Vol. 7, No. 4
Family Practice ©Oxford University Press 1990
Printed in Great Britain
Selections from Current Literature: Mammographic Breast Cancer Screening PETER J. BOWERS
THE HEALTH INSURANCE PLAN (HIP) OF GREATER NEW YORK STUDY Shapiro S, Venet W, Strax P, Venet L. Current Results of the Breast Cancer Screening Randomized Trial: The Health Insurance Plan (HIP) of Greater New York Study. In: Day NE, Millar AB, eds. Screening for Breast Cancer. Toronto, Ontario: Hans Huber Publishers; 1988: 3-15. The HIP Study was a benchmark in screening for cancer trials and greatly enhanced our understanding of self-selection, lead and length time bias. Women (members of HIP) aged 40 to 64 years at entry were randomized to control and intervention groups of approximately 31 000 each. Intervention group women were invited to have physical examinations of their breasts and two view (cephalocaudal and lateral) mammography yearly, for four years. About 67% of those invited for breast examination and mammogTaphy presented for the first screening round. Of the total intervention group about 50% completed three or more screening examinations. Control group women received their usual medical care. The outcome measure of crucial importance was the number of deaths due to breast cancer in the intervention and control groups. Cause of death was determined by review of the death certificate and hospital and physicians' records. Data identifying the patient, and whether the woman was in the control or intervention group, were eliminated from all reviewed records. To avoid dilution of the results by deaths due to breast cancers occurring after screening had concluded, results have included only those women whose breast cancers were diagnosed within 5 years, 7 years or 10
Department of Family Medicine, State University of New York at Stony Brook, Stony Brook, NY 11794-8461, USA.
333
Downloaded from http://fampra.oxfordjournals.org/ at University of Sussex on September 9, 2015
from mammography. Other, non-randomized studies7"10 have purported to demonstrate a benefit in favour of women receiving mammography. For this issue of selections from Current Literature the three randomized controlled trials, as well as several related reports and commentaries (some critical), are reviewed.
Eleven major US medical organizations, including the American Academy of Family Physicians, American Medical Association and National Cancer Institute, recently endorsed new guidelines for the earlier detection of breast cancer.1 The organizations agreed that: (i) asymptomatic women should begin having screening mammography by age 40 years, (ii) asymptomatic women, aged 40 to 49 years, should undergo annual clinical breast examination and screening mammograms every one to two years. (iii) asymptomatic women 50 years and older should have both clinical breast examination and screening mammograms yearly. The report of the US Preventive Services Task Force2 differs in that mammography is not recommended routinely for screening for breast cancer in women under 50 years, or those over 75 years of age. This group recommends mammography every 'one to two years' for those aged 50 to 75 years. Additionally, clinical breast examination is recommended annually for all women over 40 years of age.2 Recommendations regarding mammography are of interest to family physicians because, at the present time, the only practical means of potentially reducing the significant numbers of deaths due to breast cancer is earlier detection (and treatment) using mammography and clinical examination of the breast. An estimated 44000 American women will die of breast cancer in 1990,3 and approximately 150 000 new cases will be diagnosed during the same period.3 Advocates of mammographic screening claim that strict adherence to the recommendations outlined above will reduce the numbers of women dying of this disease. These claims rest, in the main, upon results of two major randomized controlled trials, the Health Insurance Plan (HIP) Study4 and the Swedish 'two county' Study.5 The only other published randomized controlled study, the 'Malmo' study,6 showed no benefit
334
FAMILY PRACTICE—AN INTERNATIONAL JOURNAL
Comment The HIP trial was the first randomized controlled trial of screening for breast cancer to report a benefit for the intervention group, and as such has been hailed as a landmark. A number of questions, however, come to mind about the trial. These concern the number of women in the trial, the determination of the cause of death from official records, and lack of an independent end point committee. An additional concern is the use of relative vs. absolute risk reduction. The exact numbers of women participating in the trial is unclear. Various papers from the original study group, and others who have analyzed their data, report different numbers of participants. A 1967 paper" gave the numbers in Table I. TABLE I
Group Study Control Initial Screening Not screened, first year
20211' 11677"
Total
31888
29694* 29 694*
*No. of women screened. **Person-years of exposure; not all women have been followed a full year.
The total number of women in the study group is at least 31888 (20 211 and 11677) according to the table. A later report from the HIP group stated 'The total number of women in each sample is 31000 .. ,'12 Shapiro reported that women'. . . were allocated to either a study or a control group consisting of about 31000 each. In a letter to the editors of Surgery'4 Shapiro wrote'.. . the source for the 31 888 figure is not stated; it is approximately 1000 higher than the sample size in the project.'14 Aron and Prorok, who analyzed the HIP data in a 1986 report, state 'the study population comprised 30 239 women who were offered an initial screen and three follow-up screens at annual intervals; the control population consists of 30 765 women.'16 Differences of one thousand or so in the denominator (the total number of women in each group) are unlikely to alter the significance of the findings. However, even small changes in the numerator (the number of deaths due to breast cancer) have the potential to reduce or eliminate the significant difference between the two groups. The obfuscation of the exact number of women in the trial is, therefore, unsettling. An explanation for these differences would be reassuring. The cause of death of women in the HIP trial was determined by a committee after a review of death certificates, and hospital and medical records. Although blinded to the identity of the woman and her randomization to control or study group, the committee was apparently not independent of the research team: the potential for bias in the classification of cause of death therefore exists. Additionally, official records, in particular death certificates, have been shown to be an inaccurate source of information as to the cause of death in breast cancer patients.6 A Swedish study6 reported a 10% discordance between the numbers of deaths due to breast cancers in official statistics and deaths classified by an independent end point committee. A misclassification of just 2% of total the number of deaths in either group of the HIP Study would account for the entire benefit. This author would have more confidence in the HIP trial results if the committee had been independent of the research team, and the cause of death verified at autopsy. Finally, placing the reduction in deaths due to breast cancer in context is important. Froom has recently written about relative and absolute risk reduction. Although the relative risk reduction of death due to breast cancer in the intervention group was an impressive 23.4%, the absolute risk reduction was just 0.145%. There were 147 breast cancer deaths in an intervention group of 31000 women, compared with 192 breast cancer deaths in the control group, over a 10-year period. The total number of deaths in the intervention group women over the same period was 2150.14 THE SWEDISH 'TWO COUNTRY STUDY' Tabar L, Fagerberg CJG, Gad A, et al. Reduction in mortality from breast cancer after mass screening with mammography: randomized trial from the breast
Downloaded from http://fampra.oxfordjournals.org/ at University of Sussex on September 9, 2015
years of entry into the trial. Results at 18 years follow-up are now available. For breast cancers diagnosed within 10 years of entry into the trial, at 10 years post entry there were 45 fewer deaths (23.4%) in intervention group women than in the control group (0.01
Family Practice ©Oxford University Press 1990
Printed in Great Britain
Selections from Current Literature: Mammographic Breast Cancer Screening PETER J. BOWERS
THE HEALTH INSURANCE PLAN (HIP) OF GREATER NEW YORK STUDY Shapiro S, Venet W, Strax P, Venet L. Current Results of the Breast Cancer Screening Randomized Trial: The Health Insurance Plan (HIP) of Greater New York Study. In: Day NE, Millar AB, eds. Screening for Breast Cancer. Toronto, Ontario: Hans Huber Publishers; 1988: 3-15. The HIP Study was a benchmark in screening for cancer trials and greatly enhanced our understanding of self-selection, lead and length time bias. Women (members of HIP) aged 40 to 64 years at entry were randomized to control and intervention groups of approximately 31 000 each. Intervention group women were invited to have physical examinations of their breasts and two view (cephalocaudal and lateral) mammography yearly, for four years. About 67% of those invited for breast examination and mammogTaphy presented for the first screening round. Of the total intervention group about 50% completed three or more screening examinations. Control group women received their usual medical care. The outcome measure of crucial importance was the number of deaths due to breast cancer in the intervention and control groups. Cause of death was determined by review of the death certificate and hospital and physicians' records. Data identifying the patient, and whether the woman was in the control or intervention group, were eliminated from all reviewed records. To avoid dilution of the results by deaths due to breast cancers occurring after screening had concluded, results have included only those women whose breast cancers were diagnosed within 5 years, 7 years or 10
Department of Family Medicine, State University of New York at Stony Brook, Stony Brook, NY 11794-8461, USA.
333
Downloaded from http://fampra.oxfordjournals.org/ at University of Sussex on September 9, 2015
from mammography. Other, non-randomized studies7"10 have purported to demonstrate a benefit in favour of women receiving mammography. For this issue of selections from Current Literature the three randomized controlled trials, as well as several related reports and commentaries (some critical), are reviewed.
Eleven major US medical organizations, including the American Academy of Family Physicians, American Medical Association and National Cancer Institute, recently endorsed new guidelines for the earlier detection of breast cancer.1 The organizations agreed that: (i) asymptomatic women should begin having screening mammography by age 40 years, (ii) asymptomatic women, aged 40 to 49 years, should undergo annual clinical breast examination and screening mammograms every one to two years. (iii) asymptomatic women 50 years and older should have both clinical breast examination and screening mammograms yearly. The report of the US Preventive Services Task Force2 differs in that mammography is not recommended routinely for screening for breast cancer in women under 50 years, or those over 75 years of age. This group recommends mammography every 'one to two years' for those aged 50 to 75 years. Additionally, clinical breast examination is recommended annually for all women over 40 years of age.2 Recommendations regarding mammography are of interest to family physicians because, at the present time, the only practical means of potentially reducing the significant numbers of deaths due to breast cancer is earlier detection (and treatment) using mammography and clinical examination of the breast. An estimated 44000 American women will die of breast cancer in 1990,3 and approximately 150 000 new cases will be diagnosed during the same period.3 Advocates of mammographic screening claim that strict adherence to the recommendations outlined above will reduce the numbers of women dying of this disease. These claims rest, in the main, upon results of two major randomized controlled trials, the Health Insurance Plan (HIP) Study4 and the Swedish 'two county' Study.5 The only other published randomized controlled study, the 'Malmo' study,6 showed no benefit
334
FAMILY PRACTICE—AN INTERNATIONAL JOURNAL
Comment The HIP trial was the first randomized controlled trial of screening for breast cancer to report a benefit for the intervention group, and as such has been hailed as a landmark. A number of questions, however, come to mind about the trial. These concern the number of women in the trial, the determination of the cause of death from official records, and lack of an independent end point committee. An additional concern is the use of relative vs. absolute risk reduction. The exact numbers of women participating in the trial is unclear. Various papers from the original study group, and others who have analyzed their data, report different numbers of participants. A 1967 paper" gave the numbers in Table I. TABLE I
Group Study Control Initial Screening Not screened, first year
20211' 11677"
Total
31888
29694* 29 694*
*No. of women screened. **Person-years of exposure; not all women have been followed a full year.
The total number of women in the study group is at least 31888 (20 211 and 11677) according to the table. A later report from the HIP group stated 'The total number of women in each sample is 31000 .. ,'12 Shapiro reported that women'. . . were allocated to either a study or a control group consisting of about 31000 each. In a letter to the editors of Surgery'4 Shapiro wrote'.. . the source for the 31 888 figure is not stated; it is approximately 1000 higher than the sample size in the project.'14 Aron and Prorok, who analyzed the HIP data in a 1986 report, state 'the study population comprised 30 239 women who were offered an initial screen and three follow-up screens at annual intervals; the control population consists of 30 765 women.'16 Differences of one thousand or so in the denominator (the total number of women in each group) are unlikely to alter the significance of the findings. However, even small changes in the numerator (the number of deaths due to breast cancer) have the potential to reduce or eliminate the significant difference between the two groups. The obfuscation of the exact number of women in the trial is, therefore, unsettling. An explanation for these differences would be reassuring. The cause of death of women in the HIP trial was determined by a committee after a review of death certificates, and hospital and medical records. Although blinded to the identity of the woman and her randomization to control or study group, the committee was apparently not independent of the research team: the potential for bias in the classification of cause of death therefore exists. Additionally, official records, in particular death certificates, have been shown to be an inaccurate source of information as to the cause of death in breast cancer patients.6 A Swedish study6 reported a 10% discordance between the numbers of deaths due to breast cancers in official statistics and deaths classified by an independent end point committee. A misclassification of just 2% of total the number of deaths in either group of the HIP Study would account for the entire benefit. This author would have more confidence in the HIP trial results if the committee had been independent of the research team, and the cause of death verified at autopsy. Finally, placing the reduction in deaths due to breast cancer in context is important. Froom has recently written about relative and absolute risk reduction. Although the relative risk reduction of death due to breast cancer in the intervention group was an impressive 23.4%, the absolute risk reduction was just 0.145%. There were 147 breast cancer deaths in an intervention group of 31000 women, compared with 192 breast cancer deaths in the control group, over a 10-year period. The total number of deaths in the intervention group women over the same period was 2150.14 THE SWEDISH 'TWO COUNTRY STUDY' Tabar L, Fagerberg CJG, Gad A, et al. Reduction in mortality from breast cancer after mass screening with mammography: randomized trial from the breast
Downloaded from http://fampra.oxfordjournals.org/ at University of Sussex on September 9, 2015
years of entry into the trial. Results at 18 years follow-up are now available. For breast cancers diagnosed within 10 years of entry into the trial, at 10 years post entry there were 45 fewer deaths (23.4%) in intervention group women than in the control group (0.01
