- Email: [email protected]
Service screening with mammography of women aged 70–74 years in Sweden
Cancer Detection and Prevention 27 (2003) 360–369
Service screening with mammography of women aged 70–74 years in Sweden Effects on breast cancer mortality Håkan Jonsson, PhD a,∗ , Sven Törnberg, MD, PhD b , Lennarth Nyström, PhD c , Per Lenner, MD, PhD a a
Department of Oncology, Umeå University, Umeå, Sweden Oncologic Centre, Karolinska Hospital, Stockholm, Sweden Epidemiology, Public Health and Clinical Medicine, Umeå University, Umeå, Sweden b
c
Accepted 24 July 2003
Abstract Since the benefit of mammography screening for women 70 years and older is unclear, the aim of the present study was to evaluate the effect on breast cancer mortality of the population-based service-screening program in Sweden inviting women 70–74 years. Among the counties with service-screening programs in Sweden which started 1986–1990 those with upper age limit 74 years were compared to counties with 69 years as upper age limit with respect to refined breast cancer mortality. Allowance was made for potential biases namely inclusion of cases diagnosed before invitation and lead time. Two methods for estimation of breast cancer mortality were used; underlying cause of death (UCD) and excess mortality. With a mean follow-up of 10.1 years a reduction of the breast cancer excess mortality was estimated at 24%. Using the underlying cause of death the corresponding result was 6%. A non-significant reduction in breast cancer mortality was found in the counties with service-screening program including the age group 70–74 years in Sweden. The estimated reduction was larger when using excess mortality compared to the use of individual underlying cause of death. © 2003 International Society for Preventive Oncology. Published by Elsevier Ltd. All rights reserved. Keywords: Breast cancer; Mortality; Mammography screening; Evaluation
1. Introduction Randomized studies have shown that screening for breast cancer with mammography reduces the breast cancer mortality [1,2], especially for women aged 50–69 years at invitation to screening. Theoretically, screening should be beneficial also for women aged 70 years or more [3,4] however, there is a lack of empirical evidence to support this statement. The only randomized screening trial that included women 70 years and older was the Swedish Two-County trial [1]. Today nationwide service-screening programs have been introduced in Sweden, Finland, The Netherlands, UK, Iceland and Luxembourg but only in Sweden and The Netherlands women aged 70 and older are invited to screening. Only the Finnish program, which invites women 50–59 years with option to continue up to 64 years, was designed to facil∗ Corresponding author. Present address: Oncological Centre, Umeå University Hospital, S-901 85 Umeå, Sweden. Tel.: +46-90-7851990; fax: +46-90-127464. E-mail address: [email protected] (H. Jonsson).
itate an evaluation [5]. Some attempts have been made to estimate the effects of service screening in Sweden. Breast cancer mortality in counties with screening programs was compared with counties with no screening [6], or compared over time [7] as well as related to surrogate measures [8]. Evaluations of programs inviting women 40–49 and 50–69 years to screening has previously been reported [9,10]. In a Dutch study, women aged 68–83 years at first invitation to service screening were studied [11]. The only randomized trial including the age group 70–74 years was the Two-County study [1]. After 20 years of follow-up [12], the relative risk (RR) was 0.76 (95% CI 0.44–1.33) and 0.73 (95% CI 0.45–1.19), respectively in each of the two counties. The study group 70–74 years at randomization were invited to screening for two rounds only. In a Dutch study the breast cancer mortality in women aged 68–83 at first invitation to screening was reduced by 20% (P > 0.05) [11]. A study in northern Sweden [6] indicated a non-significant 17% reduction in excess mortality related to breast cancer in a service-screened population in comparison with an unscreened population for women 70–74 years.
0361-090X/$30.00 © 2003 International Society for Preventive Oncology. Published by Elsevier Ltd. All rights reserved. doi:10.1016/S0361-090X(03)00131-4
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
361
However, no reference period was used, hence some of the reduction may be explained by differences in baseline mortality in that study. When the first results from the Two-County trial were published [1], the National Board of Health and Welfare in Sweden issued guidelines for mammography screening [13] recommending a lower age limit not below 40 years and not over 50 years, and an upper age limit not below 69 years, and not over 74 years. Consequently the age group 50–69 years was covered in all counties where screening was introduced. Service screening started in Sweden in 1986 and had in 1997 been introduced in all 23 counties. Out of these, 13 counties have 74 years as an upper limit for invitation to screening. The age limits were based on local political decisions and availability of resources. When studying the effect of screening in age-group 70–74 years two main questions are of importance: (1) Does screening per se in women aged 70–74 years reduce breast cancer mortality? and (2) Does screening in this age group, as a continuation of screening of women 50–69 years, cause a further reduction of the breast cancer mortality? It seems adequate to answer the question one first and if the answer is positive, to continue with the second question. However, the first question is only a theoretical one as no screening program has a lower age limit of 70 years. Therefore, the relevant issue should be to investigate how much screening in age 70–74 years adds to the reduction of breast cancer mortality. In this study, we have addressed this latter question. The aim of the present study was to estimate the effect on breast cancer mortality of the population-based service-screening program in Sweden inviting women with 74 years as upper limit for invitation in comparison with counties where the upper limit was 69 years.
2. Materials and methods All organized mammography screening in Sweden is population-based. Screening programs are usually organized at county level. However, some of the counties were divided into sub-regions that differed concerning starting time and age limits. We therefore had to divide those counties into smaller geographical areas to obtain homogeneous geographical units. From the potential study population consisting of all 23 counties, we excluded counties or parts of counties where randomized studies had been performed (Kopparberg, Östergötland, Stockholm (south) and the cities of Göteborg and Malmö). Also the county of Gävleborg was excluded because population-based screening was introduced there already in 1974. Areas where the upper age limit for invitation to screening was 74 years constituted the study population. For comparison, we used a control group based on the geographical areas where the upper limit for invitation was 69 years (Table 1). In the county of Skaraborg an upper age limit of 74 years was used for the first 1.5 years but thereafter it was changed to 69 years, thus also
Fig. 1. Map of municipalities in Sweden illustrating areas included, in the study group (dark gray), in the control group (light gray) and not included (white).
Skaraborg was excluded from this study. To ensure a reasonable follow-up time only the geographical areas where screening started early (1986–1990) were included. Excluded were therefore also the counties or parts of counties; Landskrona (part of Skåne), Värmland, Älvsborg (north), Västerbotten, Jämtland and Gotland. The map in Fig. 1 shows areas included in the study and control groups. In 1987, the mean number of women 70–74 years was 83,830
362 Table 1 Female population 70–74 years (1987, mid-year), time for screening start, years of follow-up in the study, number of person-years, cumulative number of breast cancer deaths and mortality rates (70–74 years at diagnosis and according to the refined mortality model) in the geographical areas for the reference and the study period, respectively Geographical area
group Jönköping Kalmar Västra Götaland Jönköping Skåne Örebro Uppsala Västra Götaland Halland Norrbotten Skåne Skåne Södermanland Västernorrland Skåne Skåne
Total Geographical areas control group Västmanland Västmanland Blekinge Blekinge Danderyd Stockholm Karolinska Stockholm Skärholmen/Huddinge Stockholm Sabbatsberg/St Göran Stockholm Kronoberg Kronoberg Total
Number of women
3121 6482 6330 4838 2937 7405 5433 6645 5758 5835 4075 3941 6299 7082 2403 5246
Starting (month/year)
Years of follow-up
8/86 10/86 11/86 4/87 4/87 10/87 10/88 11/88 1/89 3/89 3/89 4/89 11/89 1/90 1/90 11/90
12 12 12 11 11 11 10 10 10 9 9 9 9 9 9 8
83830 6165 4012 6783 4815 7609 8081 4143 41608
10/86 3/88 8/89 8/89 8/89 3/90 8/90
12 10 9 9 9 8 8
Reference period Person-years × 1000
Study period Cumulative number of breast cancer deaths
52 111 98 67 45 113 74 93 78 69 51 51 80 93 29 59
17 33 18 23 18 37 20 20 25 14 17 14 21 29 10 11
1163
327
91 56 86 55 89 108 49
26 11 20 14 24 25 21
534
141
Cumulative number of breast cancer deaths per 100,000 390 355 219 379 444 362 271 216 320 184 300 245 237 280 309 149
342 195 210 231 242 186 345
Person-years × 1000
Cumulative number of breast cancer deaths
55 113 113 78 48 120 81 97 89 75 54 54 87 93 33 61
14 29 34 28 13 30 34 15 17 19 21 12 18 22 10 9
1251
325
109 58 95 72 107 89 50
29 6 30 12 33 26 10
580
146
Cumulative number of breast cancer deaths per 100,000 303 309 360 396 298 275 418 154 190 227 350 200 187 214 274 119
319 104 283 151 277 233 159
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Geographical areas study Eksjö/Nässjö Kalmar Bohus Jönköping/Ryhov Trelleborg Örebro Uppsala Älvsborg (south) Halland Norrbotten Helsingborg Lund Södermanland Västernorrland Ängelholm Kristianstad
County
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
for time of invitation within each geographical area. For the calculation of person-years in the population, aggregated data were used. Information on month and year of start for each screening program, progression of the building-up of screening activities within geographical areas, age groups invited and intervals between screening rounds were obtained from the screening centers via a questionnaire. For each breast cancer case, date of diagnosis and date and cause of death for the deceased subjects were obtained from the nationwide Swedish Cancer Registry. Breast cancer mortality data aggregated by calendar year, municipality and 5-year age groups were obtained from the Swedish Cause of Death Registry and population data aggregated by municipality, calendar year and 5-year age groups from Statistics Sweden. The mean follow-up time (weighted by the population) was 10.1 years (range 8–12) in the study cohorts and 9.3 years in the control cohorts (range 8–12). By definition, the mean follow-up times were the same in the reference period (10.1 years). Considering a mean time for the first round of 27.6 months the mean individual follow-up time, with consideration to both entry in the cohort and first invitation to screening, was estimated at 8.1 years. A breast cancer case was defined as a case of invasive adenocarcinoma of the breast (site code = 174 in International Classification of Diseases, ICD-9; and histo-pathological code = 096 according to WHO/HS/CAN/24.1) diagnosed at age 70–74 years during the reference or the study period (Fig. 2). If a woman had two breast cancer diagnoses in one of the periods 1976–1988 or 1986–1998, the second cancer was excluded. A breast cancer death was defined as a breast cancer case with breast cancer as the underlying cause of death (UCD) and reported to the Cause of Death Registry not later than 31 December 1998.
90
in the study population and 41,608 in the control population. Based on the information from the screening centers the mean screening interval (weighted by the population) was estimated at 22.8 months in the study population while it was 23.1 months in the control population (screening up to 69 years). The corresponding figures for the duration of the first screening round was 27.6 and 24.5 months for the study population and the control population, respectively. We have studied two time periods, a study period, in which screening was introduced 1986–1998, and a reference period 1976–1988. The time for start of screening in the study population varied between August 1986 and November 1990 (Table 1) and was in average (weighted by the population) mid-July 1988. In the control population screening started between October 1986 and August 1990. The average start time was the end of March 1989. Each geographical area in the study and the control populations constituted separate cohorts. During the study period the cohorts consisted of all women 65–74 years old at the first invitation to screening. Similar cohorts were also defined in the reference period 10 years before the start of screening for respective area, with the same age criterion. Women below 70 years at start were included when they reached the age of 70. Breast cancer cases in the study cohorts were accrued in age 70–74 years at diagnosis from screening start and forward (Fig. 2). Accrual of breast cancer cases and follow-up was defined in the same way for the cohorts in the reference period with start 10 years ahead of the start of screening. The cohorts in the study period were followed through 1998 and the cohorts in the reference period through 1988. Individual information was used regarding date, age and residence at breast cancer diagnosis, date of death and cause of death for the breast cancer cases. As we had no access to individual screening data we had to make an approximation
363
65
70
75
Age
80
85
Follow-up Accrual
1986
1988
1990
1992
1994
1996
1998
2000
Calendar time (1 Jan)
Fig. 2. Illustration of the inclusion and follow-up in the refined mortality model. Example is given with cases who died from breast cancer in a cohort where invitation to screening started January 1987. The plus sign denotes breast cancer diagnosis and squares symbolize death from breast cancer. Deaths valid in the refined model are denoted by filled squares.
364
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
In the Swedish cancer register cancer cases are not included based on death certificates only. It is therefore possible that a deceased case was reported with breast cancer as underlying cause to the Cause of Death Registry but not as a breast cancer case to the Cancer Registry. In a study [14], it was found that 1% of the deaths aged 40–74 years in the Swedish cause of death register reported with breast cancer as underlying cause of death were not reported to the cancer register. Age specific breast cancer mortality for women 70–84 years at death, was plotted for the study and the control group. This was based on the total number of breast cancer deaths each year, here referred to as total breast cancer mortality. To avoid including cases with a breast cancer diagnosis before start of screening and below 70 years of age the so-called refined breast cancer mortality was used [6,15]. In the refined mortality model breast cancer deaths are included only if breast cancer was diagnosed at age 70–74 years after screening start in the study period or after the corresponding start times in the reference period (Fig. 2). Refined breast cancer mortality cannot be interpreted for different years of follow-up in the same way as total breast cancer mortality. During the follow-up women were between 70 and 86 years of age. There might be uncertainty in the coded underlying cause of death among old women in these ages due to competing diseases. We therefore used two different outcome measures; excess breast cancer mortality [6] and breast cancer as the underlying cause of death. For each deceased case, UCD was used as it is recorded in the cause of death register. Excess mortality is estimated by (O − E)/N where O is the total number of deaths among the breast cancer cases, and E is the expected number of deaths among the breast cancer cases. It is derived as the population total mortality times the number of person-years among the breast cancer cases. N is the total number of person-years in the studied group. The variance of the ln(RR) can be estimated by Vˆ (ln(RR)) = O0 /(O0 − E0 )2 + O1 /(O1 − E1 )2 , where the indexes correspond to the study group (1) and the control group (0). Cumulative refined breast cancer mortality per 100,000 was computed with the mean number of person-years as denominator (person-years divided by years of follow-up). Cumulative relative risks were estimated. In order to adjust for possible geographical differences in breast cancer mortality between the study group and the control group the relative risk for the study period was divided by the relative risk for the reference period. This ratio is the relative risk due to invitation to screening assuming multiplicative effects between the groups and the time periods. The refined breast cancer mortality was also analyzed in a multiplicative Poisson model with the number of breast cancer deaths (UCD) as the dependent variable and follow-up year (3-year intervals), age during follow-up (5-year classes), geographical area and time period (study or reference) as covariates, all categorical [16]. The excess mortality was analyzed in the same way but with the
number of deaths among the breast cancer cases as the dependent variable [10]. The screening effect was measured by a dummy variable set to 1 for the study group cohorts in the study period and to 0 elsewhere. The logarithm of person-years defined by the cohorts and the follow-up was taken as offset. Inclusion and lead time bias may be inherent in this observational study. 2.1. Inclusion bias One reason for dilution of the results stems from difficulties in defining the studied population at the beginning of the follow-up since it takes the first screening round to progressively include all the women within a screening area. We included all incident breast cancer cases after start of invitation to screening within a geographical area in the study group, implicating the inclusion of an unknown number of cases diagnosed before invitation to screening during the first round. This will of course lead to a dilution of the potential benefits of screening, since a number of cases not yet invited were mixed with the already invited women. The magnitude of the problem was estimated by simulation [10] using the fact that the mean screening interval was 28 months in the first screening round. For a given 28-month calendar period we assumed a random time point for invitation to be uniformly distributed over (0, 28) months for each woman with a breast cancer diagnosis. For this sample, the cumulative breast cancer mortality in the given period can be estimated for the women who had a breast cancer diagnosis before invitation to screening. Based on the breast cancer cases aged 70–74 years and diagnosed in the two periods: December 1981 to March 1984 and April 1984 to July 1986 for the study and control groups combined, 200 replicates from each period were simulated. The cumulative refined breast cancer mortality for the cases diagnosed before invitation was estimated. Based on the length of follow-up and the size of the population the expected number of breast cancer deaths was estimated for each geographical area. These were added to a total of 76.8 (24% (76.8/325) of the observed number of breast cancer deaths (UCD) in the study cohorts during the study period). If the relative risk adjusted for the reference period is formulated as OC /EC , where OC is the observed number of breast cancer deaths and EC is the corresponding expected number without screening, we can perform an adjustment of the relative risk for inclusion bias by (OC − ψ)/((OC /RR) − ψ). Here, ψ denotes the expected number of breast cancer deaths in women with a breast cancer diagnosis before invitation to screening during the first screening round. 2.2. Lead time bias In the present study, age was defined as age at diagnosis, implicating that a lead time bias might have been introduced. The purpose of mammography screening is early detection
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
of breast cancer. Women who were diagnosed with and died from breast cancer in the study areas during the study period could get their diagnosis in an earlier stage of the disease than in the control areas. For example a case in the group invited to screening may be classified as belonging to the age group below 75 years while a corresponding woman in the control group may be 75 years or more at diagnosis, even though she had an otherwise comparable cancer and died from it at the same time. Since both the study group and the control group were followed from screening start and the ages 50–69 years were covered in all areas there should not be any lead time bias around the age limit 70 years. Five counties which started screening late (between April 1993 and May 1997) for women up to 69 years were used as control group (Gotland, Värmland, Västerbotten, Jämtland and northern Älvsborg). The geographical areas in the study group were followed from screening start and the control group was followed from April 1988. All areas were followed for 8 years. The estimated mean survival time for the women who died from breast cancer was 2.52 years in the study group and 2.42 years in the control group. There was 0.1 years shorter survival among the cases who died from breast cancer in the control population than in the study population. Thus, five (1.7%) of the women in the study group who died from breast cancer were older than 74.9 years at diagnosis during the study period. They should have been 75 years or older if the diagnosis had not been made earlier by mammography assuming the same lead time for every woman. All calculations were done using the program S-Plus [17].
3. Results For the study and the control group, the annual age-specific breast cancer mortality in age group 70–84 years are shown in Fig. 3. From 1974 to 1998 there has been a decrease in the breast cancer mortality for both the study and the control group. However, it was more pronounced in the study group during the study period. 3.1. Refined excess mortality During the study period 1986–1998 there were 683 deaths among breast cancer patients observed in the study cohorts and 256 in the control cohorts (Table 2). Based on the total age and county specific mortality and the person-years among the breast cancer cases, totally 12,056 in the study cohorts and 3651 in the control cohorts, the expected number of deaths was 371.7 and 105.5, respectively. This gives the excess number (O − E) of deaths, 311.3 and 150.5, respectively. The cumulative breast cancer excess mortality per 100,000 at 12 years in the study period was 299 and 311 for the study group and the control group, respectively (Fig. 4B). The relative risk for the study group compared to the control group was 0.96 and the 95% CI 0.73–1.25. In the reference period, the figures were 382 and 328, respectively (Fig. 4A). The relative risk in the reference period was 1.17. Hence, the RR adjusted for the reference period was 0.82 and the 95% CI 0.57–1.19 (Table 3). The data were also fitted in a multiplicative Poisson model (Table 4). Three covariates were found to be statistically significant, namely 3-year follow-up intervals (P < 0.001), age
100
150
Mortality 70-84 years, Study group Mortality 70-84 years, Control group
0
50
Mortality/100,000
365
1975
1980
1985
1990
1995
Year
Fig. 3. Annual breast cancer mortality per 100,000 (UCD), 1974–1998 for women 70–84 years at death in the study group and the control group. In the control group, a part of the target population for the Stockholm trial (14% of the control group) had to be included. The trial, which invited women aged up to 65 years, started 1981 and invited its control group from 1985.
366
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Table 2 Number of deaths, person-years, expected and excess number of deaths in the breast cancer cases and person-years in all women in the cohorts (70–74 years at diagnosis and according to the refined mortality model) Group
Study cohorts Control cohorts Study cohorts Control cohorts
Period
Breast cancer cases
Reference Reference Study Study
All women
Total number of deaths
Person-years × 1000
Expected number of deaths
Excess number of deaths
Total number of person-years × 1000
611 258 683 256
6.9 3.3 12.1 3.7
241.0 112.4 371.7 105.5
370.0 145.6 311.3 150.5
1162.8 533.4 1251.3 580.1
Fig. 4. Cumulative number of breast cancer deaths per 100,000 (refined mortality) for women 70–74 years at diagnosis for the study group (S) and the control group (C) by year since start of follow-up. Excess mortality: (A) reference period, (B) study period Breast cancer mortality; (UCD): (C) reference period, (D) study period. Table 3 Summary of results on refined breast cancer mortality using excess mortality and underlying cause of death (UCD) as outcome measures Model
RR
95% CI
Adj RRa
Cumulative excess mortality Excess mortality, Poisson
0.82 0.84
0.57–1.19 0.59–1.19
0.76 0.78
Cumulative UCD UCD, Poisson
0.97 0.96
0.73–1.28 0.73–1.28
0.94 0.93
RR: relative risk, CI: confidence interval, UCD: underlying cause of death a RR adjusted for inclusion bias and lead time bias.
at follow-up (P < 0.001) and geographical area (P = 0.02). The estimated relative risk due to invitation to screening from this model was 0.84 (95% CI 0.59–1.19) (Table 3). 3.2. Refined breast cancer mortality based on underlying cause of death (UCD) During the study period 1986–1998, there were 325 breast cancer deaths (UCD) observed in the study cohorts and 146 in the control cohorts. The cumulative number of breast can-
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
367
Table 4 Summary of fitting multiplicative Poisson models to estimate refined breast cancer mortality using excess mortality and underlying cause of death (UCD) as outcome measures Cause of death
Model
Excess mortality 1 2 3 4 5 UCD 1 2 3 4 5
Terms included in the model
Degree of freedom (d.f.)
Deviance
Compared models
Null Years of follow-up 1 + age at follow-up 2 + time period 3 + geographical area 4 + screening Null Years of follow-up 1 + age at follow-up 2 + time period 3 + geographical area 4 + screening
365 362 360 359 337 336 365 362 360 359 337 336
508.6 453.3 410.9 407.6 371.0 370.0 523.6 447.7 394.5 393.7 358.6 358.5
– Null–1 1–2 2–3 3–4 4–5 – Null–1 1–2 2–3 3–4 4–5
cer deaths and cumulative breast cancer mortality by geographical area and period are given in Table 1 and the cumulative breast cancer mortality for the study group and the control group in the two time periods are illustrated in Fig. 4C–D. The cumulative breast cancer mortality per 100,000 at 12 years was 312 and 302 in the study period for the study group and the control group, respectively. The corresponding figures for the reference period were 337 and 317, respectively. Hence, the adjusted relative risk in the screening group was 0.97 and the 95% CI 0.73–1.28 (Table 3). The UCD data were also fitted in a multiplicative Poisson model (Table 4). Three covariates were found to be statistically significant, namely years of follow-up (P < 0.001), age at follow-up (P < 0.001) and geographical area (P = 0.004). The estimated relative risk due to invitation to screening from this model was 0.97 (95% CI 0.73–1.28) (Table 3). 3.3. Adjustment for inclusion and lead time biases The expected number of breast cancer deaths in the study group due to cases diagnosed before invitation was estimated at 76.8. This gives an adjustment of the RR based on excess mortality from 0.82 to 0.77. The corresponding adjustment for the UCD changed the RR from 0.97 to 0.96. Allowance for lead time bias reduced the RR 1.7% to 0.76 and 0.94 for excess mortality and UCD, respectively (Table 3).
4. Discussion In the present study, the effect of service screening of age group 70–74 years on a routine basis has been studied. We are not aware of any obvious differences in quality indicators such as attendance rates, recall rates, screening intervals, cancer detection rates, etc. between service screening and the randomized trials in this age group [18,19]. There are,
Difference in deviance
Difference in d.f.
P-value
55.3 42.4 3.2 36.6 1.0
3 2 1 22 1
<0.001 <0.001 0.07 0.02 0.32
75.9 53.3 0.8 35.1 0.1
3 2 1 22 1
<0.001 <0.001 0.37 0.04 0.79
however some other possible sources of bias, relevant for studies of service screening, not mentioned earlier, to take into consideration. If women in the control group had a mammogram on their own initiative, there is a possibility that the observed effect on mortality reduction could have been diluted. Opportunistic screening in Sweden is a phenomenon that mainly appears in large cities [20]. Stockholm is the only large city in the control group and constitutes 66% of the control group. We believe that opportunistic screening is less frequent among older than among younger women. The total breast cancer mortality time trends (Fig. 3) were similar in both groups except for a lower level in the study group from the late 1980s. However, the refined mortality indicates, for both outcome measures, a difference during the reference period towards a higher breast cancer mortality in the study group. The impact of adjustment for refined mortality in the reference period was substantial for the estimates of relative risk. Therefore, we believe that the adjustment was justified although the confidence intervals became wider. The difference in the reference period could indicate different background mortality in the study and control groups. The discrepancy between the number of breast cancer deaths in the defined cohorts (refined mortality) and the more commonly used total mortality for 70–84 or 75–84 years at death is considerable. In a study of the effects of screening in age 50–69 years on breast cancer mortality [10] only 27% of the deaths in the total mortality of women 50–79 years at death consisted of cases valid in the refined mortality model after 11 years of follow-up. Thus, especially when evaluating screening effects in certain age intervals, total mortality can be insufficiently discriminative for effects of screening. We have used two methods in measuring breast cancer mortality; individual underlying cause of death coded by the National Cause of Death Registry and excess mortality. Using individual data, it can be difficult in many cases to decide whether breast cancer is an underlying cause of death. The degree of complication increases by increasing age due
368
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Fig. 5. Ratio of the number of breast cancer deaths (UCD) (numerator) and excess mortality (EM) (denominator) by age groups in the invited group (IG) and the control group (CG) from the paper by Larsson et al. [21].
to intercurrent diseases [14]. Excess mortality compares observed and expected mortality among breast cancer patients. It is therefore a measure of all mortality caused from breast cancer. In the latest overview of all Swedish randomized studies these two measures were compared [21]. In the age group 70–74 years at randomization a 16% non-significant excess breast cancer mortality reduction was seen after a mean individual follow-up of 11–12 years. The corresponding relative risk was 1.05 when using UCD, thus a difference of 21%. The corresponding difference for the age group 50–69 years was 3–4%. In the present study, the difference between the two measures was 10–13%. A possible explanation for these differences could be the higher incidence of breast cancer in the study group due to screening. For some of the deceased breast cancer cases in a population it may be difficult to decide whether the underlying cause of death was breast cancer or not. However, for some of these cases the decision by the clinician would be breast cancer. On the other hand, if the same population had been screened the number of breast cancer cases would have been increased. If among this increment some deceased subjects also had caused decision problems there will be a possibility for a differential bias of overestimation of the number of breast cancer deaths in the study population. Using data from the paper mentioned above [21] the ratio between the breast cancer mortality (UCD) and the excess mortality from breast cancer for different ages at randomization is shown in Fig. 5 for the invited and the control group, respectively. The breast cancer mortality (UCD) was higher in the study group for all ages and the difference was most pronounced in the age group 70–74 years. This suggests that breast cancer mortality using UCD may be less valid when comparing a screened group with a group who had not been screened especially in women over 70 years. Thus, the relative risks using UCD might be biased.
To summarize, with a mean screening interval of 22.8 months and with a mean follow-up time of 10.1 years of the county based Swedish service-screening program for women of age 70–74 years, the reduction of the breast cancer excess mortality was estimated to be 18% when compared to the control group. Using the breast cancer mortality based on individual cause of death the corresponding result was 3%. Adjusting for the inclusion of cases in the study cohorts diagnosed before invitation and for lead time bias, the reduction was 24 and 6% for excess mortality and breast cancer mortality (UCD), respectively (Table 3).
Acknowledgements The present study was supported by the Swedish Cancer Society and the European Commission. The authors are indebted to the responsible radiologists in the respective screening centers who gave us valuable and necessary information by kindly answering the questionnaire; A.-M. Berggren, P. Bordas, S. Cederblom, J.-O. Englund, B. Epstein, E. Frodis, B. Heddson, H. Håkansson, J. Johanson, M. Kubista, M. Löfgren, T. Mathiesen, Z. von Pàlffej, L.F. Samuelsson, H. Svensson, P.-Å. Svensson, M. Tholin, E. Thurfjell, K. Yde.
References [1] Tabár L, Fagerberg CJG, Gad A, et al. Reduction in mortality from breast cancer after mass screening with mammography: randomised trial from the Breast Cancer Screening Working Group of the Swedish National Board of Health and Welfare. Lancet 1985;1:829–32. [2] Nyström L, Rutqvist L, Wall S, et al. Breast cancer screening with mammography: overview of Swedish randomised trials. Lancet 1993;341:973–8.
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369 [3] Hwang ES, Cody III HS. Does the proven benefit of mammography extend to breast cancer patients over age 70? South Med J 1998;91:522–6. [4] Kopans DB. Screening mammography in women over age 65. J. Gerontol. 1992;47(Spec. No.):59–62. [5] Hakama M, Pukkala E, Söderman B, et al. Implementation of screening as a public health policy: issues in design and evaluation. J Med Screen 1999;6:209–16. [6] Lenner P, Jonsson H. Excess mortality from breast cancer in relation to mammography screening in northern Sweden. J Med Screen 1997;4:6–9. [7] Tabár L, Vitak B, Chen HH, et al. Beyond randomized controlled trials. Cancer 2001;91:1724–31. [8] Thurfjell EL, Lindgren JAÅ. Breast cancer survival rates with mammographic screening: similar favourable survival rates for women younger and those older than 50 years. Radiology 1996;201:421–6. [9] Jonsson H, Törnberg S, Nyström L, et al. Service screening with mammography in Sweden: evaluation of effects from screening in age 40–49 years on the breast cancer mortality. Acta Oncol 2000;39:617– 23. [10] Jonsson H, Nyström L, Törnberg S, et al. Service screening with mammography of women aged 50–69 years in Sweden: effects on mortality from breast cancer. J Med Screen 2001;8:152–60. [11] van Dijck JA, Verbeek AL, Beex LV, et al. Breast-cancer mortality in a non-randomized trial on mammographic screening in women over age 65. Int J Cancer 1997;70:164–8. [12] Tabár L, Vitak B, Chen HH, et al. The Swedish Two-County trial twenty years later: updated mortality results and new insights
[13]
[14]
[15]
[16] [17] [18]
[19] [20]
[21]
369
from long-term follow-up. Radiol Clin North Am 2000;38:625– 51. NBHW. Mammographic screening for early detection of breast cancer (in Swedish). Stockholm: The National Board of Health and Welfare; 1986. Nyström L, Larsson L-G, Rutqvist LE, et al. Determination of cause of death among cancer cases in the Swedish randomized mammography screening trials: a comparison between official statistics and validation by an endpoint committee. Acta Oncol 1995;34:145–52. Hakama M, Pukkala E, Heikkila M, et al. Effectiveness of the public health policy for breast cancer screening in Finland: population based cohort study. Br Med J 1997;314:864–7. Breslow NE, Day NE. The design and analysis of cohort studies. Lyon: IARC; 1987. Venables WN, Ripley BD. Modern applied statistics with S-PLUS, 2nd ed. New York: Springer; 1997. Lidbrink EK, Törnberg SA, Azavedo EM, et al. The general mammography screening program in Stockholm. Acta Oncol 1994;33:353–8. Thurfjell E. Mammography screening: methods and diagnostic results. Thesis. Uppsala: Uppsala, Sweden; 1995. Törnberg S. A survey on mammography examinations performed in private clinics during the period 1992–1993 (in Swedish). Stockholm: Socialstyrelsen; 1994. Larsson LG, Nyström L, Wall S, et al. The Swedish randomised mammography screening trials: analysis of their effect on the breast cancer related excess mortality. J Med Screen 1996;3:129– 32.
Service screening with mammography of women aged 70–74 years in Sweden Effects on breast cancer mortality Håkan Jonsson, PhD a,∗ , Sven Törnberg, MD, PhD b , Lennarth Nyström, PhD c , Per Lenner, MD, PhD a a
Department of Oncology, Umeå University, Umeå, Sweden Oncologic Centre, Karolinska Hospital, Stockholm, Sweden Epidemiology, Public Health and Clinical Medicine, Umeå University, Umeå, Sweden b
c
Accepted 24 July 2003
Abstract Since the benefit of mammography screening for women 70 years and older is unclear, the aim of the present study was to evaluate the effect on breast cancer mortality of the population-based service-screening program in Sweden inviting women 70–74 years. Among the counties with service-screening programs in Sweden which started 1986–1990 those with upper age limit 74 years were compared to counties with 69 years as upper age limit with respect to refined breast cancer mortality. Allowance was made for potential biases namely inclusion of cases diagnosed before invitation and lead time. Two methods for estimation of breast cancer mortality were used; underlying cause of death (UCD) and excess mortality. With a mean follow-up of 10.1 years a reduction of the breast cancer excess mortality was estimated at 24%. Using the underlying cause of death the corresponding result was 6%. A non-significant reduction in breast cancer mortality was found in the counties with service-screening program including the age group 70–74 years in Sweden. The estimated reduction was larger when using excess mortality compared to the use of individual underlying cause of death. © 2003 International Society for Preventive Oncology. Published by Elsevier Ltd. All rights reserved. Keywords: Breast cancer; Mortality; Mammography screening; Evaluation
1. Introduction Randomized studies have shown that screening for breast cancer with mammography reduces the breast cancer mortality [1,2], especially for women aged 50–69 years at invitation to screening. Theoretically, screening should be beneficial also for women aged 70 years or more [3,4] however, there is a lack of empirical evidence to support this statement. The only randomized screening trial that included women 70 years and older was the Swedish Two-County trial [1]. Today nationwide service-screening programs have been introduced in Sweden, Finland, The Netherlands, UK, Iceland and Luxembourg but only in Sweden and The Netherlands women aged 70 and older are invited to screening. Only the Finnish program, which invites women 50–59 years with option to continue up to 64 years, was designed to facil∗ Corresponding author. Present address: Oncological Centre, Umeå University Hospital, S-901 85 Umeå, Sweden. Tel.: +46-90-7851990; fax: +46-90-127464. E-mail address: [email protected] (H. Jonsson).
itate an evaluation [5]. Some attempts have been made to estimate the effects of service screening in Sweden. Breast cancer mortality in counties with screening programs was compared with counties with no screening [6], or compared over time [7] as well as related to surrogate measures [8]. Evaluations of programs inviting women 40–49 and 50–69 years to screening has previously been reported [9,10]. In a Dutch study, women aged 68–83 years at first invitation to service screening were studied [11]. The only randomized trial including the age group 70–74 years was the Two-County study [1]. After 20 years of follow-up [12], the relative risk (RR) was 0.76 (95% CI 0.44–1.33) and 0.73 (95% CI 0.45–1.19), respectively in each of the two counties. The study group 70–74 years at randomization were invited to screening for two rounds only. In a Dutch study the breast cancer mortality in women aged 68–83 at first invitation to screening was reduced by 20% (P > 0.05) [11]. A study in northern Sweden [6] indicated a non-significant 17% reduction in excess mortality related to breast cancer in a service-screened population in comparison with an unscreened population for women 70–74 years.
0361-090X/$30.00 © 2003 International Society for Preventive Oncology. Published by Elsevier Ltd. All rights reserved. doi:10.1016/S0361-090X(03)00131-4
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
361
However, no reference period was used, hence some of the reduction may be explained by differences in baseline mortality in that study. When the first results from the Two-County trial were published [1], the National Board of Health and Welfare in Sweden issued guidelines for mammography screening [13] recommending a lower age limit not below 40 years and not over 50 years, and an upper age limit not below 69 years, and not over 74 years. Consequently the age group 50–69 years was covered in all counties where screening was introduced. Service screening started in Sweden in 1986 and had in 1997 been introduced in all 23 counties. Out of these, 13 counties have 74 years as an upper limit for invitation to screening. The age limits were based on local political decisions and availability of resources. When studying the effect of screening in age-group 70–74 years two main questions are of importance: (1) Does screening per se in women aged 70–74 years reduce breast cancer mortality? and (2) Does screening in this age group, as a continuation of screening of women 50–69 years, cause a further reduction of the breast cancer mortality? It seems adequate to answer the question one first and if the answer is positive, to continue with the second question. However, the first question is only a theoretical one as no screening program has a lower age limit of 70 years. Therefore, the relevant issue should be to investigate how much screening in age 70–74 years adds to the reduction of breast cancer mortality. In this study, we have addressed this latter question. The aim of the present study was to estimate the effect on breast cancer mortality of the population-based service-screening program in Sweden inviting women with 74 years as upper limit for invitation in comparison with counties where the upper limit was 69 years.
2. Materials and methods All organized mammography screening in Sweden is population-based. Screening programs are usually organized at county level. However, some of the counties were divided into sub-regions that differed concerning starting time and age limits. We therefore had to divide those counties into smaller geographical areas to obtain homogeneous geographical units. From the potential study population consisting of all 23 counties, we excluded counties or parts of counties where randomized studies had been performed (Kopparberg, Östergötland, Stockholm (south) and the cities of Göteborg and Malmö). Also the county of Gävleborg was excluded because population-based screening was introduced there already in 1974. Areas where the upper age limit for invitation to screening was 74 years constituted the study population. For comparison, we used a control group based on the geographical areas where the upper limit for invitation was 69 years (Table 1). In the county of Skaraborg an upper age limit of 74 years was used for the first 1.5 years but thereafter it was changed to 69 years, thus also
Fig. 1. Map of municipalities in Sweden illustrating areas included, in the study group (dark gray), in the control group (light gray) and not included (white).
Skaraborg was excluded from this study. To ensure a reasonable follow-up time only the geographical areas where screening started early (1986–1990) were included. Excluded were therefore also the counties or parts of counties; Landskrona (part of Skåne), Värmland, Älvsborg (north), Västerbotten, Jämtland and Gotland. The map in Fig. 1 shows areas included in the study and control groups. In 1987, the mean number of women 70–74 years was 83,830
362 Table 1 Female population 70–74 years (1987, mid-year), time for screening start, years of follow-up in the study, number of person-years, cumulative number of breast cancer deaths and mortality rates (70–74 years at diagnosis and according to the refined mortality model) in the geographical areas for the reference and the study period, respectively Geographical area
group Jönköping Kalmar Västra Götaland Jönköping Skåne Örebro Uppsala Västra Götaland Halland Norrbotten Skåne Skåne Södermanland Västernorrland Skåne Skåne
Total Geographical areas control group Västmanland Västmanland Blekinge Blekinge Danderyd Stockholm Karolinska Stockholm Skärholmen/Huddinge Stockholm Sabbatsberg/St Göran Stockholm Kronoberg Kronoberg Total
Number of women
3121 6482 6330 4838 2937 7405 5433 6645 5758 5835 4075 3941 6299 7082 2403 5246
Starting (month/year)
Years of follow-up
8/86 10/86 11/86 4/87 4/87 10/87 10/88 11/88 1/89 3/89 3/89 4/89 11/89 1/90 1/90 11/90
12 12 12 11 11 11 10 10 10 9 9 9 9 9 9 8
83830 6165 4012 6783 4815 7609 8081 4143 41608
10/86 3/88 8/89 8/89 8/89 3/90 8/90
12 10 9 9 9 8 8
Reference period Person-years × 1000
Study period Cumulative number of breast cancer deaths
52 111 98 67 45 113 74 93 78 69 51 51 80 93 29 59
17 33 18 23 18 37 20 20 25 14 17 14 21 29 10 11
1163
327
91 56 86 55 89 108 49
26 11 20 14 24 25 21
534
141
Cumulative number of breast cancer deaths per 100,000 390 355 219 379 444 362 271 216 320 184 300 245 237 280 309 149
342 195 210 231 242 186 345
Person-years × 1000
Cumulative number of breast cancer deaths
55 113 113 78 48 120 81 97 89 75 54 54 87 93 33 61
14 29 34 28 13 30 34 15 17 19 21 12 18 22 10 9
1251
325
109 58 95 72 107 89 50
29 6 30 12 33 26 10
580
146
Cumulative number of breast cancer deaths per 100,000 303 309 360 396 298 275 418 154 190 227 350 200 187 214 274 119
319 104 283 151 277 233 159
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Geographical areas study Eksjö/Nässjö Kalmar Bohus Jönköping/Ryhov Trelleborg Örebro Uppsala Älvsborg (south) Halland Norrbotten Helsingborg Lund Södermanland Västernorrland Ängelholm Kristianstad
County
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
for time of invitation within each geographical area. For the calculation of person-years in the population, aggregated data were used. Information on month and year of start for each screening program, progression of the building-up of screening activities within geographical areas, age groups invited and intervals between screening rounds were obtained from the screening centers via a questionnaire. For each breast cancer case, date of diagnosis and date and cause of death for the deceased subjects were obtained from the nationwide Swedish Cancer Registry. Breast cancer mortality data aggregated by calendar year, municipality and 5-year age groups were obtained from the Swedish Cause of Death Registry and population data aggregated by municipality, calendar year and 5-year age groups from Statistics Sweden. The mean follow-up time (weighted by the population) was 10.1 years (range 8–12) in the study cohorts and 9.3 years in the control cohorts (range 8–12). By definition, the mean follow-up times were the same in the reference period (10.1 years). Considering a mean time for the first round of 27.6 months the mean individual follow-up time, with consideration to both entry in the cohort and first invitation to screening, was estimated at 8.1 years. A breast cancer case was defined as a case of invasive adenocarcinoma of the breast (site code = 174 in International Classification of Diseases, ICD-9; and histo-pathological code = 096 according to WHO/HS/CAN/24.1) diagnosed at age 70–74 years during the reference or the study period (Fig. 2). If a woman had two breast cancer diagnoses in one of the periods 1976–1988 or 1986–1998, the second cancer was excluded. A breast cancer death was defined as a breast cancer case with breast cancer as the underlying cause of death (UCD) and reported to the Cause of Death Registry not later than 31 December 1998.
90
in the study population and 41,608 in the control population. Based on the information from the screening centers the mean screening interval (weighted by the population) was estimated at 22.8 months in the study population while it was 23.1 months in the control population (screening up to 69 years). The corresponding figures for the duration of the first screening round was 27.6 and 24.5 months for the study population and the control population, respectively. We have studied two time periods, a study period, in which screening was introduced 1986–1998, and a reference period 1976–1988. The time for start of screening in the study population varied between August 1986 and November 1990 (Table 1) and was in average (weighted by the population) mid-July 1988. In the control population screening started between October 1986 and August 1990. The average start time was the end of March 1989. Each geographical area in the study and the control populations constituted separate cohorts. During the study period the cohorts consisted of all women 65–74 years old at the first invitation to screening. Similar cohorts were also defined in the reference period 10 years before the start of screening for respective area, with the same age criterion. Women below 70 years at start were included when they reached the age of 70. Breast cancer cases in the study cohorts were accrued in age 70–74 years at diagnosis from screening start and forward (Fig. 2). Accrual of breast cancer cases and follow-up was defined in the same way for the cohorts in the reference period with start 10 years ahead of the start of screening. The cohorts in the study period were followed through 1998 and the cohorts in the reference period through 1988. Individual information was used regarding date, age and residence at breast cancer diagnosis, date of death and cause of death for the breast cancer cases. As we had no access to individual screening data we had to make an approximation
363
65
70
75
Age
80
85
Follow-up Accrual
1986
1988
1990
1992
1994
1996
1998
2000
Calendar time (1 Jan)
Fig. 2. Illustration of the inclusion and follow-up in the refined mortality model. Example is given with cases who died from breast cancer in a cohort where invitation to screening started January 1987. The plus sign denotes breast cancer diagnosis and squares symbolize death from breast cancer. Deaths valid in the refined model are denoted by filled squares.
364
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
In the Swedish cancer register cancer cases are not included based on death certificates only. It is therefore possible that a deceased case was reported with breast cancer as underlying cause to the Cause of Death Registry but not as a breast cancer case to the Cancer Registry. In a study [14], it was found that 1% of the deaths aged 40–74 years in the Swedish cause of death register reported with breast cancer as underlying cause of death were not reported to the cancer register. Age specific breast cancer mortality for women 70–84 years at death, was plotted for the study and the control group. This was based on the total number of breast cancer deaths each year, here referred to as total breast cancer mortality. To avoid including cases with a breast cancer diagnosis before start of screening and below 70 years of age the so-called refined breast cancer mortality was used [6,15]. In the refined mortality model breast cancer deaths are included only if breast cancer was diagnosed at age 70–74 years after screening start in the study period or after the corresponding start times in the reference period (Fig. 2). Refined breast cancer mortality cannot be interpreted for different years of follow-up in the same way as total breast cancer mortality. During the follow-up women were between 70 and 86 years of age. There might be uncertainty in the coded underlying cause of death among old women in these ages due to competing diseases. We therefore used two different outcome measures; excess breast cancer mortality [6] and breast cancer as the underlying cause of death. For each deceased case, UCD was used as it is recorded in the cause of death register. Excess mortality is estimated by (O − E)/N where O is the total number of deaths among the breast cancer cases, and E is the expected number of deaths among the breast cancer cases. It is derived as the population total mortality times the number of person-years among the breast cancer cases. N is the total number of person-years in the studied group. The variance of the ln(RR) can be estimated by Vˆ (ln(RR)) = O0 /(O0 − E0 )2 + O1 /(O1 − E1 )2 , where the indexes correspond to the study group (1) and the control group (0). Cumulative refined breast cancer mortality per 100,000 was computed with the mean number of person-years as denominator (person-years divided by years of follow-up). Cumulative relative risks were estimated. In order to adjust for possible geographical differences in breast cancer mortality between the study group and the control group the relative risk for the study period was divided by the relative risk for the reference period. This ratio is the relative risk due to invitation to screening assuming multiplicative effects between the groups and the time periods. The refined breast cancer mortality was also analyzed in a multiplicative Poisson model with the number of breast cancer deaths (UCD) as the dependent variable and follow-up year (3-year intervals), age during follow-up (5-year classes), geographical area and time period (study or reference) as covariates, all categorical [16]. The excess mortality was analyzed in the same way but with the
number of deaths among the breast cancer cases as the dependent variable [10]. The screening effect was measured by a dummy variable set to 1 for the study group cohorts in the study period and to 0 elsewhere. The logarithm of person-years defined by the cohorts and the follow-up was taken as offset. Inclusion and lead time bias may be inherent in this observational study. 2.1. Inclusion bias One reason for dilution of the results stems from difficulties in defining the studied population at the beginning of the follow-up since it takes the first screening round to progressively include all the women within a screening area. We included all incident breast cancer cases after start of invitation to screening within a geographical area in the study group, implicating the inclusion of an unknown number of cases diagnosed before invitation to screening during the first round. This will of course lead to a dilution of the potential benefits of screening, since a number of cases not yet invited were mixed with the already invited women. The magnitude of the problem was estimated by simulation [10] using the fact that the mean screening interval was 28 months in the first screening round. For a given 28-month calendar period we assumed a random time point for invitation to be uniformly distributed over (0, 28) months for each woman with a breast cancer diagnosis. For this sample, the cumulative breast cancer mortality in the given period can be estimated for the women who had a breast cancer diagnosis before invitation to screening. Based on the breast cancer cases aged 70–74 years and diagnosed in the two periods: December 1981 to March 1984 and April 1984 to July 1986 for the study and control groups combined, 200 replicates from each period were simulated. The cumulative refined breast cancer mortality for the cases diagnosed before invitation was estimated. Based on the length of follow-up and the size of the population the expected number of breast cancer deaths was estimated for each geographical area. These were added to a total of 76.8 (24% (76.8/325) of the observed number of breast cancer deaths (UCD) in the study cohorts during the study period). If the relative risk adjusted for the reference period is formulated as OC /EC , where OC is the observed number of breast cancer deaths and EC is the corresponding expected number without screening, we can perform an adjustment of the relative risk for inclusion bias by (OC − ψ)/((OC /RR) − ψ). Here, ψ denotes the expected number of breast cancer deaths in women with a breast cancer diagnosis before invitation to screening during the first screening round. 2.2. Lead time bias In the present study, age was defined as age at diagnosis, implicating that a lead time bias might have been introduced. The purpose of mammography screening is early detection
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
of breast cancer. Women who were diagnosed with and died from breast cancer in the study areas during the study period could get their diagnosis in an earlier stage of the disease than in the control areas. For example a case in the group invited to screening may be classified as belonging to the age group below 75 years while a corresponding woman in the control group may be 75 years or more at diagnosis, even though she had an otherwise comparable cancer and died from it at the same time. Since both the study group and the control group were followed from screening start and the ages 50–69 years were covered in all areas there should not be any lead time bias around the age limit 70 years. Five counties which started screening late (between April 1993 and May 1997) for women up to 69 years were used as control group (Gotland, Värmland, Västerbotten, Jämtland and northern Älvsborg). The geographical areas in the study group were followed from screening start and the control group was followed from April 1988. All areas were followed for 8 years. The estimated mean survival time for the women who died from breast cancer was 2.52 years in the study group and 2.42 years in the control group. There was 0.1 years shorter survival among the cases who died from breast cancer in the control population than in the study population. Thus, five (1.7%) of the women in the study group who died from breast cancer were older than 74.9 years at diagnosis during the study period. They should have been 75 years or older if the diagnosis had not been made earlier by mammography assuming the same lead time for every woman. All calculations were done using the program S-Plus [17].
3. Results For the study and the control group, the annual age-specific breast cancer mortality in age group 70–84 years are shown in Fig. 3. From 1974 to 1998 there has been a decrease in the breast cancer mortality for both the study and the control group. However, it was more pronounced in the study group during the study period. 3.1. Refined excess mortality During the study period 1986–1998 there were 683 deaths among breast cancer patients observed in the study cohorts and 256 in the control cohorts (Table 2). Based on the total age and county specific mortality and the person-years among the breast cancer cases, totally 12,056 in the study cohorts and 3651 in the control cohorts, the expected number of deaths was 371.7 and 105.5, respectively. This gives the excess number (O − E) of deaths, 311.3 and 150.5, respectively. The cumulative breast cancer excess mortality per 100,000 at 12 years in the study period was 299 and 311 for the study group and the control group, respectively (Fig. 4B). The relative risk for the study group compared to the control group was 0.96 and the 95% CI 0.73–1.25. In the reference period, the figures were 382 and 328, respectively (Fig. 4A). The relative risk in the reference period was 1.17. Hence, the RR adjusted for the reference period was 0.82 and the 95% CI 0.57–1.19 (Table 3). The data were also fitted in a multiplicative Poisson model (Table 4). Three covariates were found to be statistically significant, namely 3-year follow-up intervals (P < 0.001), age
100
150
Mortality 70-84 years, Study group Mortality 70-84 years, Control group
0
50
Mortality/100,000
365
1975
1980
1985
1990
1995
Year
Fig. 3. Annual breast cancer mortality per 100,000 (UCD), 1974–1998 for women 70–84 years at death in the study group and the control group. In the control group, a part of the target population for the Stockholm trial (14% of the control group) had to be included. The trial, which invited women aged up to 65 years, started 1981 and invited its control group from 1985.
366
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Table 2 Number of deaths, person-years, expected and excess number of deaths in the breast cancer cases and person-years in all women in the cohorts (70–74 years at diagnosis and according to the refined mortality model) Group
Study cohorts Control cohorts Study cohorts Control cohorts
Period
Breast cancer cases
Reference Reference Study Study
All women
Total number of deaths
Person-years × 1000
Expected number of deaths
Excess number of deaths
Total number of person-years × 1000
611 258 683 256
6.9 3.3 12.1 3.7
241.0 112.4 371.7 105.5
370.0 145.6 311.3 150.5
1162.8 533.4 1251.3 580.1
Fig. 4. Cumulative number of breast cancer deaths per 100,000 (refined mortality) for women 70–74 years at diagnosis for the study group (S) and the control group (C) by year since start of follow-up. Excess mortality: (A) reference period, (B) study period Breast cancer mortality; (UCD): (C) reference period, (D) study period. Table 3 Summary of results on refined breast cancer mortality using excess mortality and underlying cause of death (UCD) as outcome measures Model
RR
95% CI
Adj RRa
Cumulative excess mortality Excess mortality, Poisson
0.82 0.84
0.57–1.19 0.59–1.19
0.76 0.78
Cumulative UCD UCD, Poisson
0.97 0.96
0.73–1.28 0.73–1.28
0.94 0.93
RR: relative risk, CI: confidence interval, UCD: underlying cause of death a RR adjusted for inclusion bias and lead time bias.
at follow-up (P < 0.001) and geographical area (P = 0.02). The estimated relative risk due to invitation to screening from this model was 0.84 (95% CI 0.59–1.19) (Table 3). 3.2. Refined breast cancer mortality based on underlying cause of death (UCD) During the study period 1986–1998, there were 325 breast cancer deaths (UCD) observed in the study cohorts and 146 in the control cohorts. The cumulative number of breast can-
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
367
Table 4 Summary of fitting multiplicative Poisson models to estimate refined breast cancer mortality using excess mortality and underlying cause of death (UCD) as outcome measures Cause of death
Model
Excess mortality 1 2 3 4 5 UCD 1 2 3 4 5
Terms included in the model
Degree of freedom (d.f.)
Deviance
Compared models
Null Years of follow-up 1 + age at follow-up 2 + time period 3 + geographical area 4 + screening Null Years of follow-up 1 + age at follow-up 2 + time period 3 + geographical area 4 + screening
365 362 360 359 337 336 365 362 360 359 337 336
508.6 453.3 410.9 407.6 371.0 370.0 523.6 447.7 394.5 393.7 358.6 358.5
– Null–1 1–2 2–3 3–4 4–5 – Null–1 1–2 2–3 3–4 4–5
cer deaths and cumulative breast cancer mortality by geographical area and period are given in Table 1 and the cumulative breast cancer mortality for the study group and the control group in the two time periods are illustrated in Fig. 4C–D. The cumulative breast cancer mortality per 100,000 at 12 years was 312 and 302 in the study period for the study group and the control group, respectively. The corresponding figures for the reference period were 337 and 317, respectively. Hence, the adjusted relative risk in the screening group was 0.97 and the 95% CI 0.73–1.28 (Table 3). The UCD data were also fitted in a multiplicative Poisson model (Table 4). Three covariates were found to be statistically significant, namely years of follow-up (P < 0.001), age at follow-up (P < 0.001) and geographical area (P = 0.004). The estimated relative risk due to invitation to screening from this model was 0.97 (95% CI 0.73–1.28) (Table 3). 3.3. Adjustment for inclusion and lead time biases The expected number of breast cancer deaths in the study group due to cases diagnosed before invitation was estimated at 76.8. This gives an adjustment of the RR based on excess mortality from 0.82 to 0.77. The corresponding adjustment for the UCD changed the RR from 0.97 to 0.96. Allowance for lead time bias reduced the RR 1.7% to 0.76 and 0.94 for excess mortality and UCD, respectively (Table 3).
4. Discussion In the present study, the effect of service screening of age group 70–74 years on a routine basis has been studied. We are not aware of any obvious differences in quality indicators such as attendance rates, recall rates, screening intervals, cancer detection rates, etc. between service screening and the randomized trials in this age group [18,19]. There are,
Difference in deviance
Difference in d.f.
P-value
55.3 42.4 3.2 36.6 1.0
3 2 1 22 1
<0.001 <0.001 0.07 0.02 0.32
75.9 53.3 0.8 35.1 0.1
3 2 1 22 1
<0.001 <0.001 0.37 0.04 0.79
however some other possible sources of bias, relevant for studies of service screening, not mentioned earlier, to take into consideration. If women in the control group had a mammogram on their own initiative, there is a possibility that the observed effect on mortality reduction could have been diluted. Opportunistic screening in Sweden is a phenomenon that mainly appears in large cities [20]. Stockholm is the only large city in the control group and constitutes 66% of the control group. We believe that opportunistic screening is less frequent among older than among younger women. The total breast cancer mortality time trends (Fig. 3) were similar in both groups except for a lower level in the study group from the late 1980s. However, the refined mortality indicates, for both outcome measures, a difference during the reference period towards a higher breast cancer mortality in the study group. The impact of adjustment for refined mortality in the reference period was substantial for the estimates of relative risk. Therefore, we believe that the adjustment was justified although the confidence intervals became wider. The difference in the reference period could indicate different background mortality in the study and control groups. The discrepancy between the number of breast cancer deaths in the defined cohorts (refined mortality) and the more commonly used total mortality for 70–84 or 75–84 years at death is considerable. In a study of the effects of screening in age 50–69 years on breast cancer mortality [10] only 27% of the deaths in the total mortality of women 50–79 years at death consisted of cases valid in the refined mortality model after 11 years of follow-up. Thus, especially when evaluating screening effects in certain age intervals, total mortality can be insufficiently discriminative for effects of screening. We have used two methods in measuring breast cancer mortality; individual underlying cause of death coded by the National Cause of Death Registry and excess mortality. Using individual data, it can be difficult in many cases to decide whether breast cancer is an underlying cause of death. The degree of complication increases by increasing age due
368
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369
Fig. 5. Ratio of the number of breast cancer deaths (UCD) (numerator) and excess mortality (EM) (denominator) by age groups in the invited group (IG) and the control group (CG) from the paper by Larsson et al. [21].
to intercurrent diseases [14]. Excess mortality compares observed and expected mortality among breast cancer patients. It is therefore a measure of all mortality caused from breast cancer. In the latest overview of all Swedish randomized studies these two measures were compared [21]. In the age group 70–74 years at randomization a 16% non-significant excess breast cancer mortality reduction was seen after a mean individual follow-up of 11–12 years. The corresponding relative risk was 1.05 when using UCD, thus a difference of 21%. The corresponding difference for the age group 50–69 years was 3–4%. In the present study, the difference between the two measures was 10–13%. A possible explanation for these differences could be the higher incidence of breast cancer in the study group due to screening. For some of the deceased breast cancer cases in a population it may be difficult to decide whether the underlying cause of death was breast cancer or not. However, for some of these cases the decision by the clinician would be breast cancer. On the other hand, if the same population had been screened the number of breast cancer cases would have been increased. If among this increment some deceased subjects also had caused decision problems there will be a possibility for a differential bias of overestimation of the number of breast cancer deaths in the study population. Using data from the paper mentioned above [21] the ratio between the breast cancer mortality (UCD) and the excess mortality from breast cancer for different ages at randomization is shown in Fig. 5 for the invited and the control group, respectively. The breast cancer mortality (UCD) was higher in the study group for all ages and the difference was most pronounced in the age group 70–74 years. This suggests that breast cancer mortality using UCD may be less valid when comparing a screened group with a group who had not been screened especially in women over 70 years. Thus, the relative risks using UCD might be biased.
To summarize, with a mean screening interval of 22.8 months and with a mean follow-up time of 10.1 years of the county based Swedish service-screening program for women of age 70–74 years, the reduction of the breast cancer excess mortality was estimated to be 18% when compared to the control group. Using the breast cancer mortality based on individual cause of death the corresponding result was 3%. Adjusting for the inclusion of cases in the study cohorts diagnosed before invitation and for lead time bias, the reduction was 24 and 6% for excess mortality and breast cancer mortality (UCD), respectively (Table 3).
Acknowledgements The present study was supported by the Swedish Cancer Society and the European Commission. The authors are indebted to the responsible radiologists in the respective screening centers who gave us valuable and necessary information by kindly answering the questionnaire; A.-M. Berggren, P. Bordas, S. Cederblom, J.-O. Englund, B. Epstein, E. Frodis, B. Heddson, H. Håkansson, J. Johanson, M. Kubista, M. Löfgren, T. Mathiesen, Z. von Pàlffej, L.F. Samuelsson, H. Svensson, P.-Å. Svensson, M. Tholin, E. Thurfjell, K. Yde.
References [1] Tabár L, Fagerberg CJG, Gad A, et al. Reduction in mortality from breast cancer after mass screening with mammography: randomised trial from the Breast Cancer Screening Working Group of the Swedish National Board of Health and Welfare. Lancet 1985;1:829–32. [2] Nyström L, Rutqvist L, Wall S, et al. Breast cancer screening with mammography: overview of Swedish randomised trials. Lancet 1993;341:973–8.
H. Jonsson et al. / Cancer Detection and Prevention 27 (2003) 360–369 [3] Hwang ES, Cody III HS. Does the proven benefit of mammography extend to breast cancer patients over age 70? South Med J 1998;91:522–6. [4] Kopans DB. Screening mammography in women over age 65. J. Gerontol. 1992;47(Spec. No.):59–62. [5] Hakama M, Pukkala E, Söderman B, et al. Implementation of screening as a public health policy: issues in design and evaluation. J Med Screen 1999;6:209–16. [6] Lenner P, Jonsson H. Excess mortality from breast cancer in relation to mammography screening in northern Sweden. J Med Screen 1997;4:6–9. [7] Tabár L, Vitak B, Chen HH, et al. Beyond randomized controlled trials. Cancer 2001;91:1724–31. [8] Thurfjell EL, Lindgren JAÅ. Breast cancer survival rates with mammographic screening: similar favourable survival rates for women younger and those older than 50 years. Radiology 1996;201:421–6. [9] Jonsson H, Törnberg S, Nyström L, et al. Service screening with mammography in Sweden: evaluation of effects from screening in age 40–49 years on the breast cancer mortality. Acta Oncol 2000;39:617– 23. [10] Jonsson H, Nyström L, Törnberg S, et al. Service screening with mammography of women aged 50–69 years in Sweden: effects on mortality from breast cancer. J Med Screen 2001;8:152–60. [11] van Dijck JA, Verbeek AL, Beex LV, et al. Breast-cancer mortality in a non-randomized trial on mammographic screening in women over age 65. Int J Cancer 1997;70:164–8. [12] Tabár L, Vitak B, Chen HH, et al. The Swedish Two-County trial twenty years later: updated mortality results and new insights
[13]
[14]
[15]
[16] [17] [18]
[19] [20]
[21]
369
from long-term follow-up. Radiol Clin North Am 2000;38:625– 51. NBHW. Mammographic screening for early detection of breast cancer (in Swedish). Stockholm: The National Board of Health and Welfare; 1986. Nyström L, Larsson L-G, Rutqvist LE, et al. Determination of cause of death among cancer cases in the Swedish randomized mammography screening trials: a comparison between official statistics and validation by an endpoint committee. Acta Oncol 1995;34:145–52. Hakama M, Pukkala E, Heikkila M, et al. Effectiveness of the public health policy for breast cancer screening in Finland: population based cohort study. Br Med J 1997;314:864–7. Breslow NE, Day NE. The design and analysis of cohort studies. Lyon: IARC; 1987. Venables WN, Ripley BD. Modern applied statistics with S-PLUS, 2nd ed. New York: Springer; 1997. Lidbrink EK, Törnberg SA, Azavedo EM, et al. The general mammography screening program in Stockholm. Acta Oncol 1994;33:353–8. Thurfjell E. Mammography screening: methods and diagnostic results. Thesis. Uppsala: Uppsala, Sweden; 1995. Törnberg S. A survey on mammography examinations performed in private clinics during the period 1992–1993 (in Swedish). Stockholm: Socialstyrelsen; 1994. Larsson LG, Nyström L, Wall S, et al. The Swedish randomised mammography screening trials: analysis of their effect on the breast cancer related excess mortality. J Med Screen 1996;3:129– 32.