Using Epidemiology to Study Medical Devices in Women
In the second session, “Methodology and Data Sources,” the first talk was mine, “Using Epidemiology to Study Medical Devices in Women.” The purpose was to give an overview of epidemiology, using device studies as examples, for the sake of audience members in other fields. Because one of my hopes is to attract some of these same types of readers to this supplement, the rest of this section describes epidemiologic concepts in some detail.
I started by defining epidemiology as the science of the occurrence of diseases in human populations and explained that epidemiologists combine quantitative skills, clinical medicine, public health orientation, and a good dose of humor to practice their craft. There are two general epidemiologic goals. The first is to describe a population; an example would be to estimate the number of U.S. breast implant recipients. The second is to assess causation, such as estimating the effect of longer continuous contact lens wear time on the risk of corneal ulcer. Readers who are interested in more details may be interested in an introductory book on epidemiology, 2 an advanced epidemiology textbook, 3 and a resource on medical device epidemiology. 4
Describe a Population
My example of estimating the number of U.S. breast implant recipients came from work based on the Medical Device Implant Supplement to the 1988 National Health Interview Survey, administered by the National Center for Health Statistics (part of the Centers for Disease Control and Prevention) and cosponsored by CDRH. This was an interview survey of a stratified random sample of households that asked about current “medical” implants in household members. For implant recipients, it collected the reason for the implant, duration of implantation, and major adverse effects. For everyone in the overall survey, the interviewer collected data on age, sex, race, adult income, and adult education level. There were 143 women in the sample who were identified as currently having a breast implant. They were calculated to represent 304,000 women in the United States having breast implants. The U.S. prevalence of women with breast implants in 1988 was therefore 0.3% (0.2–0.4%). 5
How does this study compare with others? Table 2 shows the results for three others 6,7 (and J. Daling, unpublished results, 1992; personal communication), which estimate a prevalence of about 1% for approximately the same time period. Therefore, it appears that the Bright et al 5 study severely underestimated the prevalence of breast implants. A likely reason is that “cosmetic” implants, or implants not known to the household questionnaire respondent, were not counted. This example illustrates a pitfall in survey design and the value of comparing multiple studies of the same question.
An association between exposure and outcome may be due to any single factor or combination of the following:
• Statistical variation (otherwise known as chance).
• Biased study design.
• Alternative explanations.
• Cause and effect.
Of these, epidemiologists expend the bulk of their effort trying to rule out biased study designs and alternative explanations as reasons for observing an association. It is worth noting here that in contrast, randomized clinical trials use patient randomization (resulting in a random distribution of both treatments and other factors) and, to the extent possible, masking, to eliminate biased study designs and alternative explanations as reasons for observing an association. Thus, the causality argument for study results is very strong. However, there are times when randomized clinical trials are impractical. It may be difficult to enroll consenting subjects, or the outcome of interest may be so rare that a trial is too expensive. Furthermore, randomized clinical trials study ideal conditions; the selection criteria for subjects are often strict, and all of the participants are on “best behavior” because they know they are being watched.
If a randomized controlled trial is not feasible, the next best study design choices are cohort and case-control designs. In the cohort study, subjects are chosen on the basis of their exposure; then the subjects are followed in time to observe occurrences of the outcome, ideally by an investigator who is masked to the exposure status of the subjects. In the case-control study, subjects are chosen on the basis of their outcome status, and then their exposure histories are obtained. In variations or combinations of the two, the experience of cases is compared with the experience of a sample of an entire, defined population.
Methodologically weaker, although easier, study designs include the cross-sectional study (in which the investigators are unsure of the relative timings of exposure and outcome), the ecological study (in which data are available for populations, rather than individuals in the populations), and the case series (in which some cases are described and case selection generally is not done on a population basis).
Assessing Causation with a Cohort Study
An intuitively obvious way to observe (what one hopes is) cause and effect is to compare the experiences of people who have had different exposures. Using a cohort study rather than a randomized clinical trial is particularly advantageous when randomization of treatment would be unethical; other considerations could include the wish to study actual, rather than idealized, behavior, or to quickly study the long-term effects of a treatment. My example is a comparison of “laparoscopic vs traditional appendectomy for suspected appendicitis.”8 All of the study subjects were adults operated for presumed acute appendicitis in the same hospital during a 4-month period. The cohorts were defined by the type of surgery: laparoscopic appendectomy (47 patients), diagnostic laparoscopy (6), laparoscopy converted to open incision (15), and open appendectomy (54). The outcomes included duration of operation, mortality, delayed discharge, and wound complications. Compared with open appendectomy, the rate ratio of wound complications for laparoscopic appendectomy was 0.57, with a 95% confidence interval (CI) of 0.21–1.56, and for laparoscopic converted to open incision, 2.5 (95% CI = 1.2–5.5). There were several issues that relate to the validity of these results:
• There was no explanation of the reason behind the initial choice of surgical type, which may be related to the outcome.
• The surgical roles of the junior and senior surgical residents differed for each surgical type, and may themselves be related to the outcome.
• The choice of surgical type and unplanned procedures were both related to gender. The authors do not report whether the complication rate was also related to gender.
This example illustrates some of the advantages (ease of conceptualization and conduct) and pitfalls (inadequate accounting of alternate explanations of the results) of cohort studies. Careful study design can sidestep most, if not all, pitfalls.
Assessing Causation with a Case Control Study
As another example, suppose one wanted to estimate the effect of longer continuous contact lens wear time on the risk of corneal ulcer. Would a randomized controlled trial be appropriate? Unfortunately, it wouldn’t, because corneal ulcers are rare and their occurrence depends on wearer behavior, which would in turn be affected by the subjects’ knowledge that they are being “watched.” As it happens, a few epidemiologic studies have already been done, out of which I selected a pair to use in the example. One is a case-control, and the other is a case-base study. Because they were designed by the same study team, many study features are similar and won’t obscure my comparison of the two designs.
The example study by Schein et al 9 defined its population as U.S. residents at risk of ulcerative keratitis who were at least 12 years old, spoke English or Spanish, and wore cosmetic soft contact lenses. The cases were from six university ophthalmology centers across the United States. They had ulcerative keratitis that had been confirmed by culture, treated with antibiotics, and verified by a corneal specialist. Of 109 eligible cases, 99 were enrolled and 86 (79%) completed the telephone interview.
Two sets of controls were selected in different ways. “Hospital controls” were selected from the same six university centers as the cases. They had conditions unrelated to lens use: minor trauma, floaters, headache, or other problems. The controls were frequency matched to cases by sex and center. Sixty-seven hospital controls enrolled and 61 (91%) completed the telephone interview. “Population controls” were chosen by, for each case, dialing the same area code, same first four digits of the case’s phone number, and a random last three digits to get eligible controls. Most cases had at least four controls. Ninety-two per cent of contacted households cooperated, and 410 population controls enrolled in the study.
The telephone interviews were used to assess exposure by asking about the type of lens, lens use and care in the prior 2 months, lens care instructions that had been received, previous use of lenses, general medical and ophthalmic history, and demographic characteristics.
The odds ratio, which estimates the relative risk, was calculated separately, using the two control groups. Using the hospital controls, the odds ratio was 4.21 (95% CI = 1.95–9.08). Using the population controls, the odds ratio was 3.90 (95% CI = 2.35–6.48). Fortunately, the odds ratios are very similar.
Assessing Causation with a Case-Base Study
Over the last 15 years, new study designs have combined features of both cohort and case-control studies. An example is the “case base” study, in which the exposure history of cases is compared with the exposure history of the “population base” that gave rise to the cases. 10 With this method, one can study a rare outcome and obtain estimates of population rates.
Poggio et al 11 estimated both the risk of corneal ulcer and the causal role of longer continuous contact lens wear time. The study population was all wearers of contact lenses in the self-contained area of Maine, Vermont, New Hampshire, Rhode Island, and Massachusetts (except Berkshire County). This area has more than 9 million residents and minimal crossing of the study area boundaries for medical care.
The study authors estimated contact lens exposure with a household telephone survey that used the random-digit dialing technique. The household respondent had to be at least 16 years old to answer questions about the age and sex of all lens wearers who were at least 12 years old. Information on the type of lens (soft extended or soft daily wear) and the reason for wear (if for cataract, excluded from study) was collected. This survey was completed for 60% of 6,990 sample households.
The corneal ulcer outcome was measured with a complete ophthalmologist survey that asked about all ulcerative keratitis cases in the 4-month study period that were treated with antibiotics. For the cases that were eligible for the study (lens wearers who were at least 12 years old), information on the type of lens (soft extended or soft daily wear) and the reason for wear (if for cataract, excluded from the study) were collected from the ophthalmologist. This survey was completed for 99% of practices in the study area.
The corneal ulcer rate was much higher among users of extended rather than daily wear lenses. That increase was expressed as the relative rate, or 20.9/4.1 = 5.15. The 95% confidence interval on the relative rate is 3.47–7.65. The relative rate approximates the relative risk.
Assessing Causation by Comparing Studies
Table 3 shows the main results from the Poggio et al 11 and Schein et al 9 studies, and two other related studies. 12,13 Epidemiologists generally derive some reassurance from a situation like this, in which independent studies show generally the same results. The association between extended lens wear and microbial corneal ulcer was strongly positive in each study, despite differing designs and slightly different study questions. The very weak effect occurring only with sterile corneal ulcer as the outcome points to a suggestion that the extended wear risk is specific to the type of ulcer. Specificity of study results, in the absence of major design flaws, is generally seen as supportive of causality.
As a point of interest, both the Schein et al 9 and Dart et al 12 studies were able to examine detailed aspects of exposure (see Table 4). Among other features, both chose to measure the effect of various lengths of lens use. Schein et al 9 found that risk increased with each category of increased length of wear, whereas Dart et al 9 found a stepped effect at 7 days. The breadth of the confidence intervals indicates that these results are statistically consistent with each other, and making a choice between the two apparent patterns would require either combining the underlying data from the two studies or a new, larger data collection effort.
When choosing between study types, one bears in mind the symmetry of strengths and weaknesses. Cohort designs have particular advantages for studying rare exposures and multiple outcomes from the same exposures, whereas case-control studies are the best choice for study questions involving rare outcomes with many possible exposures to be examined. If neither situation applies, or if both the exposure and outcome are rare, then combination designs, such as the case-base type, may offer significant advantages.
Other Presentations on Methodology and Data Sources
The next presentation was on “Methods of Health Technology Assessment.” Vivian Coates explained that these methods include information synthesis, expert opinion, group judgment, epidemiologic or other observational studies, clinical trials, and laboratory tests. Each method offers its own advantages and disadvantages. She reported that at her organization, ECRI, technology assessment is an information synthesis based on a systematic analysis of data from published, peer-reviewed clinical literature, gray literature (eg, meeting abstracts and proprietary reports), registries, and administrative databases. Her colleagues develop evidence models and evidence tables, use quantitative meta-analysis, create computerized decision models for cost-effectiveness analysis, conduct both internal and external review, and use an external audit committee. She gave examples of recent analyses that have been done for topics of particular interest to women and emphasized that “findings are not written in stone, and must be re-examined periodically in light of new findings.” She also said that the “assessment of medical technology is often limited by the lack of strong evidence for improved patient outcomes and cost-effectiveness relative to alternatives,” the small number of controlled trials, and flawed study designs and reporting of study results. “In spite of this, evidence-based technology assessment is essential for deciding what works best for patients.”
The presentation on data sources is published in this supplement. 14 It was followed by a talk on “Methods of Device Evaluation” by Katherine Detre, of the University of Pittsburgh.
Deborah Blum and her colleagues, of the CDRH Office of Surveillance and Biometrics, in a poster, discussed the difficulties of using adverse medical device event reports submitted to the FDA to study gender-related issues arising from the very low frequency with which gender is noted on the report.
Karen Witkin and Rosanne McTyre of the Weinberg Group, presented a poster on methodologic issues related to evaluating implant safety and performance and discussed the analysis of product complaints, explant retrieval and analysis, and clinical follow-up studies. Another poster, by Kathleen Neill of Georgetown University, reported on focus groups that were used to learn about women’s incentives, barriers, expectations from the research team, and disincentives to participating in clinical trials. 15
Two of the presentations, on home pregnancy tests 16 and prenatal ultrasound, 17 are included in this supplement. Mary D’Alton, of New England Medical Center, gave a talk on telemedicine, “the use of telecommunications technology to provide medical services across a geographical separation,” in obstetrics. She described the sonography system based at the New England Medical Center in Boston, MA, and presented results from two published studies. 18,19 She predicted that both academic and commercial networks will grow, especially as the following potential problems are resolved: licensure for interpreting images sent across state lines, credentialing, antitrust laws, medicolegal liability, and patient confidentiality and privacy.
Julian Parer, University of California San Francisco, presented “Findings from Studies of Intrapartum Fetal Monitoring, and the Need for New Technologies,” which has been published as a review article. 20 An independent poster by Sonia Swayze of the CDRH Office of Surveillance and Biometrics summarized shortcomings of electronic fetal monitors associated with serious birth outcomes that have been reported to FDA. 21
There were two talks on cervical cytology screening by David Wilbur 22 and Tom Sedlacek. 23 Unfortunately, Dr. Sedlacek died before his paper could be updated, so a brief update by Max Robinowitz is included in this supplement. 24 A talk by Alula Hadgu (Centers for Disease Control and Prevention), “Bias in the Evaluation of Medical Diagnostic Devices for Detecting Infectious Diseases,” was based on one of his publications. 25
Tanya Spirtos, of Stanford University, addressed the treatment of menorrhagia. She described the etiology, the impact of menorrhagia on women and the health care system, diagnostic approach, and the current treatment course, based on scientific literature. 26–29 After detailing the effectiveness and deficiencies of current therapies, she described the evolution and outcomes of endometrial ablation technologies: hysteroscopic, rollerball, heated saline, laser, phototherapy, microwave, radiotherapy hyperthermia, and thermal transfer balloon, also based on publications. 30–36
A poster by Sonia Swayze showed that toxic shock syndrome associated with tampons is still being reported to FDA.
The systematic collection of data to be used for research will hopefully lead to improvements in breast cancer detection by mammography. Bonnie Yankaskas, of the University of North Carolina at Chapel Hill, described the motivation, data sources, structure, and administration of the Carolina Mammography Registry. The pathology information from biopsies is entered into a linked database to enable the study of practice patterns and health outcomes in the community. Many of the methodologic, validation, and administrative issues have since been published. 37–40
Rachel Brem, of Johns Hopkins Hospital, discussed the limitations of mammography as a motivation for developing enhanced or new methods. She reviewed the types of minimally invasive breast biopsy: fine-needle aspiration, core needle biopsy with stereotactic and ultrasound guidance, stereotactically guided vacuum-assisted directed biopsy, and Advanced Breast Biopsy Instrumentation for en mass percutaneous excision of tissue. She also discussed computer-aided diagnosis of film mammograms, digital mammography, magnetic resonance imaging, and nuclear imaging. She has since published studies of some of these topics. 41–44
A poster by Richard Kisielewski and his CDRH colleagues displayed a summary of reports to FDA of residual metal shavings and fragments associated with core biopsy needles, along with a laboratory analysis that apparently narrowed the problem to models with a side-notch angle of 90 degrees. 45,46
The presentations on silicone gel breast implants have been published. The first one, “Breast Implants and Breast Cancer: Risk, Stage at Diagnosis and Survival,” has been reported in two papers. 47,48 The second talk, “Epidemiology of Breast Implants,” by Lori Brown, is included in this supplement. 49 She also presented a poster that has since been published elsewhere. 50
At the poster session, Kathleen Neill presented an update of a previously published paper on decision making by women contemplating breast reconstruction. 51 Another poster, by Martha A. Embrey et al (Breast Implant Public Health Project, LLC), reported results from interviewing 25 women with breast implants about their perceptions. A third poster (Arthur Brawer of Monmouth Medical Center), “Amelioration of systemic disease after removal of silicone breast implants,” reported on the experience of 156 breast implant recipients with systemic disease, before and after breast implant removal.
After the first two talks of this session, on the epidemiology of osteoporosis 52 and “Use of Bone Mineral Density to Predict Fracture Risk,”53 Peter Steiger of Hologic, Inc, presented “Bone Densitometry for the Evaluation of the Patient at Risk for Osteoporosis.” After discussing the basic epidemiology and physiology of osteoporosis, he talked about considerations for bone site selection for assessment and the history of bone assessment techniques. Biochemical markers and their measurement parameters, as well as considerations for the appropriate use of T-scores and Z-scores for monitoring patients, were also presented.
The final talk in this session, presented by CDRH statisticians, was on the statistical methodology for comparing two diagnostic modalities, the examples being ultrasound and x-ray bone densitometry. The “ROC” curve (first developed for radio and radar technology) for any diagnostic test is plotted as (1 − specificity) on the x-axis and sensitivity on the y-axis. Plotting the curves for two diagnostic tests on one graph allows a fair comparison of the relative strengths of each. In the example given, the curves were very similar when (1 − specificity) and the sensitivity were both low. But when each of them were high, ultrasound performed better. Several methods for statistically comparing the areas under the curves were presented.
The last topical session was on cardiovascular devices. Two of the presentations are included in these proceedings. 54,55 Judith Kramer, of Duke Clinical Research Institute, talked about results from the National Cardiovascular Network, which had 19 contributing high-volume institutions. The researchers found that between 1994 and 1996, balloon and atherectomy procedures declined, whereas stent placement increased, for each sex. Overall, the higher acute death rate of women after procedures was explained by their higher burden of risk factors. A substudy of postmenopausal women found that those who used estrogen experienced less restenosis after percutaneous coronary interventions. 56
The poster by Elizabeth Kennard, of the University of Pittsburgh, and her colleagues, presented a comparison of stenting with balloon angioplasty among women enrolled in a registry.
Other Issues of Concern to Women
Joan Todd, of the CDRH Office of Surveillance and Biometrics, presented a poster that summarized a publication on the dangers of hospital bed side rails. 57 Elderly women with low body weight and cognitive impairment are apparently at the highest risk of death and injury from bed side rail entrapment.
The epidemiology of medical devices in women is a broad topic. Although this conference was the first to draw attention to this research area as a whole, there has been a strong history of work in individual specialties (obstetrics, gynecology, breast health, osteoporosis, and cardiovascular devices) of particular concern to women. However, much remains to be done. I encourage readers to consider these areas: epidemiologists to study medical devices of particular concern to women, clinical specialists already studying devices to also consider epidemiologic research in their areas, and women’s health researchers to include device issues in their studies. The confluence of these areas should enable better health for women.
1. United States Congress. Federal Food, Drug, and Cosmetic Act
. Sec. 201. [Title 21, United States Code 321]. Definitions. Paragraph (h)- 1998.
2. Ahlbom A, Norell S. Introduction to Modern Epidemiology. Chestnut Hill, MA: Epidemiology Resources, 1984.
3. Rothman KJ, Greenland S. Modern Epidemiology.
2nd ed. Philadelphia: Lippincott-Raven, 1998.
4. Bright RA. Special methodological issues in pharmacoepidemiology studies of devices. In: Strom BL, ed. Pharmacoepidemiology. 3rd ed. New York: John Wiley and Sons, 2000.
5. Bright RA, Jeng LL, Moore RM. National survey of self-reported breast implants: 1988 estimates. J Long Term Eff Med Implants 1993; 3: 81–89.
6. Cook RR, Delongchamp RR, Woodbury M, Perkins LL, Harrison MC. The prevalence of women with breast implants in the United States: 1989. J Clin Epidemiol 1996; 48: 519–525.
7. Gabriel SE, O’Fallon WM, Beard CM, Kurland LT, Woods JE, Melton LJ 3rd. Trends in the utilization of silicone breast implants, 1964–1991, and methodology for a population-based study of outcomes. J Clin Epidemiol 1995; 48: 527–537.
8. Schirmer BD, Schmieg RE Jr, Dix J, Edge SB, Hanks JB. Laparoscopic vs
traditional appendectomy for suspected appendicitis. Am J Surg 1993: 165: 670–675.
9. Schein OD, Glynn RJ, Poggio EC, Seddon JM, Kenyon KR. The relative risk of ulcerative keratitis among users of daily-wear and extended wear soft contact lenses: a case-control study. N Engl J Med 1989; 321: 773–778.
10. Miettinen OS. Theoretical Epidemiology. New York: John Wiley and Sons, 1985.
11. Poggio EC, Glynn RJ, Schein OD, et al
. The incidence of ulcerative keratitis among users of daily-wear and extended-wear soft contact lenses. N Engl J Med 1989; 321: 779–783.
12. Dart JK, Stapleton F, Minassian D. Contact lenses and other risk factors in microbial keratitis. Lancet 1991; 338: 650–653.
13. Bates AK, Morris RJ, Stapleton F, Minassian DC, Dart JK. “Sterile” corneal infiltrates in contact lens wearers. Eye 1989; 3: 803–810.
14. Torrence ME. Data sources: use in the epidemiologic study of medical devices. Epidemiology 2001; 13: S10–S14.
15. Neill KM, Chessa F. Recruitment and retention of women in nontherapeutic clinical trials. Appl Nurs Res 1998; 11: 148–151.
16. Ehrenkranz JRL. Home and point of care pregnancy tests: a review of the technology. Epidemiology 2001; 13: S15–S18.
17. Marinac-Dabic D, Krulewitch CJ, Moore RM. The safety of prenatal ultrasound exposure in human studies. Epidemiology 2001; 13: S19–S22.
18. Malone FD, Nores JA, Athanassiou A, et al
. Validation of fetal telemedicine as a new obstetric imaging technique. Am J Obstet Gynecol 1997; 177: 626–631.
19. Nores J, Malone FD, Athanassiou A, Craigo SD, Simpson LL, D’Alton ME. Validation of first-trimester telemedicine as an obstetric imaging technology: a feasibility study. Obstet Gynecol 1997; 90: 353–356.
20. Parer JT, King T. Fetal heart rate monitoring: is it salvageable? Am J Obstet Gynecol 2000; 182: 982–987.
21. Swayze SC. Electronic fetal monitoring: are you monitoring mother or fetus? Nursing 1998; 28: 20.
22. Wilbur DC, Norton MK. The primary screening clinical trials of the TriPath AutoPap®
system. Epidemiology 2001; 13: S30–S33.
23. Sedlacek TV. Cost effectiveness in new technology in cervix cancer screening. Epidemiology 2001; 13: S26–S29.
24. Robinowitz M. Introduction to and update of “Cost effectiveness in new technology in cervix cancer screening.” Epidemiology 2001; 13: S23–S25.
25. Hadgu A. Bias in the evaluation of DNA-amplification tests for detecting Chlamydia trachomatis
. Stat Med 1997; 16: 1391–1399.
26. Hallberg L, Hogdahl AM, Nilsson L, Rybo G. Menstrual blood loss—a population study: variation at different ages and attempts to define normality. Acta Obstet Gynecol Scand 1966; 45: 320–351.
27. Rosenfeld JA. Chronic pelvic pain: an integrated approach. Am Fam Phys
28. Sculpher MJ, Dwyer N, Byford S, Stirrat GM. Randomised trial comparing hysterectomy and transcervical endometrial resection: effect on health related quality of life and costs two years after surgery. Br J Obstet Gynaecol 1996; 103: 142–149.
29. Carlson KJ, Nichols DH, Schiff I. Indications for hysterectomy. N Engl J Med 1993; 328: 856–860.
30. Neuwirth RS, Amin HK. Excision of submucus fibroids with hysteroscopic control. Am J Obstet Gynecol 1976; 126: 95–99.
31. Goldrath MH, Fuller TA, Segal S. Laser photovaporization of endometrium for the treatment of menorrhagia. Am J Obstet Gynecol 1981; 140: 14–19.
32. DeCherney A, Polan ML. Hysteroscopic management of intrauterine lesions and intractable uterine bleeding. Obstet Gynecol 1983; 61: 392–397.
33. Vancaillie TG. Electrocoagulation of the endometrium with the ball-end resectoscope. Obstet Gynecol 1989; 74 (3 pt 1):425–427.
34. Brill AI. What is the role of hysteroscopy in the management of abnormal uterine bleeding? Clin Obstet Gynecol 1995; 38: 319–345.
35. O’Connor H, Magos A. Endometrial resection for the treatment of menorrhagia. N Engl J Med 1996; 335: 151–156.
36. Overton C, Maresh MJ. Audit of currently available endometrial ablative techniques. Baillieres Clin Obstet Gynaecol 1995; 9: 357–372.
37. Kwok RK, Yankaskas BC. The use of census data for determining race and education as SES indicators: a validation study. Ann Epidemiol 2001; 11: 171–177.
38. Rosenberg RD, Yankaskas BC, Hunt WC, et al
. Effect of variations in operational definitions on performance estimates for screening mammography. Acad Radiol 2000; 7: 1058–1068.
39. Carney PA, Geller BM, Moffett H, et al
. Current medicolegal and confidentiality issues in large, multicenter research programs. Am J Epidemiol 2000; 152: 371–378.
40. Heinzen MT, Yankaskas BC, Kwok RK. Comparison of woman-specific vs
breast-specific data for reporting screening mammography performance. Acad Radiol 2000; 7: 232–236.
41. Brem RF, Schoonjans JM, Sanow L, Gatewood OM. Reliability of histologic diagnosis of breast cancer with stereotactic vacuum-assisted biopsy. Am Surg 2001; 67: 388–392.
42. Brem RF, Schoonjans JM. Radiologist detection of microcalcifications with and without computer-aided detection: a comparative study. Clin Radiol 2001; 56: 150–154.
43. Schoonjans JM, Brem RF. Sonographic appearance of ductal carcinoma in situ
diagnosed with ultrasonographically guided large core needle biopsy: correlation with mammographic and pathologic findings. J Ultrasound Med 2000; 19: 449–457.
44. Revelon G, Sherman ME, Gatewood OM, Brem RF. Focal fibrosis of the breast: imaging characteristics and histopathologic correlation. Radiology 2000; 216: 255–259.
45. Finder CA, Kisielewski RW, Kedas AM. Residual metal shavings and fragments associated with large-core biopsy needles: a follow-up. Radiology 1998; 208: 833–834.
46. Kedas AM, Byrd LJ, Kisielewski RW. Residual metal shavings and fragments associated with large-core breast biopsy. Radiology 1996; 200: 585.
47. Deapen DM, Bernstein L, Brody GS. Are breast implants anticarcinogenic? A 14-year follow-up of the Los Angeles Study. Plast Reconstr Surg 1997; 99: 1346–1353.
48. Deapen D, Hamilton A, Bernstein L, Brody GS. Breast cancer stage at diagnosis and survival among patients with prior breast implants. Plast Reconstr Surg 2000; 105: 535–540.
49. Brown SL. Epidemiology of silicone-gel breast implants. Epidemiology 2001; 13: S34–S39.
50. Brown SL, Parmentier CM, Woo EK, Vishnuvajjala RL, Headrick ML. Silicone gel breast implant adverse event reports to the Food and Drug Administration, 1984–1995. Public Health Rep 1998; 113: 535–543.
51. Neill KM, Armstrong N, Burnett CB. Choosing reconstruction after mastectomy: a qualitative analysis. Oncol Nurs Forum 1998; 25: 743–750.
52. Scott JC. Epidemiology of osteoporosis. Rev Rhum Engl Ed 1997; 64 (suppl 6): 10S–14S.
53. Black DM. Screening and treatment in the elderly to reduce osteoporotic fracture risk. Br J Obstet Gynaecol 1996; 103 (suppl 13): 2–8.
54. Kennard ED. New device angioplasty in women: outcomes of planned and unplanned use. Epidemiology 2001; 13: S40–S45.
55. Lansky AJ, Mehran R, Dangas G, et al
. New device angioplasty in women: clinical outcome and predictors in a 7372-patient registry. Epidemiology 2001; 13: S46–S51.
56. O’Brien JE, Peterson ED, Keeler GP, et al
. Relation be-tween estrogen replacement therapy and restenosis after percutaneous coronary interventions. J Am Coll Cardiol 1996; 28: 1111–1118.
© 2002 Lippincott Williams & Wilkins, Inc.
57. Todd JF, Ruhl CE, Gross TP. Injury and death associated with hospital bed side-rails: reports to the US Food and Drug Administration from 1985 to 1995. Am J Public Health
1997;87:1675–1677. (Published correction appears in Am J Public Health