Overdiagnosis in mammography screening for breast cancer in Sweden and Norway Senior Statistician Per-Henrik Zahl, MA MD PhD Norwegian Institute of Public Health Professor Jan Mæhlen, MD PhD Department of Pathology, Ullevål University Hospital July 29th, 2004
The aim of mammography screening is to diagnose the tumour when it is small and before tumour spread occurs since such small stage I tumours have favourable prognosis. For staging of breast cancer see http://www.breastcancer.org/dia_pict_staging_pf.html. Relative survival rates are calculated for Norwegian women diagnosed in the period 1992-96 ( based on approximately 10 000 cancer diagnosis). The idea that early diagnosis is beneficial is based on common assumptions including: breast cancer at stage I never regress, with time an increasing proportion of stage I cancers progress to a more advanced stage leading to clinical symptoms. From these assumptions one may deduce that early diagnosis (by screening) will lead to an incidence increase for stage I disease and an incidence decline for stage II-IV disease.
Mammogramme with small breast cancer (T) Many small tumours (stage I) are detected at the mammography screening Mammogramme with small breast cancer (T) Macroscopic picture of excised tumor (T) T T http://medlib.med.utah.edu/WebPath/
Background Prior to the introduction of mammography screening in Sweden and Norway the breast cancer incidence increased less than 1% annually (Rostgaard et al, Stat Med 2001) Coinciding with the introduction of mammography screening the incidence of breast cancer in the invited age group 50-69 years increased about 50% (Zahl, Strand, Mæhlen BMJ 2004) Sweden started several large mammography screening trials in the late 1970’s, and in 1986 they started a nation-wide screening programme for breast cancer. Norway started mammography screening in 1996. All women aged 50-69 years are invited to mammography screening in both countries. In addition, about 50% of women aged 40-49 and 70-74 years are invited to mammography screening in Sweden. Opportunistic screening was frequent in the capital area of Norway (Oslo and Akershus county) before 1996 (where about 20% of the population lives). More than 92% of women aged 50-69 years have been invited to screening since 1991 in Sweden, and about 40% of women in Norway were invited to mammography screening in Norway in the period 1996-99. The annual increase in breast cancer incidence was less than 1% before the mid 1970 in Sweden. After the mid 1970’s the breast cancer incidence increased more than 1% annually and the incidence increase was mainly in the age group invited to screening in Sweden. At the same the annual increase in breast cancer incidence in Norway was less than 1%. Norway started mammography 10 years after Sweden. The incidence increases in first, second and third screening round were 82%, 54% and 56%, respectively.
Ductal breast carcinoma Lobular breast carcinoma Several histological types of infiltrating breast cancer exist, see http://medlib.med.utah.edu/WebPath/ORGAN.html#1. Generally the clinical tumour stage is far more important for the prognosis than the histological type.
Definition: Overdiagnosis is the diagnosis of disease that (without screening) would not have given symptoms during the lifetime of the patient Overdiagnosed invasive tumours may be slowly growing tumours, or tumours that normally undergo spontaneous regression at an early stage (probably before the tumours disseminate).
May these three factors explain the incidence increase? There was an underlying increase in breast cancer incidence rates in the periods before the screening programmes started. In the first screening round a large number of slow-growing tumours that would have occurred in the future are diagnosed earlier. In the following screening rounds some excess of breast cancer will persist due to a combined effect of lead-time and incidence increase with age. The underlying annual increase in breast cancer rates in Norway and Sweden before the screening programmes started was 1% or less. For example, the incidence increase in the age group 0-49 years was 0.8% in the period 1990-2001 in Norway. In first screening round many cancers are detected that normally would have been detected later. In second and later screening rounds, the incidence rate in the age group invited to screening is expected to be higher than before the screening programme started due to a combined effect of earlier diagnosis and that the breast cancer incidence is increasing with age.
While the Swedish mammography screening programme started in 1986, the Norwegian programme started in 1996. The slide shows the per cent of women in the age group 50-69 years that were invited in organised mammography screening programmes in Sweden (blue) and Norway (red). In addition some opportunistic (non-organized) screening has occurred in both countries.
Estimation of the level of overdiagnosis Overdiagnosis is calculated as the incidence increase in the screened age group minus the subsequent incidence decline that occurs when these women reach the age when they are no longer invited to screening. If no overdiagnosis occurs, this difference should be zero. The level of overdiagnosis can also be estimated as the proportion of the total cancer cases that would not have been diagnosed in the absence of screening. Overdiagnosis is essentially the incidence increase due to screening and may be defined statistically. Earlier diagnosis will always lead to an incidence increase and some overdiagnosis. Overdiagnosis cannot (at present) be defined on the individual level.
During the introduction of mammography screening in Sweden the breast cancer incidence increased by nearly 50% in the screened age group. No similar fall in breast cancer incidence in women above age 70 years has emerged. Age-specific incidence of invasive breast cancer in Sweden in the period 1971-2002. The annual incidence increase in the period 1971-1985 was about 1%. The incidence increase in 1976-77 is probably due the start of the mammography screening trials. After 1986 when the organized screening started, a marked incidence increase occurred in the invited age group 50-69 years. More than 90% of women aged 50-69 years have been invited since 1991. The incidence increase in the last part of the 1990’s may be due partly to improved techniques (double picture, computer assisted reading and the use of quality manuals in the screening programme) and partly to the start of mammography screening in the last counties. Up to 2002 no similar fall in breast cancer incidence in older women has emerged. If the women above 69 years are grouped into 5-years age groups, no significant incidence reduction occurred in women aged 70-74 years (in several Swedish counties women in the age group 70-74 have been invited to screening), while a 12% incidence decline occurred in women aged 75-79 years in 1997-2002.
Overdiagnosis in the Swedish screening programme The incidence increase is 50% in the age group 50-69 years. The incidence decline is 0% in the age group 70-74 years and 12% in the age group 75-79 years. This decrease compensates less than 3% of the incidence increase in the screened age group. No incidence decline is expected among women above age 79 years. We conclude that nearly all of the incidence increase in Sweden are caused by cancers that would not have been diagnosed in the absence of screening.
During the introduction of mammography screening in Norway the breast cancer incidence increased >50% in the screened age groups. Only an 11 % reduction in breast cancer incidence occurred in previously screened women aged 70-74 yrs in 2000-1. Age-specific incidence of invasive breast cancer in Norway in the period 1971-2001. The annual increase in the period 1971-1995 was about 0.6%. In 1996 a mammography screening programme was started in four major counties (Akershus, Oslo, Rogaland and Hordaland – the AORH-counties) where about 40% of the population lives. First screening round was in 1996-7; the second screening round in 1998-9 and the third screening round was in 2000-1. Compared to the period 1991-5 the breast cancer incidence in the age group 50-69 years was 80% higher during the first (prevalence-) screening. Compared to the period 1991-5 the breast cancer incidence in the age group 50-69 years in the second and third screening round was increased by 54% and 52% respectively. Note that screening started several years later in the non-AORH-counties. The increase in breast cancer incidence only occurred for stage I disease. No decline in the incidence of stage II-IV has been observed in the Norwegian (or any other) screening programme.
Table 1
Overdiagnosis in the Norwegian screening programme (AORH-counties) The incidence increase is 52-54% in the age group 50-69 years. The incidence decline is 13% in the age group 70-74 years. This decline compensates only 3% of the increase in the screened age group. No incidence decline is expected in women above age 74 years. We conclude that nearly all of the incidence increase in Norway are caused by cancers that would not have been diagnosed in the absence of screening.
Incidence rates among those attending and those not attending screening in the age group 50-69 years in four Norwegian counties Incidence of invasive breast cancer for the age group 50-69 years in the AORH-counties in Norway. The first column shows the incidence rate in the period 1991-5; i.e. before women were invited to mammography screening. The second column shows the incidence rate for those not attending screening in the six first year of the screening program. The two columns to the right show incidence rates for those attending the second and third screening round (the incidence rate in the prevalence screening in 1996-97 is not shown). Ductal carcinoma in situ (DCIS) has not been included in the figure. If we add the invasive cancer rate and the DCIS rate, then the combined incidence rate is 400; women who attend screening have doubled their risk to be diagnosed with breast cancer.
Definition of interval cancer Interval cancer is breast cancer detected clinically in the time interval between two mammography screenings. The incidence rate of interval cancer in the Norwegian screening programme has been 190 cases per 100 000 per 2 year (i.e. close to 50% of the predicted rate in the absence of screening). Compared to the incidence rate of cancer in the absence of screening, the rate of interval cancer is 30% in the first year and 70% in the second year after screening. The expected rate of invasive breast cancer over at two year period for the age group 50-69 years in Norway is 400 in the absence of mammography screening. The observed cumulative rate of invasive breast cancer in a 2 year period after a mammography screening in Norway is 190. The absolute number of interval cancers is very similar in different national screening programmes (for example in the Dutch screening programme the rate of interval cancer was 180 cases per 100 000 per 2 year). For un-known reasons the rate of interval cancers in the Swedish Two-County Study was only half of this number. After a group of women have been screened their rate of breast cancer is typically reduced by 70% the first year, 30% the second year. The rate is reduced by only 15% the third year and tend to the background level after four years. Therefore, time between screening rounds is 2 or 3 years.
Cancer incidence in those leaving the screening program at age 69 years reflects the interval cancer incidence If the incidence reduction is 70% in the first year, 30% in the second year, 15 % in the third year, 5% in the fourth year and return to the background in the fifth year, the overall reduction in age group 70-74 year is 13% (assuming 75% attendance rate at the last screening at age 68-69 years).
Lead-time Lead-time is defined as the time the diagnosis is brought forward by the screening. The average lead-time in the randomized trials has been estimated to about 4 years. When we adjust for overdiagnosis in the Norwegian programme, we estimated lead-time to be 1.2 years. The high rate of interval cancer at the end of a screening interval support the idea that the lead-time is short. If the tumour can disseminate not only in the beginning of the tumours life but during the whole the life time, then earlier diagnosis will lead to a reduction in number of cancers with metastases. Long average lead-time will then be associated with a lower rate of cancers diagnosed with metastases (which may lead to a reduction in mortality). The average lead-times in the randomized trials have been estimated to be about 4 years. However, these calculations are based on the assumption that a: all tumours grow and b: that the slow-growing tumours have very low sensitivity. Multiple screening rounds will then empty the reservoir of slow-growing tumours and lead to a strong decline in incidence rates when women are not invited to screening any more. When the randomized screening trials were stopped, mammography screening started immediately in the control group. Therefore, these studies cannot be used to calculate incidence rates when women are not invited to mammography screening. The calculation of lead-time in the randomized trails is therefore very speculative because they have not observed the lead-time.
Is it sensible to expect a mortality reduction when The average lead time in the Norwegian screening programme after adjusting for overdiagnosis is much smaller than in the screening trials; 1.2 years vs. 4 years. The rate of interval cancer in the Norwegian screening programme is much higher than in the screening trials (the rate in the Norwegian programme is twice the rate in the Swedish WE-study). Our estimated average lead-time is based on incidence rates after 2 screening rounds (Norway) or 5 screening rounds or more (Sweden). We compare with the expected incidence rates when we calculate the average lead-time, and have adjusted for an underlying incidence increase by using regression analysis modelling the underlying incidence increase. Because the observed average lead-time in the screening programme is smaller than in the randomized screening trials and that the rate of interval cancers is higher, it is likely that reduction in breast cancer mortality rates will be small than in the randomized screening trials (mortality reduction in the screening trials varies between 0 and 30%).
As shown in the next slide none of these phenomena have been observed. In the randomized screening trials the mortality decline was observed after four years. Since Sweden started mammography screening more than ten years before Norway one would expect: In the 1990s the decline in the Swedish breast cancer mortality rates should be substantially larger than the decline in the Norwegian rates. In Sweden the breast cancer mortality rates should decline substantially more in the age group 55-74 years than in the other age groups. As shown in the next slide none of these phenomena have been observed.
Age-adjusted breast cancer mortality rates in Norway and Sweden in 1978-2001 for the age groups 45-54, 55-74 and 75-84 years Breast cancer mortality rates are slowly falling in both countries (the annual decline is about 1% after 1990). There is no significant differences in trends between Sweden and Norway.
What is known about overdiagnosed cancers in other locations? Screening for prostate cancer leads to 30% overdiagnosis (Etzioni et al, JNCI 2002). In this organ sub-clinical cancer is very frequent at autopsy (prevalence 50% after age 80 years, Liavag I, Harbitz TB, Haugen OA. Latent carcinoma of the prostate. Recent Results Cancer Res. 1972;39:131-7). Screening of children for neuroblastoma leads to substantial overdiagnosis. For such tumours spontaneous regression without treatment is the most likely outcome, therefore these patients do not benefit from earlier diagnosis (Schilling et al, N Eng J Med 2002).
After three biennial screenings in Norway the accumulated breast cancer incidence was substantially higher than the accumulated breast cancer incidence after prevalence screening of previously un-screened women. This observation suggests that in the absence of screening many small and localized invasive breast cancers undergo spontaneous regression. Some invasive cancers are naturally regressing. This is well documented for neuroblastoma in children but has also been observed in malignant melanoma and other adult cancers.
Summary Overdiagnosis of breast cancer in national mammography screening programmes is substantial. In Sweden the decline in breast cancer mortality rates after 10-15 years of screening is much smaller than expected from the randomized trials. In fact, a very similar small decline in breast cancer mortality has also occurred in Norway before the screening started. Our results show that many small tumours detected by mammography behave as benign lesions although they fulfill the histological criteria for invasive cancers. We propose that such lesions normally undergo spontaneous regression.