The Breast xxx (2014) 1e6
Contents lists available at ScienceDirect
The Breast journal homepage: www.elsevier.com/brst
Original article
Breast cancer screening halves the risk of breast cancer death: A case-referent study Ellen Paap a, b, André L.M. Verbeek a, Anita A.M. Botterweck c, Heidi J. van Doorne-Nagtegaal c, Mechli Imhof-Tas d, e, Harry J. de Koning f, Suzie J. Otto f, Linda de Munck c, Annemieke van der Steen g, Roland Holland b, Gerard J. den Heeten b, h, Mireille J.M. Broeders a, b, * a
Department for Health Evidence, Radboud University Medical Center, Nijmegen, The Netherlands National Expert and Training Centre for Breast Cancer Screening, Nijmegen, The Netherlands c Comprehensive Cancer Centre the Netherlands, Utrecht, The Netherlands d Department of Radiology, Radboud University Medical Center, Nijmegen, The Netherlands e Screening Program Early Detection of Breast Cancer in the Eastern Part of the Netherlands, Nijmegen, The Netherlands f Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, The Netherlands g Screening Program Early Detection of Breast Cancer in the South-West Part of the Netherlands, Vlaardingen, The Netherlands h Department of Radiology, Academic Medical Centre, University of Amsterdam, The Netherlands b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 19 November 2013 Received in revised form 3 March 2014 Accepted 3 March 2014
Large-scale epidemiologic studies have consistently demonstrated the effectiveness of mammographic screening programs, however the benefits are still subject to debate. We estimated the effect of the Dutch screening program on breast cancer mortality. In a large multi-region case-referent study, we identified all breast cancer deaths in 2004 and 2005 in women aged 50e75 who had been invited for screening (cases). Cases were individually matched to referents from the population invited to screening. Conditional logistic regression was used to estimate the odds ratio (OR) of breast cancer death according to individual screening history. The OR was adjusted for self-selection bias using regional correction factors for the difference in baseline risk for breast cancer death between screened and unscreened women. A total of 1233 cases and 2090 referents were included in this study. We found a 58% reduction in breast cancer mortality in screened versus unscreened women (adjusted OR ¼ 0.42, 95% CI 0.33e0.53). Screening, i.e. early detection and treatment, has resulted in a substantial reduction in breast cancer mortality, indicating that the Dutch breast cancer screening program is highly effective. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Case-referent study Breast cancer Screening Effectiveness Mammography
Introduction Screening mammography is effective in reducing breast cancer mortality for women aged 50e69, as demonstrated in several breast cancer screening trials in the 1970s and 1980s [1]. On the basis of these trial results, mammographic service screening programs have been implemented in many Western countries in the 1990s to reduce the burden of breast cancer mortality. Over the last decade, large-scale epidemiologic studies have consistently demonstrated the effectiveness of these screening programs [2].
* Corresponding author. Department for Health Evidence, Radboud University Medical Center, P.O. Box 9101, 6500 HB Nijmegen, The Netherlands. Tel.: þ31 24 361 9132; fax: þ31 24 361 3505. E-mail address: [email protected] (M.J.M. Broeders).
Nevertheless, there are researchers who question the benefits, focusing in particular on whether the mortality reduction is large enough to justify the harms of screening [3,4]. To reliably assess the effect of screening on breast cancer mortality using observational research designs, individual data directly linking a woman’s cause of death and her screening history are preferred [5]. The case-referent approach offers an efficient methodological framework for the identification of screening participation during the time period before diagnosis of the case and its matched referent [6]. Case-referent studies have been shown to produce results that are similar to those from randomized controlled trials and which do not yield a systematic bias in the effect estimates [7,8]. However, some researchers are not convinced that case-referent studies are reliable for evaluating the impact of screening [9e11]. Their main criticism in the use of case-referent studies is that self-selection bias could distort the relation
http://dx.doi.org/10.1016/j.breast.2014.03.002 0960-9776/Ó 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
2
E. Paap et al. / The Breast xxx (2014) 1e6
between screening mammography and breast cancer mortality, because of the comparison between screened and not screened women [12]. In this study, we estimate the current benefit of the Dutch population-based screening program by comparing breast cancer mortality in screened versus unscreened women. To adjust for selfselection bias, we apply regional correction factors estimated from the implementation phase of the Dutch program using the incidence-based mortality (IBM) method [13]. Methods Dutch breast cancer screening program The population-based breast cancer screening program was gradually implemented in the Netherlands from 1989 onwards. Dutch screening policy recommended bilateral mammography for all women aged 50 to 69, with a biennial screening interval. Coverage of the target population, that is, the percentage of eligible women annually invited, increased from 11% in 1990, to 69% in 1993, and to full population capacity in 1996. In 1997, the upper age limit was extended to 75 years with full coverage in 2001. In 2009, more than 1.1 million women received an invitation to participate in the screening program, and over 900,000 women were screened (mean overall attendance rate of 81.3%) [14]. Between 1989 and 2009, the organization of the screening program was executed by nine regional screening organisations, of which five participated in our study: Stichting Kankerpreventie en escreening Limburg (SKsL), Stichting Kankerpreventie IKA (SKP IKA), Stichting Bevolkingsonderzoek Noord-Nederland (BBNN), Stichting Vroege Opsporing Kanker Oost-Nederland (SVOKON) and Stichting Bevolkingsonderzoek Borstkanker Zuidwest Nederland (SBBZWN). These screening regions cover more than half of the target population for screening in the Netherlands and were selected based on the geographical distribution of rural and urban areas and early and late implementation of the screening program. The screening organizations hold screening registries with individual data on invitation, participation and screening outcomes for all women in the target population of the screening program. Mammograms performed outside the screening program are not registered in the screening databases. Cases and referents We applied the case-referent design to evaluate the effect of mammographic screening, i.e. early detection and treatment, on breast cancer mortality [5]. In the last decades, increasing attention has been paid to the possible use of this study design for the evaluation of service screening impact in the routine health care environment [15]. We prefer the term case-referent study to the more commonly used term caseecontrol study because the uptake of screening in the case group of breast cancer deaths is referred to the probability of having been screened in the population from which the cases originate [16]. The source population in this multi-region study included all women aged 50e75 years who received at least one invitation to the service screening program in the five participating screening regions. Cases originated from the source population and were defined as women who died from breast cancer in 2004 or 2005. To identify the cases, we linked all members of the source population to the Netherlands Cancer Registry and identified breast cancer patients who died in 2004 or 2005. Since the cause of death is not registered in the Cancer Registry, we subsequently linked this group to the Cause of Death Registry of Statistics Netherlands. Based on the cause of death reported on the death certificates, we
identified the patients who had died from breast cancer. Coding of breast cancer as the underlying cause of death by Statistics Netherlands has been shown to be reliable in 95% of the investigated cases [17]. In addition, the Health Council of the Netherlands concluded in 2002 that the use of death certificates is valid for use as an end point for the evaluation of the screening program, as the misclassification rate is modest and is not affected by mode of detection [18e21]. For each case, we retrieved date of death, date of diagnosis, date of birth and the complete screening history (invitations and examinations). For women who had been invited for screening but never participated, linkages between screening registry and cancer registry were allowed but only with a very strict privacy procedure. Approval for these linkages was obtained from all contributing registries and data records were made available as anonymized records only. Cases were only included if their date of diagnosis was after the date of their first invitation for screening. In the case series of breast cancer deaths, we ascertained whether women were screened or not screened before breast cancer diagnosis and calculated the odds of having been screened in this period. To interpret the screening odds in the case group, we also calculated the screening odds in a reference group. For each case, referents were sampled from the source population and their complete screening history was included in the database. For the first two screening organizations included in this study, SKsL and BBNN, one referent was matched to each case. In the other three organizations it was possible to include two referents for each case. The referents were matched for year of birth and area of residence. To ensure that the referent had the same opportunity for participating in screening as the matched case, they had to be free of breast cancer at the time they received the invitation to screening. Further, referents had to be alive at the time of death of their case following the principles of incidence density sampling [22e24]. The case and the matched referent(s) formed a case-referent set. Screening history Screening participation can only influence breast cancer death if the screening examination is performed in the period when the breast cancer is potentially detectable on the mammogram and before symptoms appear [25]. The exact duration of this period is unknown for the individual patient, but is most likely to be within a 4-year period before clinical breast cancer diagnosis, based on estimates of lead time for breast cancer diagnosis [6,25]. We therefore set the time frame for screening participation at a maximum of two screening rounds preceding the diagnosis of the case. With a biennial screening policy, this period includes two consecutive screening invitations, that is, the index invitation e the most recent invitation before diagnosis of the case e and the screening invitation preceding the index. In the case-referent set, the referent was also assigned an index invitation from the same screening round from which the index invitation of the case was selected. Both the case and the referent are classified as screened if they participated in the screening examination following their index invitation and/ or the invitation in the screening round before the index invitation. For screen-detected cases, this was the screening examination at which breast cancer was detected. Self-selection bias Since participation in service screening is voluntary, selfselection may be present when women who decide to participate in screening have a different baseline risk of breast cancer mortality to those who decide not to accept the invitation for screening. This could be due, for example, to differences in ethnicity, a history of
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
E. Paap et al. / The Breast xxx (2014) 1e6
relatives with breast cancer, or socioeconomic circumstances [26,27]. Duffy et al. defined the quantity Dr used as a correction factor as follows: ‘Dr is the relative risk of BC death for noncompliers compared with an uninvited comparison group’ [28]. Applying Dr in the Dutch screening program requires estimation of this factor in the Dutch environment since self-selection factors are known to vary across countries/regions. To correct for potential self-selection bias in our study, we calculated a correction factor for each region from the implementation phase of the Dutch national screening program using the incidence-based mortality (IBM) method [29]. A detailed description of the estimation of the regional correction factors can be found in Paap et al. [13]. In short, an incidence-based mortality (IBM) rate was calculated by dividing the number of breast cancer deaths resulting from incident breast cancer cases diagnosed in a specific period with the accompanying number of person-years. Data was gathered from women eligible for invitation to screening during the implementation period of the screening program (1990e1995). During this implementation phase, the female population in the targeted age group shifted from an uninvited population to an invited population. This is the only period where we were able to identify (1) women invited and screened, (2) women invited but not screened, and (3) a unique contemporaneous group of women eligible for screening but not yet invited. We calculated the IBM rate for women not invited to the screening program and for those invited, but not screened. The numerator of the IBM rates included breast cancer deaths of women diagnosed with breast cancer in the years 1990e1995. We followed the breast cancer patients for breast cancer death for a maximum of 10 years after date of diagnosis. The breast cancer patients were divided in two groups according to their screening status: not invited for screening before diagnosis or invited and not screened before diagnosis. In order to define the denominator of the breast cancer mortality rates, we calculated person-years for the period 1990e1995 using aggregated data on the total number of invited women, the total number of screened women, and the age-specific total population. In order to divide the person-years in a group of not invited and a group of not screened person-years, we calculated the number of person-years during the implementation period of the screening in each municipality separately. The municipality is the smallest possible unit of which the starting date of screening was known. The correction factor is the rate ratio of breast cancer death for not screened versus not yet invited women. We used the correction factors per region to adjust for a difference in baseline risk for not screened versus screened women [28]. Further details about the calculation of the correction factors can be found in Paap et al. [13]. Statistical analysis All statistical analyses were performed using SAS statistical software, version 9.2 (SAS Institute Inc., Cary, NC). To estimate the effect of screening on breast cancer mortality, we used conditional logistic regression to calculate odds ratios (OR) and the 95% confidence intervals (CIs), taking into account the matching variables year of birth and area of residence [30]. The OR is the odds of having been screened vs. not screened in the case series of breast cancer deaths, compared with the odds in the reference group from which the cases theoretically originate. As such, the OR is the breast cancer mortality in screened women divided by the breast cancer mortality in not screened women [16]. However, the protective element in screening is the treatment that follows the screening test, when the test is truly positive. Therefore, the odds ratio in a case-referent study measures the combination of early detection followed by appropriate treatment [31]. In this study, we first conducted analyses for each of the five regions separately. Second, to take self-selection bias into account,
3
we adjusted the regional ORs with the formula developed by Duffy et al., because of heterogeneity between the regional correction factors was present [13,28]. Finally, we pooled the regional corrected ORs to provide a national estimate for breast cancer mortality reduction due to screening, using the inverse variance method [32]. Results The number of screening invitations varied between the five participating regions, ranging from 133,816 in the SKsL region to 356,973 in the IKA region (Table 1). The overall participation rate was 80.8%, with the lowest participation reported in the more urban SBBZWN region (77.6%) and the highest in the more rural BBNN region (84.6%). A total of 1233 cases and 2090 referents were included in this study. Of the cases, 49% were aged 49e59 at the index invitation, 38% aged 60e69 and 13% aged 70e75. Because cases and referents were matched for year of birth, these proportions were the same for the referents. The mean age at diagnosis for the cases was 61.9 years (standard deviation ¼ 7.3 years). Most cases were diagnosed in the years 2001e2005 (53%) compared to 36% in the years 1996e2000 and 11% in the years 1990e1995. Without correction for self-selection bias, the overall OR showed a breast cancer mortality reduction of 52% (OR ¼ 0.48 [0.40 to 0.58]) for screened women compared to not screened women (Table 1). The regional ORs, without correction for self-selection, ranged from 33% (0.67 [0.42 to 1.08]) mortality reduction in the BBNN region to 73% (0.27 [0.12 to 0.62]) in the SKsL region. In the SBBZWN, SKsL and SVOKON regions the correction factor was not different from one, indicating no presence of self-selection in those regions. The correction factor for BBNN was 0.64 [0.46 to 0.90] and for IKA 0.77 [0.63 to 0.93], both indicating a lower baseline risk in women who do not attend screening versus screened women. Table 2 presents an example of the correction for self-selection using details for the BBNN region. Pooling of the regional adjusted ORs resulted in a breast cancer mortality reduction of 58% (0.42 [0.33 to 0.53], Table 1). Discussion This study is the first multi-region study using individual (patient) data to provide an estimate of the effect of screening, i.e. early detection and treatment, on breast cancer mortality in the Netherlands. Our results demonstrate that the risk of dying from breast cancer is more than halved in screened women vs. not screened, taking the effect of self-selection into account (0.42 [0.33 to 0.53]). A unique strength of this study is that we used regionspecific correction factors to adjust for self-selection bias. In an ancillary study, we showed that self-selection only has a minor influence on the estimate of breast cancer screening effectiveness from caseecontrol studies in the Netherlands [13]. We chose to focus our study on women who died recently, i.e. in 2004 or 2005, in order to obtain an estimate of the effectiveness of current service screening in a steady state situation. Although not confirmed in every country [33], it is generally thought that an optimal screening effect cannot be achieved in the first years after the implementation of a screening program [34]. In those years prevalent screen-detected cases, i.e. women diagnosed in the first round of screening or in a woman’s first test at a subsequent round cases, dominate the screening program [29]. By including cases who died in 2004 or 2005, the proportion of women diagnosed in the first years after the start of a screening program is small which results in an estimate of the current benefit of the screening program.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
4
E. Paap et al. / The Breast xxx (2014) 1e6
Table 1 Background information on the participating regions, and odds ratios with and without correction for self-selection bias for the Netherlands and the five screening regions separately.
Total BBNN IKA SKsL SBBZWN SVOKON a
Number of invitations (2004 and 2005)
Attendance rate (%) (2004 and 2005)
Cases/referents (N)
Odds ratio previous and index invitation (95%-CI)
Correction factor for self-selection (95%-CI) [13]
Odds ratio adjusted for self-selection bias (95%-CI)
1,256,688 299,256 356,973 133,816 296,051 170,592
80.8 84.6 78.8 83.8 77.6 81.9
1233/2090 258/258 343/686 118/118 330/660 184/368
0.48 0.67 0.52 0.27 0.44 0.46
0.64 0.77 0.92 1.08 1.08
0.42 0.40 0.38 0.24 0.49 0.51
(0.40e0.58) (0.42e1.08) (0.38e0.73) (0.12e0.62) (0.32e0.60) (0.30e0.72)
(0.46e0.90) (0.63e0.93) (0.65e1.30) (0.82e1.43) (0.85e1.37)
(0.33e0.53)a (0.22e0.74) (0.25e0.57) (0.10e0.62) (0.30e0.78) (0.30e0.87)
Pooled odds ratio based on the five regional adjusted odds ratios.
We demonstrate a breast cancer mortality reduction of 58% (0.42 [0.33 to 0.53] for women participating in screening. In other countries, 25%e65% breast cancer mortality reductions have been found when comparing screened women to not screened women [15]. In addition, a meta-analysis of observational studies on population-based screening in Europe reported that the relative reduction in breast cancer mortality for women who actually participated in screening was 38% (0.62 [0.56 to 0.69]) based on incidence-based mortality studies and 48% (0.52 [0.42 to 0.65], corrected for self-selection) based on caseecontrol studies [2]. Using slightly different selection criteria, another meta-analysis of published caseecontrol studies found a mortality reduction of 49% (0.51 [0.46e0.55], not corrected for self-selection) [33]. In the Netherlands, regional estimates of the impact of screening from caseecontrol studies have consistently demonstrated the effectiveness of the Dutch screening program in those regions [35e37]. In line with these regional studies, this multi-region study confirms that the centrally organized Dutch population-based screening program is highly effective in reducing breast cancer mortality. Our study shows a larger mortality reduction than the results from the breast cancer screening trials from the 1970s and 1980s as well as the more recent conducted cohort studies or IBM studies [8]. These differences are largely explained by differences in the comparison groups: invited versus not invited women in contrast to screened versus not screened women. The randomized controlled trials showed mortality reductions of 20e30% based on intention to treat analyses, thus comparing the breast cancer mortality in invited women with not-invited women. In addition, cohort studies or IBM studies mainly compare invited versus not invited women as well. In case-referent studies screened women are compared with not screened women, thus resulting in a larger effect estimate of screening than the RCTs, cohort or IBM-studies. This comparison between screened and unscreened women is performed in the group of invited women only. Based on an attendance rate of 80% in the Netherlands, translation of our effect in those actually screened to an intention to treat analysis would result in a reduction equivalent to 48% (95% CI 39%e56%) [28]. The current European best estimate ranges from 25% (incidence-based mortality studies) to 31% (caseecontrol studies) for women invited for screening [2]. Thus,
Table 2 Example of adjustment of the odds ratio (OR) for self-selection bias in the BBNN region. Formula of Duffy [28]
p j Dr/(1-(1-p)Dr)
Where j ¼ unadjusted odds ratio (95% CI) p ¼ attendance rate of screening Dr ¼ correction factor for self-selection (95% CI) OR adjusted for self-selection bias (95%CI)
0.67 0.85 0.64 0.85 0.40
(0.42e1.08) (0.46e0.90) 0.67 0.64/(1 0.15 0.64) (0.22e0.74)
our effect is still larger than the results of the RCTs and cohort or IBM studies, which may be explained by the centralized organization of the Dutch screening program including a centralized technical and medical quality assurance program and audits [38]. When using a case-referent study to estimate the effectiveness of a screening program, strenuous efforts need to be made to correct for any potential bias in order to obtain a valid result. Selfselection is generally considered the most difficult form of bias to deal with in the context of case-referent studies on cancer screening. Because participation in screening is voluntary, selection factors related to both the likelihood of being screened and the risk of dying from breast cancer may confound the estimates of effectiveness. Only a limited number of studies so far have used country or region-specific estimates for self-selection [28,35,39e42]. The earliest estimate, a correction factor of 1.36, was derived from randomized controlled trial data [28] and has been shown to differ from the current service screening situation where correction factors currently range from 0.84 to 1.17 [15]. In this large multi-region study we found that even within one country, self-selection factors differed by region. However, our estimate for self-selection in the SBBZWN region was consistent with the estimate derived using a slightly different method in another caseecontrol study in this region [35]. Otto et al. used individual data on breast cancer mortality in nonparticipants from their study period 1990 to 2003 and aggregated data on contemporaneous groups of nonparticipants and uninvited women in the implementation period (1990e1995) from the same region. Their correction factor of 1.11 was remarkably similar to the 1.08 in this study [43]. Another approach to deal with differences in background risk for women screened and not screened would be stratification on risk [44]. Age is strongly related with both the occurrence of breast cancer death and screening participation. In this study we matched for age, thereby correcting for the influence of this confounder. Other than age we had no background information available on other possible risk factors for breast cancer death in the screening databases. In an ancillary study, we therefore developed realistic scenarios for the prevalence and strength of risk factors on screening and not screened groups and explored the impact of residual confounding bias (details published elsewhere) [45]. The results of this study demonstrated that residual confounding, after adjustment for age, by risk factors like mammographic density, socioeconomic status and obesity, only played a minor role in the evaluation of screening programs [45]. For example, if a risk factor with a relative risk of four is present in 10% of the screened women compared to 20% in the not screened women, the OR of 0.35 would change to 0.43 (details published elsewhere) [37,45]. Although we cannot rule out the existence of residual confounding completely, we believe that any other confounding factors will not have had a substantial influence on the estimated benefit of the Dutch screening program. In addition, a similar finding is reported by Nickson et al., who found that their OR varied very little in any of the sensitivity
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
E. Paap et al. / The Breast xxx (2014) 1e6
analyses they performed and concluded that “they were unable to identify biases that could negate their finding” [33]. Nickson et al. adjusted their caseecontrol study to potential confounders e socioeconomic status and rurality e which made a negligible difference to their estimates (adjusted OR 0.48, unadjusted OR 0.50). External adjustment to assess bias due to unobserved confounders for 2 other potential confounders (family history and hormone replacement therapy use), also changed their results negligible (OR 0.49 and 0.45, respectively) [46]. A limitation of the current study is that the correction factors for self-selection are derived from the implementation period for screening, more than two decades ago. These historical estimates are also valid for current screening practice if self-selection, i.e. the differential between not screened and not invited women, has not changed over the years. We believe this is a reasonable assumption given the consistently high participation rate (around 80%) in the Dutch national screening program. In addition, more than ninety percent of the women participating in the screening program is a regular participant [47]. Deriving a more recent estimate for selfselection is not straightforward, since a well-defined control group is no longer available once nation-wide screening is offered to all women in the target group. Another limitation of this study is that we could not correct for any potential influence of opportunistic screening, i.e. mammograms undertaken outside the screening program for screening purpose. The extent of opportunistic screening in the Netherlands is not known, but has been demonstrated in some other countries [48,49]. In conclusion, our findings show that the current Dutch breast cancer screening program has achieved a substantial reduction in breast cancer mortality. Case-referent studies have been used since the beginning of the introduction of screening. In fact, Sasco et al. suggested in 1986 that a routine caseecontrol assessment of the functioning of a mass screening program could be, or even should be, an integral part of ongoing evaluation [50]. Moreover, several studies corroborated that well-designed observational studies produce results that are similar to those from randomized controlled trials and do not yield a systematic bias in the effect estimates [7,8]. However, it is clear from the current scientific debate that some researchers are not convinced that caseecontrol studies can be considered reliable designs for evaluating the impact of screening [9e12]. This study was part of a larger project to assess the validity of the case-referent design as a monitoring and evaluation tool. We have shown that careful attention to the design phase of the case-referent study, taking full benefit of state-of-the-art methodological insights regarding issues such as selection of cases and referents, definition of exposure (i.e. screening history) [15], and explicitly addressing biases [13,45] results in a valid estimate of screening effectiveness [5]. Conflict of interest statement H.J.K. has received funding from SCOR Global Life SE (SGL). G.J.H. received payment from Philips Health care for an invitational lecture in Dubai. The authors reported no other financial interests related to this research. Acknowledgments This study was supported by a research grant from the Dutch Cancer Society (KUN 2006-3571). We thank Stichting Kanker preventie en escreening Limburg (SKsL), Stichting Kankerpreventie IKA (SKP IKA), Stichting Bevolkingsonderzoek Noord-Nederland (BBNN), Stichting Vroege Opsporing Kanker Oost-Nederland (SVOKON), Stichting Bevolkingsonderzoek Borstkanker Zuidwest Nederland (SBBZWN), Netherlands Cancer Registry and Statistics Netherlands for providing the data.
5
References [1] Nelson HD, Tyne K, Naik A, Bougatsos C, Chan BK, Humphrey L. Screening for breast cancer: an update for the U.S. Preventive Services Task Force. Ann Intern Med 2009;151(10):727e42. [2] Broeders M, Moss S, Nystrom L, Njor S, Jonsson H, Paap E, et al. The impact of mammographic screening on breast cancer mortality in Europe: a review of observational studies. J Med Screen 2012;19(Suppl. 1):14e25. [3] Jorgensen KJ, Keen JD, Gotzsche PC. Is mammographic screening justifiable considering its substantial overdiagnosis rate and minor effect on mortality? Radiology 2011 Sep;260(3):621e7. [4] McPherson K. Screening for breast cancerebalancing the debate. BMJ 2010;340:c3106. [5] Verbeek ALM, Broeders MJM. Evaluation of cancer service screening: case referent studies recommended. Stat Methods Med Res 2010;19:487e505. [6] Broeders MJM, Verbeek ALM. Mammographic screening only matters in the detectable preclinical period of breast cancer [letter]. J Med Screen 2005;12(2):107. [7] Concato J, Shah N, Horwitz RI. Randomized, controlled trials, observational studies, and the hierarchy of research designs. N Engl J Med 2000;342(25): 1887e92. [8] Demissie K, Mills OF, Rhoads GG. Empirical comparison of the results of randomized controlled trials and case-control studies in evaluating the effectiveness of screening mammography. J Clin Epidemiol 1998;51(2):81e91. [9] Corder AP. Screening for breast cancer. “Mega-trial” needed. BMJ 2010;341: c4453. [10] Harris R, Yeatts J, Kinsinger L. Breast cancer screening for women ages 50 to 69 years a systematic review of observational evidence. Prev Med 2011 Sep;53(3):108e14. [11] Jorgensen KJ. Flawed methods explain the effect of mammography screening in Nijmegen. Br J Cancer 2011 Aug 9;105(4):592e3 [author reply 4e5]. [12] Autier P, Boniol M. Mammography screening and breast cancer mortalitye letter. Cancer Epidemiol Biomarkers Prev 2012 May;21(5):869 [author reply 70e1]. [13] Paap E, Verbeek ALM, Puliti D, Broeders MJM, Paci E. Minor influence of selfselection bias on the effectiveness of breast cancer screening in case-control studies in the Netherlands. J Med Screen 2011;18:142e6. [14] National Evaluation Team for Breast Cancer Screening. NETB interim report 2011. Main results 2008e2009 breast cancer screening programme in the Netherlands. Rotterdam: Dept. of Public Health, Erasmus MC, University Medical Center Rotterdam; 2011. [15] Paap E, Verbeek ALM, Puliti D, Paci E, Broeders MJM. Breast cancer screening case-control study design: impact on breast cancer mortality. Ann Oncol 2011;22(4):863e9. [16] Miettinen O. Estimability and estimation in case-referent studies. Am J Epidemiol 1976 Feb;103(2):226e35. [17] Harteloh P, de Bruin K, Kardaun J. The reliability of cause-of-death coding in The Netherlands. Eur J Epidemiol 2010 Aug;25(8):531e8. [18] Garne JP, Aspegren K, Balldin G. Breast cancer as cause of deathea study over the validity of the officially registered cause of death in 2631 breast cancer patients dying in Malmo, Sweden 1964e1992. Acta Oncol 1996;35(6): 671e5. [19] Goldoni CA, Bonora K, Ciatto S, Giovannetti L, Patriarca S, Sapino A, et al. Misclassification of breast cancer as cause of death in a service screening area. Cancer Causes Control 2009;20(5):533e8. [20] Nyström L, Larsson LG, Rutqvist LE, Lindgren A, Lindqvist M, Ryden S. Determination of cause of death among breast cancer cases in the Swedish randomized mammography screening trials; a comparison between official statistics and validation by an endpoint committee. Acta Oncol 1995;34:145e52. [21] Health Council of the Netherlands. Usefulness of population breast cancer screening (in Dutch). The Hague: Health Council of the Netherlands; 2002. publication no. 2002/03. [22] Greenland S, Thomas DC. On the need for the rare disease assumption in caseecontrol studies. Am J Epidemiol 1982 Sep;116(3):547e53. [23] Vandenbroucke JP, Pearce N. Caseecontrol studies: basic concepts. Int J Epidemiol 2012 Oct;41(5):1480e9. [24] Vandenbroucke JP, Pearce N. Incidence rates in dynamic populations. Int J Epidemiol 2012 Oct;41(5):1472e9. [25] Weiss NS, McKnight B, Stevens NG. Approaches to the analysis of case-control studies of the efficacy of screening for cancer. Am J Epidemiol 1992;135(7): 817e23. [26] Roder D, Houssami N, Farshid G, Gill G, Luke C, Downey P, et al. Population screening and intensity of screening are associated with reduced breast cancer mortality: evidence of efficacy of mammography screening in Australia. Breast Cancer Res Treat 2008;108(3):409e16. [27] Zackrisson S, Andersson I, Manjer J, Janzon L. Non-attendance in breast cancer screening is associated with unfavourable socio-economic circumstances and advanced carcinoma. Int J Cancer 2004;108(5):754e60. [28] Duffy SW, Cuzick J, Tabar L, Vitak B, Chen THH, Yen MF, et al. Correcting for non-compliance bias in case-control studies to evaluate cancer screening programmes. Appl Stat 2002;51(Part 2):235e43. [29] Paci E, Duffy SW, Giorgi D, Zappa M, Crocetti E, Vezzosi V, et al. Quantification of the effect of mammographic screening on fatal breast cancers: the Florence Programme 1990e96. Br J Cancer 2002;87(1):65e9.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
6
E. Paap et al. / The Breast xxx (2014) 1e6
[30] Rothman K, Greenland S, Lash T. Modern epidemiology. Philadelphia: Wolters Kluwer Health/Lippincott Williams &Wilkins; 2008. [31] Weiss NS. Application of the case-control method in the evaluation of screening. Epidemiol Rev 1994;16(1):102e8. [32] Egger M, Davey Smith G, Altman DG. Systematic reviews in health care: metaanalysis in context. 2nd ed. London: BMJ Books; 2001. [33] Nickson C, Mason KE, English DR, Kavanagh AM. Mammographic screening and breast cancer mortality: a case-control study and meta-analysis. Cancer Epidemiol Biomarkers Prev 2012 Sep;21(9):1479e88. [34] Fielder HM, Warwick J, Brook D, Gower-Thomas K, Cuzick J, Monypenny I, et al. A case-control study to estimate the impact on breast cancer death of the breast screening programme in Wales. J Med Screen 2004;11(4):194e8. [35] Otto SJ, Fracheboud J, Verbeek AL, Boer R, Reijerink-Verheij JC, Otten JD, et al. Mammography screening and risk of breast cancer death: a population-based case-control study. Cancer Epidemiol Biomarkers Prev 2012 Jan;21(1):66e73. [36] Paap E, Holland R, den Heeten GJ, van Schoor G, Botterweck AAM, Verbeek ALM, et al. A remarkable reduction of breast cancer deaths in screened versus unscreened women: a case-referent study. Cancer Causes Control 2010;21(10):1569e73. [37] van Schoor G, Moss SM, Otten JD, Donders R, Paap E, den Heeten GJ, et al. Increasingly strong reduction in breast cancer mortality due to screening. Br J Cancer 2011;104(6):910e4. [38] Holland R, Rijken H, Hendriks JHCL. The Dutch population-based mammography screening: 30-year experience. Breast Care 2007;2:12e8. [39] Gabe R, Tryggvadottir L, Sigfusson BF, Olafsdottir GH, Sigurdsson K, Duffy SW. A case-control study to estimate the impact of the Icelandic population-based mammography screening program on breast cancer death. Acta Radiol 2007;48(9):948e55. [40] Olsen AH, Njor SH, Vejborg I, Schwartz W, Dalgaard P, Jensen MB, et al. Breast cancer mortality in Copenhagen after introduction of mammography screening: cohort study. BMJ 2005;330(7485):220.
[41] Puliti D, Miccinesi G, Collina N, De Lisi V, Federico M, Ferretti S, et al. Effectiveness of service screening: a case-control study to assess breast cancer mortality reduction. Br J Cancer 2008;99(3):423e7. [42] Swedish Organised Service Screening Evaluation Group. Reduction in breast cancer mortality from the organised service screening with mammography: 2. Validation with alternative analytic methods. Cancer Epidemiol Biomarkers Prev 2006;15(1):52e6. [43] Otto SJ, Fracheboud J, Verbeek AL, Boer R, Reijerink-Verheij J, Otten JD, et al. Mammography screening and breast cancer mortality e response. Cancer Epidemiol Biomarkers Prev 2012;21:870e1. [44] Elmore JG, Reisch LM, Barton MB, Barlow WE, Rolnick S, Harris EL, et al. Efficacy of breast cancer screening in the community according to risk level. J Natl Cancer Inst 2005;97(14):1035e43. [45] van Schoor G, Paap E, Broeders MJM, Verbeek ALM. Residual confounding after adjustment for age: a minor issue in breast cancer screening effectiveness. Eur J Epidemiol 2011;26:585e8. [46] Nickson C, Mason KE, English DR, Kavanagh AM. Screening and breast cancer mortality e response. Cancer Epidemiol Biomarkers Prev 2012;21(12):2276e7. [47] Health Council of the Netherlands. Population screening of breast cancer: expectations and developments (in Dutch). The Hague: Health Council of the Netherlands; 2014. Publication no. 2014/01. [48] Hoff SR, Klepp O, Hofvind S. Asymptomatic breast cancer in non-participants of the national screening programme in Norway: a confounding factor in evaluation? J Med Screen 2012;19(4):177e83. [49] Vanier A, Leux C, Allioux C, Billon-Delacour S, Lombrail P, Molinie F. Are prognostic factors more favorable for breast cancer detected by organized screening than by opportunistic screening or clinical diagnosis? A study in Loire-Atlantique (France). Cancer Epidemiol 2013;37(5):683e7. [50] Sasco AJ, Day NE, Walter SD. Case-control studies for the evaluation of screening. J Chronic Dis 1986;39(5):399e405.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
Contents lists available at ScienceDirect
The Breast journal homepage: www.elsevier.com/brst
Original article
Breast cancer screening halves the risk of breast cancer death: A case-referent study Ellen Paap a, b, André L.M. Verbeek a, Anita A.M. Botterweck c, Heidi J. van Doorne-Nagtegaal c, Mechli Imhof-Tas d, e, Harry J. de Koning f, Suzie J. Otto f, Linda de Munck c, Annemieke van der Steen g, Roland Holland b, Gerard J. den Heeten b, h, Mireille J.M. Broeders a, b, * a
Department for Health Evidence, Radboud University Medical Center, Nijmegen, The Netherlands National Expert and Training Centre for Breast Cancer Screening, Nijmegen, The Netherlands c Comprehensive Cancer Centre the Netherlands, Utrecht, The Netherlands d Department of Radiology, Radboud University Medical Center, Nijmegen, The Netherlands e Screening Program Early Detection of Breast Cancer in the Eastern Part of the Netherlands, Nijmegen, The Netherlands f Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, The Netherlands g Screening Program Early Detection of Breast Cancer in the South-West Part of the Netherlands, Vlaardingen, The Netherlands h Department of Radiology, Academic Medical Centre, University of Amsterdam, The Netherlands b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 19 November 2013 Received in revised form 3 March 2014 Accepted 3 March 2014
Large-scale epidemiologic studies have consistently demonstrated the effectiveness of mammographic screening programs, however the benefits are still subject to debate. We estimated the effect of the Dutch screening program on breast cancer mortality. In a large multi-region case-referent study, we identified all breast cancer deaths in 2004 and 2005 in women aged 50e75 who had been invited for screening (cases). Cases were individually matched to referents from the population invited to screening. Conditional logistic regression was used to estimate the odds ratio (OR) of breast cancer death according to individual screening history. The OR was adjusted for self-selection bias using regional correction factors for the difference in baseline risk for breast cancer death between screened and unscreened women. A total of 1233 cases and 2090 referents were included in this study. We found a 58% reduction in breast cancer mortality in screened versus unscreened women (adjusted OR ¼ 0.42, 95% CI 0.33e0.53). Screening, i.e. early detection and treatment, has resulted in a substantial reduction in breast cancer mortality, indicating that the Dutch breast cancer screening program is highly effective. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Case-referent study Breast cancer Screening Effectiveness Mammography
Introduction Screening mammography is effective in reducing breast cancer mortality for women aged 50e69, as demonstrated in several breast cancer screening trials in the 1970s and 1980s [1]. On the basis of these trial results, mammographic service screening programs have been implemented in many Western countries in the 1990s to reduce the burden of breast cancer mortality. Over the last decade, large-scale epidemiologic studies have consistently demonstrated the effectiveness of these screening programs [2].
* Corresponding author. Department for Health Evidence, Radboud University Medical Center, P.O. Box 9101, 6500 HB Nijmegen, The Netherlands. Tel.: þ31 24 361 9132; fax: þ31 24 361 3505. E-mail address: [email protected] (M.J.M. Broeders).
Nevertheless, there are researchers who question the benefits, focusing in particular on whether the mortality reduction is large enough to justify the harms of screening [3,4]. To reliably assess the effect of screening on breast cancer mortality using observational research designs, individual data directly linking a woman’s cause of death and her screening history are preferred [5]. The case-referent approach offers an efficient methodological framework for the identification of screening participation during the time period before diagnosis of the case and its matched referent [6]. Case-referent studies have been shown to produce results that are similar to those from randomized controlled trials and which do not yield a systematic bias in the effect estimates [7,8]. However, some researchers are not convinced that case-referent studies are reliable for evaluating the impact of screening [9e11]. Their main criticism in the use of case-referent studies is that self-selection bias could distort the relation
http://dx.doi.org/10.1016/j.breast.2014.03.002 0960-9776/Ó 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
2
E. Paap et al. / The Breast xxx (2014) 1e6
between screening mammography and breast cancer mortality, because of the comparison between screened and not screened women [12]. In this study, we estimate the current benefit of the Dutch population-based screening program by comparing breast cancer mortality in screened versus unscreened women. To adjust for selfselection bias, we apply regional correction factors estimated from the implementation phase of the Dutch program using the incidence-based mortality (IBM) method [13]. Methods Dutch breast cancer screening program The population-based breast cancer screening program was gradually implemented in the Netherlands from 1989 onwards. Dutch screening policy recommended bilateral mammography for all women aged 50 to 69, with a biennial screening interval. Coverage of the target population, that is, the percentage of eligible women annually invited, increased from 11% in 1990, to 69% in 1993, and to full population capacity in 1996. In 1997, the upper age limit was extended to 75 years with full coverage in 2001. In 2009, more than 1.1 million women received an invitation to participate in the screening program, and over 900,000 women were screened (mean overall attendance rate of 81.3%) [14]. Between 1989 and 2009, the organization of the screening program was executed by nine regional screening organisations, of which five participated in our study: Stichting Kankerpreventie en escreening Limburg (SKsL), Stichting Kankerpreventie IKA (SKP IKA), Stichting Bevolkingsonderzoek Noord-Nederland (BBNN), Stichting Vroege Opsporing Kanker Oost-Nederland (SVOKON) and Stichting Bevolkingsonderzoek Borstkanker Zuidwest Nederland (SBBZWN). These screening regions cover more than half of the target population for screening in the Netherlands and were selected based on the geographical distribution of rural and urban areas and early and late implementation of the screening program. The screening organizations hold screening registries with individual data on invitation, participation and screening outcomes for all women in the target population of the screening program. Mammograms performed outside the screening program are not registered in the screening databases. Cases and referents We applied the case-referent design to evaluate the effect of mammographic screening, i.e. early detection and treatment, on breast cancer mortality [5]. In the last decades, increasing attention has been paid to the possible use of this study design for the evaluation of service screening impact in the routine health care environment [15]. We prefer the term case-referent study to the more commonly used term caseecontrol study because the uptake of screening in the case group of breast cancer deaths is referred to the probability of having been screened in the population from which the cases originate [16]. The source population in this multi-region study included all women aged 50e75 years who received at least one invitation to the service screening program in the five participating screening regions. Cases originated from the source population and were defined as women who died from breast cancer in 2004 or 2005. To identify the cases, we linked all members of the source population to the Netherlands Cancer Registry and identified breast cancer patients who died in 2004 or 2005. Since the cause of death is not registered in the Cancer Registry, we subsequently linked this group to the Cause of Death Registry of Statistics Netherlands. Based on the cause of death reported on the death certificates, we
identified the patients who had died from breast cancer. Coding of breast cancer as the underlying cause of death by Statistics Netherlands has been shown to be reliable in 95% of the investigated cases [17]. In addition, the Health Council of the Netherlands concluded in 2002 that the use of death certificates is valid for use as an end point for the evaluation of the screening program, as the misclassification rate is modest and is not affected by mode of detection [18e21]. For each case, we retrieved date of death, date of diagnosis, date of birth and the complete screening history (invitations and examinations). For women who had been invited for screening but never participated, linkages between screening registry and cancer registry were allowed but only with a very strict privacy procedure. Approval for these linkages was obtained from all contributing registries and data records were made available as anonymized records only. Cases were only included if their date of diagnosis was after the date of their first invitation for screening. In the case series of breast cancer deaths, we ascertained whether women were screened or not screened before breast cancer diagnosis and calculated the odds of having been screened in this period. To interpret the screening odds in the case group, we also calculated the screening odds in a reference group. For each case, referents were sampled from the source population and their complete screening history was included in the database. For the first two screening organizations included in this study, SKsL and BBNN, one referent was matched to each case. In the other three organizations it was possible to include two referents for each case. The referents were matched for year of birth and area of residence. To ensure that the referent had the same opportunity for participating in screening as the matched case, they had to be free of breast cancer at the time they received the invitation to screening. Further, referents had to be alive at the time of death of their case following the principles of incidence density sampling [22e24]. The case and the matched referent(s) formed a case-referent set. Screening history Screening participation can only influence breast cancer death if the screening examination is performed in the period when the breast cancer is potentially detectable on the mammogram and before symptoms appear [25]. The exact duration of this period is unknown for the individual patient, but is most likely to be within a 4-year period before clinical breast cancer diagnosis, based on estimates of lead time for breast cancer diagnosis [6,25]. We therefore set the time frame for screening participation at a maximum of two screening rounds preceding the diagnosis of the case. With a biennial screening policy, this period includes two consecutive screening invitations, that is, the index invitation e the most recent invitation before diagnosis of the case e and the screening invitation preceding the index. In the case-referent set, the referent was also assigned an index invitation from the same screening round from which the index invitation of the case was selected. Both the case and the referent are classified as screened if they participated in the screening examination following their index invitation and/ or the invitation in the screening round before the index invitation. For screen-detected cases, this was the screening examination at which breast cancer was detected. Self-selection bias Since participation in service screening is voluntary, selfselection may be present when women who decide to participate in screening have a different baseline risk of breast cancer mortality to those who decide not to accept the invitation for screening. This could be due, for example, to differences in ethnicity, a history of
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
E. Paap et al. / The Breast xxx (2014) 1e6
relatives with breast cancer, or socioeconomic circumstances [26,27]. Duffy et al. defined the quantity Dr used as a correction factor as follows: ‘Dr is the relative risk of BC death for noncompliers compared with an uninvited comparison group’ [28]. Applying Dr in the Dutch screening program requires estimation of this factor in the Dutch environment since self-selection factors are known to vary across countries/regions. To correct for potential self-selection bias in our study, we calculated a correction factor for each region from the implementation phase of the Dutch national screening program using the incidence-based mortality (IBM) method [29]. A detailed description of the estimation of the regional correction factors can be found in Paap et al. [13]. In short, an incidence-based mortality (IBM) rate was calculated by dividing the number of breast cancer deaths resulting from incident breast cancer cases diagnosed in a specific period with the accompanying number of person-years. Data was gathered from women eligible for invitation to screening during the implementation period of the screening program (1990e1995). During this implementation phase, the female population in the targeted age group shifted from an uninvited population to an invited population. This is the only period where we were able to identify (1) women invited and screened, (2) women invited but not screened, and (3) a unique contemporaneous group of women eligible for screening but not yet invited. We calculated the IBM rate for women not invited to the screening program and for those invited, but not screened. The numerator of the IBM rates included breast cancer deaths of women diagnosed with breast cancer in the years 1990e1995. We followed the breast cancer patients for breast cancer death for a maximum of 10 years after date of diagnosis. The breast cancer patients were divided in two groups according to their screening status: not invited for screening before diagnosis or invited and not screened before diagnosis. In order to define the denominator of the breast cancer mortality rates, we calculated person-years for the period 1990e1995 using aggregated data on the total number of invited women, the total number of screened women, and the age-specific total population. In order to divide the person-years in a group of not invited and a group of not screened person-years, we calculated the number of person-years during the implementation period of the screening in each municipality separately. The municipality is the smallest possible unit of which the starting date of screening was known. The correction factor is the rate ratio of breast cancer death for not screened versus not yet invited women. We used the correction factors per region to adjust for a difference in baseline risk for not screened versus screened women [28]. Further details about the calculation of the correction factors can be found in Paap et al. [13]. Statistical analysis All statistical analyses were performed using SAS statistical software, version 9.2 (SAS Institute Inc., Cary, NC). To estimate the effect of screening on breast cancer mortality, we used conditional logistic regression to calculate odds ratios (OR) and the 95% confidence intervals (CIs), taking into account the matching variables year of birth and area of residence [30]. The OR is the odds of having been screened vs. not screened in the case series of breast cancer deaths, compared with the odds in the reference group from which the cases theoretically originate. As such, the OR is the breast cancer mortality in screened women divided by the breast cancer mortality in not screened women [16]. However, the protective element in screening is the treatment that follows the screening test, when the test is truly positive. Therefore, the odds ratio in a case-referent study measures the combination of early detection followed by appropriate treatment [31]. In this study, we first conducted analyses for each of the five regions separately. Second, to take self-selection bias into account,
3
we adjusted the regional ORs with the formula developed by Duffy et al., because of heterogeneity between the regional correction factors was present [13,28]. Finally, we pooled the regional corrected ORs to provide a national estimate for breast cancer mortality reduction due to screening, using the inverse variance method [32]. Results The number of screening invitations varied between the five participating regions, ranging from 133,816 in the SKsL region to 356,973 in the IKA region (Table 1). The overall participation rate was 80.8%, with the lowest participation reported in the more urban SBBZWN region (77.6%) and the highest in the more rural BBNN region (84.6%). A total of 1233 cases and 2090 referents were included in this study. Of the cases, 49% were aged 49e59 at the index invitation, 38% aged 60e69 and 13% aged 70e75. Because cases and referents were matched for year of birth, these proportions were the same for the referents. The mean age at diagnosis for the cases was 61.9 years (standard deviation ¼ 7.3 years). Most cases were diagnosed in the years 2001e2005 (53%) compared to 36% in the years 1996e2000 and 11% in the years 1990e1995. Without correction for self-selection bias, the overall OR showed a breast cancer mortality reduction of 52% (OR ¼ 0.48 [0.40 to 0.58]) for screened women compared to not screened women (Table 1). The regional ORs, without correction for self-selection, ranged from 33% (0.67 [0.42 to 1.08]) mortality reduction in the BBNN region to 73% (0.27 [0.12 to 0.62]) in the SKsL region. In the SBBZWN, SKsL and SVOKON regions the correction factor was not different from one, indicating no presence of self-selection in those regions. The correction factor for BBNN was 0.64 [0.46 to 0.90] and for IKA 0.77 [0.63 to 0.93], both indicating a lower baseline risk in women who do not attend screening versus screened women. Table 2 presents an example of the correction for self-selection using details for the BBNN region. Pooling of the regional adjusted ORs resulted in a breast cancer mortality reduction of 58% (0.42 [0.33 to 0.53], Table 1). Discussion This study is the first multi-region study using individual (patient) data to provide an estimate of the effect of screening, i.e. early detection and treatment, on breast cancer mortality in the Netherlands. Our results demonstrate that the risk of dying from breast cancer is more than halved in screened women vs. not screened, taking the effect of self-selection into account (0.42 [0.33 to 0.53]). A unique strength of this study is that we used regionspecific correction factors to adjust for self-selection bias. In an ancillary study, we showed that self-selection only has a minor influence on the estimate of breast cancer screening effectiveness from caseecontrol studies in the Netherlands [13]. We chose to focus our study on women who died recently, i.e. in 2004 or 2005, in order to obtain an estimate of the effectiveness of current service screening in a steady state situation. Although not confirmed in every country [33], it is generally thought that an optimal screening effect cannot be achieved in the first years after the implementation of a screening program [34]. In those years prevalent screen-detected cases, i.e. women diagnosed in the first round of screening or in a woman’s first test at a subsequent round cases, dominate the screening program [29]. By including cases who died in 2004 or 2005, the proportion of women diagnosed in the first years after the start of a screening program is small which results in an estimate of the current benefit of the screening program.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
4
E. Paap et al. / The Breast xxx (2014) 1e6
Table 1 Background information on the participating regions, and odds ratios with and without correction for self-selection bias for the Netherlands and the five screening regions separately.
Total BBNN IKA SKsL SBBZWN SVOKON a
Number of invitations (2004 and 2005)
Attendance rate (%) (2004 and 2005)
Cases/referents (N)
Odds ratio previous and index invitation (95%-CI)
Correction factor for self-selection (95%-CI) [13]
Odds ratio adjusted for self-selection bias (95%-CI)
1,256,688 299,256 356,973 133,816 296,051 170,592
80.8 84.6 78.8 83.8 77.6 81.9
1233/2090 258/258 343/686 118/118 330/660 184/368
0.48 0.67 0.52 0.27 0.44 0.46
0.64 0.77 0.92 1.08 1.08
0.42 0.40 0.38 0.24 0.49 0.51
(0.40e0.58) (0.42e1.08) (0.38e0.73) (0.12e0.62) (0.32e0.60) (0.30e0.72)
(0.46e0.90) (0.63e0.93) (0.65e1.30) (0.82e1.43) (0.85e1.37)
(0.33e0.53)a (0.22e0.74) (0.25e0.57) (0.10e0.62) (0.30e0.78) (0.30e0.87)
Pooled odds ratio based on the five regional adjusted odds ratios.
We demonstrate a breast cancer mortality reduction of 58% (0.42 [0.33 to 0.53] for women participating in screening. In other countries, 25%e65% breast cancer mortality reductions have been found when comparing screened women to not screened women [15]. In addition, a meta-analysis of observational studies on population-based screening in Europe reported that the relative reduction in breast cancer mortality for women who actually participated in screening was 38% (0.62 [0.56 to 0.69]) based on incidence-based mortality studies and 48% (0.52 [0.42 to 0.65], corrected for self-selection) based on caseecontrol studies [2]. Using slightly different selection criteria, another meta-analysis of published caseecontrol studies found a mortality reduction of 49% (0.51 [0.46e0.55], not corrected for self-selection) [33]. In the Netherlands, regional estimates of the impact of screening from caseecontrol studies have consistently demonstrated the effectiveness of the Dutch screening program in those regions [35e37]. In line with these regional studies, this multi-region study confirms that the centrally organized Dutch population-based screening program is highly effective in reducing breast cancer mortality. Our study shows a larger mortality reduction than the results from the breast cancer screening trials from the 1970s and 1980s as well as the more recent conducted cohort studies or IBM studies [8]. These differences are largely explained by differences in the comparison groups: invited versus not invited women in contrast to screened versus not screened women. The randomized controlled trials showed mortality reductions of 20e30% based on intention to treat analyses, thus comparing the breast cancer mortality in invited women with not-invited women. In addition, cohort studies or IBM studies mainly compare invited versus not invited women as well. In case-referent studies screened women are compared with not screened women, thus resulting in a larger effect estimate of screening than the RCTs, cohort or IBM-studies. This comparison between screened and unscreened women is performed in the group of invited women only. Based on an attendance rate of 80% in the Netherlands, translation of our effect in those actually screened to an intention to treat analysis would result in a reduction equivalent to 48% (95% CI 39%e56%) [28]. The current European best estimate ranges from 25% (incidence-based mortality studies) to 31% (caseecontrol studies) for women invited for screening [2]. Thus,
Table 2 Example of adjustment of the odds ratio (OR) for self-selection bias in the BBNN region. Formula of Duffy [28]
p j Dr/(1-(1-p)Dr)
Where j ¼ unadjusted odds ratio (95% CI) p ¼ attendance rate of screening Dr ¼ correction factor for self-selection (95% CI) OR adjusted for self-selection bias (95%CI)
0.67 0.85 0.64 0.85 0.40
(0.42e1.08) (0.46e0.90) 0.67 0.64/(1 0.15 0.64) (0.22e0.74)
our effect is still larger than the results of the RCTs and cohort or IBM studies, which may be explained by the centralized organization of the Dutch screening program including a centralized technical and medical quality assurance program and audits [38]. When using a case-referent study to estimate the effectiveness of a screening program, strenuous efforts need to be made to correct for any potential bias in order to obtain a valid result. Selfselection is generally considered the most difficult form of bias to deal with in the context of case-referent studies on cancer screening. Because participation in screening is voluntary, selection factors related to both the likelihood of being screened and the risk of dying from breast cancer may confound the estimates of effectiveness. Only a limited number of studies so far have used country or region-specific estimates for self-selection [28,35,39e42]. The earliest estimate, a correction factor of 1.36, was derived from randomized controlled trial data [28] and has been shown to differ from the current service screening situation where correction factors currently range from 0.84 to 1.17 [15]. In this large multi-region study we found that even within one country, self-selection factors differed by region. However, our estimate for self-selection in the SBBZWN region was consistent with the estimate derived using a slightly different method in another caseecontrol study in this region [35]. Otto et al. used individual data on breast cancer mortality in nonparticipants from their study period 1990 to 2003 and aggregated data on contemporaneous groups of nonparticipants and uninvited women in the implementation period (1990e1995) from the same region. Their correction factor of 1.11 was remarkably similar to the 1.08 in this study [43]. Another approach to deal with differences in background risk for women screened and not screened would be stratification on risk [44]. Age is strongly related with both the occurrence of breast cancer death and screening participation. In this study we matched for age, thereby correcting for the influence of this confounder. Other than age we had no background information available on other possible risk factors for breast cancer death in the screening databases. In an ancillary study, we therefore developed realistic scenarios for the prevalence and strength of risk factors on screening and not screened groups and explored the impact of residual confounding bias (details published elsewhere) [45]. The results of this study demonstrated that residual confounding, after adjustment for age, by risk factors like mammographic density, socioeconomic status and obesity, only played a minor role in the evaluation of screening programs [45]. For example, if a risk factor with a relative risk of four is present in 10% of the screened women compared to 20% in the not screened women, the OR of 0.35 would change to 0.43 (details published elsewhere) [37,45]. Although we cannot rule out the existence of residual confounding completely, we believe that any other confounding factors will not have had a substantial influence on the estimated benefit of the Dutch screening program. In addition, a similar finding is reported by Nickson et al., who found that their OR varied very little in any of the sensitivity
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
E. Paap et al. / The Breast xxx (2014) 1e6
analyses they performed and concluded that “they were unable to identify biases that could negate their finding” [33]. Nickson et al. adjusted their caseecontrol study to potential confounders e socioeconomic status and rurality e which made a negligible difference to their estimates (adjusted OR 0.48, unadjusted OR 0.50). External adjustment to assess bias due to unobserved confounders for 2 other potential confounders (family history and hormone replacement therapy use), also changed their results negligible (OR 0.49 and 0.45, respectively) [46]. A limitation of the current study is that the correction factors for self-selection are derived from the implementation period for screening, more than two decades ago. These historical estimates are also valid for current screening practice if self-selection, i.e. the differential between not screened and not invited women, has not changed over the years. We believe this is a reasonable assumption given the consistently high participation rate (around 80%) in the Dutch national screening program. In addition, more than ninety percent of the women participating in the screening program is a regular participant [47]. Deriving a more recent estimate for selfselection is not straightforward, since a well-defined control group is no longer available once nation-wide screening is offered to all women in the target group. Another limitation of this study is that we could not correct for any potential influence of opportunistic screening, i.e. mammograms undertaken outside the screening program for screening purpose. The extent of opportunistic screening in the Netherlands is not known, but has been demonstrated in some other countries [48,49]. In conclusion, our findings show that the current Dutch breast cancer screening program has achieved a substantial reduction in breast cancer mortality. Case-referent studies have been used since the beginning of the introduction of screening. In fact, Sasco et al. suggested in 1986 that a routine caseecontrol assessment of the functioning of a mass screening program could be, or even should be, an integral part of ongoing evaluation [50]. Moreover, several studies corroborated that well-designed observational studies produce results that are similar to those from randomized controlled trials and do not yield a systematic bias in the effect estimates [7,8]. However, it is clear from the current scientific debate that some researchers are not convinced that caseecontrol studies can be considered reliable designs for evaluating the impact of screening [9e12]. This study was part of a larger project to assess the validity of the case-referent design as a monitoring and evaluation tool. We have shown that careful attention to the design phase of the case-referent study, taking full benefit of state-of-the-art methodological insights regarding issues such as selection of cases and referents, definition of exposure (i.e. screening history) [15], and explicitly addressing biases [13,45] results in a valid estimate of screening effectiveness [5]. Conflict of interest statement H.J.K. has received funding from SCOR Global Life SE (SGL). G.J.H. received payment from Philips Health care for an invitational lecture in Dubai. The authors reported no other financial interests related to this research. Acknowledgments This study was supported by a research grant from the Dutch Cancer Society (KUN 2006-3571). We thank Stichting Kanker preventie en escreening Limburg (SKsL), Stichting Kankerpreventie IKA (SKP IKA), Stichting Bevolkingsonderzoek Noord-Nederland (BBNN), Stichting Vroege Opsporing Kanker Oost-Nederland (SVOKON), Stichting Bevolkingsonderzoek Borstkanker Zuidwest Nederland (SBBZWN), Netherlands Cancer Registry and Statistics Netherlands for providing the data.
5
References [1] Nelson HD, Tyne K, Naik A, Bougatsos C, Chan BK, Humphrey L. Screening for breast cancer: an update for the U.S. Preventive Services Task Force. Ann Intern Med 2009;151(10):727e42. [2] Broeders M, Moss S, Nystrom L, Njor S, Jonsson H, Paap E, et al. The impact of mammographic screening on breast cancer mortality in Europe: a review of observational studies. J Med Screen 2012;19(Suppl. 1):14e25. [3] Jorgensen KJ, Keen JD, Gotzsche PC. Is mammographic screening justifiable considering its substantial overdiagnosis rate and minor effect on mortality? Radiology 2011 Sep;260(3):621e7. [4] McPherson K. Screening for breast cancerebalancing the debate. BMJ 2010;340:c3106. [5] Verbeek ALM, Broeders MJM. Evaluation of cancer service screening: case referent studies recommended. Stat Methods Med Res 2010;19:487e505. [6] Broeders MJM, Verbeek ALM. Mammographic screening only matters in the detectable preclinical period of breast cancer [letter]. J Med Screen 2005;12(2):107. [7] Concato J, Shah N, Horwitz RI. Randomized, controlled trials, observational studies, and the hierarchy of research designs. N Engl J Med 2000;342(25): 1887e92. [8] Demissie K, Mills OF, Rhoads GG. Empirical comparison of the results of randomized controlled trials and case-control studies in evaluating the effectiveness of screening mammography. J Clin Epidemiol 1998;51(2):81e91. [9] Corder AP. Screening for breast cancer. “Mega-trial” needed. BMJ 2010;341: c4453. [10] Harris R, Yeatts J, Kinsinger L. Breast cancer screening for women ages 50 to 69 years a systematic review of observational evidence. Prev Med 2011 Sep;53(3):108e14. [11] Jorgensen KJ. Flawed methods explain the effect of mammography screening in Nijmegen. Br J Cancer 2011 Aug 9;105(4):592e3 [author reply 4e5]. [12] Autier P, Boniol M. Mammography screening and breast cancer mortalitye letter. Cancer Epidemiol Biomarkers Prev 2012 May;21(5):869 [author reply 70e1]. [13] Paap E, Verbeek ALM, Puliti D, Broeders MJM, Paci E. Minor influence of selfselection bias on the effectiveness of breast cancer screening in case-control studies in the Netherlands. J Med Screen 2011;18:142e6. [14] National Evaluation Team for Breast Cancer Screening. NETB interim report 2011. Main results 2008e2009 breast cancer screening programme in the Netherlands. Rotterdam: Dept. of Public Health, Erasmus MC, University Medical Center Rotterdam; 2011. [15] Paap E, Verbeek ALM, Puliti D, Paci E, Broeders MJM. Breast cancer screening case-control study design: impact on breast cancer mortality. Ann Oncol 2011;22(4):863e9. [16] Miettinen O. Estimability and estimation in case-referent studies. Am J Epidemiol 1976 Feb;103(2):226e35. [17] Harteloh P, de Bruin K, Kardaun J. The reliability of cause-of-death coding in The Netherlands. Eur J Epidemiol 2010 Aug;25(8):531e8. [18] Garne JP, Aspegren K, Balldin G. Breast cancer as cause of deathea study over the validity of the officially registered cause of death in 2631 breast cancer patients dying in Malmo, Sweden 1964e1992. Acta Oncol 1996;35(6): 671e5. [19] Goldoni CA, Bonora K, Ciatto S, Giovannetti L, Patriarca S, Sapino A, et al. Misclassification of breast cancer as cause of death in a service screening area. Cancer Causes Control 2009;20(5):533e8. [20] Nyström L, Larsson LG, Rutqvist LE, Lindgren A, Lindqvist M, Ryden S. Determination of cause of death among breast cancer cases in the Swedish randomized mammography screening trials; a comparison between official statistics and validation by an endpoint committee. Acta Oncol 1995;34:145e52. [21] Health Council of the Netherlands. Usefulness of population breast cancer screening (in Dutch). The Hague: Health Council of the Netherlands; 2002. publication no. 2002/03. [22] Greenland S, Thomas DC. On the need for the rare disease assumption in caseecontrol studies. Am J Epidemiol 1982 Sep;116(3):547e53. [23] Vandenbroucke JP, Pearce N. Caseecontrol studies: basic concepts. Int J Epidemiol 2012 Oct;41(5):1480e9. [24] Vandenbroucke JP, Pearce N. Incidence rates in dynamic populations. Int J Epidemiol 2012 Oct;41(5):1472e9. [25] Weiss NS, McKnight B, Stevens NG. Approaches to the analysis of case-control studies of the efficacy of screening for cancer. Am J Epidemiol 1992;135(7): 817e23. [26] Roder D, Houssami N, Farshid G, Gill G, Luke C, Downey P, et al. Population screening and intensity of screening are associated with reduced breast cancer mortality: evidence of efficacy of mammography screening in Australia. Breast Cancer Res Treat 2008;108(3):409e16. [27] Zackrisson S, Andersson I, Manjer J, Janzon L. Non-attendance in breast cancer screening is associated with unfavourable socio-economic circumstances and advanced carcinoma. Int J Cancer 2004;108(5):754e60. [28] Duffy SW, Cuzick J, Tabar L, Vitak B, Chen THH, Yen MF, et al. Correcting for non-compliance bias in case-control studies to evaluate cancer screening programmes. Appl Stat 2002;51(Part 2):235e43. [29] Paci E, Duffy SW, Giorgi D, Zappa M, Crocetti E, Vezzosi V, et al. Quantification of the effect of mammographic screening on fatal breast cancers: the Florence Programme 1990e96. Br J Cancer 2002;87(1):65e9.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
6
E. Paap et al. / The Breast xxx (2014) 1e6
[30] Rothman K, Greenland S, Lash T. Modern epidemiology. Philadelphia: Wolters Kluwer Health/Lippincott Williams &Wilkins; 2008. [31] Weiss NS. Application of the case-control method in the evaluation of screening. Epidemiol Rev 1994;16(1):102e8. [32] Egger M, Davey Smith G, Altman DG. Systematic reviews in health care: metaanalysis in context. 2nd ed. London: BMJ Books; 2001. [33] Nickson C, Mason KE, English DR, Kavanagh AM. Mammographic screening and breast cancer mortality: a case-control study and meta-analysis. Cancer Epidemiol Biomarkers Prev 2012 Sep;21(9):1479e88. [34] Fielder HM, Warwick J, Brook D, Gower-Thomas K, Cuzick J, Monypenny I, et al. A case-control study to estimate the impact on breast cancer death of the breast screening programme in Wales. J Med Screen 2004;11(4):194e8. [35] Otto SJ, Fracheboud J, Verbeek AL, Boer R, Reijerink-Verheij JC, Otten JD, et al. Mammography screening and risk of breast cancer death: a population-based case-control study. Cancer Epidemiol Biomarkers Prev 2012 Jan;21(1):66e73. [36] Paap E, Holland R, den Heeten GJ, van Schoor G, Botterweck AAM, Verbeek ALM, et al. A remarkable reduction of breast cancer deaths in screened versus unscreened women: a case-referent study. Cancer Causes Control 2010;21(10):1569e73. [37] van Schoor G, Moss SM, Otten JD, Donders R, Paap E, den Heeten GJ, et al. Increasingly strong reduction in breast cancer mortality due to screening. Br J Cancer 2011;104(6):910e4. [38] Holland R, Rijken H, Hendriks JHCL. The Dutch population-based mammography screening: 30-year experience. Breast Care 2007;2:12e8. [39] Gabe R, Tryggvadottir L, Sigfusson BF, Olafsdottir GH, Sigurdsson K, Duffy SW. A case-control study to estimate the impact of the Icelandic population-based mammography screening program on breast cancer death. Acta Radiol 2007;48(9):948e55. [40] Olsen AH, Njor SH, Vejborg I, Schwartz W, Dalgaard P, Jensen MB, et al. Breast cancer mortality in Copenhagen after introduction of mammography screening: cohort study. BMJ 2005;330(7485):220.
[41] Puliti D, Miccinesi G, Collina N, De Lisi V, Federico M, Ferretti S, et al. Effectiveness of service screening: a case-control study to assess breast cancer mortality reduction. Br J Cancer 2008;99(3):423e7. [42] Swedish Organised Service Screening Evaluation Group. Reduction in breast cancer mortality from the organised service screening with mammography: 2. Validation with alternative analytic methods. Cancer Epidemiol Biomarkers Prev 2006;15(1):52e6. [43] Otto SJ, Fracheboud J, Verbeek AL, Boer R, Reijerink-Verheij J, Otten JD, et al. Mammography screening and breast cancer mortality e response. Cancer Epidemiol Biomarkers Prev 2012;21:870e1. [44] Elmore JG, Reisch LM, Barton MB, Barlow WE, Rolnick S, Harris EL, et al. Efficacy of breast cancer screening in the community according to risk level. J Natl Cancer Inst 2005;97(14):1035e43. [45] van Schoor G, Paap E, Broeders MJM, Verbeek ALM. Residual confounding after adjustment for age: a minor issue in breast cancer screening effectiveness. Eur J Epidemiol 2011;26:585e8. [46] Nickson C, Mason KE, English DR, Kavanagh AM. Screening and breast cancer mortality e response. Cancer Epidemiol Biomarkers Prev 2012;21(12):2276e7. [47] Health Council of the Netherlands. Population screening of breast cancer: expectations and developments (in Dutch). The Hague: Health Council of the Netherlands; 2014. Publication no. 2014/01. [48] Hoff SR, Klepp O, Hofvind S. Asymptomatic breast cancer in non-participants of the national screening programme in Norway: a confounding factor in evaluation? J Med Screen 2012;19(4):177e83. [49] Vanier A, Leux C, Allioux C, Billon-Delacour S, Lombrail P, Molinie F. Are prognostic factors more favorable for breast cancer detected by organized screening than by opportunistic screening or clinical diagnosis? A study in Loire-Atlantique (France). Cancer Epidemiol 2013;37(5):683e7. [50] Sasco AJ, Day NE, Walter SD. Case-control studies for the evaluation of screening. J Chronic Dis 1986;39(5):399e405.
Please cite this article in press as: Paap E, et al., Breast cancer screening halves the risk of breast cancer death: A case-referent study, The Breast (2014), http://dx.doi.org/10.1016/j.breast.2014.03.002
