Figure 1: MetaWorks systematic review process
 |
|
|
The Agency for Healthcare Research and Quality (AHRQ, formerly the Agency for Health Care Policy and Research, AHCPR), through its Evidence-based Practice Centers (EPCs), sponsors the development of evidence reports and technology assessments to assist public- and private-sector organizations in their efforts to improve the quality of health care in the United States. The reports and assessments provide organizations with comprehensive, science-based information on common, costly medical conditions and new health care technologies. The EPCs systematically review the relevant scientific literature on topics assigned to them by AHRQ and conduct additional analyses when appropriate prior to developing their reports and assessments.
To bring the broadest range of experts into the development of evidence reports and health technology assessments, AHRQ encourages the EPCs to form partnerships and enter into collaborations with other medical and research organizations. The EPCs work with these partner organizations to ensure that the evidence reports and technology assessments they produce will become building blocks for health care quality improvement projects throughout the Nation. The reports undergo peer review prior to their release.
AHRQ expects that the EPC evidence reports and technology assessments will inform individual health plans, providers, and purchasers as well as the health care system as a whole by providing important information to help improve health care quality.
We welcome written comments on this evidence report. They may be sent to: Director, Center for Practice and Technology Assessment, Agency for Healthcare Research and Quality, 6010 Executive Blvd., Suite 300, Rockville, MD 20852.
| Robert Graham, M.D. | John M. Eisenberg, M.D. |
| Director, Center for Practice | Director, Agency for Healthcare |
| And Technology Assessment | Research and Quality |
| The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality or the U.S. Department of Health and Human Services of a particular drug, device, test, treatment, or other clinical service. |
We are grateful to the following individuals for their diverse contributions throughout the course of developing this report: Elenie Chadbourne, Rebecca Cintron, Janet Connelly, Matthew Guldin, Veronica Ludensky, Luba Nalysnyk, Christopher Schmid, Sandy Schwartz, Marya Zilberberg, members of the TEP, peer reviewers, representatives from Kaiser, and representatives from AHRQ.
Over 170,000 women are diagnosed with breast cancer yearly in the United States, incurring enormous individual and societal costs. The objective of this systematic review is to assess the quantity and quality of published evidence regarding specific current issues of diagnosis and management of women with breast disease.
Literature published from January 1, 1994 to September 15, 1999 was searched using Medline and Current Contents® databases. These searches were supplemented by manually reviewing bibliographies of all accepted studies and review articles.
Accepted studies included observational studies, randomized controlled trials (RCTs), non-randomized controlled trials (nRCTs), and uncontrolled case series (UCSs). All accepted publications were required to address the prospectively identified areas of interest and have a total sample size of at least 10 patients. Only English language studies were accepted.
Pertinent data were extracted from accepted studies by one researcher, and reviewed by a second. Studies were evaluated for quality and level of evidence. The data were summarized and synthesized qualitatively. A panel of multidisciplinary clinical experts provided recommendations throughout the project.
The database includes 51 studies (30,178 patients) regarding breast symptoms and risk factors, 20 studies (3,501 patients) pertinent to lobular carcinoma in situ (LCIS) or atypical hyperplasia (AH), and 39 studies (5,900 patients) regarding sentinel node biopsy.
Incomplete reporting of outcomes in patients with different risk factors precludes assessment of the relationship between risk factors, breast symptoms, and cancer incidence.
When managed by observation alone, LCIS was associated with a 4.2-9.3 percent incidence of cancer within 5 years and 7.7-26.3 percent incidence in studies that followed patients for at least 5 years. AH was associated with a 3.7-19.3 percent incidence of cancer within 5 years, and 13.6-33.6 percent incidence in longer-term studies. Selective estrogen receptor modulator (SERM) therapy with Tamoxifen markedly decreased the incidence of breast cancer following a diagnosis of LCIS or AH. After excisional biopsy was done, approximately half of the atypical ductal hyperplasias (ADHs) diagnosed by stereotactic core needle biopsy (SCBX) were changed, usually to a more worrisome diagnosis.
Regardless of tumor location, size, or history of breast surgery, sentinel node biopsy had a false negative rate of 1.9 to 3.3 percent. Sentinel nodes were positive for metastatic disease in approximately one third of cases.
The best available evidence suggests that breast symptoms are evaluated initially by clinical breast exam and imaging study, with supplemental studies when the diagnosis is unclear. There is no evidence to support modifying the work-up of breast symptoms or mammographic abnormalities based on risk factors other than age. Strong evidence supports the necessity of performing excisional biopsy following SCBX diagnosis of ADH, as excisional biopsy often leads to a change in diagnosis. Preliminary evidence strongly suggests that Tamoxifen therapy markedly decreases the incidence of cancer following a diagnosis of LCIS or AH, but it is associated with increased risk of endometrial cancer, thromboembolic disease, and other complications. While studies to date strongly suggest that sentinel node biopsy is successful in most breast cancer patients, long-term cancer outcomes and survival data are required before sentinel node biopsy could be considered the standard of care.
This document is in the public domain and may be used and reprinted without permission except those copyrighted materials noted for which further reproduction is prohibited without the specific permission of copyright holders.
Levine C, Armstrong K, Chopra S, et al. Diagnosis and management of specific breast abnormalities. Evidence Report/Technology Assessment No. 33 (Prepared by MetaWorks, Inc., Boston, MA under Contract No. 290-97-0016). AHRQ Publication No. 01-E046. Rockville, MD: Agency for Healthcare Research and Quality. September 2001.
Each year, more than 170,000 women are diagnosed with breast cancer in the United States alone, and it is the most common form of cancer found in women worldwide. The average lifetime risk of breast cancer in a female infant born in the United States in the year 2000 is 12 percent, or one in eight. The most common cause of malpractice litigation is the missed or delayed diagnosis of breast cancer. Clearly, breast disease is a major concern for women, and it has a substantial effect on both individual and societal health care resources.
When a woman presents to her health care provider with a breast symptom, the initial management will nearly always include, at a minimum, a clinical breast exam (CBE) and mammogram or ultrasound, depending on the age of the patient. Controversies abound concerning the appropriate steps after the initial imaging study. It is unclear whether further management should be influenced by patients' risk factors for breast cancer. Other areas of controversy include management of abnormal mammograms, management of certain pathological diagnoses, and whether all women with breast cancer require a full axillary lymph node dissection (ALND).
These are but a few of the numerous unresolved issues regarding the diagnosis and management of breast disease. In an effort to answer some of these questions, the Agency for Healthcare Research and Quality (AHRQ) directed MetaWorks, Inc., Boston, MA, to undertake a systematic review of the English-language literature since 1994 pertinent to women with breast disease. This topic was nominated for review by Kaiser Permanente.
This review focuses on women with breast signs, symptoms, or mammographic abnormalities for whom specific risk factors for breast cancer are reported. Additional areas of interest include specific biopsy findings and initial management after breast cancer diagnosis.
The MetaWorks investigators have assembled an evidence base that should be useful to health care providers in developing evidence-based strategies to guide breast disease management. It also will be useful to investigators planning new clinical trials and others making regulatory decisions. This synthesis of the best available and most recent evidence is intended to serve as an information resource for local decisionmakers and developers of practice guidelines and recommendations. It focuses attention on gaps in the literature and areas that warrant future research.
This report presents the results of a systematic review of published studies of adult patients who were evaluated for breast symptoms. The following key questions guided this review:
What are the recommendations for evaluation of breast symptoms, mammographic findings and other suspicious findings based on menstrual status, use of hormone replacement therapy (HRT), pregnancy, age, and family history?
How are lobular carcinoma in situ (LCIS) and atypical hyperplasia (AH) managed?
How are nonpalpable lesions and calcifications managed?
What are the indications for sentinel node biopsy?
What costs are associated with diagnosis and management of breast disease as outlined above?
MetaWorks investigators applied methods derived from the evolving science of review research. The review followed a work plan that had been developed and shared with AHRQ, Kaiser Permanente, and a technical expert panel (TEP) that included gynecologists, family practitioners, internists, medical and surgical oncologists, radiologists, and a consumer representative who is a breast cancer survivor.
The work plan outlined the methods to be used for the literature search, study eligibility criteria, data elements for extraction, and methodological strategies employed both to minimize bias and maximize precision during the process of data collection and synthesis.
The published English-language literature was searched from January 1, 1994 to September 15, 1999, and the retrieval cutoff date was March 15, 2000. Different Medline search strategies were employed for each question.
Question 1 (management of symptomatic breast disease and suspicious findings): "breast neoplasms" AND "diagnosis. "
Question 2 (management of LCIS and AH): "breast neoplasms" AND ("diagnosis" OR "pathology" AND ("carcinoma in situ" OR "carcinoma, infiltrating duct" OR "carcinoma, intraductal, noninfiltrating" OR "carcinoma, lobular" OR "hyperplasia".
Question 3 (management of nonpalpable lesions and calcifications): "breast neoplasm" AND ("diagnosis" OR "pathology" OR "radiology" AND "mammography" OR "nonpalpable" OR "calcifications" OR "microcalcifications".
Question 4 (indications for sentinel node biopsy): "breast neoplasm" AND "lymph nodes" AND "pathology. "
Question 5 (costs): "breast neoplasms" AND ("costs" OR "cost analysis" OR "economic").
The Current Contents CD-ROM database was searched (breast cancer and diagnosis) to the same cutoff date. These electronic searches were supplemented by a manual search of the reference lists of all accepted articles, review articles from 1999, and relevant Internet sites.
Abstracts were screened using predefined exclusion criteria, then full papers were reviewed and assessed for fit with inclusion criteria. Only English-language studies involving a minimum of 10 patients were eligible for inclusion. Only randomized controlled trials (RCTs), nonrandomized controlled trials, uncontrolled case series (UCSs), and observational studies were accepted.
After initial screening of abstracts, it became clear that the inclusion criteria needed to be refined in order to develop evidence that addressed the specific questions that had been posed. Questions 1 and 3 are actually facets of the same question; namely, what is the management of patients with risk factors for breast cancer who present with abnormal clinical or mammographic findings? Thus, the refined inclusion criteria for questions 1 and 3 included:
Reporting of risk factors (age, menstrual status, pregnancy history, HRT use, or family history).
Description of suspicious findings (palpable lesion, nipple discharge, or mammographic findings).
Diagnosis of cancer at time of presentation and/or subsequently.
The inclusion criteria for question 2 included:
Diagnosis of LCIS and/or AH.
Management options (observation, excisional biopsy, magnetic resonance imaging [MRI], selective estrogen receptor modulator [SERM] therapy, bilateral mastectomy).
Diagnosis of cancer at time of presentation and/or subsequently.
Initially, the review of AH was limited to atypical lobular hyperplasia (ALH). All studies that involved only atypical ductal hyperplasia (ADH) were rejected. Because of the scant literature available regarding ALH, the focus of this review was expanded to include both ALH and ADH.
Inclusion criteria for question 4 included:
Indication for sentinel node biopsy (tumor size, tumor location, absence of palpable axillary nodes, no history of breast surgery).
Method of sentinel node identification (vital blue dye, radiocolloid mapping, or both).
Results of biopsy and comparison with gold standard (axillary node dissection).
The only criteria for acceptance for question 5 were:
Acceptance of study for one of the other questions.
Discussion of costs, in U.S. dollars.
Studies that involved only screening populations were rejected because the questions for this review were focused on patients with clinical or mammographic abnormalities in addition to risk factors. Populations of interest would not be represented in screening studies, which by definition consist of asymptomatic patients.
For questions 1 and 3, studies that involved only cancer patients were rejected because the questions for this review aimed to determine the incidence of cancer in patients with specific findings, not to determine the prevalence of specific findings in cancer patients.
For question 2, studies of patients who had cancer concurrently with LCIS or AH were rejected, as it would be impossible to determine whether these patients' outcomes were related to their cancer or their LCIS/AH.
Rejection of studies with cancer populations obviously did not apply to question 4, as sentinel node biopsy would be done only in patients with a diagnosis of cancer.
Relevant data from all accepted studies were entered onto data extraction forms (DEFs) designed specifically for this project. Results that required extrapolation from graphs or derivations from figures were not captured due to concerns about accuracy. All data elements were extracted by one investigator and reviewed by a second investigator; 100 hundred percent agreement between the two reviewers was required prior to entry of data elements into the database. At least one physician reviewed all data elements extracted from every study.
All accepted studies were evaluated for quality by using the previously published methods of Level of Evidence and the Jadad Quality Score Assessment.
The elements extracted from each study varied, depending on the question. In general, the information captured included date of publication, location and type of study, primary objective of study, number of patients with various risk factors and clinical or mammographic findings, management of the abnormal findings, number of patients diagnosed with cancer, and any statistical measures reported. No quantitative analyses were performed beyond descriptive statistics to summarize findings.
A group of 11 peer reviewers, drawn from consumer groups and professional organizations, was assembled to review and provide suggestions for the draft final report of this project. Their comments, in addition to those of the TEP, were incorporated into the final report.
109 studies (k) plus 11 kinship studies.
39,560 patients (n).
Study designs: interventional 69 (UCS [k=66], nRCT [k=2], RCT [k=1]); observational 40 (retrospective [k=33], prospective [k=6], cross-sectional [k=1]).
Study location: North America (k=69), Europe (k=33), other (k=7).
Risk factors were commonly reported, but age was the only risk factor consistently reported in association with symptoms and cancer diagnosis.
There was no evidence to support modifying the workup based on risk factors other than age.
The only age-related modification reported was ultrasound for younger women (specific age not reported) and mammogram for older women.
Within 5 years after LCIS diagnosis, 4.2-9.3 percent of patients were diagnosed with breast cancer (k=5, n=1,014 [of which 53 were diagnosed with cancer], overall incidence = 5.2 percent). In studies that followed patients for greater than 5 years, the incidence of cancer was 7.7-26.3 percent (k=3, n=421 [of which 77 were diagnosed with cancer], overall incidence = 22.3 percent).
Within 5 years after AH diagnosis, 3.7-19.3 percent of patients were reported to have developed breast cancer (k=3, n=752 [of which 48 were diagnosed with cancer], overall incidence = 6.4 percent). In studies that followed patients for greater than 5 years, the incidence of cancer was 13.6-33.6 percent (k=3, n=425 [of which 83 were diagnosed with cancer], overall incidence = 19.5 percent).
When the diagnosis of ADH was made by stereotactic core biopsy (SCBX), it generally was followed by open surgical biopsy to confirm the diagnosis and rule out concurrent carcinoma, LCIS, or ductal carcinoma in situ (DCIS). Forty-two percent of the ADH diagnoses were changed as a result of excisional biopsy. Most changes were to DCIS or invasive cancer, although approximately 20 percent changed to more benign diagnoses. Only one study reported the incidence of diagnosis change after ALH (in one of four patients, the diagnosis of ALH was changed to LCIS).
Although data are available from only one study of 13,175 patients, those patients with LCIS or AH who received tamoxifen therapy had a markedly decreased subsequent incidence of cancer, compared with patients who were treated by observation alone (1.9 percent vs. 4.4 percent in LCIS, 0.5 percent vs. 3.7 percent in AH). However, the risks associated with tamoxifen therapy must be considered, including an increased risk of endometrial cancer and thromboembolic disease.
Sentinel node biopsy had a false negative rate of 1.9-3.3 percent.
While both methods of sentinel node detection were effective, the combination of vital blue dye and radiocolloid mapping was more sensitive than either method alone.
Successful sentinel node detection was independent of tumor location, tumor size, and history of breast surgery.
Approximately one-third of sentinel lymph nodes were positive for metastatic disease. These patients required full axillary lymph node dissection (ALND) to determine the extent of metastatic disease. The remaining two-thirds of patients, however, could have been spared this invasive procedure at a cost of missing metastatic disease in 2 to 3 percent of women with cancer.
Long-term studies regarding the efficacy and safety of sentinel node biopsy are lacking.
Only six of the accepted studies addressed cost.
Information was too disparate to draw any conclusions regarding the actual costs of various interventions and/or the long-term cost savings resulting from their use.
Additional research is needed to more thoroughly examine risk factors and breast symptoms and how they relate to cancer diagnoses. Standardization of reporting results is essential, particularly with regard to reporting numbers of patients and not just numbers of lesions. Investigators should report baseline risk factors, presenting symptoms, and followup data pertaining to which patients developed cancer.
Future research also should be undertaken to identify additional or new risk factors. If all of this information were available, a comprehensive assessment of risk could be calculated. This, in turn, could lead to development of a risk model that could be used by doctors and patients to assess breast cancer risk. Such a model could be an adjunct to models (such as the Gail model) that currently are in use, and it could include not only risk factors but also breast symptoms and mammographic findings.
Further studies should be done to confirm and extend the promising initial data that supports prophylactic SERM therapy for patients with AH or LCIS. Although sentinel node biopsy reportedly is effective in most patients, future studies should attempt to identify differences in sensitivities based on tumor size, location, and history of breast surgeries. Additionally, variations in success rates may depend on the experience of the surgeon performing the procedure and the extent of the pathological investigation. These factors should be addressed in future studies. Long-term cancer outcomes and survival data are required before sentinel node biopsy can be recommended for breast cancer patients.
The management of breast disease is changing, with important new developments in genetic susceptibility and promising new imaging techniques. These changes will have a major impact on the diagnosis and management of breast disease in the future.
Breast cancer is the most common cancer found in women worldwide (Ford, Marcus, and Lum, 1999). Each year, over 170,000 women are diagnosed with breast cancer in the United States alone (Armstrong, Eisen, and Weber, 2000). The average lifetime risk of breast cancer in a female baby born in the United States in the year 2000 is 12 percent, or one in eight. The most common cause of malpractice litigation is the missed or delayed diagnosis of breast cancer (Schootman, Myers-Geadelmann, and Fuortes, 2000). Clearly, breast disease places a major burden on both individual and societal healthcare resources.
When a woman presents to her health care provider with a breast symptom, the initial management will nearly always include, at a minimum, a clinical breast exam (CBE) and mammogram or ultrasound, depending on the age of the patient. Controversies abound concerning the steps after the initial imaging study. It is unclear whether further management should be influenced by patients' risk factors for breast cancer. Further areas of controversy include management of abnormal mammograms, management of certain pathological diagnoses, and whether all women with breast cancer require a full axillary lymph node dissection (ALND).
These are but a few of the numerous unanswered questions regarding the diagnosis and management of breast disease. The purpose of this report is to present the published evidence regarding these issues. In order to harness the highest quality evidence supporting the management of breast disease, MetaWorks has performed a systematic review of the recent literature, consisting of articles published in English from January 1, 1994 to September 15, 1999.
This review of management of breast disease was nominated by Kaiser Permanente, Northern California, and a Task Order was commissioned by the Agency for Healthcare Research and Quality (AHRQ). Evaluation of screening asymptomatic patients, development of clinical practice guidelines or specific clinical recommendations are beyond the scope of this project. This evidence base should, however, be useful to health care providers in developing evidence-based strategies to guide breast disease management. It will be useful also to those planning new clinical trials and making regulatory decisions. Additionally, this evidence base may readily be updated as the literature evolves.
The following questions were posed by AHRQ in this Task Order. An introductory review of the issues underlying each set of questions is provided as a prelude to a presentation of the methods and results of this systematic review.
Questions 1 and 3: What are the recommendations for evaluation of breast symptoms, mammographic findings and other suspicious findings based on menstrual status, hormone replacement therapy (HRT), pregnancy, age, and family history? What is the management of nonpalpable lesions and calcifications?
The focus of these questions was to assess the impact of risk factors upon evaluation of women with breast symptoms or abnormal mammograms. It is well established that certain factors increase a woman's risk of developing breast cancer (Velentgas and Daling, 1994). These include age > 50 [relative risk (RR) 6.5], postmenopausal status, use of HRT (RR 1.0-1.5), nulliparity or late age at first pregnancy (age > 30, RR 1.3-2.2), and family history of breast cancer (first degree relative, RR 1.4-13.6; second degree relative, RR 1.5-1.8) (Armstrong, Eisen, and Weber, 2000). Management decisions cannot be based solely on risk factors, as from 50 to 90 percent of women with breast cancer have no identifiable risk factors for the disease (Askins, 1999; Bruzzi, Green, Byar, et al., 1985; Madigan, Ziegler, Benichous, et al., 1995). The Steering Committee on Clinical Practice Guidelines for the Care and Treatment of Breast Cancer states that when a woman presents with a breast lump or suspicious change in breast texture, her risk factors should be noted, but presence or absence of risk factors should not influence decisions regarding further workup (Margolese, Cantin, and Bouchard, 1998).
The frequently quoted figure of "one in eight" or a 12 percent risk of an individual woman developing breast cancer is a lifetime estimate and assumes the woman will live to age 85 (Ford, Marcus, and Lum, 1999). Age is second only to history of breast cancer as the most important risk factor for development of breast cancer in women in the United States (Overmoyer, 1999). While debate continues concerning the optimal age at which to begin screening mammography, the steady increase of breast cancer incidence with age has led most American professional societies to recommend continuing yearly mammographic screening for as long as a woman remains healthy. (Ford, Marcus, and Lum, 1999; Basset, Hendrick, Bassford, et al., 1994).
Family history is a known risk factor for breast cancer. The degree of risk, however, varies depending upon the age of the patient, the closeness of the relative(s) with breast cancer, the age(s) at which the relative(s) developed breast cancer, the number of relatives with breast cancer, and the number of relatives with other gynecologic or nongynecologic cancers. In the Nurses' Health Study (Colditz, Willett, Hunter, et al., 1993), women whose mother and/or sister had breast cancer before the age of 40 were reported to have the highest risk of developing breast cancer [RR 2.5, 95 percent confidence interval (CI) 1.5-4.2], compared with women with a negative family history. Evaluation of women in the Iowa Women's Health Study (Sellers, Mink, Cerhan, et al., 1997) showed that use of HRT did not significantly impact the risk of breast cancer in women with a positive family history.
The effect of pregnancy on the incidence of breast cancer appears to be influenced by the age at first pregnancy, age at menarche, and age at menopause. Risk of breast cancer increases with duration of estrogen exposure (Ford, Marcus, and Lum, 1999); therefore, women with longer periods of uninterrupted estrogen exposure are at higher risk for developing cancer. This includes women with early menarche (before age 12; RR 1.1-1.3) or late menopause (after age 55; RR 2.0). Nulliparous women and women who deliver their first child after the age of 30 also have increased risk of subsequent breast cancer (RR 1.3-1.9). The impact of pregnancy is not completely understood, as the short-term risk of breast cancer rises for up to 10 years after delivery (Ford, Marcus, and Lum, 1999; Lambe, Hsieh, Trichopoulos, et al., 1994).
Until recently, the perceived benefits of HRT for most women [decreased risks of osteoporosis (Felson, Zhang, Hannan, et al., 1993), coronary heart disease (Falkeborn, Persson, Adami, et al., 1992; Grady, Rubin, Petitti, et al., 1992; Hong, Romm, Reagan, et al., 1992; Psaty, Heckbert, Atkins, et al., 1994), stroke (Finucane, Madans, Bush, et al., 1993), menopausal symptoms (Hammond, 1996), and hyperlipidemia (Writing Group for the PEPI Trial. 1995)] were believed to outweigh the risks [increased risks of endometrial and breast cancer (Colditz, Hankinson, Hunter, et al., 1995; Grady, Rubin, Petitti, et al., 1992)]. Deciding whether to take HRT was particularly difficult for women with a significant risk of breast cancer, especially if they were also at increased risk for osteoporosis or heart disease. Recent studies, including the Heart and Estrogen/Progestin Replacement Study (HERS) (Hulley, Grady, Bush, et al., 1998) and the Estrogen Replacement and Atherosclerosis (ERA) study (Herrington, 2000; Interventional Cardiology, 2000) have shown an increase in short-term cardiovascular risk for postmenopausal women with preexisting cardiovascular disease who take HRT. These studies have further complicated the risk/benefit profile of HRT and will likely decrease the number of women taking HRT, at least until further studies are done. Current vs. previous use of HRT, unopposed estrogen vs. combination estrogen and progesterone (Stanford, Weiss, Voigt, et al., 1995; Speroff, 1996), and duration of HRT are factors that need to be considered in assessing a woman's risk of developing breast cancer (Schairer, Lubin, Troisi, et al., 2000; Gapstur, Morrow, and Sellers, 1999). The degree to which HRT increases the risk of breast cancer remains controversial.
There are numerous other potential risk factors for breast cancer, such as smoking, diet, alcohol, age of menarche, lactation, and genetic factors. Recent studies have shown that HER-2/neu gene overexpression in benign breast disease is associated with an increased risk of subsequent breast cancer (Stark, Hulka, Joens, et al., 2000). These areas were not listed in the questions posed by AHRQ; therefore, they were not covered in this report. They could, however, be included in an update of this systematic review.
Several models predict an individual woman's risk of breast cancer, the two most common being the Gail model (Gail, Brinton, Byar, et al., 1989), and the Claus model (Claus, Risch, and Thompson, 1994). The Gail model was derived from data from the 280,000 women in the Breast Cancer Detection Demonstration Project (BCDDP). This model uses multivariate logistic regression to combine several risk factors and provide an individualized estimate of relative risk. The risk factors assessed include age, reproductive history, maternal family history, and history of benign breast disease. The Gail model has been validated in a prospective, randomized clinical trial (Costantino, Gail, Pee, et al., 1999); however, opinions vary regarding the utility of this tool (Daly, Lerman, Ross, et al., 1996; Ford, Marcus, and Lum, 1999). It may overestimate the risk associated with benign breast biopsies and underestimate paternal familial risk. Its applicability for minority populations is unconfirmed (Morrow, 2000). A revised model calculates incidence rates for African American and Caucasian women separately but does not allow calculation of risk for other ethnic groups, such as Asian or Hispanic women.
The Claus model, on the contrary, calculates risk based solely on family history. These tools do not provide a complete picture of an individual patient's risk of developing breast cancer. They are used mainly by genetic counselors, to educate their patients (Ford, Marcus, and Lum, 1999). These models, however, do not integrate risk factors with breast symptoms.
The triple test (TT) is a risk model that incorporates breast signs and clinical findings (Vetto, Pommier, Schmidt, et al., 1996). It consists of physical examination, mammography, and fine needle aspiration. A modified triple test (MTT) was developed for younger women, in which ultrasound was substituted for mammogram. When all three elements of the test are concordant (all indicating benign or all indicating malignant), the TT has been shown to have high sensitivity. While the TT and MTT may be useful clinical tools, one limitation is that they do not take risk factors into account.
As the number of women having screening mammography and the sensitivity of mammography have increased, more breast cancers are now detected by screening mammography. Cancers detected by mammographic screening tend to be found at an earlier stage, when they are not yet palpable and are assumed to be more easily curable (Sickles, 2000). However, the limited specificity of mammography leads to false positive mammograms and unnecessary biopsies. As the technology for breast cancer detection continues to improve, it is hoped that this trend toward earlier detection will continue and that specificity will improve.
When a patient presents with a breast abnormality, she will usually have a mammogram as soon as possible; not only to evaluate the area in question, but to look also for synchronous lesions elsewhere in the breasts (Bassett, Hendrick, Bassford, et al., 1994). Mammograms are less sensitive in women under the age of 40, due to the likelihood of denser breast tissue. If a woman under the age of 40 presents with a breast abnormality, depending on her exam and risk factor analysis, the type of imaging performed may be a modality other than mammography, such as ultrasound (Vetto, Pommier, Schmidt et al., 1996).
Radiologists report mammogram results in a variety of ways. The Breast Imaging Reporting and Data System (BI-RADS) terminology was developed for the purpose of standardizing mammogram reports (Liberman, Abramson, Squires, et al., 1998; see Appendix A). However, many studies report mammographic findings in more descriptive terms, such as calcifications, masses, parenchymal distortions, and densities. The degree of suspicion the radiologist reads in the mammographic lesion will highly influence further evaluation of the patient. Recommendations may range from immediate biopsy to conservative management and repeat mammogram at a specified time. One cannot rely solely on the mammogram results, however, as up to 10 to 15 percent of patients with breast cancer will have no visible abnormality on mammogram (Lumachi, Marzola, Zucchetta, et al., 1999). Other imaging modalities, such as digital mammography and magnetic resonance imaging (MRI), have increased the sensitivity of breast imaging. Their role in evaluating breast abnormalities is currently an area of intense interest for researchers (Simonetti, Cossu, Montanaro, et al., 1998).
In summary, our objective was to integrate risk factors with symptoms and/or mammographic findings to inform management decisions.
Question 2: What is the management of lobular carcinoma in situ and atypical hyperplasia?
The incidence of lobular carcinoma in situ (LCIS) is difficult to estimate, because it is generally an incidental finding and is therefore likely to be underdiagnosed (Zurrida, Bartoli, Galimberti, et al., 1996). Increasing use of core needle biopsy has lead to increased frequency of this diagnosis (Gabriel, 1999). The incidence of LCIS in biopsies that are otherwise benign is estimated at 0.5 to 3.8 percent (Lishman and Lakhani, 1999).
Consensus for management of LCIS has changed dramatically over the years. First described in 1941, LCIS was felt to be a premalignant lesion that would progress to invasive lobular carcinoma. Ipsilateral mastectomy was recommended for all patients (Foote and Stewart, 1941; Gump, Dinne, and Schwartz, 1998).
After patients with LCIS were followed for up to 25 years, it became apparent that most women with LCIS did not develop invasive cancer. Patients who did develop cancer did not necessarily develop it at the same site or even in the same breast as the LCIS. The estimated relative risk of LCIS patients (compared with patients without LCIS) developing cancer ranges from 5.9 to 12.0 (Bodian, Perzin, and Lattes, 1996). Most authorities now consider LCIS to be a proliferative disorder that identifies women at risk of developing invasive carcinoma but not a malignant or premalignant process in its own right (Bodian, Perzin, and Lattes, 1996). Some authors favor the term lobular neoplasia, rather than LCIS, to emphasize the fact that progression to malignancy is not inevitable (Carson, Sanchez-Forgach, and Stomper, 1994).
Current strategies for management of LCIS include observation, selective estrogen receptor modulator (SERM) therapy, radiation, ipsilateral mastectomy with contralateral biopsy, or bilateral mastectomy (Gump, Dinne, and Schwartz, 1998). Radiation and ipsilateral mastectomy are not generally recommended for management of LCIS, but in a survey sent to oncologists in 1996, 60 of the 541 respondents reported recommending radiation to their patients, and 180 discussed unilateral mastectomy (Gump, Dinne, and Schwartz, 1998). Given the wide range of potential treatments, it is essential to amass data to enable clinicians to make treatment recommendations based on evidence.
Ductal carcinoma in situ (DCIS), on the contrary, is considered to be a preinvasive lesion and is generally treated with local excision, often followed by breast irradiation (Ernster, Barclay, and Kerlikowske, 2000; White, Levine, Gustafson et al., 1995). Management of DCIS is not a focus of this report.
Atypical hyperplasia (AH) is reported in 3-4 percent of biopsies performed by stereotactic core biopsy (Symmans, Weg, Gross, et al., 1999). Lesions containing AH are often histologically heterogeneous; therefore, sampling error may cause failure to diagnose cancer. Surgical biopsy allows histologic assessment of a larger volume of tissue, thus leading to more cancer diagnoses (Liberman, Cohen, Dershaw, et al., 1995). The prevalence of carcinoma at surgical biopsy after diagnosis of AH at stereotactic core biopsy (SCBX) is estimated at 33-87 percent (Brown, Wall, Christensen, et al., 1998). Due to the high incidence of coexistent carcinoma, it is recommended that excisional biopsy be done after SCBX diagnoses of atypical ductal hyperplasia (ADH) (Bassett, Winchester, Caplan, et al., 1997).
AH is considered to be a risk factor for breast cancer due to its association with increased risk of subsequent cancer in either breast (Moore, Hargett, Hanks, et al., 1997). The overall incidence of breast cancer after a biopsy of AH ranges from 5.1 to 9.8 percent (Fowble, Hanlon, Patchefsky, et al., 1998). Some studies distinguish between ADH and atypical lobular hyperplasia (ALH). When these two entities are considered separately, the RR in AH patients, compared with patients without AH, is reported as 4.7 for ADH and 5.8 for ALH (Fowble, Hanlon, Patchefsky, et al., 1998; Sparano and Mocharnuk, 1999).
Question 4: What are the indications for sentinel node biopsy?
After the diagnosis of breast cancer has been made, there remain many unanswered questions regarding subsequent evaluation and treatment. The status of regional lymph nodes is the most important prognostic factor for recurrence and survival of a patient with breast cancer (Albertini, Lyman, Cox, et al., 1996; Giuliano, Haigh, Brennan, et al., 2000; Veronesi, Paganelli, Galimberti, et al., 1997). Complete axillary lymph node dissection (ALND) is the gold standard for precise staging and prognostication for women with breast cancer (Cody, 1999), and the results are used to guide treatment decisions (Liberman, Sama, Susnik, et al., 1999). ALND is an invasive procedure, however, with significant potential for morbidity, including lymphedema and permanent numbness (Edge and Hurd, 1999). Chronic lymphedema occurs in 6 to 37 percent of patients after ALND (Giuliano, Haigh, Brennan, et al., 2000; McMasters, Tuttle, Carlson, et al., 2000). Chronic lymphedema has not been reported after sentinel lymph node dissection alone (Giuliano, Haigh, Brennan, et al., 2000). If ALND could be avoided in patients who have a low pretest probability of metastatic disease, significant physical, psychological, and monetary savings could result.
The sentinel lymph node (SLN) hypothesis is that the sentinel nodes are the first to drain solid tumors, and that the histologic status of the sentinel nodes is predictive of the status of the regional nodes (Liberman, Cody, Hill, et al., 1999). Sentinel node biopsy was first performed in 1977 to identify the first lymph node draining penile carcinoma (Cabanas, 1977), thereby avoiding deep inguinal node dissection. The procedure was subsequently used successfully to identify sentinel lymph nodes in cutaneous melanoma. It was first performed in breast cancer in 1993. The major unresolved issues concern whether SLN represents a safe alternative to axillary node dissection, how to identify optimal candidates for sentinel node biopsy, and which technique should be used.
It has been generally accepted that the best candidates for SLN biopsy are patients with small, solitary primary tumors (stages T1 or T2, see Appendix B), and clinically negative lymph nodes (Cody, 1999). Prior breast surgery may distort the architecture of the breast, decreasing the sensitivity of the procedure. Presence of a large hematoma or seroma similarly impedes SLN detection (Miner, Shriver, Jaques, et al., 1999). Tumor location is also a factor, because inner quadrant tumors may drain to the internal mammary, rather than the axillary lymph nodes. Metastases to the internal mammary nodes may not be detected by SLN biopsy, but they likewise would not be detected by ALND (Albertini, Lyman, Cox, et al., 1996; Cox, Pendas, Cox, et al., 1998).
The techniques used for sentinel node detection include vital blue dye, radiocolloid mapping, and a combination of dye plus isotope (Liberman, Cody, Hill, et al.,1999). The sensitivity and specificity of SLN dissection vary for different surgeons and improve as surgeons gain experience with the procedure (Gulec, Moffat, Carroll, et al., 1998). Current guidelines suggest that surgeons perform at least 20 to 30 SLN biopsies before they consider performing SLN biopsy without ALND (McMasters, Tuttle, Carlson, et al., 2000).
If SLN biopsy becomes standard of care, the only women who would need to undergo ALND would be those with positive sentinel nodes or unidentified sentinel nodes. Thus, the goal in reviewing the sentinel node literature is to identify which patients are optimal candidates for sentinel node biopsy.
Question 5: What are the costs associated with diagnosis and management of breast disease, as outlined above?
Cost-effectiveness analysis is not straightforward in the diagnosis and management of breast disease. The costs associated with medical consultations, surgical evaluations, mammography, other imaging modalities, and the numerous types of biopsies are widely disparate in different settings. Newer technologies, such as genetic testing, further increase costs. Long-term studies are required to determine whether these interventions are cost-effective in improving outcomes.
MetaWorks investigators used systematic review methods derived from the evolving science of review research (Mulrow, Cook, and Davidoff, 1997; Sacks, Berrier, Reitman, et al., 1987). These methods were generally applied according to standard operating procedures at MetaWorks and are displayed in Figure 1
.
A Task Order, containing the five questions described above, was developed by Kaiser Permanente, submitted to AHRQ, then presented to MetaWorks. From this Task Order, MetaWorks researchers developed a Work Plan, which was then reviewed by AHRQ, Kaiser Permanente, and the Technical Expert Panel (TEP). The work plan outlined the methods to be used for the literature search, study eligibility criteria, data elements for extraction, and methodological strategies to minimize bias and maximize precision during the process of data extraction and synthesis.
After a preliminary literature review, causal pathways relevant to the above five questions were developed. Questions 1 and 3 are facets of the same question; namely, what is the management of patients with risk factors for breast cancer who present with abnormal clinical or mammographic findings? They are therefore included in the same causal pathway. Questions 2 and 4 each have their own causal pathway. Question 5 does not have a pathway; data are derived from questions 1 through 4 that address cost.
These pathways were not designed to be clinical practice guidelines or algorithms for patient care decisions. They were constructed solely for use as guides during the systematic review for this project, and with the expectation that they might change as the project developed.
The published literature was searched from January 1, 1994 to September 15, 1999, using Medline and Current Contents® databases. A manual search was performed of the bibliographies of all publications accepted for inclusion into the evidence base. In addition, the bibliographies of selected review articles published in 1999 were searched for potentially relevant citations. The search cut-off date was September 15, 1999, and the retrieval cut-off date was March 15, 2000. The Medline search included the following search strategy, back to 1994:
For Question 1 (management of symptomatic breast disease and suspicious findings):
explode Medical Subject Heading (MeSH) "breast neoplasms"
AND Topic Subheading (TS) "diagnosis"
AND NOT foreign language
For Question 2 (management of LCIS and AH):
explode MeSH "breast neoplasms" AND TS "diagnosis" OR TS "pathology"
AND explode MeSH "carcinoma in situ" OR
explode MeSH "carcinoma, infiltrating duct" OR
explode MeSH "carcinoma, intraductal, noninfiltrating" OR
explode MeSH "carcinoma, lobular" OR
explode MeSH "hyperplasia"
AND NOT foreign language
For Question 3 (management of nonpalpable lesions and calcifications):
explode MeSH "breast neoplasm" AND (TS "diagnosis" OR TS "pathology" OR TS "radiology") AND explode MeSH "mammography"
OR Textword (TW) "nonpalpable" OR TW "calcifications" OR TW "microcalcifications"
AND NOT foreign language
For Question 4 (indications for sentinel node biopsy):
explode MeSH "lymph nodes" AND TS "pathology"
AND explode MeSH "breast neoplasm" AND TS "pathology"
AND NOT foreign language
For Question 5 (costs):
explode MeSH "breast neoplasms"
AND explode MeSH "costs and cost analysis"
OR MeSH "breast neoplasms" AND TS "economic"
AND NOT foreign language
The search of the Current Contents® CD-ROM database employed the following key word (KW) search terms:
Keyword (KW) breast cancer
AND KW diagnosis
LIMIT to English language
LIMIT to articles
LIMIT to Current Contents® clinical medicine
All citations and abstracts resulting from the above searches in Medline and Current Contents® were downloaded and printed at MetaWorks.
To assist with the development of the evidence base, pertinent articles from the following Internet sites were reviewed: The Cochrane Collaboration (http://www.cochrane.org), National Guidelines Clearinghouse (NGC; http://www.guideline.gov), Medscape (http://www.medscape.com), Oncolink (http://www.oncolink.com), CancerNet (http://cancernet.nci.nih.gov), the American College of Obstetricians and Gynecologists (ACOG; http://www.acog.org) and the American Society of Clinical Oncology (ASCO; http://www.asco.org).
During Level I screening, all abstracts were downloaded, reviewed and evaluated for the following exclusion criteria:
Reviews, meta-analyses, letters, case reports, editorials, and commentaries.
Abstracts.
Pharmacokinetic and pharmacodynamic studies.
Animal or in vitro studies.
Studies written in languages other than English.
Full articles were retrieved for all abstracts passing Level I screening. The articles then underwent Level II screening, which consisted of evaluating the articles for the following inclusion criteria:
Study designs: observational (prospective, retrospective, and cross sectional), or interventional [randomized controlled trials (RCTs), non-randomized controlled trials (non-RCTs), and uncontrolled case series (UCS)].
Adult female patients undergoing diagnosis or management of breast disease.
Studies addressing:
any diagnostic test to establish or support a diagnosis of breast disease in a woman presenting with a breast lump, nipple discharge, thickening, or specific findings on biopsy or mammography OR
any intervention used in the management of breast disease in women presenting with a breast lump, nipple discharge, thickening, or specific findings on biopsy or mammography OR
specific management of LCIS, AH, non-palpable lesions and calcifications OR
sentinel node biopsy OR
Costs associated with diagnosis and/or management of breast disease.
At least 10 patients as total sample size.
Studies where results for one patient population of interest can be separated from results from other populations.
Due to the complexity of the questions and the broad scope of the Level II accepted studies, an additional topic assessment and refinement was prepared in which the causal pathways were used to further refine the inclusion criteria. Level III screening sheets were developed, and studies that had been accepted at Levels I and II underwent Level III screening.
For questions 1 and 3, the inclusion criteria included the reporting of:
Risk factors (age, menopausal status, pregnancy history, HRT use, or family history) AND
Description of suspicious findings (palpable lesions, nipple discharge, or mammographic findings) AND
Diagnosis of cancer made at the time of presentation and/or subsequently.
For question 2, the level III inclusion criteria included the reporting of:
Diagnosis of LCIS and/or AH AND
Management options (observation), MRI, SERM therapy, bilateral mastectomy) AND
Diagnosis of cancer made at the time of presentation and/or subsequently.
For question 4, the level III inclusion criteria included the reporting of:
Indication for sentinel node biopsy (tumor size, tumor location, absence of palpable axillary nodes, no history of breast surgery) AND
Method of sentinel node identification (vital blue dye, radiocolloid mapping, both) AND
Results of biopsy and comparison with gold standard (axillary node dissection).
For question 5, any study that had been accepted for one of the previous questions in which costs were reported in U.S. dollars was accepted.
Studies that consisted entirely of screening populations were rejected, because the questions for this review were focused on patients with clinical or mammographic abnormalities, in addition to risk factors. Populations of interest would not be represented in screening studies, which, by definition, consist of asymptomatic patients. These studies were excluded only when the cohort of patients with abnormal mammograms could not be identified. Studies that included follow-up information regarding asymptomatic patients with abnormalities found on mammography were, therefore, accepted.
For Questions 1 and 3, studies that consisted entirely of cancer populations were rejected, because the questions for this review aimed to determine the incidence of cancer in patients with specific findings, not to determine the prevalence of specific findings in cancer patients.
For Question 2, studies of patients who had cancer concurrently with LCIS or AH were rejected, as it would be impossible to determine whether these patients' outcomes were related to their cancer or their LCIS/AH.
Rejection of studies with cancer populations obviously did not apply to Question 4, as sentinel node biopsy would be done only in patients with a diagnosis of cancer.
After the accepted studies were determined, kinship studies were identified. These were studies in which the same patient population was reported in more than one study. "Parent" studies were assigned, which contained primary data. "Child" studies contained supplemental information, such as follow-up data or additional analyses. Data elements were extracted from the parent studies, and supplemented by information presented in kin studies, when appropriate.
All eligible studies were rated for both quality and level of evidence at the time of data extraction. Two established methods: 1) the Jadad method, and 2) the Level of Evidence method were used.
Key data from each eligible study were extracted by a researcher recording data from original reports onto a data extraction form (DEF), and reviewed by a second researcher checking all DEF fields against the original report. Differences were resolved prior to data entry. In all cases, at least one physician reviewed each study. DEFs were designed in advance and pilot tested on a small sample of eligible studies. The pilot test allowed for necessary edits to the DEF to be made prior to implementation on all studies. Dual review of all data served to reduce error and bias in the data extraction process. The data were then entered into MetaWorks' relational database of clinical studies, MetaHubTM. Key data elements sought for extraction from each study included:
Citation
Publication date
Accrual year(s)
Study duration
Study design
Industry sponsorship (sponsor name or not reported)
Level of evidence
Quality score
Total number of patients analyzed
Geographic location
Questions addressed
Quality of life
Primary objective of study
Mention of cost
Beyond these data elements, the data extraction forms differed for each question. For Questions 1 and 3, the following elements were captured:
Method of detection (clinical, mammographic, or both)
Method of biopsy
Risk factors:
Age
Menopausal status
Family history
Pregnancy history
Use of hormone replacement therapy
Clinical findings:
Patient-detected
Clinician-detected
Palpable mass or thickening
Nipple discharge
Pain
Other
Mammographic results:
Calcifications
Masses
Parenchymal distortion
Densities
BI-RADS categories
MRI results
Ultrasound results
Scintigraphy results
Biopsy results
For each of these categories, the number of patients who presented with the symptom and the number who were diagnosed with cancer were both captured. Odds ratios, relative risks, sensitivity, specificity, and accuracy were also captured, when reported.
For Question 2, the following elements were captured:
LCIS or AH (ADH or ALH)
Method of biopsy
Comparison of initial biopsy results with excisional biopsy results
Management
Observation
Yearly mammogram
MRI
SERM therapy
Radiation
Local excision
Bilateral mastectomy
Ipsilateral mastectomy
Ipsilateral mastectomy with contralateral biopsy
Chemotherapy
For each category, captured elements included the number of patients with LCIS or AH and the number who subsequently developed cancer. Risk factors and time of cancer diagnosis after LCIS/AH diagnosis were also captured when reported.
For Question 4 (indications for sentinel node biopsy), the following data elements were captured:
Method of sentinel node detection
Vital blue dye
Radiocolloid mapping
Both
Number of patients who had axillary node dissection
Criteria for sentinel node biopsy
Tumor size
Metastases
History of breast surgery
Tumor location
Number of patients in whom sentinel node detection was attempted
Number of patients in whom sentinel node was detected
Number of patients with sentinel node(s) positive for metastatic disease
Number of patients with axillary node(s) positive for metastatic disease
Sentinel node biopsy statistics
Accuracy
Sensitivity
Specificity
Positive predictive value (PPV)
Negative predictive value (NPV)
Lymphoscintigraphy (LSG) has been used to define unanticipated patterns of lymphatic drainage in melanoma patients, but has not been helpful in breast cancer patients (Linehan, Hill, Tran, et al., 1999). Information regarding LSG was, therefore, not extracted from papers that were accepted for this report.
For Question 5, information was infrequently available. It was, therefore, not captured on DEFs. The intention was to summarize data for question 5 in this report.
Only clearly reported aggregate results were extracted from studies. Results that would require extrapolations from graphs or derivations from figures were not captured, due to the potential inaccuracy of reading precise results from graphs or diagrams.
Data were entered from the DEFs into MetaHub, MetaWorks' relational database of clinical trials. At the time each DEF was entered, 100 percent of the data elements were checked back against the originals. In addition, a 20 percent random sample of data in the completed database was checked against the DEFs by the quality control (QC) group. Error rates in excess of 2 percent of QC-checked data would have triggered a 100 percent recheck of all data elements entered into the database. However, the 100 percent recheck was not necessary in this case, due to a low error rate.
No statistical analyses were planned beyond basic descriptive statistics used to summarize data. When studies reported summary statistics, they were captured and reported in this summary. Recommendations for supplemental analyses are provided.
Six people made up the TEP, including two medical oncologists, one surgical oncologist, one internist, one radiologist, and one consumer representative. They all received copies of the work plan and its revisions, causal pathways, topic refinements, study listings, data listings, and draft report. When TEP members provided feedback, MetaWorks investigators reviewed their comments, and applied them as deemed appropriate. Additionally, during the course of the project, monthly conference calls were instituted with the topic nominator (Kaiser Permanente), the AHRQ Task Order Officer (TOO), and the coinvestigator from Leonard Davis Institute (LDI). During these conference calls, project updates were provided and issues of concern were addressed.
A group of 11 peer reviewers was assembled to review a draft version of this report. The panel was composed of medical, surgical, and radiation oncologists, a family practitioner, a gynecologist, a pathologist, and a breast cancer patient. All reviewers were asked to complete a list of questions about the format and content of the report and were also invited to provide additional comments in writing. Responses were received from 10 of the 11 peer reviewers and 4 of the 6 TEP members. All of these responses were reviewed and, where appropriate, incorporated into the final report.
In the following results, "k" refers to the number of studies, and "n" refers to the number of patients.
The numbers of abstracts obtained from searches in Medline and Current Contents® are displayed in Figure 2
. The primary search in Medline (search window: 1994-1999) yielded 7,575 abstracts, and the primary search in Current Contents® (search window: 1994-1999) yielded 1,762. When these two groups of citations were merged, 1,147 duplicates were removed resulting in 8,190 unique citations for Level I screening. Bibliographies of all accepted studies were searched for potentially relevant articles. Citations of interest were referenced against a database containing abstracts downloaded from electronic searches of Medline and Current Contents®. All citations identified in the database had already undergone Level I screening at an earlier stage in the project. Citations that were not present in the database were retrieved for Level II and Level III screening.
). Seventy of the 1,015 potentially eligible papers were not available in any of the three major Boston medical libraries and were not retrieved by the retrieval cut-off date.
During Level III screening of full-text papers, 319 were rejected, and 89 additional studies were rejected during data extraction. The final set of studies is composed of 109 primary and 11 linked or "child" studies.
The screening strategies were reviewed a priori with the TEP, TOO, and Kaiser Permanente. After the draft report was submitted to the TEP and peer reviewers, three of the reviewers expressed concern regarding the decision to exclude studies consisting entirely of screening populations. They pointed out that most mammographic abnormalities are found in asymptomatic women, i.e., screening populations. However, papers in which the cohort of patients with abnormal mammograms could be identified and followed were accepted for this project.
With the exception of one study (Fisher, Constantino, Wickerham, et al., 1998), which was graded as Level I evidence, all studies were classified as reporting Level III (scale of I - V) evidence. Quality score can be calculated only for randomized controlled trials (RCTs), therefore only one study (Fisher, Constantino, Wickerham, et al., 1998) was rated for this parameter; its quality score was three out of five possible points.
Two studies reported industry sponsorship, and two authors mentioned relationships with industry but did not indicate if the studies were sponsored by industry.
Questions 1 & 3 Results: What are the recommendations for evaluation of breast symptoms, mammographic findings and other suspicious findings based on menstrual status, hormone replacement therapy (HRT), pregnancy, age, and family history?
Lack of standardization of risk factor reporting prevents definitive answers. Many studies reported presence of risk factors and breast symptoms. Relatively few studies associated the presence of risk factors (other than age) and symptoms with the eventual diagnosis of breast cancer. Furthermore, studies of breast disease suffer from lack of standardized reporting formats. Specifically, studies reported results in terms of numbers of patients or numbers of lesions. Results could not be combined. The evidence shows that most patients who presented with a breast symptom underwent clinical breast examination and imaging study, regardless of their risk factors. Further imaging modalities, type of biopsy, and performance of biopsy by surgeon vs. radiologist were not impacted by risk factor analysis. While risk factors may be recorded during patient encounters, the presence of a breast symptom or mammographic abnormality overrides the existence of any risk factors in the work-up of each individual patient.
Therefore, the evidence does not support modifying the evaluation of breast symptoms based on risk factors other than age. The only age modification that is mentioned is the use of ultrasound, instead of mammography, for younger women.
Fifty of the 51 studies contained only women. One study of 70 patients (Ozdemir, Oznur, Vural, et al., 1997) included two males and did not discriminate between the outcomes of men and women. Thus, the database is composed of 30,176 women and two men.
An initial requirement for accepting a study was that the results must be reported in units of patients rather than numbers of breasts or numbers of lesions. However, it rapidly became apparent that numerous informative studies would be excluded; therefore, this requirement was removed. All studies reported the number of patients enrolled, but when the results were given, they were often given in terms of lesions, and it was not possible to accurately convert these to numbers of patients. Data were captured in the units that were reported. Thirty-three studies reported results in terms of numbers of patients [n=20,230, of which 4,261 (21.1 percent) developed cancer], 17 studies used numbers of lesions [n=9,766 lesions, of which 2,004 (20.5 percent) led to a diagnosis of cancer], and one study reported numbers of episodes of patients reporting to a clinic with a new breast symptom [Barton, Elmore, and Fletcher, 1999; n=539 episodes, of which 24 (4.4 percent) led to a diagnosis of cancer].
The initial plan was to capture the number of patients with each risk factor and each clinical or mammographic result, the number who were diagnosed with breast cancer at the time of presentation, and the number who developed cancer at a later date. During data extraction, it became apparent that the timing of the cancer diagnosis was infrequently reported; therefore, this approach was abandoned. In most studies, the duration of follow-up was not reported. Therefore, the results in this report for questions 1 and 3 include all patients in whom cancer was diagnosed as a result of a particular clinical or mammographic finding, whether the diagnosis was made at the time of presentation or subsequently.
Diagnosis was made by various biopsy methods. The most commonly reported biopsy method was excisional, or open biopsy (k=19, n=2,506). Fourteen studies each reported needle localization biopsy results (n=3,669) and fine needle aspiration [(FNA), n=2,594]. Six studies reported use of SCBX (n=2,145). Thirteen studies did not report the biopsy method (n=12,272). The number of studies and patients listed here do not equal the number of studies and patients in the database, because some studies reported more than one biopsy technique, some studies did not report how many of their patients underwent biopsies, and some studies did not report any biopsy technique.
Very few studies reported risk factors (other than age) in relation to breast abnormalities and cancer incidence.
For any symptom or abnormal finding, patients age 50 or older were much more likely to be diagnosed with cancer than patients younger than 50. When each symptom or mammographic abnormality was considered separately, the numbers of patients stratified by age were too small to make valid generalizations to larger populations.
Menopausal status is not specifically listed in these results. For the purpose of data collection, postmenopausal or perimenopausal women were categorized as age 50 or older, and premenopausal women were categorized as younger than 50. From the papers surveyed, it was usually not possible to distinguish patients with natural vs. surgical menopause.
Forty-three studies reported cancer incidence for patients > 50 vs. < 50. This includes one study (Bianchi, Palli, Ciatto, et al., 1995), in which the age demarcation was > 53 vs. < 53, and one study (Markopoulos, Kakisis, Kouskos, et al., 1999), in which the age demarcation was > 50 vs. < 50. For studies that reported in terms of patients (k=29), 23.6 percent of patients > 50 were diagnosed with cancer, compared with 9.5 percent of those < 50 years of age. For studies that reported in terms of lesions (k=14), 22.6 percent of patients > 50 were diagnosed with cancer, compared with 16.3 percent of those < 50. These totals refer to patients who presented with any clinical finding or mammographic abnormality.
Only two studies stratified the number of patient-detected lesions according to the ages at which the patients presented. In Wakefield and Powis, (1995), eighteen patients > 50 detected a symptom, and none of them were diagnosed with cancer, compared with 82 patients < 50 years of age, three of whom were diagnosed with cancer (3.7 percent). In Barton, Elmore, and Fletcher, (1991), 11 of 172 women in their 40s were diagnosed with cancer (6.4 percent), compared with 6 of 136 women in their 50s (4.4 percent), three of 68 women who presented in their 60s (4.4 percent), and 3 of 36 women who presented in their 70s (8.3 percent). Combining the results from these two studies, 12 of 258 patients > 50 with patient-detected lesions were diagnosed with cancer (4.7 percent), compared with 14 of 254 patients < 50 (5.5 percent).
One study (Sardanelli, Melani, Ottonello, et al., 1998) stratified the number of clinician-detected lesions according to the age of the patients (11 patients > 50 had a clinician-detected lesion, 100 percent were diagnosed with cancer, while 4 patients < 50 had a clinician-detected lesion, 75 percent were diagnosed with cancer).
Eleven studies stratified patients by age when reporting information about palpable masses. Nine of the eleven studies reported cancer incidence; six in relation to numbers of patients and three in relation to numbers of lesions. Of patients with palpable masses, 34.6 percent of those > 50 (n=382) were diagnosed with cancer, compared with 8.7 percent of those < 50 years of age (n=1,338). Of lesions in patients with palpable masses, 54.0 percent in patients > 50 were cancerous (n=137), compared with 26.7 percent of lesions in patients < 50 (n=180).
Nipple discharge was reported by age distinctions in three studies, and only two reported cancer incidence, which was 7.0 percent for patients > 50 and 3.8 percent for patients < 50. One study (Lee, Petrakis, Wrensch, et al., 1994) reported the odds ratios (ORs) of breast cancer risk factors in patients with nipple discharge. For patients > 60 years of age, the OR was 15.6, compared with 2.4 for patients aged 50-54, and 2.2 for patients aged 40-49. No studies characterized patients' ages when reporting cancer diagnoses in women with breast pain.
Five studies reported masses on mammograms according to age. Two studies reported this by number of patients (13 of 13 patients > 50 had cancer, compared with eight of eight patients < 50). Three studies reported this by number of lesions [370 of 888 lesions in patients > 50 were cancerous (41.7 percent), compared with 125 of 879 lesions in patients < 50 (14.2 percent)].
In the studies that reported BI-RADS classification by age group, three studies reported that 19 of 31 patients > 50 with BI-RADS 4 readings had cancer (61.3 percent), compared with 11 of 23 patients < 50 (47.8 percent). Two studies reported that 45 of 48 patients > 50 with BI-RADS 5 readings had cancer (93.8 percent), compared with 33 of 35 patients < 50 (94.3 percent).
Family history was reported in eight studies, but most did not distinguish between first- or second-degree relatives. One study (Brendlinger, Robinson, Sylvest, et al., 1994) reported the number of patients with cancer who had a positive family history, but it did not report the number of patients with a positive family history who did not have cancer. Of the remaining seven studies, 584 of 1,476 patients with a positive family history were diagnosed with cancer (39.6 percent), compared with 1,575 of 4,642 percent of patients with a negative family history (33.9 percent). One study (Byrne, Schairer, Wolfe, et al., 1995) reported that women with a family history of breast cancer in a first-degree relative had an OR of 1.79 [95 percent confidence interval (CI) 1.6-2.1], compared with women with no family history.
Clinical findings were rarely reported in relation to family history. No studies reported patient-detected lesions, clinician-detected lesions, palpable mass, or breast pain with respect to family history. Only one study (Lee, Petrakis, Wrensch, et al., 1994) reported nipple discharge with respect to family history. Of 87 patients with a nipple discharge and positive family history, five were diagnosed with cancer (5.7 percent, OR 1.5), compared with 12 of 316 patients with a nipple discharge and negative family history (3.8 percent). No studies reported mammographic findings in relation to family history.
Only two studies reported parity. Byrne, Schairer, Wolfe, et al., (1995) related risk factors, breast density, and odds ratio adjusted for the percent of breast area with dense appearance on mammogram. Nulliparous women had the highest odds ratio (1.63, CI 1.3-2.1) and the highest proportion of women with high breast density (42 percent). Lee, Petrakis, Wrensch, et al., (1994) reported that seven of 156 nulliparous patients with nipple discharge were diagnosed with breast cancer (4.5 percent), compared with 10 of 252 parous patients with nipple discharge (4.0 percent, OR 0.9).
HRT was frequently mentioned, but only two studies reported the combination of HRT use, abnormal findings, and cancer diagnosis. Harkins, Tartter, Hermann, et al., (1994) reported the number of patients taking and not taking HRT and how many of each developed cancer. The study did not, however, report clinical or mammographic abnormalities of the patients based on their HRT use.
In Harvey (1999), 699 patients took HRT, compared with 434 patients who never used HRT. This study reported the number of patients in each group who presented with palpable masses or pain, but it did not go on to report the number of patients with symptoms who were diagnosed with cancer. It did report cancer incidence in terms of BI-RADS scores; of the 699 patients on HRT, 13 received a BI-RADS score of 4 or 5, and four of these patients (29 percent) were diagnosed with cancer. Of the 434 non-HRT users, 11 received a BI-RADS score of 4 or 5, and seven of these patients (54 percent) were diagnosed with cancer.
These examples are illustrative of the limitations in the database; it is rich with baseline data, but followup information is lacking. Alternatively, followup information is presented without correlation to baseline symptoms or risk factors. In all cases, the small number of patients limits their applicability to the general population.
Only four studies reported a "patient-detected" symptom; of these, only two studies reported the number of patients diagnosed with cancer (two of 167 patients, or 22.2 percent). Three studies reported "clinician-detected" signs, and 50 of 236 patients (21.2 percent) were diagnosed with cancer.
Seventeen studies reported the presence of palpable masses. Of these 17 studies, 11 reported results in terms of patients. One study did not report the number of patients diagnosed with cancer. In the remaining 10 studies, 303 of 2,027 patients (14.9 percent) with palpable masses developed cancer. Six studies reported results in terms of lesions, of which 358 of 1,094 lesions (32.7 percent) were cancerous. Barton, Elmore, Fletcher, et al., (1999) reported that the likelihood ratio of a palpable mass leading to a cancer diagnosis was 65.
Six studies reported the presence of nipple discharge. Four were reported in terms of patients (18 of 570, or 3.2 percent were diagnosed with cancer), and although two were reported in terms of lesions, only one study reported cancer incidence, in 1 of 67 (3.0 percent) of episodes, for a likelihood ratio of 16 (Barton, Elmore, Fletcher, et al., 1999).
Five studies reported breast pain, but only two reported cancer incidence; 7 of 216 (3.2 percent) of patients (Kerin, O'Hanlon, Khalid, et al., 1997) and four of 221 (1.8 percent) of episodes (Barton, Elmore, Fletcher, et al., 1999). Barton, Elmore, Fletcher, et al., (1999) also reported that the likelihood ratio for pain leading to a diagnosis of cancer was 10.
Questions 1 and 3 Results (continued): What is the management of nonpalpable lesions and calcifications?
The data indicate that patients who present with palpable masses are much more likely to be diagnosed with cancer than patients who present with nonpalpable masses, nipple discharge or breast pain. Microcalcifications or clustered calcifications on mammography were associated with 26.0 to 44.8 percent incidence of cancer. BI-RADS 1, 2, and 3 readings very rarely lead to cancer diagnoses, while 4 and 5 are associated with a cancer incidence of 38.5 to 94.0 percent.
Five studies reported clustered calcifications. Two were reported in terms of patients (13 of 29, or 44.8 percent were diagnosed with cancer), while three were reported in terms of lesions (207 of 796, or 26.0 percent were cancerous).
Fifteen studies reported mammographic findings of masses. It was not clear whether these were palpable masses or masses that were detected only by mammography. Eight were reported in terms of patients, but only six reported which patients were diagnosed with cancer (83 of 260 patients, or 31.9 percent). Seven studies reported masses in terms of lesions (672 of 2,514, or 26.7 percent were cancerous).
Three studies reported mammographic findings of masses plus calcifications. Two were reported in terms of lesions (30 of 96, or 31.3 percent were diagnosed with cancer), and one was reported in terms of patients (zero of three were cancerous).
Seven studies reported mammographic findings of parenchymal distortion. Four were reported in terms of patients (22 of 79, or 27.8 percent were diagnosed with cancer), and three were reported in terms of lesions (9 of 50, or 18.0 percent were cancerous).
For the purposes of this report, all densities are considered together. Focal symmetric, diffuse increased, nodular, irregular, and unspecified densities are all included in this category. Four studies reported numbers of patients with densities, but only three reported the number of patients who received a diagnosis of cancer (five of 39, or 12.8 percent). One study reported that six lesions were densities; two of these were cancerous (33.3 percent).
Mammographic results were infrequently reported in terms of BI-RADS scores. When the specific term "BI-RADS" was not used, but the description of the mammogram reports coincided with the BI-RADS terminology, the results were converted into BI-RADS categories. For example, "negative" became BI-RADS 1. This was an attempt to standardize results from different studies and was rarely necessary. Some studies grouped BI-RADS 4 and 5 together and are noted as such in this report. As BI-RADS readings are for entire mammograms, the numbers of lesions are considered to be equivalent to the numbers of patients.
Three studies reported BI-RADS 1, but only two reported incidence of cancer diagnoses (7 of 163, or 4.3 percent). BI-RADS 2 was reported in two studies, but only one reported incidence of cancer diagnosis (19 of 883, or 2.2 percent).
Four studies reported patients with BI-RADS 3 mammographic readings, of which cancer diagnoses were reported for three studies (22 of 2,034, or 1.1 percent incidence). Five studies reported patients with BI-RADS 4 readings. Three of these studies reported the number of patients diagnosed with cancer (30 of 54, or 55.6 percent).
Four studies reported patients with BI-RADS 5 readings. Two of these studies reported the number of patients diagnosed with cancer (78 of 83, or 94.0 percent). Three studies reported patients with BI-RADS 4 + 5 readings, in which 117 of 304 (38.5 percent) were diagnosed with cancer.
Combining the studies with BI-RADS 4, 5, or 4 + 5, in which cancer diagnoses were reported, there were eight studies (n=441, of which 225 patients (51.0 percent) were diagnosed with cancer).
Question 2 Results: What is the management of lobular carcinoma in situ (LCIS) and atypical hyperplasia (AH)?
Within 5 years after LCIS diagnosis, 4.2-9.3 percent of patients in the database were diagnosed with breast cancer. In studies that followed patients for greater than 5 years, the incidence of cancer was 7.7-26.3 percent. Incidence of cancer in LCIS patients varied widely, depending on treatment. The lowest incidence of cancer was seen in women who underwent bilateral mastectomy (zero), closely followed by women who took tamoxifen (1.9 percent), then observation, ipsilateral mastectomy, and local excision (9.3 percent).
Within 5 years after AH diagnosis, 3.7-19.3 percent of patients were reported to develop breast cancer. In studies that followed patients for greater than 5 years after AH diagnosis, the incidence of cancer was 13.6-33.6 percent.
The evidence indicates that when SCBX yields a diagnosis of atypical ductal hyperplasia, an excisional biopsy of the lesion should be performed. Nearly 50 percent of patients with an original diagnosis of ADH had a change in diagnosis as a result of excisional biopsy. There was not enough information regarding ALH patients in the database to be able to make a similar assessment.
Although data are available from only one study, SERM therapy with tamoxifen appears to have a profound impact in decreasing the incidence of breast cancer following LCIS or AH. Enthusiasm for these results must be tempered, however, by recognition of the risks associated with SERM therapy, such as increased risk of endometrial cancer and thromboembolic disease.
Nineteen studies were performed in North America, and one was performed in Europe. Twelve studies were observational, including nine retrospective (n=705) and three prospective (n=420) studies. There were eight interventional studies, consisting of seven UCS (n=357) and one RCT (n=2,019).
The number of studies described below is greater than the total number of studies for Question 2 because three studies included two populations (ALH and ADH, or LCIS and AH); these studies therefore are cited twice, although the patients are counted only once.
One study (Raju and Vertes, 1996) also was included in the database for Question 1. This study involved 20 patients, but the results for one patient were not included in the database, as this patient had a prior history of breast cancer.
Burbank (1997) described 18 patients with ADH. Eight of these patients were described also in Jackman, Burbank, Parker, et al., (1997) and therefore were not captured from Burbank's study.
Eight of the ten studies with ADH patients compared initial biopsy results with excisional biopsy results. Excisional biopsy was performed on 314 (77.9 percent) of the 403 patients who were given a diagnosis of ADH. This led to a change in diagnosis in 133 (42.4 percent) of patients, with 26 (19.5 percent) being diagnosed with cancer, and 76 (57.1 percent) receiving a diagnosis of DCIS. The remainder of changes were LCIS (four patients; 3.0 percent) or benign (27 patients; 20.3 percent). The change in diagnosis was even more pronounced when patients were stratified by biopsy type. Those who underwent SCBX (k=7, n=206) had a 56.3 percent incidence of diagnosis change, compared with 27.8 percent diagnosis change for patients who underwent vacuum-assisted biopsy (k=2, n=108).
Only one study of ALH patients (Liberman, Cody, Hill, et al., 1999) compared initial biopsy results with excisional biopsy results. Of the seven patients in the study, four underwent an excisional biopsy, leading to a change in diagnosis in just one patient (ALH was changed to LCIS).
One study (Fisher, Costantino, Wickerham, et al., 1998) examined the effect of tamoxifen therapy on patients with AH. Patients were not separated according to ALH vs. ADH. In this study, 1,193 patients with atypical hyperplasia were followed for up to 5 years. Of the 614 patients who were treated with placebo, 23 developed breast cancer (3.7 percent), compared with three of the 579 patients who were treated with tamoxifen (0.5 percent).
While long-term, confirmatory studies are necessary, the above results suggest that tamoxifen therapy is highly efficacious in decreasing the incidence of breast cancer following a diagnosis of LCIS or AH.
Question 4 Results: What are the indications for sentinel node biopsy?
Regardless of the technique of sentinel node identification (vital blue dye or radiocolloid mapping), sentinel nodes are detected in the vast majority of cases, and are positive in approximately one-third of cases. Both vital blue dye and radiocolloid mapping work well individually, but combining them results in a higher yield of sentinel node detection. False negative sentinel nodes occur in 2 to 3 percent of cases.
The limited data regarding types of tumors most amenable to this procedure suggest that tumor size, location, and history of prior breast surgery do not have major impact on the utility of sentinel node biopsy.
The decrease in morbidity associated with full axillary node dissection is encouraging; however, the risk of false negative sentinel nodes cannot be ignored. The decision regarding sentinel node biopsy vs. full ALND should be made by the patient and the clinician, after full discussion of the benefits and risks. Before SLN biopsy can be recommended as a routine procedure, long-term data are needed, to show that cancer outcome and survival are not impaired. Further attention should be paid to the impact of surgeon experience and extent of pathological investigation of the tissue.
To evaluate indications for SLN biopsy, data regarding tumor size and location were captured, in addition to information regarding patients' prior history of breast surgery (Evidence Tables 10 and 11). Baseline data regarding these characteristics were frequently available, but information regarding which patients had sentinel nodes identified was not. The majority of studies did not report the impact of tumor size, location, or prior breast surgery on success rate in identifying SLN. In studies that did report this information, the success rate was over 90 percent for almost all categories. In general, smaller tumors were associated with higher success rate, and tumors in the upper outer quadrant showed a better success rate than those in the upper inner quadrant.
In most studies, there was no significant difference between success rates for patients with prior breast surgery vs. no prior surgery. One study (Feldman, Krag, McNally, et al., 1999) reported that 4 of 21 patients with positive axillary nodes had false-negative sentinel nodes. All four of these patients had prior large excisional biopsies. While the overall sensitivity of sentinel node biopsy was 81 percent in this study, it was only 60 percent for patients with prior excisional biopsies.
Full axillary lymph node dissection (ALND) was done on all patients in 27 studies (n=3,182) and in approximately 30 percent of patients in the remaining 12 studies. In 2 of these 12 studies, the number of patients who received a full ALND was not reported. Both vital blue dye and radiocolloid mapping were used to identify the sentinel lymph nodes (SLN) in the majority of studies (k=20, n=3,866), vital blue dye was used alone in seven studies (n=506), and radiocolloid mapping was used alone in 12 studies (n=1,524). The sum of these numbers does not equal the total number of patients in all of the studies because not all patients in the combination studies ended up receiving both dyes.
Most studies reported the number of patients in whom the sentinel node was detected, the number of sentinel nodes that were positive for cancer, the number of positive axillary nodes, and the number of false negative (FN) sentinel nodes. In the studies in which ALND was not performed on all patients, it was not always possible to distinguish which patients did receive a full ALND. Therefore, the results presented in the following summary are from only the studies in which full ALND was performed on all patients.
From the 27 studies in which all patients underwent ALND, at least one sentinel node was detected in 2,909 of the 3,182 patients studied (91.4 percent). Sentinel nodes were positive for metastatic disease in 1,089 patients (37.4 percent). Axillary nodes were positive for metastatic disease in 1,184 patients. A total of 84 FN sentinel nodes was reported, for a false negative rate of 2.9 percent. Adding the number of FN sentinel nodes to the number of positive sentinel nodes does not yield the number of positive axillary nodes because some of the positive axillary nodes were in patients in whom sentinel nodes were not detected.
Most studies reported results of SLN biopsy in patients with carcinomas only, but two studies included patients with DCIS also. Bass, Duaway, Mahatme, et al., (1999) included 150 patients with DCIS, of which 11 had positive SLNs. Cox, Pendas, Cox, et al., (1998) included 87 patients with DCIS, of which four had positive SLNs. Combining these data elements, 15 of 237 (6.3 percent) patients with DCIS had positive SLN biopsies. This brings into question the accuracy of the DCIS diagnosis.
The percentage of FN procedures varied not only by biopsy method, but also by calculation technique. Roumen, Valkenburg, and Geuskens, (1997) argued that the number of FN biopsies should be compared with the number of true positive ALNDs, rather than comparing the false negatives to the full number of patients studied. This does not take into account the number of true negatives. Another method of calculation would be dividing the false negatives by the total negatives. Both of these methods of calculation would result in higher false negative rates.
Ideally, comparisons of sensitivities of sentinel node biopsy between patients with different tumor characteristics could be calculated; however, the database is lacking sufficient information. Additional useful information would be comparison of success rates in sentinel node biopsy based on amount of experience by the surgeon and yield of biopsy results based on extent of pathological exploration.
Two recent studies which were published after this database was collected provide further evidence in support of sentinel node biopsy (Giuliano, Haigh, Brennan, et al., 2000; McMasters, Tuttle, Carlson, et al., 2000). In the two studies, 939 patients underwent SLN biopsy followed by full ALND. Sentinel nodes were successfully identified in 841 of the 939 patients (89.6 percent). The false-negative rate is consistent with that reported in this database. Long-term, multicenter trials currently are underway to evaluate the impact of SLN biopsy on cancer outcomes and survival (Giuliano, Haigh, Brennan, et al., 2000). If the long-term data are as promising as the short-term data appear to be, SLN biopsy may become the standard of care for nodal staging of women with breast cancer.
Question 5 Results: What are the costs associated with diagnosis and management of breast disease, as outlined above?
Given the wide variation in methods of presenting cost information, it is not possible to answer this question using this database.
Each of the studies that evaluated cost described it very differently, making comparison difficult. Costs ranged widely across different hospitals and are quoted from different years. Some studies factored in the amount that would be covered by insurance, while others did not. Costs were reported per patient or per episode, some included costs of ancillary services while others reported strictly the cost of the procedures, and others did not differentiate exactly what costs they were describing. It is difficult to summarize this information in a useful manner.
The strengths of the review presented here include the clear definition of the research question, adherence to an explicit research protocol developed prior to the analysis, the comprehensive nature of the data search (employing both computer databases and manual bibliography searches, resulting in the inclusion of all relevant published materials), consensus between two reviewers of all data elements prior to entry into the database, and a quality control review of every element of this report.
Another primary strength of this evidence base derives from the collaboration of multidisciplinary researchers who participated in its development. It was compiled by investigators who are skilled in employing highly systematic and unbiased methods to collect, review and synthesize data from published clinical literature. Throughout the course of this project, there was frequent input from the coinvestigator and the TEP. In addition, the final draft was evaluated by a panel of peer reviewers.
The major limitations of this review are those related to weaknesses of the available published literature on the management of breast disease specifically linking risk factors, symptoms, and cancer incidence. The database enabled researchers to develop partial answers for Questions 2 and 4 (LCIS, AH, and sentinel node biopsy), however the answers for Questions 1, 3, and 5 (risk factors, breast abnormalities, cost) are limited by the paucity of data.
The major difficulties associated with compiling this evidence report were due to the variations in methods of reporting data. Many studies were rejected because they reported risk factors or breast symptoms but not both. In accepted studies, it was often difficult to ascertain whether authors were referring to numbers of patients or numbers of breast lesions. Studies which did report both risk factors and symptoms often reported cancer incidence only for the entire population. While mammographic results were almost always reported, variations in reporting the results made it impossible to combine the data in a useful way. This was unexpected, given that widespread use of the BI-RADS nomenclature could have prevented this problem. Other areas in which reporting style varied greatly included biopsy methods, timing of subsequent cancer diagnoses, other followup information, types of atypical hyperplasia, and information regarding patients who underwent sentinel node biopsy.
This systematic review of diagnosis and management of breast disease identified 109 studies published between 1994 and 1999 that met the prospectively determined inclusion criteria. The literature suffers from a lack of standardization of the terms used in reporting information regarding breast disease. The most glaring example of this is the number of studies that referred to numbers of lesions rather than numbers of patients. If all studies reported numbers of patients, comparison of results from different studies would be greatly facilitated.
The following conclusions may be drawn from this systematic review.
Questions 1 and 3 (risk factors, abnormal clinical or mammographic findings):
The best available evidence suggests that breast symptoms are evaluated by clinical breast exam and mammogram, with supplemental studies when the diagnosis is unclear.
There is no evidence to support modifying the work-up based on risk factors other than age.
It was hoped that a risk model, similar to the Gail model, could be created from the data amassed during this project, based not only on risk factors, but also on breast symptoms. This is not possible, however, because individual patient data, rather than aggregate data, are required. If a model could be constructed using aggregate data, age is the only risk factor that could be included, as sufficient information regarding other risk factors is lacking.
Question 2 (management of LCIS and AH):
Excisional biopsy is essential following SCBX diagnosis of ADH.
Although data are limited, selective estrogen receptor modulator therapy with tamoxifen appears to markedly decrease the incidence of cancer following a diagnosis of LCIS or AH.
There is no evidence to support ipsilateral mastectomy after a diagnosis of LCIS.
Question 4 (sentinel node biopsy):
Sentinel node biopsy is successful in most breast cancer patients and could save millions of women from the more invasive procedure of axillary node dissection. However, the risk of false-negative results must be recognized, in addition to the potential variability of results depending on surgical experience and extent of pathological investigation.
The following recommendations would enable researchers to generate useful data to support answers for the questions posed in this report:
Questions 1 and 3 (risk factors, abnormal clinical or mammographic findings):
Perform prospective studies that report patients' risk factors, symptoms, cancer incidence, time of cancer diagnosis, and duration of followup. The current literature reports some, but not all, of these elements. Complete information is essential to build an accurate assessment of the relationship between breast cancer risk factors, breast symptoms, and cancer incidence. If this information were available, risk-factor assessment could be used to accurately assess pre-test probability of malignancy and could possibly affect management of patients with breast symptoms.
Standardize reporting of risk factors.
For family history, the number of first degree and more remote relatives with cancer, as well as the age at which cancer was diagnosed are important. Simply reporting the presence or absence of family history is inadequate to support meaningful analysis.
Regarding parity, the number of pregnancies and the age at first delivery should be reported. Additionally, age at menarche and age at menopause are needed to estimate duration of estrogen exposure.
With reference to HRT, the type of hormone (unopposed estrogen vs. combination therapy), dosages, and duration of therapy are important elements to report.
Genetic alterations, including HER-2/neu gene expression in benign breast disease, should be further evaluated with regard to subsequent risk of breast cancer.
Standardize reporting of mammographic results. The BI-RADS classification is useful and, if generally adopted, could enable researchers to perform valid inter-study comparisons.
Evaluate role of digital mammography.
Summary Receiver Operating Characteristics (ROC) curves could be prepared to summarize the sensitivities and specificities of mammography and ultrasound. The sensitivities and specificities from each study could be weighted for its sample size (Irwig, Tosteson, Gatsonis, et al., 1994). The validity of such comparisons may be questioned, however, due to possible variations in mammographic equipment, technique, and operator skill.
Important new developments in genetic susceptibility for breast cancer and promising new imaging techniques are likely to have a major impact on the diagnosis and management of breast disease. Information on these topics should be collected and included in an update of this systematic review.
Consider expanding the literature search to include other, less well established risk factors, such as diet, exercise, and alcohol use. Given the limited data on established risk factors, however, it is likely that the information on additional risk factors would be even more sparse.
Consider expanding the search to include foreign language articles.
Question 2 (management of LCIS and AH):
Evidence supports tamoxifen therapy as cancer prophylaxis for patients with either LCIS or AH. Further studies need to evaluate the risks and benefits of tamoxifen compared with other SERMs.
The appropriate duration of SERM therapy needs to be established.
Further studies should address whether SERMs should be recommended for all patients with AH or LCIS diagnoses or only for patients with additional risk factors.
Further studies should make an attempt to distinguish the different types of AH to determine whether the subsequent cancer incidence differs between ADH and ALH.
Question 4 (sentinel node biopsy):
The existing short-term data support performing sentinel node biopsy on most patients with newly diagnosed breast cancer. Long-term data are required, however, before this procedure can be recommended in place of full axillary dissection.
Further studies should attempt to clarify optimal use of sentinel node biopsy by determining differences in sensitivities based on tumor size, location, and history of breast surgeries. Most studies in this database give baseline information on these characteristics but do not follow up by linking the characteristics with the outcome of sentinel node detection.
Evaluate use of sentinel node biopsy for DCIS.
Compare incidence and severity of lymphedema after SLN biopsy vs. ALND.
Correlate surgeon's amount of experience with success rate for SLN biopsy.
Determine appropriate degree of pathological dissection of SLN biopsy to maximize identification of malignancy.
Given the large volume of information continuing to be published regarding breast cancer, semiannual updates are recommended to keep this evidence base current.
This systematic review has led to an evidence base that contains a wealth of data regarding diagnosis and management of breast disease. The relational database could be provided with a navigational software interface that permits easy filtering and exporting for analysis. The evidence base provides a valuable opportunity to test recommendations from clinical practice guidelines against the weight of the best available evidence.
American College of Radiology (ACR)
Breast Imaging Reporting and Data System
(BI-RADS)
| Category | Assessment | Description |
| 1 | Negative | There is nothing on which to comment |
| 2 | Benign finding | A definitely benign finding is described |
| 3 | Probably benign finding | Very high probability of being benign; short term follow-up is recommended to establish stability. |
| 4 | Suspicious abnormality | Not characteristic, but has a reasonable probability of being malignant; biopsy should be urged. |
| 5 | Highly suggestive of malignancy | High probability of being cancer; appropriate action should be taken. |
From: Liberman L, Abramson AF, Squires FB, et al. The breast imaging reporting and data system: Positive predictive value of mammographic features and final assessment categories. Am J Roentgenol. 1998; 171:35-40. Used with permission.
Tumor Stages
| T0 | No evidence of primary tumor |
| T1 | Tumor size < 2 cm |
| T2 | Tumor size > 2 - 5 cm |
| T3 | Tumor size > 5 cm |
| T4 | Tumor extended to chest wall or skin |
From: American Cancer Society. Breast Cancer: Detection and Symptoms. http://www3.cancer.org/cancerinfo/load_cont.asp?ct=5&st=ds#stage (10 Sept. 1999). Used with permission.
Free Full text in PMC]Free Full text in PMC]Free Full text in PMC]Free Full text in PMC]Free Full text in PMC]Free Full text in PMC]
