecting BRCA1/2 mutation carriers at a range of threshold values (0% to 100% probability) on the y-axis against 1 minus specificity (the false-positive rate) for that threshold value on the x-axis. The areas under the ROCs were calculated as a measure of the level of discrimination achieved by each model and were compared using the method described by DeLong et al.22 Proportions were compared using the
Description of the Study Population Probands from 257 families underwent genetic testing for mutations in the BRCA1 and BRCA2 genes between January 1997 and June 2003. Data was collected on 5,650 individuals from these families. The baseline characteristics of the study sample are listed in Table 3. In 11 families, the proband was found to have a variant of unknown significance in at least one of the genes, and the families were excluded from further analysis. Of the remaining 246 families, 37 (15%) were of Ashkenazi ancestry. The median age of the proband at testing was 52 years (standard deviation, 12.5 years; range, 28 to 94 years), and 238 of the probands (97%) were female. The average number of family members per pedigree was 22 (standard deviation, 7.9 members) and was not significantly different between the mutation-positive and mutation-negative families (P = .37). A BRCA1 mutation was detected in 33 probands (13.4%), a BRCA2 mutation was detected in 34 probands (13.8%), and no mutation was detected in 179 probands (72.8%).
Comparison of Methods Used to Select Probands for Mutation Testing A broad range of pretest mutation probability was found by all the models, with a strong weighting towards low- and moderate-risk families, reflecting the spectrum of families referred to the Family Cancer Centre. Examples of the distribution of scores using the BRCAPRO program and the FHAT scoring system are shown in Figure 1. Table 4 lists the sensitivity and specificity with which the models predicted the result of the proband’s mutation testing using a 10% mutation probability as a threshold. The overall ability of each method to discriminate between mutation-positive and -negative families at any testing threshold is best summarized by an ROC. The ROC for each method is shown in Figure 2, and the areas under the curves (AUCs) are listed in Table 5. The highest discrimination between mutation positive and negative was shown by the combined (BRCA1 and BRCA2) BRCAPRO score, with an AUC ROC of 0.78 (95% CI, 0.72 to 0.85; Fig 2C). The combined BRCAPRO statistic out-performed the attempts by the program to assign separate mutation probabilities for the individual genes, and all of the following data presented for BRCAPRO are for the combined BRCA1 and BRCA2 tool. A pair-wise comparison of the accuracy of each method is shown in Table 6. The differences in discrimination between BRCAPRO and the Frank model and Couch model were not statistically significant when predicting a mutation in either the BRCA1 or BRCA2 genes. The performance of the two clinical scoring methods was comparatively reduced. The accuracy of both the FHAT and the Manchester Score was significantly less than the accuracy of BRCAPRO for predicting any mutation and significantly less than the accuracy of both BRCAPRO and the Frank model for predicting BRCA1 mutations when all families were included. In the case of the FHAT, this was mainly the result of a lack of specificity. The patient group used to develop the Manchester Score did not contain Ashkenazi Jewish families. When these families are excluded and the analysis is repeated, the performance of the Manchester Score improves to a sensitivity at the 10% threshold of 81%, a specificity of 58%, and an AUC ROC of 0.71 (95% CI, 0.63 to 0.79) for any mutation (BRCA1: AUC ROC = 0.78; 95% CI, 0.67 to 0.88; BRCA2: AUC ROC = 0.69; 95% CI, 0.59 to 0.78). The removal of the 37 Ashkenazi families did not alter the rank order of the methods, but the reduction in the number of families meant that pair-wise comparisons between methods were no longer statistically significant.
The application of clinical criteria (Adelaide Criteria) resulted in a specificity of 37% and a sensitivity of 81%. The poor discrimination of these criteria is emphasized by the finding that five of the families carrying mutations in BRCA1 and eight families carrying mutations in BRCA2 did not meet any of the clinical criteria.
Inclusion of Pathology Data Improves Mutation Prediction
For the most effective management of hereditary breast and ovarian cancer families and the best utilization of health care resources, it is important to accurately identify the group of individuals at genuinely high risk of carrying BRCA1 and BRCA2 mutations. Information on family history and the age at which the cancers are diagnosed is widely used by clinicians to select families for testing. Clinical criteria that take these factors into account are able to select for families at high risk but are less effective at excluding mutation-negative families, resulting in low specificity. As a result, in the reported patient series where families were defined as high risk on the basis of varying criteria (summarized by Antoniou et al2) and screened for BRCA1 and BRCA2 mutations, the proportion of families found to carry a mutation ranged from 2% to 40%, with an average pick-up rate of only 5.5%.2 Euhus et al23 directly compared assessment by an objective statistical method (the BRCAPRO program) with the clinical assessment of a group of experienced genetic counselors. Although the ability of the clinicians to identify mutation-positive families was not significantly different from the statistical method, they were less likely to exclude a mutation-negative family from testing (specificity for the counselors = 16%; specificity for BRCAPRO = 32%; P = .004). In our study, a set of clinical guidelines for selecting families for gene testing also suffered from being overinclusive, resulting in a low overall accuracy. Using these criteria, only a little more than one third of families without a mutation could be excluded from testing. Adapting clinical criteria into a simple scoring system does not completely overcome a lack of specificity. The FHAT score was the least specific method and ranked lowest in terms of overall accuracy, whereas the Manchester Score was intermediate in accuracy. Each of the remaining methods showed an improved ability to discriminate between the mutation-positive and -negative groups. In this study, the combined BRCA1 and BRCA2 mutation probability generated by the BRCAPRO program achieved the greatest accuracy. When a 10% probability is applied to data generated by BRCAPRO, 61% of the families without a mutation could be excluded from testing. Fourteen of the 67 families in which a mutation was found would also have been excluded. The majority of families (65% to 83%) that would have been incorrectly excluded at this threshold for testing, under all the models, had mutations in the BRCA2 gene. One family with a BRCA2 mutation would not have been selected for testing, using a 10% mutation probability threshold (or the equivalent FHAT score or Manchester Score), by any of the models examined. The possibility was considered that the low scores assigned to some mutation-positive families may have been a result of the proband not being the person at the highest individual risk in the family or a result of affected family members being excluded from the assessment because of the limitations of the models. However, when the pedigrees of these families were reviewed, neither of these possibilities was found to be the explanation in any case. Instead, these families were principally BRCA2 families with genuinely unremarkable family histories, reflecting the lower penetrance of mutations in this gene. The accuracy of the methods compared in this study is similar to that reported previously. De la Hoya et al18 examined four different logistic regression methods and compared their performance to the Frank model12 in a group of 109 Spanish families. The AUC ROCs reported for these methods ranged from 0.77 to 0.82, with no statistically significant difference between the methods. Similarly, Euhus et al23 found an AUC ROC of 0.71 (95% CI, 0.71 to 0.72) for the BRCAPRO program when compared with the results of mutation testing in 148 families. In their description of the Manchester Score, Evans et al9 compared its performance with BRCAPRO, the Frank model, and the Couch model in a group of 175 probands who had been tested for BRCA1 and BRCA2 mutations. In this study, the accuracy of the Manchester score, as measured by the AUC ROC, was greater than for the other methods in the comparison, although this only reached a statistically significant level for the prediction of BRCA2 mutations alone. The authors demonstrated an accurate performance of the Manchester Score considering the unremarkable nature of the family history that is found in many BRCA2-positive families (BRCA2 AUC ROC = 0.82; 95% CI, 0.74 to 0.90). In our study, this level of accuracy for the Manchester Score was not replicated, and this method did not outperform other models, as has been reported.9 Empirical scoring methods have the advantage for clinical application of providing the most rapid evaluation of each family, taking only 1 to 2 minutes. In contrast, the computerized models require 10 to 15 minutes of data entry, although this may not be significant for clinics where computerized data entry of pedigrees is already part of standard clinical practice. For the computerized methods, the quality of data entry is also critical to the performance of the methods. This is particularly true for BRCAPRO because unaffected family members are included in the assessment. A consistent approach to the problem of individuals with missing data that inevitably arises in clinical practice (such as the use of population average figures) is essential to allow this model to perform properly. Previous studies using the BRCAPRO model have not detailed how they resolved this issue, and the noninclusion of unaffected individuals may contribute to the unusually poor performance of the BRCAPRO program in some studies.9 All the models examined in this study have limitations. Only FHAT and the Manchester Score take account of other malignancies known to be associated with BRCA gene mutations, such as pancreatic or prostatic carcinomas, and none of the methods are able to incorporate the effect of risk-reducing strategies such as prophylactic mastectomy or oophorectomy. The methods provide a mechanism for prioritizing families for mutation testing, rather than producing individual patient risk assessment figures. In all cases, correct application of the methods remains dependent on the experience and clinical judgement of the clinician or counselor. This is particularly the case with respect to the decision to test smaller families for which less family history is available. An alternative approach to predicting the presence of a BRCA gene mutation has focused on the pathologic features of the cancers observed in hereditary breast and ovarian cancer families. The association of specific histomorphologic features with familial forms of breast cancer is well established.17,24-28 A small number of studies have shown that histopathologic data can increase the sensitivity of personal or family history as used to identify BRCA1 gene carriers.25,29,30 A broad pathologic classification in the form of grade and hormone receptor status was applied in this study. This information is available for most breast cancers without requiring specialist pathologic review and, in historical cases, can be frequently obtained from existing medical records or cancer registries. The distribution of grade and receptor status was consistent with previous reports in the literature.17,25,26 Using this distribution as a set of conditional probabilities to adjust the family history–based assessment of the mutation risk avoids introducing any bias or subjectivity that could occur if a derived set of pathologic criteria or scoring system was used. Incorporating three aspects of the breast cancer pathology (grade, estrogen receptor status, and progesterone receptor status) for the proband alone using the simple Bayesian formula improved our ability to discriminate between mutation-positive and -negative families, in the case of BRCA1 families, to a high level (a testing threshold of 10% corresponds to a sensitivity of 95% and a specificity of 70%). This improvement was a result of the distinct pathologic profile of BRCA1-related breast cancers. Incorporating pathology data did not significantly improve selection of BRCA2 families. Six (21%) of 28 BRCA2 carriers were missed despite using this strategy. Some BRCA2 carriers belong to families with unremarkable cancer histories, and these individuals will always be difficult to identify. The findings of the study are limited by the sensitivity of the mutation testing. Although all exons of both genes were examined, some pathologic missense mutations in exon 11 of BRCA1 could have escaped detection, and large-scale deletions or rearrangements certainly would not have been detected. Such mutations may have contributed to the eight (5%) of 160 mutation-negative families in which the most accurate model predicted a 90% or greater probability of a mutation. Our study demonstrates that, in the setting of the routine practice of a familial cancer clinic, formal probabilistic models provide significantly greater accuracy in the selection of families for gene testing than the use of clinical criteria or scoring methods. The accuracy is further enhanced by incorporating information on the pathology of breast cancers occurring in the families into the model. The combination of BRCAPRO adjusted for the pathology resulted in a highly accurate method of great utility for selecting families for BRCA1 mutation testing. Greater clinical input is required when selecting families for BRCA2 testing because of the lower penetrance of malignancies associated with these mutations, which results in less remarkable family histories, and the lack of a distinct profile of breast cancer pathology.
The authors indicated no potential conflicts of interest.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Claus EB, Schildkraut JM, Thompson WD, et al: The genetic attributable risk of breast and ovarian cancer. Cancer 77:2318-2324, 1996[CrossRef][Medline] 2. Antoniou A, Pharoah PDP, Narod S, et al: Average risks of breast and ovarian cancer associated with BRCA1 or BRCA2 mutations detected in case series unselected for family history: A combined analysis of 22 studies. Am J Hum Genet 72:1117-1130, 2003[CrossRef][Medline] 3. Evans DGR, Bulman M, Young K, et al: Sensitivity of BRCA1/2 mutation testing in 466 breast/ovarian cancer families. J Med Genet 40:e107, 2003 4. American Society of Clinical Oncology: Statement of the American Society of Clinical Oncology: Genetic testing for cancer susceptibility. J Clin Oncol 14:1730-1736, 1996 5. Eccles DM, Evans DGR, Mackay J: Guidelines for a genetic risk based approach to advising women with a family history of breast cancer. J Med Genet 37:203-209, 2000 6. National Health and Medical Research Council, National Breast Cancer Centre, New South Wales: Advice about familial aspects of breast cancer and ovarian cancer: A guide for health professionals. http://www.genetics.com.au/pdf/pubs/boguidelines.pdf 7. Tang NLS, Choy KW, Pang CP, et al: Prevalence of breast cancer predisposition gene mutations in Chinese women and guidelines for genetic testing. Clin Chim Acta 313:179-185, 2001[CrossRef][Medline] 8. Gilpin CA, Carson N, Hunter AG: A preliminary validation of a family history assessment form to select women at risk for breast or ovarian cancer for referral to a genetics center. Clin Genet 58:299-308, 2000[CrossRef][Medline] 9. Evans DGR, Eccles DM, Rahman N, et al: A new scoring system for the chances of identifying a BRCA1/2 mutation outperforms existing models including BRCAPRO. J Med Genet 41:474-480, 2004 10. Evans DGR, Lalloo F, Wallace A, et al: An update on the Manchester Scoring System for BRCA1 and BRCA2 testing. J Med Genet 42:e39, 2005 11. Frank TS, Deffenbaugh AM, Reid JE, et al: Clinical characteristics of individuals with germline mutations in BRCA1 and BRCA2: Analysis of 10,000 individuals. J Clin Oncol 20:1480-1490, 2002 12. Frank TS, Manley SA, Olopade OI, et al: Sequence analysis of BRCA1 and BRCA2: Correlation of mutations with family history and ovarian cancer risk. J Clin Oncol 16:2417-2425, 1998[Abstract] 13. Couch FJ, DeShano ML, Blackwood MA, et al: BRCA1 mutations in women attending clinics that evaluate the risk of breast cancer. N Engl J Med 336:1409-1415, 1997 14. Berry DA, Parmigiani G, Sanchez J, et al: Probability of carrying a mutation of breast-ovarian cancer gene BRCA1 based on family history. J Natl Cancer Inst 89:227-238, 1997 15. Parmigiani G, Berry D, Aguilar O: Determining carrier probabilities for breast cancer-susceptibility genes BRCA1 and BRCA2. Am J Hum Genet 62:145-158, 1998[CrossRef][Medline] 16. Berry DA, Iversen ES Jr, Gudbjartsson DF, et al: BRCAPRO validation, sensitivity of genetic testing of BRCA1/BRCA2, and prevalence of other breast cancer susceptibility genes. J Clin Oncol 20:2701-2712, 2002 17. Lakhani SR, Van De Vijver MJ, Jacquemier J, et al: The pathology of familial breast cancer: Predictive value of immunohistochemical markers estrogen receptor, progesterone receptor, HER-2, and p53 in patients with mutations in BRCA1 and BRCA2. J Clin Oncol 20:2310-2318, 2002 18. De la Hoya M, Díez O, Pérez-Segura PJ, et al: Pre-test prediction models of BRCA1 or BRCA2 mutation in breast/ovarian families attending familial cancer clinics. J Med Genet 40:503-510, 2003 19. Gross E, Arnold N, Goette J, et al: A comparison of BRCA1 mutation analysis by direct sequencing, SSCP and DHPLC. Hum Genet 105:72-78, 1999[CrossRef][Medline] 20. Wagner T, Stoppa-Lyonnet D, Fleischmann E, et al: Denaturing-high performance liquid chromatography detects reliably BRCA1 and BRCA2 mutations. Genomics 62:369-376, 1999[CrossRef][Medline] 21. Garvin AM: A complete protein truncation test for BRCA1 and BRCA2. Eur J Hum Genet 6:226-234, 1998[CrossRef][Medline] 22. DeLong ER, DeLong DM, Clark-Pearson DL: Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach. Biometrics 44:837-845, 1988[CrossRef][Medline] 23. Euhus DM, Smith KC, Robinson L, et al: Pretest prediction of BRCA1 or BRCA2 mutation by risk counselors and the computer model BRCAPRO. J Natl Cancer Inst 94:844-851, 2002 24. Lakhani SR: The pathology of familial breast cancer: Morphological aspects. Breast Cancer Res 1:31-35, 1999[CrossRef][Medline] 25. Breast Cancer Linkage Consortium: Pathology of familial breast cancer: Differences between breast cancers in carriers of BRCA1 or BRCA2 mutations and sporadic cases. Lancet 349:1505-1510, 1997[CrossRef][Medline] 26. Lakhani SR, Jacquemier J, Sloane JP, et al: Multifactorial analysis of differences between sporadic breast cancers and cancers involving BRCA1 and BRCA2 mutations. J Natl Cancer Inst 90:1138-1145, 1998 27. Osin P, Crook T, Powles T, et al: Hormone status of in-situ cancer in BRCA1 and BRCA2 mutation carriers. Lancet 351:1487, 1998[Medline] 28. Foulkes WD, Stefansson IM, Chappuis PO, et al: Germline BRCA1 mutations and a basal epithelial phenotype in breast cancer. J Natl Cancer Inst 95:1482-1485, 2003 29. Eisinger F, Noguès C, Guinebretière J-M, et al: Novel indications for BRCA1 screening using individual clinical and morphological features. Int J Cancer 84:263-267, 1999[CrossRef][Medline] 30. Cortesi L, Turchetti D, Bertoni C, et al: Comparison between genotype and phenotype identifies a high-risk population carrying BRCA1 mutations. Genes Chromosomes Cancer 27:130-135, 2000[CrossRef][Medline] Submitted March 30, 2005; accepted November 18, 2005. This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2006 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
2 test, and continuous variables were compared using t tests between mutation-carrying and noncarrying families. Performance measures (sensitivity and specificity) between the models were compared in a pair-wise fashion using the Wilcoxon signed rank test. Statistical analysis was performed using SAS version 8.0 (SAS Institute, Cary, NC). 
