The sensitivity, specificity, and positive predictive value of screening ...

45 downloads 64 Views 165KB Size Report
racy of screening mammography varies according to symptomatic ... have a personal history of breast cancer. ... disease that is the object of screening”,1 62% of.
J Med Screen 2000;7:105–110

105

The sensitivity, specificity, and positive predictive value of screening mammography and symptomatic status A M Kavanagh, G G Giles, H Mitchell, J N Cawson

Australian Research Centre in Sex, Health and Society, La Trobe University, 215 Franklin St (first floor), Melbourne, Victoria, Australia 3000 A M Kavanagh, senior research fellow Cancer Epidemiology Centre, Anti-Cancer Council of Victoria, 100 Drummond St, Victoria, 3053, Australia G G Giles, director and professor Victorian BreastScreen Registry, PO Box 542, Carlton South 3053, Victoria, Australia H Mitchell, director and epidemiologist St Vincents BreastScreen, St Vincents Hospital, 41 Victoria Parade, Fitzroy 3065, Australia J N Cawson, director and radiologist Correspondence to: Dr Kavanagh email: [email protected] Accepted for publication 11 May 2000

Abstract Objective—To examine whether the accuracy of screening mammography varies according to symptomatic status reported at the time of screening. Setting—Victoria, Australia, where free biennial screening is provided to women aged 40 and older. Methods—We examined the sensitivity, specificity, and the positive predictive value of screening mammography by symptom status in 106 826 women from Victoria, who attended for first round mammography in 1994 and who did not have a personal history of breast cancer. Symptomatic status was divided into the following categories: asymptomatic; significant symptoms, if the woman reported a breast lump and/or blood stained or watery nipple discharge; and other symptoms, if reported. Unconditional logistic regression modelling was used to adjust for age, use of hormone replacement therapy (HRT), and family history. Results—Sensitivity was lower for women with other symptoms (60.0%) than asymptomatic women (75.6%), or women with significant symptoms (80.8%). Specificity was lower for women with significant symptoms (73.7%) than asymptomatic women (94.9%), or women with other symptoms (95.4%). Among women who had invasive cancer detected during screening interval, women with other symptoms were more likely to get a false negative result (odds ratio 1.79, 95% confidence interval 1.03 to 3.04) than asymptomatic women, after adjusting for age, use of HRT, and family history. Conclusion—The lower sensitivity in women with other symptoms requires further investigation. Possible explanations include increased breast density and poor image quality. The high sensitivity in women with significant symptoms is probably due to more cautious radiological practice, which has also resulted in low specificity in this group. (J Med Screen 2000;7:105–110) Keywords: mammography; breast neoplasms; positive predictive value

Despite the fact that screening can be defined as “the examination of asymptomatic people to classify them as likely, or unlikely, to have the disease that is the object of screening”,1 62% of Australian pamphlets promoting mammo-

www.jmedscreen.com

graphic screening do not indicate that screening is for asymptomatic women.2 In Victoria, Australia, promotional material indicates that the programme is for women without breast symptoms or problems. In Australia, where a national mammographic screening programme (BreastScreen Australia) was introduced in the early 1990s, significant symptoms are defined as a breast lump and/or blood stained or watery nipple discharge. The proportion of women who reported significant symptoms at screening varied from 0.7% to 7.5% across the states and territories.3 States and territories have diVerent policies for women with significant symptoms: some states do not discourage women with significant symptoms from participating, while other states do not accept them into their programme. In Victoria, women with symptoms are accepted but not encouraged to attend for screening. Since August 1995, women with significant symptoms are recalled for assessment even if their mammography is normal. Women with other symptoms (symptoms not classified as significant) are recalled for assessment if there is a mammographic abnormality, otherwise a letter is sent to the woman and her nominated general practitioner (GP), advising that she visit her doctor for a physical examination of her reported symptoms. In the United Kingdom, it is suggested that women report symptoms to their GP, however, many women with symptoms attend for screening (Patnick, J. Personal communication, 1998) and there are guidelines issued to radiographers about screening women with symptoms.4 In the Netherlands, the information leaflets do not state that women with symptoms should not attend for screening. The screening programme screens these women and makes follow up contact with the GPs (Schouten, L. Personal communication, 1998). If we accept women with symptoms into BreastScreen Victoria, it is appropriate to assess the performance of screening mammography for these women. To evaluate the accuracy of the screening programme in women with symptoms, we examined the sensitivity, specificity, and positive predictive value of the screening programme according to symptom status for women who attended for first round screening in 1994. Women with symptoms (significant and other symptoms) may be at higher risk of breast cancer, but the sensitivity and specificity of screening mammography are independent of the underlying incidence of breast cancer. However, sensitivity

106

Kavanagh, Giles, Mitchell, et al

and specificity are conceptually related: if the test conditions remain constant, an increase in sensitivity will result in a reduction in specificity, and the corollary is also true.1 Therefore, the sensitivity and specificity will only vary if the characteristics of the women with symptoms aVect the performance of screening mammography, or if there are particular policies in place (as is the case in Victoria) for the care of women with symptoms that favour an increase in specificity or sensitivity at the expense of the other. Methods The study population consists of all women who attended BreastScreen Victoria for screening mammography for the first time between 1 January 1994 and 31 December 1994. We excluded women who reported a personal history of breast cancer and women who were not resident in Victoria at the time of screening. DETAILS OF THE PROGRAMME

Details of the programme have been described before.5 Briefly, BreastScreen oVers free biennial, two view mammography through an organised screening service to women aged 40 and older. Women aged between 50 and 69 are the target group, and those who are listed on the electoral roll and who have never attended for screening are sent an invitation. Women aged 50 to 74 receive a reminder for rescreening 23 months after their previous screen if they have not made a further appointment. Although screening is biennial, women may attend for rescreening from 21 months after their previous screen. The films are read independently by two radiologists. If they disagree over whether to recall, the films are either reviewed by the two radiologists and a consensus reached, or a third reader makes the decision. Women are then recommended for routine rescreen or are referred for assessment. Assessment may include clinical examination, further radiographs, ultrasound, or biopsy. After assessment, women may have a cancer diagnosed, be recommended for routine rescreening, or be recommended for further assessment (early review). SELF REPORTED DATA

At the time women attend for screening, they complete a questionnaire giving details of previous mammography, family history of breast cancer, date of birth, use of hormone replacement therapy (HRT), and symptomatic status. For family history, women are asked if their mother, a daughter, or a sister had breast cancer, and the age at diagnosis. We define a strong family history as having a first degree relative diagnosed with breast cancer before the age of 50. Any other family history is recorded as “other family history”. In 1994, women were asked if they had a current breast lump, a current blood stained or watery nipple discharge, and to detail any other symptoms. It is diYcult to judge the importance of many of the self reported symptoms, as the questionnaire may prompt some women to report symptoms

www.jmedscreen.com

that are part of their normal breast cycle, and that had not been troublesome, and other women may fail to report significant symptoms. Nonetheless, for this analysis we refer to “significant symptoms” as a current breast lump and/or blood stained or watery discharge, and “other symptoms” as any other symptom, including breast pain and tenderness. POLICY FOR WOMEN WITH SIGNIFICANT SYMPTOMS

It is BreastScreen Victoria policy that a woman who reports a breast lump that has been present for less than than 12 months, which has not been investigated by a doctor, and/or a current blood stained or watery nipple discharge is judged to have highly suspicious symptoms and is recalled for assessment irrespective of the mammographic findings. This practice was first adopted in some services in 1992, although data regarding the length of time the lump had been present, and whether they had seen a doctor about the lump, were not routinely collected at all services until 11 August 1995. Therefore, the policy was not routinely implemented until August 1995. Before that time, some women with significant symptoms but normal mammography were referred back to their GPs (Sanders, P. BreastScreen Victoria Coordination Unit. Personal communication). In 1994, only 0.5% of women who had assessment were recalled because of significant symptoms without evidence of a mammographic abnormality.6 If women who had significant symptoms and a normal mammogram were referred to their GP for further investigation and were found to have invasive cancer this was called an interval cancer for the purposes of this study. CLASSIFICATION OF SUBJECTS

Subjects were classified into four categories: true negative, false positive, true positive, and false negative. True negative cases were women who did not develop cancer during the screening interval and who were not referred for assessment. False positive cases were women referred for assessment following screening, who did not have invasive cancer diagnosed. True positives were cases of invasive breast cancer diagnosed as a result of screening. False negatives were women who developed invasive breast cancer during the interval between attending for screening and the next recommended screening attendance (usually 24 months). Cancer cases only included invasive cancers because ductal carcinoma in situ is predominantly a screen detected lesion and will inflate estimates of sensitivity. The screening interval was defined as 24 months from the screening examination, except in the unusual circumstance where a woman died or moved between states. It is unlikely that migration out of the state would aVect the estimates of the interval cancer rate and sensitivity because less than 2% of people from Victoria move out of the state per year.7 Interval cancers were identified by linking the Victorian BreastScreen Registry records of

107

Value of screening mammography and symptomatic status Table 1 Symptom status according to age, hormone replacement therapy (HRT) status, and family history (%) No symptoms (n = 96420) (%)

Other symptoms (n = 7637) (%)

Significant symptoms (n = 2769) (%)

Age group* 40 to 49 50 to 59 60 to 69 70 to 79 80 and older

16456 (83.8) 40313 (91.1) 30431 (92.8) 8453 (91.2) 767 (87.8)

2060 (10.5) 2883 (6.5) 1939 (5.9) 668 (7.2) 87 (10.0)

1112 (5.7) 1079 (2.4) 414 (1.3) 144 (1.6) 20 (2.3)

Using HRT* No Yes

25924 (90.9) 70235 (88.7)

2408 (6.7) 5203 (8.2)

898 (2.4) 1865 (3.1)

Family history No family history Other family history Strong family history

87661 (90.3) 5706 (90.6) 3053 (89.2)

6919 (7.1) 452 (7.2) 266 (7.8)

2530 (2.6) 137 (2.2) 102 (3.0

* ÷2 Test p < 0.00001.

eligible women with the Victorian Cancer Registry records for all invasive breast cancers diagnosed between 1 January 1994 and 31 December 1996. Details of the matching process have been described before.5 Breast cancers are fast tracked by the cancer registry to ensure all cases are registered within six months of the end of the previous year. This match took place in September 1997. If invasive breast cancer was diagnosed at a subsequent screen, performed 21 to 24 months after a woman’s first screen, this was considered an interval cancer (false negative) if a breast lump or blood stained or watery nipple discharge in the aVected breast was reported at the time of the second screen.5 STATISTICAL METHOD

Sensitivity is the number of women with screen detected invasive cancer (true positive) divided by the sum of the screen detected (true positive) and interval (false negative) invasive cases. Specificity is the number of true negative cases divided by the true negatives and false positives. These measures of sensitivity and specificity are not direct measures of the sensitivity and specificity of the mammography per se because they are also dependent on the screening interval (sensitivity) and screening policies, such as the policy for women with significant symptoms. They are also not strictly measures of programme sensitivity or programme specificity because they are limited to first round attenders over a 2 year period, and measures of programme validity should be taken on the target population women over the duration of the programme.8 The positive predictive value is the likelihood of having invasive cancer if one is recalled for assessment (true positive cases

divided by the sum of the true positive and false positive cases). Sensitivity, specificity, and positive predictive value are expressed as percentages. Asymptotic 95% confidence intervals were calculated around these percentages, assuming a binomial distribution. Unconditional logistic regression modelling was used to control for potential confounders (age, HRT status, and family history). Age and family history might be associated with symptoms status and have been related to the accuracy of screening mammography.9–12 Women on HRT might be more likely to experience breast pain and tenderness (other symptoms),13 and HRT use has been associated with a reduction in the in the accuracy of screening mammography.14–21 Age was fitted as a continuous variable, and family history (no family history, other family history, and strong family history) and HRT use (yes or no) as categorical variables. We modelled the eVect of symptom status on the following: the proportion of false negatives among women who were diagnosed with cancer during the screening interval (analysis of sensitivity); the proportion of false positives among women who were not diagnosed with cancer during the screening interval (analysis of specificity); and the proportion of false positives among women referred for assessment (analysis of positive predictive value). After adjusting for confounders, we tested for interactions between symptomatic status and age, HRT status, and family history. A 5% significance level was used for all statistical tests. Results There were 106 826 women in this cohort. Table 1 shows the characteristics of the cohort by symptomatic status at screening. Most women (90.3%) were asymptomatic. Only 2.6% reported a breast lump or blood stained or watery nipple discharge (significant symptoms) and 7.1% reported other symptoms. Women aged between 40 to 49 years, and 80 and older were more likely to have other or significant symptoms than women aged between 50 and 79 years. Significant and other symptoms were more common in women who were using HRT than those who were not. Of the 65 women who reported other symptoms and had invasive cancer (screen detected or interval), 41 reported pain, discomfort, or tenderness; four reported inverted nipples (including one woman who also reported pain); and the remainder reported a variety of other symptoms including lumpy breasts, thickening, burning, and itching.

Table 2 Mammographic outcomes and sensitivity, specificity, and positive predictive value for women attending first round screening in 1994 according to breast symptom status (n = 106 826)

True negatives False positives True positives False negatives Screen detected cancer rate per 1000 (95% confidence interval (CI)) Interval cancer rate per 1000 (95% CI) Specificity (%) (95% CI) Sensitivity (%) (95% CI) Positive predictive value (%) (95% CI)

No symptoms

Other symptoms

Significant symptoms

90859 4906 495 160

7220 352 39 26

1988 708 59 14

5.1 (4.7 to 5.6) 1.7 (1.4 to 1.9) 94.9 (94.7 to 95.0) 75.6 (72.3 to 78.9) 9.2 (8.4 to 9.9)

www.jmedscreen.com

5.1 (3.5 to 6.7) 3.4 (2.1 to 4.7) 95.4 (94.9 to 95.8) 60.0 (48.1 to 71.9) 10.0 (7.0 to 12.9)

21.0 (15.7 to 26.3) 5.0 (2.4 to 7.6) 73.7 (72.1 to 75.4) 80.8 (71.8 to 89.9) 7.7 (5.8 to 9.6)

108

Kavanagh, Giles, Mitchell, et al

Table 3 Multiple logistic regression analyses of sensitivity, specificity, and positive predictive value Unadjusted odds ratio (95% confidence levels)

Adjusted odds ratio (95% confidence levels)*

Analysis of sensitivity Asymptomatic Other symptoms Significant symptoms

1.0 2.06 (1.22 to 3.49) 0.73 (0.40 to 1.35)

1.0 1.79 (1.03 to 3.14) 0.53 (0.28 to 1.00)

Analysis of specificity Asymptomatic Other symptoms Significant symptoms

1.0 0.91 (0.81 to 1.01) 6.57 (2.65 to 16.28)

1.0 0.89 (0.80 to 1.00) 6.36 (5.00 to 7.19)

Analysis of positive predictive value Asymptomatic Other symptoms Significant symptoms

1.0 0.91 (0.65 to 1.27) 1.22 (0.92 to 1.62)

1.0 0.75 (0.53 to 1.07) 0.90 (0.67 to 1.19)

*Adjusted for age, family history, and hormone replacement therapy status.

Table 2 shows the outcome of screening according to symptomatic status at screening. Sensitivity was lower in women with other symptoms than in asymptomatic women or women with significant symptoms. The sensitivity for women in the other symptoms category who reported breast pain or tenderness (n = 41) was 68.3% (95% confidence interval (CI) 54% to 83%). Specificity was more than 20% lower in women with significant symptoms than in asymptomatic women, or women with other symptoms. There were no statistically significant diVerences in the positive predictive values between any of the symptom categories. Table 3 shows the results of the logistic regression modelling. Among women who had cancer diagnosed during the screening interval, women with other symptoms were more likely to have a false negative screen. This eVect was slightly attenuated when we controlled for the confounding eVects of age, HRT status, and family history (adjusted odds ratio 1.79, 95% CI 1.03 to 3.14). After adjusting for confounding, women with significant symptoms were less likely to have a false negative screen (odds ratio 0.53, 95% CI 0.28 to 1.00) although the result was marginally insignificant. When compared with asymptomatic women, the risk of a false positive mammogram in women who did not have breast cancer was decreased in women with other symptoms (adjusted odds ratio 0.89, 95% CI 0.80 to 1.00) and increased in women with significant symptoms (adjusted odds ratio 6.36, 95% CI 5.00 to 7.19). There were no statistically significant diVerences in the risk of a false positive among women recalled for assessment between the diVerent symptom categories. None of the two way interactions between symptomatic status and age, HRT status, and family history were statistically significant in any of the logistic regression models. Discussion Our major finding is that the sensitivity of screening mammography for women with “other symptoms” is nearly 16% lower than asymptomatic women. Another significant finding is that women with significant symp-

www.jmedscreen.com

toms were more likely to get a false positive screening outcome (adjusted odds ratio 6.36, 95% CI 5.00 to 7.19), than women who did not have symptoms. Women with significant symptoms were less likely to have a false negative mammogram in the two year screening interval (odds ratio 0.53, 95% CI 0.28 to 1.00) than asymptomatic women. The reduction in sensitivity in the “other symptoms” category relative to asymptomatic women cannot be explained by a diVerence in the cancer prevalence in these two groups, because sensitivity and specificity are independent of the prevalence of disease. As the specificity is only slightly higher in the “other symptom” group, this is unlikely to account for the reduction in sensitivity we describe. Most women with “other symptoms” had breast pain or tenderness. Discomfort at the time of mammography may prevent adequate compression of the breast resulting in poor image quality.22 In order to obtain an adequate image of the breast, it is advised that compression be applied until the breast is taut.23 Firm compression of the breast is necessary for a number of technical reasons. Firstly, it immobilises the breast and eliminates motion blurring; it decreases the thickness of the breast and produces a uniform breast thickness that decreases scatter radiation and reduces the radiation dose; and by minimising the distance between breast structures and the film, and spreading the tissues apart, the resolution and sharpness of the image are optimised. One of the factors that has been related to pain or discomfort experienced at the time of mammography is pre-existing breast pain. Two studies have not found a significant association between breast tenderness within three days of mammography and discomfort at mammography.24 25 However, these were small studies: Cockburn et al had only seven women with pre-existing breast tenderness; and Polous and Rickard did not report how many women had pre-existing breast pain. Two other studies reported that women who have pre-existing breast pain are more likely to experience discomfort at the time of mammography.26 27 Stomper et al studied 1847 women, of whom 367 reported pre-existing breast tenderness. BreastScreen radiographers are taught to use firm compression during mammography to a level the women can comfortably tolerate, after an explanation of why compression is important in obtaining a good quality mammogram. If there is discomfort or pain, the process is aborted and further explanation and discussion is entered into until the woman is happy to proceed. Better compression can often be obtained with good explanation even if there is considerable discomfort. At the time of the study, the number of radiographers who had been fully trained by BreastScreen was limited, and it might be that less attention was paid to obtaining good compression with the woman’s informed consent. McNicholas et al found that breast pain was more common in women who had increased mammographic density, and high mammographic density has been associated with a

109

Value of screening mammography and symptomatic status

reduction in the sensitivity of screening.28 29 Previous studies have shown that use of HRT is associated with higher breast density for age,30–32 and breast pain has also been associated with use of HRT.13 However, we controlled for the eVect of HRT in our logistic regression analysis. It is particularly important to apply adequate compression in women who have high breast density to enable the separation of possibly superimposed densities,33 and for other technical reasons. In order to assess whether the image quality or breast density could explain the lower sensitivity one could blindly review mammographic films from women with “other symptoms” and compare these with asymptomatic women. Women with breast pain and tenderness comprise 63% of all women who were in the other symptom category and had a sensitivity that was slightly higher (68.3%) than that for the entire category (60%). The remainder of the women in the other symptoms category reported symptoms such as inverted nipples, lumpy breasts, thickening, burning, and itching. It is possible that these symptoms may also cause discomfort at the time of mammography, impairing the capacity of the radiographer to achieve a high quality image. The other major finding of this analysis is that women with significant symptoms had a significantly lower specificity than asymptomatic women due to the increased rate of recall for assessment. Many women with significant symptoms were recalled for clinical assessment even if their mammogram was normal. In addition, if radiologists know that a woman has reported significant symptoms at the time they read the mammogram, they might have a higher index of suspicion, and detect abnormalities that they would otherwise miss. And when they do detect an abnormality, they might be more likely to recall that woman for assessment than an asymptomatic women with the same abnormality. When the BreastScreen policy of referring all women with significant symptoms for clinical assessment regardless of symptomatic status at screening is routinely implemented, the specificity should drop to 0%, and the sensitivity will be even higher. Therefore, the increased rate referral of women with significant symptoms for clinical assessment probably explains the high sensitivity we describe. The uniform policy of referring all women with significant symptoms for clinical assessment regardless of screening outcome will decrease the positive predictive value of mammography. The positive predictive value would fall from 7.7% to between 2.1% (if no additional cancers were detected by this policy) and 2.6% (if all interval cancers were detected at clinical assessment). This low positive predictive value draws into question the value of performing screening mammography in this group. In conclusion, we have found a statistically significant reduction in the sensitivity of screening mammography in women with “other symptoms”, which were predominantly breast pain and tenderness. Possible explanations for this reduction might be inadequate breast com-

www.jmedscreen.com

pression leading to poor image quality or increased breast density. Further research is required to establish the reasons for this finding. However, it is important to inform women presenting with pain or tenderness that a satisfactory mammographic image requires a reasonable amount of compression of the breast. We are grateful for the input, support, and guidance of the Victorian BreastScreen Coordination Unit, particularly Pauline Sanders and Onella Stagoll. We are also indebted to Helen Farrugia, Vicky Higgins, Ester Cukier, and Georgina Marr for their assistance. Finally, we acknowledge the work and commitment of the staV of the screening and assessment units in Victoria, Australia. 1 Morrison A. Screening in chronic disease. New York: Oxford University Press, 1992. 2 Ward JE, Slaytor EK. Outcome measures of an Australian breast-screening program. Med J Aust 1999;170:142–3. 3 Kavanagh A, Amos A, Marr G. The ascertainment and reporting of interval cancers within the BreastScreen Australia Program. Sydney: National Breast Cancer Centre, 1999. 4 NHS Breast Screening Program. Information and advice for radiographers. NHSBSP, 1997. 5 Kavanagh AM, Mitchell H, Farrugia H, et al. Monitoring interval cancers in an Australian mammographic screening programme. J Med Screen 1999;6:139–43. 6 BreastScreen Victoria. Annual Statistical Report 1994. Melbourne, 1995. 7 Australian Bureau of Statistics. Australian Demographic Statistics. Canberra: Australian Bureau of Statistics, 1999. 8 Hakama M. Screening. In: Holland W, Detels R, Knox G, eds. Oxford textbook of public health. Oxford: Oxford University Press, 1991:91–106. 9 Kaizer L, Fishell EK, Hunt JW et al. Ultrasonographically defined parenchymal patterns of the breast: relationship to mammographic patterns and other risk factors for breast cancer. Br J Radiol 1988;61:118–24. 10 Kerlikowske K, Grady D, Barclay J et al. Positive predictive value of screening mammography by age and family history of breast cancer. JAMA 1993;270:2444–50. 11 Kerlikowske K, Grady D, Barclay J et al. EVect of age, breast density and family history on the sensitivity of first screening mammography. JAMA 1996;276:33–8. 12 Peer PG, Verbeek AL, Straatman H et al. Age-specific sensitivities of mammographic screening for breast cancer. Breast Cancer Res Treat 1996;38:153–60. 13 Callantine MR, Martin PL, Bolding OT et al. Micronized 17 beta-estradiol for oral estrogen therapy in menopausal women. Obstet Gynecol 1975;46:37–41. 14 Laya MB, Larson EB, Taplin SH et al. EVect of estrogen replacement therapy on the specifity and sensitivity of screening mammography. J Natl Cancer Inst 1996;88:643– 9. 15 Kavanagh AM, Mitchell H, Giles GG. Hormone replacement therapy and accuracy of mammographic screening. Lancet 2000;355:270–4. 16 Litherland JC, Evans AJ, Wilson AR. The eVect of hormone replacement therapy on recall rate in the National Health Service Breast Screening Programme. Clinic Radiol 1997; 52:276–9. 17 Thurfjell EL, Holmberg LH, Persson IR. Screening mammography: sensitivity and specificity in relation to hormone replacement therapy. Radiology 1997;203:339– 41. 18 Cohen ML, Johnson A, Shekdar J. EVect of hormone replacement therapy on cancer detection by mammography [letter]. Lancet 1997;349:1624. 19 Litherland J, Stallard S, Hole D et al. The eVect of hormone replacement therapy on the sensitivity of screening mammograms. Clin Radiol 1999;54:285–8. 20 Rosenberg RD, Hunt WC, Williamson MR et al. EVects of age, breast density, ethinicity, and estrogen replacement therapy on screening mammographic sensitivity and cancer stage at diagnosis: review of 183 134 screening mammograms in Albuquerque, New Mexico. Radiology 1998;209: 511–8. 21 Seradour B, Esteve J, Heid P et al. Hormone replacement therapy and screening mammography: analysis of the results in the Bouches du Rhone programme. J Med Screen 1999;6:99–102. 22 McNicholas MM, Heneghan JP, Milner MH et al. Pain and increased mammographic density in women receiving hormone replacement therapy: a prospective study. AJR Am J Roentgenol 1994;163:311–5. 23 Elkund GW. Mammographic compression: science or art? Radiology 1991;181:339–41. 24 Cockburn J, Cawson J, Hill D et al. An analysis of reported discomfort caused by mammographic X-ray amongst attenders at an Australian pilot breast screening program. Australas Radiol 1992;36:115–9. 25 Poulos A, Rickard M. Compression in mammography and the perception of discomfort. Australas Radiol 1997;41: 247–52. 26 Stomper PC, Van Voorhis BJ, Ravnikar VA et al. Mammographic changes associated with postmenopausal hormone replacement therapy: a longitudinal study. Radiology 1990; 174:487–90.

110

Kavanagh, Giles, Mitchell, et al 27 Brew MD, Billings JD, Chisolm RJ. Mammography and breast pain. Australas Radiol 1989;33:335–6. 28 Ma L, Fishell E, Wright B et al. Case-control study of factors associated with failure to detect breast cancer by mammography. J Natl Cancer Instit 1992;84:781–5. 29 Van Gils CH, Otten JD, Verbeek AL et al. EVect of mammographic breast density on breast cancer screening performance: a study in Nijmegen, The Netherlands. J Epidemiol Community Health 1998;52:267–71. 30 Bergkvist L, Tabar L, Adami HO et al. Mammographic parenchymal patterns in women receiving noncontra-

www.jmedscreen.com

ceptiveceptive estrogen treatment. Am J Epidemiol 1989; 130: 503–10. 31 Leung W, Goldberg F, Zee B, et al. Mammographic density in women on postmenopausal hormone replacement therapy. Surgery 1997;122:669–74. 32 Rand T, Heytmanek G, Seifert M, et al. Mammography in women undergoing hormone replacement therapy. Possible eVects revealed at routine examination. Acta Radiol 1997;38:228–31. 33 Elkund GW. Compression: the mammographer’s friend or foe? Mammography today 1995:220–2.