- Email: [email protected]
Validating the IBIS and BOADICEA Models for Predicting Breast Cancer Risk in the Iranian Population
Accepted Manuscript Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population Mahshid Ghoncheh, MSc, Fatane Ziaee, MD, Manoochehr Karami, PhD, Jalal Poorolajal, MD, PhD PII:
S1526-8209(17)30010-1
DOI:
10.1016/j.clbc.2017.01.003
Reference:
CLBC 570
To appear in:
Clinical Breast Cancer
Received Date: 5 July 2016 Revised Date:
24 December 2016
Accepted Date: 9 January 2017
Please cite this article as: Ghoncheh M, Ziaee F, Karami M, Poorolajal J, Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population, Clinical Breast Cancer (2017), doi: 10.1016/j.clbc.2017.01.003. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Original article
Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population
RI PT
Mahshid Ghoncheh (MSc) Department of Epidemiology, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Cancer Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran Email: [email protected] Fatane Ziaee (MD) Cancer Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran Email: [email protected]
SC
Manoochehr Karami (PhD) Department of Epidemiology, School of Public Health, Social Determinants of Health Research Center, Hamadan University of Medical Sciences, Hamadan, Iran Email: [email protected]
M AN U
Jalal Poorolajal (MD, PhD): Corresponding author Department of Epidemiology, School of Public Health, Noncommunicable Diseases Research Center, Hamadan University of Medical Sciences, Hamadan, Iran Email: [email protected] Zip code: 65157838695 Tel: +98 81 38380090 Fax: +98 81 38380509
Word count Abstract: 249 Main text: 1822
TE D
Running title IBIS and BOADICEA models for predicting breast cancer risk
AC C
EP
Author Contributions • Mahshid Ghoncheh contributed to study conception and design, acquisition of data, analysis and interpretation of data, and critical revision. • Fatane Ziaee contributed to acquisition of data and critical revision. • Manoochehr Karami contributed to study design and critical revision. • Jalal Poorolajal contributed to study conception and design, analysis and interpretation of data, and drafting of manuscript.
1
ACCEPTED MANUSCRIPT
Micro abstract
The validity of the IBIS model was evaluated for predicting breast cancer risk in Iranian population. We performed a case-control study, enrolling 1633 Iranian women. We indicated that the discrimination of cases and non-cases based on IBIS model was better than BOADICEA model for
AC C
EP
TE D
M AN U
SC
RI PT
Iranian population, although the discrimination of the both models was relatively low.
2
ACCEPTED MANUSCRIPT
Original article
Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population Abstract Background: Several approaches have been suggested for incorporating risk factors to predict the
RI PT
future risk of breast cancer. The Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) and International Breast Cancer Intervention Study (IBIS) are among these approaches. We compared the performance of these models in discriminate between cases and non-case in the Iranian population.
SC
Patients and Methods: We performed a case-control study in Tehran, from November 2015 to April 2016, and enrolled 1633 women aged 24 to 75 years, including 506 cases of breast cancer,
M AN U
916 population-based controls, and 211 clinic-based controls. We calculated and compared the risk of breast cancer predicted by the IBIS and BOADICEA models and the logistic regression model. For model discrimination, we computed the area under the Receiver Operating Characteristic (ROC).
Results: The risk of breast cancer predicted by IBIS model was higher than BOADICEA model, but lower than logistic model. The area under the ROC plots indicated that the logistic regression
TE D
model showed better discrimination between cases and non-cases (71.53%) compared to IBIS model (49.36%) and BOADICEA model (35.84%). Based on the Pierson correlation coefficient, the correlation between IBIS and BOADICEA models was much stronger than the correlation between
EP
IBIS and logistic models (0.3884 and 0.1639, respectively). Conclusion: The IBIS model discriminated cases and non-cases better than BOADICEA model in
AC C
the Iranian population. However, the discrimination of the logistic regression model, which included a larger array of familial, genetic, and personal risk factors, was better than the two models.
Keywords: Breast Neoplasms; Risk Assessment; Statistical Models; Case-Control Studies
3
ACCEPTED MANUSCRIPT
Introduction
By far, breast cancer is the most common invasive cancer diagnosed in women both in the developed and the developing countries1. Breast cancer incidence varies greatly worldwide. Although it is supposed to be a common cancer in the developed countries, about 58% of all breast cancer deaths occur in less developed countries2. Several modifiable and non-modifiable factors are associated with breast cancer, including older
RI PT
age3, genetics4,5, late age at first childbirth and lack of breastfeeding6, early age at menarche7, late menopause8, oral contraceptives and hormone replacement therapy9-11, obesity12, alcohol13, smoking14, and radiation15.
SC
Several studies have evaluated the association between subsets of genetic and individual factors and the incidence risk of breast cancer. However, there is a need to incorporate these risk factors and predict the overall risk of developing breast cancer. Different approaches have been suggested for
M AN U
incorporating risk factors to predict the risk of breast cancer. The Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) is a breast cancer risk prediction model which was developed by Cunningham
16
and was described by Lee et al
17
.
BOADICEA is presented as a web-based computer program that is used to estimate the future risks of developing breast or ovarian cancer in women according to their family history. Another
TE D
approach is International Breast Cancer Intervention Study (IBIS) which was introduced by Tyrer et al18. They provided this tool for research purposes only and did not recommend it for clinical decisions yet19. This model is adapted for UK and Sweden populations only. However, it is not clear how these models work in other populations in the world. In this study, we employed the IBIS
population.
AC C
Methods
EP
and BOADICEA models to predict the risk of developing breast cancer among Iranian women
We carried out this case-control study in Tehran, the capital city of Iran, from November 2015 to April 2016. We enrolled 1633 native Iranian women aged 24 to 75 years, including 506 women with confirmed breast cancer as the case group, 916 women without known breast cancer from the general population as the population-based control group, and 211 women without known breast cancer who attended to the Cancer Research Center for other reasons as the clinic-based control group. The Research Council of Hamadan University of Medical Sciences approved this study. Since we did not carry out any intervention, we just took participants' implicit (verbal) consent to participate in the study and answer to our questions
4
ACCEPTED MANUSCRIPT
We selected cases among patients with breast cancer, newly diagnosed pathologically based on International Classification of Diseases for Oncology 3rd edition sites C50.0–C50.9)20 and registered with the Cancer Research Center database, affiliated with Shahid Beheshti University of Medical Sciences. Cases were attended to this center from different parts of the city, including hospitals, laboratories, and clinics. We excluded patients with unknown pathology or those who out-migrated or died. We selected clinic-based controls among individuals who attended to the
RI PT
same center for breast diseases other than breast cancer or suspected to have breast cancer, but eventually ruled out pathologically. We extracted data from the medical records and obtained additional data, not included in the medical records, through interviews with cases and controls after describing the objectives of the study for them and taking their verbal consent.
SC
We selected population-based controls from the general population of Tehran and its suburbs. In order to increase the generalizability of the results, we divided the city into five regions, including
M AN U
north, south, east, west, and center. Then, we selected a probability sample of the population from each region through the door to door approach.
The data collection tool was a questionnaire included age, weight, height, age at menarche, age at first delivery, number of children, menopausal status, and presence or absence of a benign breast disease, history of breast cancer in the first-degree relatives (mother or sister) or other relatives, history of hormone replacement therapy (HRT) including estrogen and/or progestin, and presence
TE D
or absence of the BRCA genes.
We used IBIS model to predict the risk of developing breast cancer. This computer-based program provides a woman's overall risk of breast cancer by incorporating genetic determinants such as the
EP
BRCA1 and BRCA2 genes, details about a developing breast and/or ovarian cancers among family members, and personal risk factors such as age, body mass index, age at menarche, parity, age at
AC C
first child, menopausal status, age at menopause, and benign breast diseases18. We used BOADICEA model to calculate the risks of breast cancer in women based on their family history. We run BOADICEA risk calculations using the latest version of BOADICEA web application (BWA v3).
We performed a multiple logistic regression analysis to estimate the risk of breast cancer as the third model. For this purpose, we generated a set of breast cancer risk factors (age, body mass index, early menarche (<13 years), the presence or absence of a benign breast disease, and a history of breast cancer in the first-degree or other relatives) based on the information obtained from cases and controls. Then, we combined risk factors with information on the odds ratio of each risk factor
5
ACCEPTED MANUSCRIPT
and calculated the observed probability of breast cancer for an individual with a given set of characteristics as follows: P(୷|୶) =
ଵ
Equ. 1
ଵାୣష(ౘబ శౘ ౮ )
Where, 'P' is the calculated probability of breast cancer for an individual. 'b0' is the intercept, that is, the estimated value of P when x = 0. 'bi' is the regression coefficient, that is, the estimated increase
RI PT
in the breast cancer risk per unit increase in the predictor variable (x). According to this equation, we modelled and combined the risks of breast cancer and calculated the individual probability of breast cancer. Then, we combined the individual risk and obtained the overall risk of breast cancer for cases and controls.
SC
We considered observed cases and non-cases as the gold standard. Then, we compared the risk of breast cancer predicted by IBIS and BOADICEA models and logistic regression analysis with the
M AN U
gold standard to determine which model could better discriminate between cases and non-cases on the basis of Receiver Operating Characteristic (ROC) area and plot. We also estimated the correlation between models using Pierson correlation coefficient.
We used to chi-square test for analysis of nominal variables. We performed all statistical analyses at a significance level of 0.05 using Stata software, version 11 (StataCorp, College Station, TX, USA). We also employed IBIS Breast Cancer Risk Evaluation Tool version 7 for predicting the breast
TE D
cancer risk.
In order to measure the validity (sensitivity and specificity) of the IBIS model, we estimated the breast cancer risk for each individual using IBIS model. Then, we set a risk threshold of 20% and
EP
considered subjects with a lifetime risk of greater than 20% for developing breast cancer positive and otherwise negative21. Finally, we estimated sensitivity and specificity of the IBIS model against
AC C
gold standard that was the cases of breast cancer confirmed pathologically and the population-based and clinic-based controls. Results
In this study, 506 cases of breast cancer were compared with 916 population-based controls and 211 clinical-based controls. The mean (SD) age of the cases was 48.37 (10.79) years and that of population-based and clinic-based controls was 42.37 (9.84) and 44.31 (10.85) years, respectively. The characteristics of the case and control groups are given in Table 1. There was a statistically significant association between breast cancer and age, body mass index (BMI), age at menarche, age at first birth, menopausal status, a positive history of a benign breast disease, a positive history of breast cancer in the first-degree or other relatives, and presence of the BRCA genes. 6
ACCEPTED MANUSCRIPT
The predicted risk of breast cancer based on the IBIS, BOADICEA, and logistic regression models is given in Table 2. There was a significant difference between the risk of breast cancer predicted by the three models for both population-based and clinic-based controls and cases of breast cancer. According to out result, the risk of breast cancer predicted by BOADICEA model was lower than that of IBIS model. On the other hand, the risk of breast cancer predicted by logistic model was much higher than that of IBIS model.
RI PT
We computed the ROC area to compare the discrimination of the IBIS, BOADICEA, and logistic models, (Figure 1). The area under ROC curve was 71.53% for the logistic model, 49.36% for the IBIS model, and 35.84% for the BOADICEA model. According to these results, the logistic regression model discriminated between cases and non-cases better than IBIS and BOADICEA
SC
models.
The scatter plot of the risk of breast cancer predicted by IBIS against the risk of breast cancer
M AN U
predicted by logistic model and BOADICEA model is shown in Figure 2 and 3, respectively. The correlation between IBIS model and BOADICEA model was much stronger than the correlation between IBIS model and logistic model. As shown in Figure 2, there was the moderate correlation between IBIS and BOADICEA models (Pierson correlation coefficient = 0.3884). However, as shown in Figure 3, the correlation between IBIS and logistic models was poor (Pierson correlation
Discussion
TE D
coefficient = 0.1639).
Our results indicated that the IBIS model, which include family history and genetic data, could discriminate cases and non-cases of breast cancer better than BOADICEA model, which is based on
EP
only family history. However, the performance of the multiple logistic regression model was better than the two models. The reason was that, logistic model included family history, genetic
AC C
information, and individual risk factors. Therefore, the larger arrays of familial, genetic, and personal risk factors the risk models have, the more performance they would have. The IBIS model was actually developed to predict individual risk of breast cancer from both familial and personal risk factors. It is an easy-to-use computed-based program. It is possible to add new factors or to change the parameters that are used in the model. It provides detailed data available on family history, particularly if the subject is associated with two or more relatives with breast cancer. This type of data would give greater power to discriminate between possible parameters18. This model has been used by several researchers to predict the breast cancer risk in various populations, but inconsistent results have been reported. Boughey et al22 evaluated the performance of the IBIS model for the prediction of 10-year risk of breast cancer in a well-defined 7
ACCEPTED MANUSCRIPT
cohort of 331 women with atypia in the USA population. The results of this study indicated that the model did not accurately distinguish, on an individual level, between women who developed invasive breast cancer and those who did not. Therefore, the authors concluded that the IBIS model significantly overestimated risk of breast cancer for women with atypia. They did not recommend the use of the model to predict 10-year breast cancer risk in women with atypical hyperplasia in the USA. Another study was conducted to examine the accuracy of the IBIS model and BOADICEA
RI PT
models for the 10-year risk prediction in 358 Jewish high-risk women. The 10-year risks assigned by BOADICEA and IBIS ranged from 0.2% to 12.6% and 0.89% to 21.7%, respectively. The authors concluded that the BOADICEA model had a better predictive value and accuracy for estimating the 10-year breast cancer risk than the IBIS model23. A third study was conducted to
SC
estimate the risk of cancer in 5318 Jewish men and women from the Washington, D.C. area and reported that the estimated risk of breast cancer among carriers was 56% (95% CI: 40% to 73%) by
M AN U
the age of 70. There were no significant differences in the risk of breast cancer between carriers of BRCA1 and BRCA2 mutations. The authors concluded that the risk of breast cancer was overestimated, although it fell below previous estimates based on subjects from high-risk families24. Beside its advantages, this model does not seem to be perfect and alternative models and strategies for arriving at the model parameters should be taken into account when predicting breast cancer risk in a given population18.
TE D
This study had a few limitations. First, we used IBIS model to predict the risk of breast cancer on the basis of data from a case-control study rather than a prospective cohort study. This issue may explain part of the low performance of the model. Second, we compared the results of the IBIS model with that of logistic regression analysis. Actually, it was not clear whether the results of
EP
logistic regression overestimated the risk of breast cancer or the results of IBIS model underestimated the risk.
AC C
Conclusion
According to our findings, IBIS model, which is based on extended family history, genetic data and personal risk factors, discriminated cases and non-cases much better than BOADICEA model, which is based on age and family history. On the other hand, the discrimination of the multiple logistic regression model, which is based on family history, genetic information, and several individual risk factors, was better than the both models. Therefore, extending current models to include a larger arrays of familial, genetic, and personal risk factors may help risk models perform better in predicting risk of breast cancer. Clinical practice points 8
•
ACCEPTED MANUSCRIPT
IBIS and BOADICEA have been developed to predict future risk of breast cancer for specific populations of the world.
•
The discrimination of these models may vary in different populations of the world and needs to be tested.
•
Models that have been based on larger arrays of risk factors may have better performance in predicting risk of breast cancer.
RI PT
Acknowledgments
This was part of the MSc thesis in Epidemiology. We would like to appreciate the Vice-Chancellor for Research and Technology of the Hamadan University of Medical Sciences for supporting this
SC
work. We also thank Prof. Mohammad Esmaeil Akbari, the head of Cancer Research Center, and his colleagues for their valuable collaborations.
M AN U
Conflict of interest statement
The authors declare that they have no conflicts of interest for this work. Sources of support
The Vice-Chancellor for Research and Technology, the Hamadan University of Medical Sciences,
Reference 1.
TE D
provided the financial resources for this work.
World Health Organization. Breast cancer: prevention and control. 2016; http://www.who.int/cancer/detection/breastcancer/en/, 27 May 2016. World Health Organization. GLOBOCAN 2012: estimated cancer incidence, mortality and
EP
2.
prevalence worldwide in 2012. Geneva: WHO; 2012. Colditz GA, Rosner B. Cumulative risk of breast cancer to age 70 years according to risk
AC C
3.
factor status: data from the Nurses' Health Study. Am J Epidemiol. 2000;152(10):950-964. 4.
Malone KE, Daling JR, Thompson JD, O'Brien CA, Francisco LV, Ostrander EA. BRCA1 mutations and breast cancer in the general population: analyses in women before age 35 years and in women before age 45 years with first-degree family history. JAMA. 1998;279(12):922-929.
5.
Ford D, Easton DF, Stratton M, et al. Genetic heterogeneity and penetrance analysis of the BRCA1 and BRCA2 genes in breast cancer families. The Breast Cancer Linkage Consortium. Am J Hum Genet. 1998;62(3):676-689.
9
6.
ACCEPTED MANUSCRIPT
Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and breastfeeding: collaborative reanalysis of individual data from 47 epidemiological studies in 30 countries, including 50302 women with breast cancer and 96973 women without the disease. Lancet. 2002;360(9328):187-195.
7.
Yeo W, Lee HM, Chan A, et al. Risk factors and natural history of breast cancer in younger
8.
RI PT
Chinese women. World J Clin Oncol. 2014;5(5):1097-1106. Key TJ, Verkasalo PK, Banks E. Epidemiology of breast cancer. Lancet Oncol. 2001;2(3):133-140. 9.
Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormonal
SC
contraceptives: collaborative reanalysis of individual data on 53 297 women with breast cancer and 100 239 women without breast cancer from 54 epidemiological studies. Lancet.
10.
M AN U
1996;347(9017):1713-1727.
Yager JD, Davidson NE. Estrogen carcinogenesis in breast cancer. N Engl J Med. 2006;354(3):270-282.
11.
Poorolajal J, Nafisi N, Akbari ME, Mahjub H, Esmailnasab N, Babaee E. Breast cancer survival analysis based on immunohistochemistry subtypes (ER/PR/HER2): a retrospective
12.
TE D
cohort study. Arch Iran Med. 2016;19(10):680-686.
Cheraghi Z, Poorolajal J, Hashem T, Esmailnasab N, Doosti Irani A. Effect of body mass index on breast cancer during premenopausal and postmenopausal periods: a meta-analysis. PloS One. 2012;7(12):e51446.
Key TJ, Reeves GK. Alcohol, diet, and risk of breast cancer. BMJ. 2016;353:i2503.
14.
Russo IH. Cigarette smoking and risk of breast cancer in women. Lancet.
EP
13.
15.
AC C
2002;360(9339):1033-1034.
Hutchinson L. Breast cancer: Radiation risk in BRCA carriers. Nat Rev Clin Oncol. 2012;9(11):611.
16.
Cunningham AP, Antoniou AC, Easton DF. Clinical software development for the Web: lessons learned from the BOADICEA project. BMC Med Inform Decis Mak. Apr 10 2012;12:30.
17.
Lee AJ, Cunningham AP, Kuchenbaecker KB, Mavaddat N, Easton DF, Antoniou AC. BOADICEA breast cancer risk prediction model: updates to cancer incidences, tumour pathology and web interface. Br J Cancer. Jan 21 2014;110(2):535-545. 10
18.
ACCEPTED MANUSCRIPT
Tyrer J, Duffy SW, Cuzick J. A breast cancer prediction model incorporating familial and personal risk factors. Stat Med. 2004;23(7):1111-1130.
19.
Cuzick J. IBIS Breast Cancer Risk Evaluation Tool. 2016; http://www.emstrials.org/riskevaluator/, 27 Jun 2016.
20.
World Health Organization. International Classification of Diseases for Oncology: ICD-O-
21.
Bevers TB, Anderson BO, Bonaccio E, et al. Breast cancer screening and diagnosis. J Natl Compr Canc Netw. 2009;7(10):1060-1096.
22.
RI PT
3. 3rd ed. Geneva: WHO; 2000.
Boughey JC, Hartmann LC, Anderson SS, et al. Evaluation of the Tyrer-Cuzick
SC
(International Breast Cancer Intervention Study) model for breast cancer risk prediction in women with atypical hyperplasia. J Clin Oncol. 2010;28(22):3591-3596. Laitman Y, Simeonov M, Keinan-Boker L, Liphshitz I, Friedman E. Breast cancer risk
M AN U
23.
prediction accuracy in Jewish Israeli high-risk women using the BOADICEA and IBIS risk models. Genet Res (Camb). 2013;95(6):174-177.
Struewing JP, Hartge P, Wacholder S, et al. The risk of cancer associated with specific mutations of BRCA1 and BRCA2 among Ashkenazi Jews. N Engl J Med.
EP
TE D
1997;336(20):1401-1408.
AC C
24.
11
ACCEPTED MANUSCRIPT
Table 1: Characteristics of breast cancer cases versus population-based and clinic-based controls
29.4 43.6 27.0
27 87 202 100 42 28
5.6 17.9 41.6 20.6 8.6 5.8
124 168 83 46
29.4 39.9 19.7 10.9
240 266
47.4 52.6
EP
AC C
a
RI PT
147 218 135
396 110
78.3 21.7
457 49
90.3 9.7
400 106
79.1 20.9
498 8
98.4 1.6
6 2 498
1.2 0.4 98.4
Hormone Replacement Therapy
12
Clinic-based control n=211 n % P value 0.001 76 36.0 65 30.8 52 24.7 18 8.5 0.022 81 38.4 90 42.6 40 19.0 0.001 15 7.6 55 27.9 48 24.4 42 21.3 19 9.7 18 9.1 0.713 55 33.5 58 35.4 34 20.7 17 10.4 0.001 134 63.5 77 36.5 0.001 56 26.5 155 73.5 0.035 179 84.8 32 15.2 0.416 161 76.3 50 23.7 0.225 210 99.5 1 0.5 0.010 0 0.0 0 0.0 211 100.0
SC
22.5 32.0 29.6 15.8
Population-based controls n=916 n % P value 0.001 383 41.8 328 35.8 152 16.6 53 5.8 0.003 347 38.0 371 41.0 196 21.4 0.001 72 8.0 208 23.1 262 29.2 185 20.6 98 10.9 72 8.0 0.017 171 23.7 264 36.6 188 26.0 98 13.6 0.001 695 75.9 221 24.1 0.001 858 93.7 58 6.3 0.001 880 96.1 36 3.9 0.001 815 89.0 101 11.0 0.096 910 99.3 6 0.7 0.001 13 1.4 0 0.0 903 98.6
M AN U
114 162 150 80
TE D
Variables Age group (year) 24-39 40-49 50-59 60-75 Body mass index (kg/m2) <24.9 25.0-29.9 ≥30.0 Age at menarche (year) ≤11 12 13 14 15 ≥16 Age at first birth (year) <20 20-24 25-29 ≥30 Menopausal status Premenopausal Postmenopausal History of a benign breast disease No Yes Mother or sister with breast cancer No Yes Other relatives with breast cancer No Yes HRTa (estrogen or progestin) No Yes BRCA genes Negative Positive Unchecked
Breast cancer cases n=506 n %
ACCEPTED MANUSCRIPT
Table 2: Predicted risk of breast cancer based on IBIS model versus logistic regression model for population-based controls, clinic-based controls and breast cancer cases 1-year risk 0.15 0.32 0.27
IBIS predicted risk % 10-year risk Life-time risk 1.95 13.06 4.06 22.12 3.17 14.42
BOADICEA Life-time risk % 9.94 9.72 9.06
Logistic Risk prediction % 30.58 53.71 44.63
AC C
EP
TE D
M AN U
SC
RI PT
Groups Population-based controls Clinic-based controls Breast cancer cases
13
ACCEPTED MANUSCRIPT
Figures title
Figure 1: Receiver operating characteristic (ROC) plot for risk of breast cancer predicted by logistic regression model. IBIS model, BOADICEA model Figure 2: Scatter plot based on the IBIS risk prediction model versus logistic risk prediction model (Pierson correlation coefficient = 0.1639)
AC C
EP
TE D
M AN U
SC
RI PT
Figure 3: Scatter plot based on the IBIS risk prediction model versus BOADICEA risk prediction model (Pierson correlation coefficient = 0.3884)
14
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
S1526-8209(17)30010-1
DOI:
10.1016/j.clbc.2017.01.003
Reference:
CLBC 570
To appear in:
Clinical Breast Cancer
Received Date: 5 July 2016 Revised Date:
24 December 2016
Accepted Date: 9 January 2017
Please cite this article as: Ghoncheh M, Ziaee F, Karami M, Poorolajal J, Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population, Clinical Breast Cancer (2017), doi: 10.1016/j.clbc.2017.01.003. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Original article
Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population
RI PT
Mahshid Ghoncheh (MSc) Department of Epidemiology, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Cancer Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran Email: [email protected] Fatane Ziaee (MD) Cancer Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran Email: [email protected]
SC
Manoochehr Karami (PhD) Department of Epidemiology, School of Public Health, Social Determinants of Health Research Center, Hamadan University of Medical Sciences, Hamadan, Iran Email: [email protected]
M AN U
Jalal Poorolajal (MD, PhD): Corresponding author Department of Epidemiology, School of Public Health, Noncommunicable Diseases Research Center, Hamadan University of Medical Sciences, Hamadan, Iran Email: [email protected] Zip code: 65157838695 Tel: +98 81 38380090 Fax: +98 81 38380509
Word count Abstract: 249 Main text: 1822
TE D
Running title IBIS and BOADICEA models for predicting breast cancer risk
AC C
EP
Author Contributions • Mahshid Ghoncheh contributed to study conception and design, acquisition of data, analysis and interpretation of data, and critical revision. • Fatane Ziaee contributed to acquisition of data and critical revision. • Manoochehr Karami contributed to study design and critical revision. • Jalal Poorolajal contributed to study conception and design, analysis and interpretation of data, and drafting of manuscript.
1
ACCEPTED MANUSCRIPT
Micro abstract
The validity of the IBIS model was evaluated for predicting breast cancer risk in Iranian population. We performed a case-control study, enrolling 1633 Iranian women. We indicated that the discrimination of cases and non-cases based on IBIS model was better than BOADICEA model for
AC C
EP
TE D
M AN U
SC
RI PT
Iranian population, although the discrimination of the both models was relatively low.
2
ACCEPTED MANUSCRIPT
Original article
Validating the IBIS and BOADICEA models for predicting breast cancer risk in Iranian population Abstract Background: Several approaches have been suggested for incorporating risk factors to predict the
RI PT
future risk of breast cancer. The Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) and International Breast Cancer Intervention Study (IBIS) are among these approaches. We compared the performance of these models in discriminate between cases and non-case in the Iranian population.
SC
Patients and Methods: We performed a case-control study in Tehran, from November 2015 to April 2016, and enrolled 1633 women aged 24 to 75 years, including 506 cases of breast cancer,
M AN U
916 population-based controls, and 211 clinic-based controls. We calculated and compared the risk of breast cancer predicted by the IBIS and BOADICEA models and the logistic regression model. For model discrimination, we computed the area under the Receiver Operating Characteristic (ROC).
Results: The risk of breast cancer predicted by IBIS model was higher than BOADICEA model, but lower than logistic model. The area under the ROC plots indicated that the logistic regression
TE D
model showed better discrimination between cases and non-cases (71.53%) compared to IBIS model (49.36%) and BOADICEA model (35.84%). Based on the Pierson correlation coefficient, the correlation between IBIS and BOADICEA models was much stronger than the correlation between
EP
IBIS and logistic models (0.3884 and 0.1639, respectively). Conclusion: The IBIS model discriminated cases and non-cases better than BOADICEA model in
AC C
the Iranian population. However, the discrimination of the logistic regression model, which included a larger array of familial, genetic, and personal risk factors, was better than the two models.
Keywords: Breast Neoplasms; Risk Assessment; Statistical Models; Case-Control Studies
3
ACCEPTED MANUSCRIPT
Introduction
By far, breast cancer is the most common invasive cancer diagnosed in women both in the developed and the developing countries1. Breast cancer incidence varies greatly worldwide. Although it is supposed to be a common cancer in the developed countries, about 58% of all breast cancer deaths occur in less developed countries2. Several modifiable and non-modifiable factors are associated with breast cancer, including older
RI PT
age3, genetics4,5, late age at first childbirth and lack of breastfeeding6, early age at menarche7, late menopause8, oral contraceptives and hormone replacement therapy9-11, obesity12, alcohol13, smoking14, and radiation15.
SC
Several studies have evaluated the association between subsets of genetic and individual factors and the incidence risk of breast cancer. However, there is a need to incorporate these risk factors and predict the overall risk of developing breast cancer. Different approaches have been suggested for
M AN U
incorporating risk factors to predict the risk of breast cancer. The Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) is a breast cancer risk prediction model which was developed by Cunningham
16
and was described by Lee et al
17
.
BOADICEA is presented as a web-based computer program that is used to estimate the future risks of developing breast or ovarian cancer in women according to their family history. Another
TE D
approach is International Breast Cancer Intervention Study (IBIS) which was introduced by Tyrer et al18. They provided this tool for research purposes only and did not recommend it for clinical decisions yet19. This model is adapted for UK and Sweden populations only. However, it is not clear how these models work in other populations in the world. In this study, we employed the IBIS
population.
AC C
Methods
EP
and BOADICEA models to predict the risk of developing breast cancer among Iranian women
We carried out this case-control study in Tehran, the capital city of Iran, from November 2015 to April 2016. We enrolled 1633 native Iranian women aged 24 to 75 years, including 506 women with confirmed breast cancer as the case group, 916 women without known breast cancer from the general population as the population-based control group, and 211 women without known breast cancer who attended to the Cancer Research Center for other reasons as the clinic-based control group. The Research Council of Hamadan University of Medical Sciences approved this study. Since we did not carry out any intervention, we just took participants' implicit (verbal) consent to participate in the study and answer to our questions
4
ACCEPTED MANUSCRIPT
We selected cases among patients with breast cancer, newly diagnosed pathologically based on International Classification of Diseases for Oncology 3rd edition sites C50.0–C50.9)20 and registered with the Cancer Research Center database, affiliated with Shahid Beheshti University of Medical Sciences. Cases were attended to this center from different parts of the city, including hospitals, laboratories, and clinics. We excluded patients with unknown pathology or those who out-migrated or died. We selected clinic-based controls among individuals who attended to the
RI PT
same center for breast diseases other than breast cancer or suspected to have breast cancer, but eventually ruled out pathologically. We extracted data from the medical records and obtained additional data, not included in the medical records, through interviews with cases and controls after describing the objectives of the study for them and taking their verbal consent.
SC
We selected population-based controls from the general population of Tehran and its suburbs. In order to increase the generalizability of the results, we divided the city into five regions, including
M AN U
north, south, east, west, and center. Then, we selected a probability sample of the population from each region through the door to door approach.
The data collection tool was a questionnaire included age, weight, height, age at menarche, age at first delivery, number of children, menopausal status, and presence or absence of a benign breast disease, history of breast cancer in the first-degree relatives (mother or sister) or other relatives, history of hormone replacement therapy (HRT) including estrogen and/or progestin, and presence
TE D
or absence of the BRCA genes.
We used IBIS model to predict the risk of developing breast cancer. This computer-based program provides a woman's overall risk of breast cancer by incorporating genetic determinants such as the
EP
BRCA1 and BRCA2 genes, details about a developing breast and/or ovarian cancers among family members, and personal risk factors such as age, body mass index, age at menarche, parity, age at
AC C
first child, menopausal status, age at menopause, and benign breast diseases18. We used BOADICEA model to calculate the risks of breast cancer in women based on their family history. We run BOADICEA risk calculations using the latest version of BOADICEA web application (BWA v3).
We performed a multiple logistic regression analysis to estimate the risk of breast cancer as the third model. For this purpose, we generated a set of breast cancer risk factors (age, body mass index, early menarche (<13 years), the presence or absence of a benign breast disease, and a history of breast cancer in the first-degree or other relatives) based on the information obtained from cases and controls. Then, we combined risk factors with information on the odds ratio of each risk factor
5
ACCEPTED MANUSCRIPT
and calculated the observed probability of breast cancer for an individual with a given set of characteristics as follows: P(୷|୶) =
ଵ
Equ. 1
ଵାୣష(ౘబ శౘ ౮ )
Where, 'P' is the calculated probability of breast cancer for an individual. 'b0' is the intercept, that is, the estimated value of P when x = 0. 'bi' is the regression coefficient, that is, the estimated increase
RI PT
in the breast cancer risk per unit increase in the predictor variable (x). According to this equation, we modelled and combined the risks of breast cancer and calculated the individual probability of breast cancer. Then, we combined the individual risk and obtained the overall risk of breast cancer for cases and controls.
SC
We considered observed cases and non-cases as the gold standard. Then, we compared the risk of breast cancer predicted by IBIS and BOADICEA models and logistic regression analysis with the
M AN U
gold standard to determine which model could better discriminate between cases and non-cases on the basis of Receiver Operating Characteristic (ROC) area and plot. We also estimated the correlation between models using Pierson correlation coefficient.
We used to chi-square test for analysis of nominal variables. We performed all statistical analyses at a significance level of 0.05 using Stata software, version 11 (StataCorp, College Station, TX, USA). We also employed IBIS Breast Cancer Risk Evaluation Tool version 7 for predicting the breast
TE D
cancer risk.
In order to measure the validity (sensitivity and specificity) of the IBIS model, we estimated the breast cancer risk for each individual using IBIS model. Then, we set a risk threshold of 20% and
EP
considered subjects with a lifetime risk of greater than 20% for developing breast cancer positive and otherwise negative21. Finally, we estimated sensitivity and specificity of the IBIS model against
AC C
gold standard that was the cases of breast cancer confirmed pathologically and the population-based and clinic-based controls. Results
In this study, 506 cases of breast cancer were compared with 916 population-based controls and 211 clinical-based controls. The mean (SD) age of the cases was 48.37 (10.79) years and that of population-based and clinic-based controls was 42.37 (9.84) and 44.31 (10.85) years, respectively. The characteristics of the case and control groups are given in Table 1. There was a statistically significant association between breast cancer and age, body mass index (BMI), age at menarche, age at first birth, menopausal status, a positive history of a benign breast disease, a positive history of breast cancer in the first-degree or other relatives, and presence of the BRCA genes. 6
ACCEPTED MANUSCRIPT
The predicted risk of breast cancer based on the IBIS, BOADICEA, and logistic regression models is given in Table 2. There was a significant difference between the risk of breast cancer predicted by the three models for both population-based and clinic-based controls and cases of breast cancer. According to out result, the risk of breast cancer predicted by BOADICEA model was lower than that of IBIS model. On the other hand, the risk of breast cancer predicted by logistic model was much higher than that of IBIS model.
RI PT
We computed the ROC area to compare the discrimination of the IBIS, BOADICEA, and logistic models, (Figure 1). The area under ROC curve was 71.53% for the logistic model, 49.36% for the IBIS model, and 35.84% for the BOADICEA model. According to these results, the logistic regression model discriminated between cases and non-cases better than IBIS and BOADICEA
SC
models.
The scatter plot of the risk of breast cancer predicted by IBIS against the risk of breast cancer
M AN U
predicted by logistic model and BOADICEA model is shown in Figure 2 and 3, respectively. The correlation between IBIS model and BOADICEA model was much stronger than the correlation between IBIS model and logistic model. As shown in Figure 2, there was the moderate correlation between IBIS and BOADICEA models (Pierson correlation coefficient = 0.3884). However, as shown in Figure 3, the correlation between IBIS and logistic models was poor (Pierson correlation
Discussion
TE D
coefficient = 0.1639).
Our results indicated that the IBIS model, which include family history and genetic data, could discriminate cases and non-cases of breast cancer better than BOADICEA model, which is based on
EP
only family history. However, the performance of the multiple logistic regression model was better than the two models. The reason was that, logistic model included family history, genetic
AC C
information, and individual risk factors. Therefore, the larger arrays of familial, genetic, and personal risk factors the risk models have, the more performance they would have. The IBIS model was actually developed to predict individual risk of breast cancer from both familial and personal risk factors. It is an easy-to-use computed-based program. It is possible to add new factors or to change the parameters that are used in the model. It provides detailed data available on family history, particularly if the subject is associated with two or more relatives with breast cancer. This type of data would give greater power to discriminate between possible parameters18. This model has been used by several researchers to predict the breast cancer risk in various populations, but inconsistent results have been reported. Boughey et al22 evaluated the performance of the IBIS model for the prediction of 10-year risk of breast cancer in a well-defined 7
ACCEPTED MANUSCRIPT
cohort of 331 women with atypia in the USA population. The results of this study indicated that the model did not accurately distinguish, on an individual level, between women who developed invasive breast cancer and those who did not. Therefore, the authors concluded that the IBIS model significantly overestimated risk of breast cancer for women with atypia. They did not recommend the use of the model to predict 10-year breast cancer risk in women with atypical hyperplasia in the USA. Another study was conducted to examine the accuracy of the IBIS model and BOADICEA
RI PT
models for the 10-year risk prediction in 358 Jewish high-risk women. The 10-year risks assigned by BOADICEA and IBIS ranged from 0.2% to 12.6% and 0.89% to 21.7%, respectively. The authors concluded that the BOADICEA model had a better predictive value and accuracy for estimating the 10-year breast cancer risk than the IBIS model23. A third study was conducted to
SC
estimate the risk of cancer in 5318 Jewish men and women from the Washington, D.C. area and reported that the estimated risk of breast cancer among carriers was 56% (95% CI: 40% to 73%) by
M AN U
the age of 70. There were no significant differences in the risk of breast cancer between carriers of BRCA1 and BRCA2 mutations. The authors concluded that the risk of breast cancer was overestimated, although it fell below previous estimates based on subjects from high-risk families24. Beside its advantages, this model does not seem to be perfect and alternative models and strategies for arriving at the model parameters should be taken into account when predicting breast cancer risk in a given population18.
TE D
This study had a few limitations. First, we used IBIS model to predict the risk of breast cancer on the basis of data from a case-control study rather than a prospective cohort study. This issue may explain part of the low performance of the model. Second, we compared the results of the IBIS model with that of logistic regression analysis. Actually, it was not clear whether the results of
EP
logistic regression overestimated the risk of breast cancer or the results of IBIS model underestimated the risk.
AC C
Conclusion
According to our findings, IBIS model, which is based on extended family history, genetic data and personal risk factors, discriminated cases and non-cases much better than BOADICEA model, which is based on age and family history. On the other hand, the discrimination of the multiple logistic regression model, which is based on family history, genetic information, and several individual risk factors, was better than the both models. Therefore, extending current models to include a larger arrays of familial, genetic, and personal risk factors may help risk models perform better in predicting risk of breast cancer. Clinical practice points 8
•
ACCEPTED MANUSCRIPT
IBIS and BOADICEA have been developed to predict future risk of breast cancer for specific populations of the world.
•
The discrimination of these models may vary in different populations of the world and needs to be tested.
•
Models that have been based on larger arrays of risk factors may have better performance in predicting risk of breast cancer.
RI PT
Acknowledgments
This was part of the MSc thesis in Epidemiology. We would like to appreciate the Vice-Chancellor for Research and Technology of the Hamadan University of Medical Sciences for supporting this
SC
work. We also thank Prof. Mohammad Esmaeil Akbari, the head of Cancer Research Center, and his colleagues for their valuable collaborations.
M AN U
Conflict of interest statement
The authors declare that they have no conflicts of interest for this work. Sources of support
The Vice-Chancellor for Research and Technology, the Hamadan University of Medical Sciences,
Reference 1.
TE D
provided the financial resources for this work.
World Health Organization. Breast cancer: prevention and control. 2016; http://www.who.int/cancer/detection/breastcancer/en/, 27 May 2016. World Health Organization. GLOBOCAN 2012: estimated cancer incidence, mortality and
EP
2.
prevalence worldwide in 2012. Geneva: WHO; 2012. Colditz GA, Rosner B. Cumulative risk of breast cancer to age 70 years according to risk
AC C
3.
factor status: data from the Nurses' Health Study. Am J Epidemiol. 2000;152(10):950-964. 4.
Malone KE, Daling JR, Thompson JD, O'Brien CA, Francisco LV, Ostrander EA. BRCA1 mutations and breast cancer in the general population: analyses in women before age 35 years and in women before age 45 years with first-degree family history. JAMA. 1998;279(12):922-929.
5.
Ford D, Easton DF, Stratton M, et al. Genetic heterogeneity and penetrance analysis of the BRCA1 and BRCA2 genes in breast cancer families. The Breast Cancer Linkage Consortium. Am J Hum Genet. 1998;62(3):676-689.
9
6.
ACCEPTED MANUSCRIPT
Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and breastfeeding: collaborative reanalysis of individual data from 47 epidemiological studies in 30 countries, including 50302 women with breast cancer and 96973 women without the disease. Lancet. 2002;360(9328):187-195.
7.
Yeo W, Lee HM, Chan A, et al. Risk factors and natural history of breast cancer in younger
8.
RI PT
Chinese women. World J Clin Oncol. 2014;5(5):1097-1106. Key TJ, Verkasalo PK, Banks E. Epidemiology of breast cancer. Lancet Oncol. 2001;2(3):133-140. 9.
Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormonal
SC
contraceptives: collaborative reanalysis of individual data on 53 297 women with breast cancer and 100 239 women without breast cancer from 54 epidemiological studies. Lancet.
10.
M AN U
1996;347(9017):1713-1727.
Yager JD, Davidson NE. Estrogen carcinogenesis in breast cancer. N Engl J Med. 2006;354(3):270-282.
11.
Poorolajal J, Nafisi N, Akbari ME, Mahjub H, Esmailnasab N, Babaee E. Breast cancer survival analysis based on immunohistochemistry subtypes (ER/PR/HER2): a retrospective
12.
TE D
cohort study. Arch Iran Med. 2016;19(10):680-686.
Cheraghi Z, Poorolajal J, Hashem T, Esmailnasab N, Doosti Irani A. Effect of body mass index on breast cancer during premenopausal and postmenopausal periods: a meta-analysis. PloS One. 2012;7(12):e51446.
Key TJ, Reeves GK. Alcohol, diet, and risk of breast cancer. BMJ. 2016;353:i2503.
14.
Russo IH. Cigarette smoking and risk of breast cancer in women. Lancet.
EP
13.
15.
AC C
2002;360(9339):1033-1034.
Hutchinson L. Breast cancer: Radiation risk in BRCA carriers. Nat Rev Clin Oncol. 2012;9(11):611.
16.
Cunningham AP, Antoniou AC, Easton DF. Clinical software development for the Web: lessons learned from the BOADICEA project. BMC Med Inform Decis Mak. Apr 10 2012;12:30.
17.
Lee AJ, Cunningham AP, Kuchenbaecker KB, Mavaddat N, Easton DF, Antoniou AC. BOADICEA breast cancer risk prediction model: updates to cancer incidences, tumour pathology and web interface. Br J Cancer. Jan 21 2014;110(2):535-545. 10
18.
ACCEPTED MANUSCRIPT
Tyrer J, Duffy SW, Cuzick J. A breast cancer prediction model incorporating familial and personal risk factors. Stat Med. 2004;23(7):1111-1130.
19.
Cuzick J. IBIS Breast Cancer Risk Evaluation Tool. 2016; http://www.emstrials.org/riskevaluator/, 27 Jun 2016.
20.
World Health Organization. International Classification of Diseases for Oncology: ICD-O-
21.
Bevers TB, Anderson BO, Bonaccio E, et al. Breast cancer screening and diagnosis. J Natl Compr Canc Netw. 2009;7(10):1060-1096.
22.
RI PT
3. 3rd ed. Geneva: WHO; 2000.
Boughey JC, Hartmann LC, Anderson SS, et al. Evaluation of the Tyrer-Cuzick
SC
(International Breast Cancer Intervention Study) model for breast cancer risk prediction in women with atypical hyperplasia. J Clin Oncol. 2010;28(22):3591-3596. Laitman Y, Simeonov M, Keinan-Boker L, Liphshitz I, Friedman E. Breast cancer risk
M AN U
23.
prediction accuracy in Jewish Israeli high-risk women using the BOADICEA and IBIS risk models. Genet Res (Camb). 2013;95(6):174-177.
Struewing JP, Hartge P, Wacholder S, et al. The risk of cancer associated with specific mutations of BRCA1 and BRCA2 among Ashkenazi Jews. N Engl J Med.
EP
TE D
1997;336(20):1401-1408.
AC C
24.
11
ACCEPTED MANUSCRIPT
Table 1: Characteristics of breast cancer cases versus population-based and clinic-based controls
29.4 43.6 27.0
27 87 202 100 42 28
5.6 17.9 41.6 20.6 8.6 5.8
124 168 83 46
29.4 39.9 19.7 10.9
240 266
47.4 52.6
EP
AC C
a
RI PT
147 218 135
396 110
78.3 21.7
457 49
90.3 9.7
400 106
79.1 20.9
498 8
98.4 1.6
6 2 498
1.2 0.4 98.4
Hormone Replacement Therapy
12
Clinic-based control n=211 n % P value 0.001 76 36.0 65 30.8 52 24.7 18 8.5 0.022 81 38.4 90 42.6 40 19.0 0.001 15 7.6 55 27.9 48 24.4 42 21.3 19 9.7 18 9.1 0.713 55 33.5 58 35.4 34 20.7 17 10.4 0.001 134 63.5 77 36.5 0.001 56 26.5 155 73.5 0.035 179 84.8 32 15.2 0.416 161 76.3 50 23.7 0.225 210 99.5 1 0.5 0.010 0 0.0 0 0.0 211 100.0
SC
22.5 32.0 29.6 15.8
Population-based controls n=916 n % P value 0.001 383 41.8 328 35.8 152 16.6 53 5.8 0.003 347 38.0 371 41.0 196 21.4 0.001 72 8.0 208 23.1 262 29.2 185 20.6 98 10.9 72 8.0 0.017 171 23.7 264 36.6 188 26.0 98 13.6 0.001 695 75.9 221 24.1 0.001 858 93.7 58 6.3 0.001 880 96.1 36 3.9 0.001 815 89.0 101 11.0 0.096 910 99.3 6 0.7 0.001 13 1.4 0 0.0 903 98.6
M AN U
114 162 150 80
TE D
Variables Age group (year) 24-39 40-49 50-59 60-75 Body mass index (kg/m2) <24.9 25.0-29.9 ≥30.0 Age at menarche (year) ≤11 12 13 14 15 ≥16 Age at first birth (year) <20 20-24 25-29 ≥30 Menopausal status Premenopausal Postmenopausal History of a benign breast disease No Yes Mother or sister with breast cancer No Yes Other relatives with breast cancer No Yes HRTa (estrogen or progestin) No Yes BRCA genes Negative Positive Unchecked
Breast cancer cases n=506 n %
ACCEPTED MANUSCRIPT
Table 2: Predicted risk of breast cancer based on IBIS model versus logistic regression model for population-based controls, clinic-based controls and breast cancer cases 1-year risk 0.15 0.32 0.27
IBIS predicted risk % 10-year risk Life-time risk 1.95 13.06 4.06 22.12 3.17 14.42
BOADICEA Life-time risk % 9.94 9.72 9.06
Logistic Risk prediction % 30.58 53.71 44.63
AC C
EP
TE D
M AN U
SC
RI PT
Groups Population-based controls Clinic-based controls Breast cancer cases
13
ACCEPTED MANUSCRIPT
Figures title
Figure 1: Receiver operating characteristic (ROC) plot for risk of breast cancer predicted by logistic regression model. IBIS model, BOADICEA model Figure 2: Scatter plot based on the IBIS risk prediction model versus logistic risk prediction model (Pierson correlation coefficient = 0.1639)
AC C
EP
TE D
M AN U
SC
RI PT
Figure 3: Scatter plot based on the IBIS risk prediction model versus BOADICEA risk prediction model (Pierson correlation coefficient = 0.3884)
14
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT