Model Predicts Risk of Breast and Gynecologic Cancers
Model Predicts Risk of Breast and Gynecologic Cancers
Using easy-to-obtain risk factors for breast, ovarian, and endometrial cancers, researchers have come up with models that can predict an individual woman’s absolute risk for developing each type of cancer. While risk models for breast and ovarian cancers have been previously developed, this is the first risk model for endometrial cancer.
Still, further development of more encompassing models or additional models for other races are needed, as the current study, published in PLOS Medicine, used data solely from white women over the age of 50. The current models are also not for use to predict risk of women with BRCA mutations or those with hereditary nonpolyposis colorectal cancer (HNPCC), state the authors.
Ruth Pfeiffer, PhD, statistician and senior investigator at the National Cancer Institute (NCI) in Bethesda, Maryland, and colleagues compiled data from two large population-based cohort studies—the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) and the National Institutes of Health–AARP Diet and Health Study (NIH-AARP). The authors combined estimated relative risks for cancer development with incidence of the cancers in the US population using Surveillance, Epidemiology, and End Results (SEER) data from the NIH. The current analysis included more than 42,000 women from the PLCO study, enrolled between 1993 and 2001, and more than 114,000 women from the NIH-AARP study, enrolled between 1995 and 1996. For each cancer type, a distinct patient cohort from each study was analyzed.
“We would like to test these models in other large cohorts to validate this tool in diverse populations and expand the models to other races and ethnicities as data become available,” said Pfeiffer.
The Nurses’ Healthy Study, which began in 1976, was used as a validation for the cancer risk model, using cohorts of more than 37,000 participants for each of the three women's cancers. The new models were also compared to results using the Breast Cancer Risk Assessment Tool developed by the NCI using the Nurses’ Health Study cohorts.
“We developed statistical models to predict risk for breast, ovarian, and endometrial cancers since these three cancer types share several hormonal and epidemiologic risk factors,” said Pfeiffer.
All three models included some overlapping and some different predictive factors based on previous studies that have shown that both the same risk factors are at play for breast as well as gynecologic cancers.
The breast cancer model includes parity (number of times a woman has given birth); estrogen, progestin, and other menopausal hormonal therapy; age at first live birth; menopausal status; age at menopause; family history of breast and ovarian cancer; as well as history of alcohol consumption, any benign breast disease, and body mass index (BMI).
“We included body mass index in our breast and endometrial models because BMI is an indicator of obesity and is associated with increased risk of both breast and endometrial cancer,” said Pfeiffer. “With rates of obesity increasing and rates of hysterectomy declining in many regions of the US, endometrial cancer incidence may rise further, and thus it is important to identify women at highest risk for this disease.”
The endometrial cancer model included age at menopause, BMI, menopause status, smoking history, use of oral contraceptives, and use of menopausal hormonal therapy.
Ten-year breast cancer risk ranged from 1.5% to 22%, 0.4% to 10.5% for endometrial cancer, and 0.3% to 0.96% for ovarian cancer.
While the breast and ovarian cancer absolute risk models predicted well the risk of women in the Nurses’ Health Study cohort, the endometrial cancer model tended to overpredict the risk of endometrial cancer for the Nurses’ Health Study cohort due to substantially lower rates of this cancer in that group of women compared to the incidence rates in the SEER database.
Models that help to predict the absolute risk of developing a cancer can help to facilitate both clinical trials that aim to test chemopreventive measures on a population at high risk for cancer and to guide decisions for individuals at a higher risk for a specific cancer. Fore example, using an absolute risk model can help a clinician make a decision to run further tests on a patient who is a high risk and presents with ambiguous symptoms that could be suggestive of cancer. Further studies using other cohorts are needed to better assess the utility of the endometrial cancer model for predicting risk of the cancer among white women in the United States.
The utility of an endometrial cancer risk model, stated the authors, could be to design intervention trials to prevent the cancer and identify the specific subpopulation of women who would benefit from such treatment.
According to Pfeiffer, risk prediction models can help clinicians in counseling patients and aid in chemopreventive and treatment decisions. “This is particularly useful because many of these options have both beneficial and harmful side effects,” said Pfeiffer.
The current study incorporates risk factors such as smoking, alcohol consumption, use of menopausal hormone therapy, and body mass index into an absolute risk model that could better influence women to modify their individual risk of breast and gynecologic cancers.
Limitations of the current models include missing data on the number of biopsies women had in both the PLCO and the NIH-AARP studies that is a predictor of breast cancer in the BCRAT risk model. The frequency of menopausal hormonal therapy may be higher in the cohort analyzed compared to its use in today’s population.
Additionally, it will be important to add data on non-white women as well as those under the age of 50 to broaden the prediction of ovarian, breast, and endometrial cancer for women in the United States.
In an accompanying editorial, Lars Holmberg, PhD, professor of cancer epidemiology at the School of Medicine in King’s College London, London, and Andrew Vickers, PhD, in the department of epidemiology and statistics at the Memorial Sloan-Kettering Cancer Center in New York, highlight that improving models for cancer risk are useful for clinical practice but also outline the challenges for such models: the need for improved models, accurate interpretation, and application in the clinic.
The goal for cancer risk models is to lead to better clinical decisions, but how to evaluate whether a model actually improves decision-making is difficult, write Holmberg and Vickers.