| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Original Article |
1 Department of Pathology, University of Vermont College of Medicine and Vermont Cancer Center, Health Science Complex, 89 Beaumont Avenue, Burlington, Vermont 05405-0068
2 Department of Biostatistics, University of Vermont College of Medicine and Vermont Cancer Center, Hills Building, Burlington, Vermont 05405
3 Department of Health Promotion Research, University of Vermont College of Medicine and Vermont Cancer Center, 1 South Prospect Street, Burlington, Vermont 05405
Correspondence: Address correspondence and reprint requests to: Donald L. Weaver, MD; E-mail: donald.weaver{at}uvm.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Methods: A population-based study was undertaken to develop and test models for predicting the probability of invasive breast cancer and/or ductal carcinoma-in-situ in 7670 women undergoing breast biopsy after mammography. Logistical prediction models were developed by using data from 6129 randomly selected women and tested with data from the remaining women.
Results: The overall cancer prevalence among women undergoing biopsy was 22.4%. Prevalence in women with mammograms highly suggestive of malignancy (category 5) was 84.6%, with minimal variation in individual cancer probabilities due to age. A total of 24.6% of women with suspicious mammograms (category 4) had cancer, but individual probability estimates ranged from .01 to .86, depending on age, presence of a lump, previous biopsy, menopausal status, and use of postmenopausal hormone therapy. These variables also influenced biopsy outcome in women with other mammography assessments (categories 0–3), but the overall prevalence was lower (8.6%), and estimated probabilities ranged from .01 to .45. When cancer was present, the probability of invasive disease was influenced by mammogram assessment category, absence of mammogram calcifications, and presence of a lump.
Conclusions: The probabilities of invasive cancer and ductal carcinoma-in-situ in women undergoing biopsy can be more accurately predicted by using clinical characteristics in addition to mammography findings. This information could potentially influence decisions regarding immediate biopsy or continued surveillance.
Key Words: Breast neoplasm • Invasive carcinoma • Ductal carcinoma-in-situ mammography • Breast imaging • Risk prediction
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Regardless of the underlying reasons for a high benign biopsy rate, a recommendation for breast biopsy induces anxiety and fear of cancer in women. Several studies have found that women who had suspicious mammograms had substantial worries about breast cancer that affected their moods and daily functioning, even after malignancy was excluded.4–6 In one study, women still had a significantly higher level of anxiety about breast cancer 18 months after a false-positive mammogram compared with women whose mammogram was truly negative.5 A better understanding of the likelihood of a diagnosis of cancer in a biopsy after breast imaging may help reduce women’s anxiety.
Many studies have been directed toward identifying risk factors for breast cancer. However, estimating breast cancer risk for an individual woman differs considerably from prediction of biopsy outcome because risk is the probability that a woman will develop breast cancer sometime in the future and not the probability that she has cancer at a particular point in time. The probability that cancer is present at the time of biopsy depends not only on a woman’s risk for the disease, but also on the factors leading to the decision to perform a biopsy, such as symptoms and mammography findings. As a consequence, a woman at high risk for breast cancer can have a biopsy with a low probability of positive results and vice versa.
Perhaps the factor most influencing whether a woman has a biopsy is the mammogram finding. The American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) standardizes reporting by stratifying mammography findings into six major categories: five are hierarchical for risk, and one category is for unresolved imaging results.7 The two categories that routinely lead to biopsy are category 5 (highly suggestive of malignancy) and category 4 (suspicious abnormality). In studies of women undergoing needle localized excision biopsy, the proportion with cancer after category 5 mammograms ranges from 81% to 97%; however, approximately 70% of needle localized excision biopsies follow the less predictive category 4 assessment, for which the cancer rate decreases sharply to 30% to 34%.8,9 Improved prediction of biopsy outcome would be particularly desirable in women undergoing biopsy after a BI-RADS category 4 mammogram.
Factors independent of the mammographic findings but specific to the individual may influence and improve prediction of biopsy outcome. Improved prediction would, it is hoped, help physicians reduce anxiety for many patients. More specific prediction may potentially, over time, modify recommendations for immediate biopsy as physicians and patients better understand the likelihood of a cancer diagnosis within subsets of mammogram classifications, thus allowing continued surveillance when the likelihood of cancer is low. In fact, the newest version of BI-RADS allows the radiologist to subdivide category 4 mammograms into three levels of perceived risk. This study examines the pathology results of women undergoing breast biopsy after mammography in Vermont from 1997 to 2001, as entered into the population-based mammography and pathology registry of the Vermont Breast Cancer Surveillance System (VBCSS).10 The relationship between biopsy diagnosis, mammography BI-RADS assessment, and several clinical characteristics specific to the woman are used to develop and evaluate a model to predict biopsy outcome.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Data Collection, Management, and Quality Assurance
All mammography and pathology information used in the study was obtained from the VBCSS. The VBCSS data collection methods have been described in detail elsewhere.10 Briefly, when a woman receives a mammogram in Vermont, she completes a relevant health history form and is asked for permission to use these data for research. The radiology staff completes the information about the imaging procedures performed and sends both the health history and radiological data to the VBCSS, either on paper or electronically. The following self-reported information from the health history was used in this study: age, education, menopausal status, postmenopausal use of hormone therapy (HT), body mass index (BMI; derived from reported height and weight), prior diagnosis of breast cancer, family history of breast cancer, prior biopsy, prior mammography, and presence of a lump at the time of mammography. Information supplied by the radiology facility included the reason for the mammogram, whether a comparison film was available at the assessment, the woman’s mammographical breast density, the presence of calcifications, and the final BI-RADS assessment.
During the period of this study, the American College of Radiology’s BI-RADS used six assessment categories: 0, "need additional imaging"; 1, "negative"; 2, "benign finding"; 3, "probably benign finding"; 4, "suspicious abnormality"; and 5, "highly suggestive of malignancy".7 If a BI-RADS assessment of 1 to 5 was assigned, it was used as the final assessment for analysis. If a mammogram had a BI-RADS category 0 assessment and additional mammographical views or ultrasound examinations were performed before biopsy, the BI-RADS assessment from the subsequent imaging was used for analysis. If no subsequent imaging was performed before biopsy, the mammogram was considered unresolved and was included in the analysis as a BI-RADS category 0 assessment.
Data from breast pathology performed in Vermont laboratories are collected on standardized forms and submitted to the VBCSS with a copy of the laboratory pathology report. In this study, pathologic outcomes were coded as benign, ductal carcinoma-in-situ (DCIS), or invasive breast cancer. Lobular carcinoma-in-situ was considered a benign biopsy outcome in this study. If both DCIS and invasive disease were present, the outcome was designated as invasive.
Mammographical and pathologic data are entered into a relational database. A quality-control system has been established according to accepted guidelines11 for timeliness, completeness, and accuracy. The Institutional Review Board at the University of Vermont approved the protocol for this project with an alteration of informed consent.
Statistical Analysis
Women were randomly allocated to 2 groups: data from 6129 women (80%) were used to develop a model for predicting biopsy outcome, and data from the remaining 1541 women were used for model validation. The comparability of these model development and validation groups was assessed by 
2 tests. In the model development group, 2 tests were also used to assess bivariate relationships between biopsy outcome and each of the health history and mammographic variables. Logistical regression was used to construct multivariate models for predicting biopsy outcome. Separate analyses were performed to predict the probability of a cancer outcome (either DCIS or invasive cancer) and to predict the probability of invasive disease in women with a cancer outcome. To avoid eliminating all women with missing data for any of the potential predictors, preliminary regressions were run to identify variables that were significantly associated with outcome after adjustment for BI-RADS assessment, which was the strongest predictor of outcome, and to assess interactions. These analyses were also performed to determine whether any response categories should be combined or whether missing responses could be combined with one of the response categories. Significant predictors and interactions from these analyses were then entered into a stepwise logistic regression analysis. In analyses to predict a cancer outcome, significant interaction terms indicated that the effects of all other predictors varied depending on BI-RADS assessment. A separate model was therefore fitted for women with a BI-RADS category 5 assessment to simplify interpretation. Models were assessed by using a modification of Pearson’s 2 goodness-of-fit test, which is appropriate when sample sizes for some cells are small, as was the case in this study.12 Individual biopsy outcome probabilities were estimated by computing the value of the model function for a specified set of predictor values and using the resulting value, f(x), in the following equation:
 |
To validate the models, predicted probabilities were computed for women in the validation group. Women were grouped according to their values for the predictors, and the expected numbers of outcomes in the groups were compared with the observed values by using the goodness-of-fit test described previously, with the appropriate degrees of freedom to reflect that the predicted values were independent of the data.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
25 kg/m2. A history of breast cancer in a first-degree relative (mother, sister, or daughter) was reported by 17.6% of the women, and 30.0% had previously had a nonmalignant biopsy result. Only 8.8% of the women did not have a high school diploma. Although the study included all women who had a mammogram within a year before biopsy, 84.4% of the mammograms were performed within 90 days before biopsy. Most (65.4%) mammograms were screening examinations, and 70.6% of the women had received another mammogram within the preceding 3 years. Of the mammograms preceding biopsy, 61.0% were categorized as either highly suggestive of malignancy (BI-RADS category 5; 8.1%) or suspicious abnormality (BI-RADS category 4; 52.9%), 10.6% were categorized as probably benign (BI-RADS category 3), and 27.1% had a negative or benign assessment (BI-RADS category 1 or 2). The remaining mammograms (1.4%) were incomplete (BI-RADS category 0), thus indicating that the women did not have additional imaging to reach a final assessment before undergoing biopsy. Most women’s biopsies (77.7%) had benign outcomes, whereas 4.8% had DCIS and 17.6% had invasive cancer.
Health history and radiological data for the 6129 women in the model development group and the 1541 women in the validation group are listed in Table 1
. As anticipated, the randomly assigned groups were very similar. They differed significantly only in BMI (P = .010): the validation group had somewhat greater proportions of women in the lowest and highest categories.
|
). Most of the associations are attributable to differences between women with cancer (either DCIS or invasive) and those with benign outcomes. Among the women with cancer, there were fewer differences between those with DCIS versus invasive disease. The prevalence of cancer at the time of biopsy increased with age and family history of breast cancer but decreased with level of education. More postmenopausal women (32.0%) than premenopausal women (12.6%) undergoing biopsy had cancer, and among postmenopausal women, those who had never taken HT had a higher prevalence of cancer (38.9%) than those currently taking HT and those who had previously taken HT (25.9% and 29.4%, respectively). BMI showed a positive association with the prevalence of invasive cancer but not with DCIS. Invasive cancer was less prevalent in women who had previously had benign biopsy results.
|
Biopsy outcome was highly associated with BI-RADS assessment category. Seventy-nine percent of women with a BI-RADS category 5 assessment had invasive cancer, and 5.5% had DCIS. Invasive cancer was much less prevalent in women with a BI-RADS category 4 assessment (16.6%), but DCIS was more prevalent (7.0%). Biopsies after other BI-RADS assessment categories were associated with a lower prevalence of both DCIS and invasive cancer. Invasive cancer was more prevalent among women with BI-RADS categories 1 and 2 (7.9% and 7.8%, respectively) than among those with categories 0 and 3 (1.3% and 5.2%, respectively).
Models to Predict Biopsy Outcome
Multivariate logistic models for estimating the probability that a woman undergoing breast biopsy has cancer (either DCIS or invasive) are listed in Table 3. Separate models were created for women with BI-RADS category 5 and BI-RADS category 0 to 4 assessments because the predictive value of the other variables differed for these two groups of women. Estimates of cancer probabilities obtained from the models are listed in Figs. 1 to 3 and can be calculated from the following equations based on the coefficients listed in Table 3. For a woman with BI-RADS category 5 assessment,
|
|
|
|
 |
For a woman with a BI-RADS assessment of category 1 to 4 or 0/incomplete,
 |
The value of the function is computed by multiplying each of the coefficients by the value of the corresponding factor. Except for age, which is expressed in years, each of the other variables takes on a value of 1 if the patient is positive for that factor and a value of 0 otherwise. The following example is the computation for a 60-year-old postmenopausal woman with a BI-RADS category 4 assessment, no history of HT, a lump, no prior benign biopsy, and calcifications.
 |
 |
The value of f(x) is then used in the following formula to calculate the estimated probability of cancer (p):
 |
The probability of cancer is highest for women undergoing biopsy after a BI-RADS category 5 assessment; age was the only variable other than assessment category that was a significant predictor of outcome for these women. The log-odds of cancer were best modeled as a quadratic function of age (Table 3
), which yielded the probability estimates in Fig. 1. The probability of a cancer outcome increases from .78 for a 40-year-old woman to approximately .89 for women aged 56 to 70 years and then decreases with age for older women. Most of the probability estimates in Fig. 1 have standard errors of approximately .02 and 95% confidence intervals of ±.04, but standard errors of prediction for women younger than 45 years are larger (e.g., the probability estimate for a 40-year-old woman has a standard error of .036). The estimated probabilities for women aged <40 years and >80 years are not shown in Fig. 1 because they are based on small numbers of women and may not be reliable.
For women with BI-RADS category 0 to 4 assessments, age, the presence of a lump, menopausal status, HT use, a prior benign biopsy, and a BI-RADS category 4 assessment were jointly predictive of outcome. There were significant interactions between the effects of age, menopausal status, and HT use, as well as between BI-RADS category 4 assessment and the presence of a lump. The model coefficients (Table 3) and the estimated probabilities (Figs. 2 and 3) indicate that for women undergoing biopsy after BI-RADS category 0 to 4 assessments, the probability that cancer is present increases with age, given any fixed set of values for the other predictors in the model. Premenopausal women <50 years of age have a lower probability of a cancer outcome than postmenopausal women of the same age, but after age 50, predicted probabilities are higher in premenopausal women. There were only 12 premenopausal women over the age of 55 in the model development group, so predicted probabilities for these women may not be valid and are therefore not presented in Figs. 2 and 3. Also, for comparability with Fig. 1, only predicted probabilities for women aged 40 to 80 years are shown, but there were sufficient numbers of younger and older women to reliably estimate their outcome probabilities as well. When women of all ages are considered, outcome probabilities are estimated to range from .01 to .86 for women with BI-RADS category 4 assessments and from .01 to .45 for women with BI-RADS category 0 to 3 assessments.
The probability of a cancer outcome is higher for women with BI-RADS category 4 assessments than for comparable women with BI-RADS category 0 to 3 assessments. After adjustment for the other variables in the model, cancer probabilities do not differ significantly among BI-RADS category 0 to 3 assessments. The presence of a lump increases the probability that cancer is present if the women has a BI-RADS category 4 assessment (Fig. 2) but is not predictive of a cancer outcome for biopsies after other BI-RADS assessments. The predicted probabilities for women with BI-RADS category 0 to 3 assessments in Fig. 3 are therefore not presented separately for women with and without lumps. A previous benign biopsy result and a history of HT slightly reduce the chance that cancer is present, regardless of BI-RADS category.
Standard errors of the probability estimates in Figs. 2 and 3 range from .004 to .044 and generally increase with the magnitude of the probability. For estimated cancer probabilities <.20, most standard errors are <.02, and most (87%) of all standard errors are
.03. Thus, the 95% confidence intervals for the predicted probabilities of cancer are within ±.06 for most women.
If cancer is present, the multivariate logistical model in Table 4 predicts the probability that it is invasive cancer. A BI-RADS category 5 assessment or the presence of a lump increases the probability that the cancer is invasive, whereas a mammographic finding of calcifications increases the probability of DCIS, especially when no lump is present. Predicted probabilities of invasive cancer obtained from this model and their associated standard errors are listed in Table 5 for all combinations of predictor values. The probabilities are conditional on cancer being present, so to obtain the unconditional probability that a woman undergoing biopsy has invasive cancer, these conditional probabilities must be multiplied by the probability that a malignancy is present. By using the previous example, if a 60-year-old postmenopausal woman with a BI-RADS category 4 assessment, no history of HT, a lump, no prior benign biopsy, and calcifications undergoes biopsy, there is a .53 chance that she has a malignancy. If she does have a malignancy, then there is a .76 chance that it is invasive and a .24 chance that it is DCIS, so the unconditional probability that she has invasive cancer at the time of biopsy is .53 x .76 = .40 (40%). Similarly, the unconditional probability of a DCIS outcome is .53 x .24 = .13 (13%).
|
|
) provided a good fit to the data from the model development group. The test for the BI-RADS category 5 model compared observed and expected numbers of cancer outcomes in 68 groups of women defined by years of age, and the P value was .859. In an analogous test for the BI-RADS category 0 to 4 models, there were 865 groups of women corresponding to each unique combination of values for the predictor variables, and the P value was .745. For the model to predict invasive cancer conditional on a biopsy outcome of cancer, there are only eight groups, one corresponding to each combination of values for the three dichotomous predictors in the model (Table 5), and the P value for the goodness-of-fit test was .836.
Similar goodness-of-fit tests were performed to ascertain how well the models predicted biopsy outcomes for 1541 women in the validation group, whose data were not used to derive the models. The results indicated that the differences between the observed and expected numbers of cancer outcomes were consistent with random variability. For the two models used to predict a cancer outcome, the P values for the goodness-of-fit tests, based on groups of women with the same values for the predictors, were .382 and .645 for the BI-RADS category 5 and BI-RADS category 0 to 4 models, respectively. For the model to predict invasive cancer conditional on a biopsy outcome of cancer, the observed numbers of invasive cancers were similar to those predicted by the probabilities in Table 5, and the P value for the goodness-of-fit test was .999.
Figure 4 illustrates the ability of the model to differentiate among women with the same BI-RADS category 4 mammographic assessment but with differing likelihoods of a cancer outcome. Women in the validation group with BI-RADS category 4 assessments were grouped according to their predicted probabilities, and the observed average proportion with cancer was plotted versus the average predicted probability within each group. Most of the observed proportions correspond closely to the predicted probabilities, and all of the 95% confidence intervals for the proportions include the average predicted probability for the group. This indicates that none of the discrepancies is larger than would be expected because of chance.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
As in other published studies, BI-RADS category 5 was highly predictive of malignancy: 85% of women had cancer on biopsy.3,8,9,13 In the model for women with category 5 assessments, the only other variable that contributed to outcome prediction was age. The model predicts a very high likelihood (78%–89%) of cancer on biopsy across the entire age range, but the highest probabilities are between ages 56 and 70 years. The model for category 5 mammograms would be unlikely to be clinically useful because of the high and relatively consistent cancer outcome within this assessment category. However, most biopsies were performed to evaluate mammographical breast lesions classified as suspicious abnormalities (BI-RADS category 4). In contrast to BI-RADS category 5, the overall cancer prevalence within category 4 was much lower (26%). As this study demonstrates, factors other than intrinsic mammographical features influence the likelihood of cancer on biopsy when suspicious abnormalities are present, and they can be used to substantially improve the prediction of a cancer outcome within category 4. Age is the most significant predictor of biopsy outcome: the likelihood of cancer increases as a woman ages. Furthermore, for any given age, the cancer probability is significantly influenced by the presence or absence of a lump, a previous biopsy, menopausal status, and any use of postmenopausal HT. Although a complex set of variables and algorithms is used to establish the mammogram assessment category, the preceding variables were determined to be useful in predicting the biopsy outcome after a commitment to biopsy was established. The model based on these variables indicates that probabilities of cancer for differing subgroups of women with BI-RADS category 4 assessments vary from 1% to 86%, and this permits more accurate prediction than the overall cancer yield of 26%.
The models presented in this article also permit the calculation of separate outcome probabilities for DCIS and invasive cancer. For example, although a premenopausal 55-year-old woman with a category 4 mammogram, no lump, a prior benign breast biopsy, and calcifications on the current mammogram has a 25% risk of cancer on biopsy (Fig. 2), the likelihood of invasive carcinoma is only 9% (.25 x .35; Table 5). Initially, patients may not distinguish between DCIS and invasive cancer. However, once health care providers explain that the two diagnoses have vastly different clinical and biological implications, specific prediction estimates may be educational and reduce anxiety in women with a moderate or high likelihood of having cancer but a lower likelihood of invasive disease.
None of the available demographic variables except age was predictive of biopsy outcome after adjustment for BI-RADS assessment, and this suggests that population differences may not substantially affect the utility of the models. However, the applicability of the models could potentially be affected by differences in practice patterns regarding mammography and biopsy, particularly radiologists’ use of BI-RADS category 4 and the associated recommendation for biopsy. For example, if radiologists in a specific area or practice tend to assign a BI-RADS category 4 in situations in which radiologists usually assign category 3, the model would overpredict the probability of cancer for category 4 mammograms in that practice. The applicability of the models could theoretically also be influenced by differences in the types of biopsies performed. The models predict the outcome of a diagnostic event (the one or more biopsies used to diagnose a breast problem) rather than an individual biopsy result, so the effect of biopsy type is likely to be small in women with BI-RADS category 4 and 5 assessments unless the choice of procedures affects diagnostic accuracy. For establishing a diagnosis of cancer, core biopsy has been shown to be an accurate alternative to excision biopsy,14,15 and in our data, fine-needle aspiration was used in only approximately 3% of women undergoing biopsy after an abnormal mammogram. However, among women with BI-RADS category 0 to 3 assessments, fine-needle aspiration was more frequent, and this may be an indicator of how often women with apparently benign cysts are referred for biopsy. In this study, 20% of biopsies performed in women with BI-RADS category 0 to 3 assessments were fine-needle aspirations. If this proportion were higher or lower in another population, the model might overpredict or underpredict the probability of cancer in that population. The models were based on all biopsies performed after mammography for a 5-year period and thus reflect the general use of BI-RADS categories by 65 radiologists with diverse training and experience from a wide variety of practice settings that included 1 university-affiliated academic practice. Thus, the models would be expected to perform well over a broad range of settings.
If the results presented in this article are used to change biopsy recommendations, it is difficult to predict exactly how these changes in recommendation might affect model performance. For example, fewer biopsies among women with a low probability of a cancer outcome would increase the overall proportion of biopsies having a cancer diagnosis but would not necessarily alter the predicted probabilities of cancer for women undergoing biopsy. Some women who postpone immediate biopsy will have a biopsy after a subsequent mammogram for either short-term follow-up or regularly scheduled screening. Any effect on the models would depend on the BI-RADS assessment and/or biopsy recommendation for that mammogram. Although theoretical scenarios can be conceived, the effects of minor variations in practice patterns are probably overshadowed by the stronger influence of age on a woman’s probability of a cancer outcome.
Prediction of biopsy outcome in a woman who either does or does not have breast cancer at the time of biopsy is distinctly different from prediction of the future development of cancer based on the presence or absence of known breast cancer risk factors. Because of this, the relationships between traditional risk factors for breast cancer and the likelihood of cancer on a biopsy after imaging were expected to be complex and not necessarily intuitive. The presence of breast cancer risk factors in a particular woman could influence decisions to biopsy in borderline cases, thus leading to oversampling of high-risk groups and a resulting lower cancer yield on biopsy —an effect opposite to what might be predicted purely on the basis of epidemiological risk. Examples of this phenomenon included the influence of HT, educational status, and breast density. Breast density is strongly associated with breast cancer risk, so it is particularly surprising that women with dense breasts were more likely to have a benign biopsy outcome than women with breasts composed entirely of fat or with scattered densities (Table 2). However, dense breasts are more difficult to assess mammographically, so women with dense breasts have a higher biopsy rate. They also tend to be younger and, hence, more likely to have a benign outcome.
The findings from this study and the prediction models presented substantially add to our understanding of the likelihood of cancer on biopsy after breast imaging. When the likelihood of cancer at the time of biopsy is predicted, a breast abnormality is already present and has usually been assigned to a mammographical assessment category. Other clinical factors such as patient age, HT use, previous benign breast biopsy, and presence or absence of a lump or calcifications are usually known but are often underused when counseling the patient and predicting likely biopsy outcome. The models presented in this article provide a more accurate tool for predicting the likelihood of cancer on biopsy after breast imaging, particularly when suspicious abnormalities are present (BI-RADS assessment category 4). The literature contains no other similar data or analysis.
Practicing clinicians and health care professionals should find the curves derived from the prediction models and presented in Figs. 2 and 3 to be most useful. The age-associated prevalence of cancer ranges widely from 1% to 86% after a category 4 mammogram and 1% to 45% after a category 3 mammogram. An individual woman’s overall risk for cancer on biopsy can be determined from the appropriate curve based on prior mammogram assessment, presence or absence of a lump and/or calcifications, menopausal status, use of postmenopausal hormones, and previous biopsy. This probability can then be multiplied by the appropriate factor in the legends to determine the probability of invasive cancer or DCIS. More specific prediction may reduce anxiety, and in situations with a low likelihood of cancer or a low likelihood of invasive cancer, a more informed decision can be made with regard to immediate biopsy, delayed biopsy, or short-interval follow-up with breast imaging.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Received for publication September 10, 2004. Accepted for publication April 3, 2005.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  E. S. Burnside, J. Davis, J. Chhatwal, O. Alagoz, M. J. Lindstrom, B. M. Geller, B. Littenberg, K. A. Shaffer, C. E. Kahn Jr, and C. D. Page Probabilistic Computer Model Developed from Clinical Data in National Mammography Database Format to Classify Mammographic Findings Radiology, June 1, 2009; 251(3): 663 - 672. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W. E. Barlow, E. White, R. Ballard-Barbash, P. M. Vacek, L. Titus-Ernstoff, P. A. Carney, J. A. Tice, D. S. M. Buist, B. M. Geller, R. Rosenberg, et al. Prospective breast cancer risk prediction model for women undergoing screening mammography. J Natl Cancer Inst, September 6, 2006; 98(17): 1204 - 1214. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
