Abstract Background Mammography screening increases incidence because cancers are detected earlier in time and because of overdiagnosis. We developed an Excel-based model to visualize the expected increase from lead-time amplified by increasing background incidence. Subsequently, we added overdiagnosis to the model. Methods We constructed two hypothetical populations of women aged 50–79 in 5-year age and calendar groups: one with screening for women aged 50–69 and one without. The user enters information on population at risk, number of breast cancers, trends in background incidence, average length of lead-time and, optionally, overdiagnosis. The model computes incidence rate ratios (IRRs) comparing incidence changes with screening to changes without in open and closed cohorts. Results We entered information from Norway from 1990 to 1994, the period preceding the gradual introduction of a national mammography screening programme. As expected, the Screening Illustrator showed prevalence peaks and compensatory drops. Only the closed cohort approach remained unaffected by increasing background incidence. The model showed a 20% sustained increase in incidence (IRR: 1.20) from lead-time and increasing background incidence in the open cohort approach for women aged 50–69. However, real life Norwegian data show a corresponding 38% increase. For the model to achieve the observed incidence, 10–14% overdiagnosis had to be added. Conclusion The observed breast cancer incidence increase in Norway after screening implementation could not be obtained from an average lead-time of 2.5 years and empirical background incidence trends, but had to incorporate overdiagnosis. Introduction The aim of mammography screening is to detect breast cancer earlier in time. Lead-time is the time interval between screening diagnosis and clinical diagnosis, had the woman not been screened.1 Introduction of mammography screening leads to a sustained increase in breast cancer incidence, because cancers are detected earlier.2 Overdiagnosis occurs when mammography screening detects breast cancers that would not have presented with symptoms during a woman’s remaining lifetime.3 Overdiagnosis adds to the sustained increase in breast cancer incidence after screening introduction. Separating the intended effect of lead-time on incidence from the excess increase due to overdiagnosis is complex. One approach to assess overdiagnosis is to follow a ‘closed cohort’ that is a group of women followed from the time they enter the screening programme and until screening ends plus the maximal length of lead-time, where screening no longer affects their incidence (figure 1). Since cancers are diagnosed earlier within a cohort, the cumulative incidence with screening will equal the cumulative incidence without screening.3 Hence, excess breast cancers at end of follow-up with screening by definition represent overdiagnosis. Another approach is to follow an ‘open cohort’ that is a group of screening-age women followed over periods of calendar time, but individual women are not necessarily followed during their entire screening history. Since some cancers that would have been diagnosed above the upper-age limit without screening are diagnosed below with screening, the average incidence in screening-age women will be higher than without screening. If the open cohort is expanded to include women above the upper-age limit corresponding to the maximal length of lead-time, the average incidence with screening may approach the incidence without, depending on trends in background incidence. Hence, overdiagnosis is more difficult to distinguish from the expected incidence increase in open cohorts. Figure 1 View largeDownload slide Different analytic approaches for estimating incidence increase with screening. The symbols indicate time of clinical diagnosis (o) and screening diagnosis (x) Figure 1 View largeDownload slide Different analytic approaches for estimating incidence increase with screening. The symbols indicate time of clinical diagnosis (o) and screening diagnosis (x) We define trends in background breast cancer incidence as the naturally occurring change in incidence over time without screening caused by environmental or biologic factors, termed period or cohort effects.4 These are usually estimated jointly as a total change over calendar time, which is known as the drift parameter in age-period-cohort models.4 Increasing background breast cancer incidence may amplify the expected incidence increase due to lead-time, because cancers from older age groups and future periods with a higher incidence are detected earlier.2 For example, assuming an average lead-time of 5 years, the observed incidence in women aged 60 years in year 2000 with screening reflects the incidence in women aged 65 in 2005 without screening. Increases in breast cancer incidence have been observed with the introduction of national mammography screening programmes.5 Some have argued that this results from lead-time effects inflated by increasing background incidence2 and the use of analytic approaches based on open instead of closed cohorts.5,6 Others have argued that the increased incidence is due to overdiagnosis7 and that open and closed cohort approaches are equally valid.8 We developed an Excel-based model, the Screening Illustrator, to visualize the increase in incidence expected from lead-time and from background incidence trends in open and closed cohort approaches. We also allowed for overdiagnosis in the model. Finally, we compare our model to the observed breast cancer increase associated with introduction of mammography screening in Norway from late 1995–2004. Methods The Screening Illustrator compares two scenarios. In the first, there is no screening and in the second, screening is introduced in the second period for women aged 50–69 years. In both scenarios, we follow a hypothetical population of women aged 50–79 in 5-year age groups and 5-year calendar periods. In this idealized model, we assume full participation that is all women within the screening age range are screened. On the first spreadsheet, the user customizes the model by entering information on age-stratified population at risk, age-stratified number of breast cancers, year of screening introduction, and percentage change in background incidence (i.e. the naturally occurring change in incidence over time without screening4) over a 5-year period without screening. The incidence increase due to increasing age is implicitly implemented, when the user enters observed age-stratified number of cancers. The increase is kept constant over subsequent calendar periods. The user enters an average length of lead-time ranging from 0 to 5 years, which is translated to a percentage of cancers moved from a later calendar period and older age group to an earlier period and younger age group, i.e. from one cell to a preceding cell. The percentage of moved cancers is calculated by dividing the average length of lead-time with the total length of 5 years. Thereby, we assume that all women move along the diagonal within each cell. The Screening Illustrator computes the absolute number of breast cancers in 5-year age groups and calendar periods in both scenarios and converts the absolute numbers to incidence rates per 100 000 person years. Incidence rates are also reported for aggregated groups according to the open or closed cohort analytic approaches and displayed in figures. The Screening Illustrator provides incidence rate ratios (IRRs) comparing changes in the screening scenario to the scenario without screening. An IRR of 1 means that there is no difference between the two scenarios, an IRR higher than 1 means that the incidence rate in the screening scenario is higher, and an IRR below 1 means that the incidence in the screening scenario is lower than in the scenario without screening. On the second spreadsheet, the user has the option to add overdiagnosis by entering a percentage that will generate a number of excess cancers in the screening scenario. The percentage is multiplied with the corresponding number of breast cancers in the scenario without screening corresponding to a period 10 years later and 10 years older than with screening. This is a pragmatic way to cause the amount of overdiagnosis to follow trends in background incidence. The Screening Illustrator then converts the percentage of excess cancers to three standardized ways of estimating overdiagnosis.9 Overdiagnosis is expressed as the number of excess cancers divided by three different denominators: number of cancers among unscreened women aged 50–79, among screened women aged 50–79 or among screened women aged 50–69. The user is able to vary the change in background incidence and the average length of lead-time to assess how various scenarios affect the overdiagnosis measures. Results To demonstrate the use of the Screening Illustrator, we entered population at risk and number of breast cancers in Norway for the period 1990–94.10,11 This period preceded the gradual introduction of a national mammography screening programme from late 1995–2004.12 We entered a background incidence increase of 3.7% (95% CI: 0.8–6.6%) per 5-year period, which has been estimated for Norway during 1978–97,13 and an average lead-time of 2.5 years, which translates to 50% of cancers being moved from one cell to the preceding cell. Expected incidence increases The Screening Illustrator outputs the incidence changes in figure 2, when following a ‘closed cohort’ of women from the time they enter the screening programme and 10 years after screening ends. Figure 2A shows how incidence is higher at screening introduction at age 50–54, because cancers from the prevalent pool of pre-clinical cancers are detected earlier, therefore termed the ‘prevalence peak’.3 Incidence stabilizes at a higher level at age 55–69, because cancers from older age groups and future calendar periods with a higher incidence are detected earlier. Incidence drops at screening termination at age 70–74, because cancers that would have been diagnosed at this age have already been detected earlier with screening, therefore termed the ‘compensatory drop’. Incidence reaches the same level as without screening at age 75–79 years, when the effect of the screening programme ceases. Figure 2B shows the cumulative incidence that during screening-age remains at a higher level than without screening, because cancers are detected at an earlier age. At screening termination, the cumulative incidence with screening drops to the same level as without screening, because the compensatory drop counterbalances the prevalence peak. Figure 2 View largeDownload slide Incidence rates (A) and cumulative incidence (B) of breast cancer per 100 000 person-years in a closed cohort followed from age 50–79 in Norway, assuming 3.7% increase in background incidence per 5 years and an average lead-time of 2.5 years. The area between the solid and dashed line (A) sums to 0, leading to IRRs of 1.00 for the closed cohort in table 1 Figure 2 View largeDownload slide Incidence rates (A) and cumulative incidence (B) of breast cancer per 100 000 person-years in a closed cohort followed from age 50–79 in Norway, assuming 3.7% increase in background incidence per 5 years and an average lead-time of 2.5 years. The area between the solid and dashed line (A) sums to 0, leading to IRRs of 1.00 for the closed cohort in table 1 The Screening Illustrator outputs the incidence changes in figure 3, when following ‘open cohorts’ over periods of calendar time. In the open cohort approach limited to screening-age women, the incidence peaks at screening introduction in 1995–99 and thereafter stabilizes at a higher level than without screening. This results from the prevalence peak stemming from all participants at screening introduction and from first-time participants at subsequent rounds. Moreover, incidence remains at a higher level because cancers from older age groups and future calendar periods with a higher incidence are detected earlier within the screening programme. In the open cohort approach including above-screening-age women, incidence also peaks in 1995–99, but thereafter declines to almost the same level as without screening because the compensatory drop in incidence among women aged 70–74 is included. However, in the open cohort approach, the compensatory drop from an older cohort substitutes for the drop the specific cohort would have experienced later in life in a closed cohort approach. In consequence, the open cohort approach is affected by trends in background incidence. If background incidence is increasing, the drop the specific cohort would have experienced later in life exceeds the current drop of an older cohort. Therefore, the contemporary drop cannot fully substitute for the future drop, and incidence remains marginally increased compared to the scenario without screening. Had the background incidence remained unchanged over calendar time, the incidence would have declined to the exact same level as without screening. Figure 3 View largeDownload slide Incidence rates of breast cancer per 100 000 person-years in open cohorts followed over calendar periods in Norway with screening introduction in 1995, 3.7% increase in background incidence per 5 years, and an average lead-time of 2.5 years. The areas between the solid and dashed lines are positive, leading to IRRs above 1 for open cohorts in table 1 Figure 3 View largeDownload slide Incidence rates of breast cancer per 100 000 person-years in open cohorts followed over calendar periods in Norway with screening introduction in 1995, 3.7% increase in background incidence per 5 years, and an average lead-time of 2.5 years. The areas between the solid and dashed lines are positive, leading to IRRs above 1 for open cohorts in table 1 The Screening Illustrator outputs IRRs in table 1 comparing changes in the screening scenario to concurrent changes in the scenario without screening. At screening introduction, the prevalence peak is apparent from the 51–64% higher incidence (IRR: 1.51–1.64) among screening-age women compared to what would have been observed without screening. In the subsequent periods, the peak continues among first-time participants aged 50–54 (IRR: 1.51). Among women aged 55–69 incidence remains at a higher level (IRR: 1.09–1.14) than without screening, because cancers from older age groups in future periods with higher incidence are diagnosed earlier in time. Among women aged 70–74, a compensatory drop of 50% (IRR: 0.50) occurs 5 years after screening introduction. In the open cohort approach limited to women aged 50–69, screening introduction leads to an initial increase of 59% in 1995–99 that stabilizes at 20% in subsequent periods. In the open cohort including women aged 50–79, no compensatory drop occurs in 1995–99 among women aged 70–74 resulting in an initial increase of 33% in 1995–99. In subsequent periods, compensatory drops from older cohorts do not fully substitute for the drop that screening-age cohorts would have experienced later in life due to the increasing background incidence, which results in a 1% sustained increase. In the closed cohort approach, the prevalence peak within each cohort is fully compensated by the later drop resulting in equal cumulative incidences with and without screening, evident in the IRRs of 1.00. In the closed cohort approach, cumulative IRRs are therefore unaffected by changes in background incidence. Table 1 Incidence rate ratios comparing screening to no screening Year of screening introduction is set to 1995, change in background incidence to 3.7% per 5-year period, and average lead-time to 2.5 years. The black frame indicates screening introduction and hatched cells prevalence peaks (age 50–69) and compensatory drops (age 70–74) Table 1 Incidence rate ratios comparing screening to no screening Year of screening introduction is set to 1995, change in background incidence to 3.7% per 5-year period, and average lead-time to 2.5 years. The black frame indicates screening introduction and hatched cells prevalence peaks (age 50–69) and compensatory drops (age 70–74) Observed incidence increases The annual breast cancer incidence in Norway has been steadily increasing over the past 50 years. During the implementation of screening from late 1995–2004 the increase was enhanced (See Supplementary figure S1).11 Because screening was introduced gradually in Norway, the prevalence peak was stretched over several years, which complicates direct comparison with the output from the Screening Illustrator. In a previous publication, we took account of the gradual introduction by using time from introduction instead of calendar year as the underlying time axis. We estimated an average incidence increase of 38% (95% CI: 29–47%) across all regions from year 3 after introduction and onwards among women aged 50–69 compared with women aged 30–49.14 This estimate is comparable to the sustained increase in subsequent rounds in the open cohort approach limited to women aged 50–69 in the Screening Illustrator. However, the Screening Illustrator only showed a 20% sustained increase in incidence among women aged 50–69 in the open cohort approach (table 1). The observed increase cannot be achieved from an average lead-time of 2.5 years and empirical increases in background incidence. In order to obtain a sustained increase of 38%, we need to turn to the second spreadsheet and generate overdiagnosis. By entering 14% excess cancers, the open cohort approach for women aged 50–69 shows a sustained increase of 39%. This corresponds to 10–14% overdiagnosis depending on the way overdiagnosis is expressed. The presence of overdiagnosis leads to increased prevalence peaks at introduction (IRR: 1.68–1.81) and among first time participants aged 50–54 (IRR: 1.68). Among women aged 55–69 incidence remains at an even higher level (IRR: 1.29–1.31) in subsequent periods. However, the compensatory drop of 50% (IRR: 0.50) is unaffected, since overdiagnosis stems from excess cancers that never would have been diagnosed without screening. The open cohort approach including women aged 50–79 shows a sustained increase of 12%, which is a combined effect of lead-time and overdiagnosis. The closed cohort approach shows an increase of 11%, exclusively due to overdiagnosis. To assess how the input parameters affect the measures of overdiagnosis, we varied the change in background incidence according to the reported 95% CI, i.e. from 0.8 to 6.6%, and the average length of lead-time from 1 to 4 years. For each scenario, we changed the percentage generating excess cancers until a sustained increase of 38–39% was obtained in the open cohort approach for women aged 50–69. Varying the change in background incidence only led to minor changes, while changing the length of lead-time led to overdiagnosis measures varying from 4–5 to 15–22% (Supplementary table S1). Discussion We have demonstrated how the Screening Illustrator can be used to separate the effects of lead-time and overdiagnosis in different analytic approaches, which makes these complex effects accessible to clinicians, educators and researchers. In the development of the Screening Illustrator, we have prioritized transparency over complexity. A strength is the ideal comparison group consisting of identical women in a counterfactual scenario without screening. Estimates from observational studies and the Screening Illustrator will differ based on the plausibility of observed comparison groups. In an earlier study of the observed incidence increase in Norway,14 we relied on uninvited women aged 30–49 to substitute for screening-age women in a scenario without screening. In the Screening Illustrator, the underlying key assumption is the average length of lead-time, which may be varied, but cannot exceed 5 years2,15 and limits the compensatory drop to women aged 70–74. Since we assume full participation, we overestimate the increase that would be observed in a real setting with a given lead-time and thereby, underestimate the level of overdiagnosis. In Norway, 76% of invited women chose to participate.12 We assume that the change in background incidence is the same for all age groups. Another simplification is the 5-year age and period grid. Altogether, this results in the incidence peak limited to the period with screening introduction followed by a decrease that stabilizes in the subsequent period. Expected incidence increases are bound by the lead-time assumption and the 5-year grid. If an average lead-time of 5–10 years was imposed by moving cancers two cells diagonally, the peak would last for two periods and thereafter stabilize. Duffy and colleagues16 chose a 1-year grid that allowed a more elaborate lead-time assumption with a certain percentage moved 1–10 years, which corresponded to an average lead-time of about 3.3 years. This resulted in a higher peak in the first period followed by a gradual decline afterwards. Although a more elaborate lead-time assumption would result in more accurate changes in incidence, the required complexity would reduce the immediately apparent lead-time effect represented by cancers being moved from one cell to a preceding cell. The Screening Illustrator suggests that in settings without age-period interactions the incidence inflation from increasing background incidence is small in open cohort approaches. Moreover, the Screening Illustrator shows how an open cohort approach limited to screening-age women results in sustained incidence increases even without overdiagnosis in the introduction and subsequent periods. Similarly, the open cohort approach including above-screening-age women results in an incidence increase in the introduction period, but when compensatory drops are included in subsequent periods, the open cohort approach performs nearly identical to the closed cohort approach. The closed cohort perfectly absorbs changes in background incidence as movement of diagnoses occurs within the cohort, and consequently no incidence increase is observed due to screening (the IRR equals one). Full follow-up of several birth cohorts does, however, require study periods of >30 years, which is rarely feasible. Thus, we find it interesting that the more feasible open cohort approach including above-screening-age women performs nearly as well, even with substantial changes in background incidence. In our application, observed increases in breast cancer incidence in Norway could not be achieved from lead-time effects and increasing background incidence alone, but had to incorporate overdiagnosis. Changing the input parameters showed how the overdiagnosis measures rely on assumptions about the average length of lead-time and change in background incidence. Although the Screening Illustrator cannot provide an exact estimate of the amount of overdiagnosis, it helps understand the interplay between lead-time and overdiagnosis. We encourage readers to employ the tool and compare their customized model to observed data available to them. Please send suggestions for modifications to the corresponding author. Updates of the Screening Illustrator will be published on http://ph.au.dk/uddannelse/software/. Funding Aarhus University funded this study and was not involved in any part of the study design, data collection, analysis or interpretation. Conflicts of interest: None declared. Key points The Screening Illustrator, an Excel-based model, illustrates how incidence increases from lead-time can be distinguished from trends in background incidence and overdiagnosis. Realistic increases in background incidence of breast cancer lead only to small increases in observed incidence in open cohort analyses. Observed incidence increases in Norway cannot be explained without incorporating overdiagnosis. The Screening Illustrator may be a useful tool for clinicians, researchers and educators to teach, understand and evaluate ongoing screening programmes. References 1 Vainio H , Bianchini F , editors. IARC Handbook on Cancer Prevention. Vol 7. Breast Cancer Screening . Lyon, France : IARC Press , 2002 . 2 Duffy SW , Lynge E , Jonsson H , et al. Complexities in the estimation of overdiagnosis in breast cancer screening . Br J Cancer 2008 ; 99 : 1176 – 8 . Google Scholar Crossref Search ADS PubMed 3 Biesheuvel C , Barratt A , Howard K , et al. Effects of study methods and biases on estimates of invasive breast cancer overdetection with mammography screening: a systematic review . Lancet Oncol 2007 ; 8 : 1129 – 38 . Google Scholar Crossref Search ADS PubMed 4 Clayton D , Schifflers E . Models for temporal variation in cancer rates. I: age-period and age-cohort models . Stat Med 1987 ; 6 : 449 – 67 . Google Scholar Crossref Search ADS PubMed 5 Puliti D , Duffy SW , Miccinesi G , et al. Overdiagnosis in mammographic screening for breast cancer in Europe: a literature review . J Med Screen 2012 ; 19 : 42 – 56 . Google Scholar Crossref Search ADS PubMed 6 Falk RS . Why are results of organised mammography screening so difficult to interpret? Tidsskr Nor Legeforen 2014 ; 134 : 1124 – 6 . Google Scholar Crossref Search ADS 7 Jørgensen KJ , Gøtzsche PC . Overdiagnosis in publicly organised mammography screening programmes: systematic review of incidence trends . BMJ 2009 ; 339 : b2587 . Google Scholar Crossref Search ADS PubMed 8 Jørgensen KJ , Kalager M , Barratt A , et al. Overview of guidelines on breast screening: why recommendations differ and what to do about it . Breast 2017 ; 31 : 261 – 9 . Google Scholar Crossref Search ADS PubMed 9 Marmot MG , Altman DG , Cameron DA , et al. The benefits and harms of breast cancer screening: an independent review . Lancet 2012 ; 380 : 1778 – 86 . Google Scholar Crossref Search ADS PubMed 10 Statistics Norway . Available at: https://www.ssb.no/en/ (13 February 2017, date last accessed). 11 Engholm G , Ferlay J , Christensen N , et al. NORDCAN: Cancer Incidence, Mortality, Prevalence and Survival in the Nordic Countries, Version 7.3 (08.07.2016). Association of the Nordic Cancer Registries. Danish Cancer Society. Available at: http://www.ancr.nu (13 February 2017, date last accessed). 12 Hofvind S , Geller B , Vacek PM , et al. Using the European guidelines to evaluate the Norwegian Breast Cancer Screening Program . Eur J Epidemiol 2007 ; 22 : 447 – 55 . Google Scholar Crossref Search ADS PubMed 13 Møller B , Weedon-Fekjaer H , Hakulinen T , et al. The influence of mammographic screening on national trends in breast cancer incidence . Eur J Cancer Prev 2005 ; 14 : 117 – 28 . Google Scholar Crossref Search ADS PubMed 14 Lousdal ML , Kristiansen IS , Møller B , Støvring H . Effect of organised mammography screening on stage-specific incidence in Norway: population study . Br J Cancer 2016 ; 114 : 590 – 6 . Google Scholar Crossref Search ADS PubMed 15 Zahl P-H , Jørgensen KJ , Gøtzsche PC . Overestimated lead times in cancer screening has led to substantial underestimation of overdiagnosis . Br J Cancer 2013 ; 109 : 2014 – 9 . Google Scholar Crossref Search ADS PubMed 16 Duffy SW , Parmar D . Overdiagnosis in breast cancer screening: the importance of length of observation period and lead time . Breast Cancer Res 2013 ; 15 : R41 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
The European Journal of Public Health – Oxford University Press
Published: Dec 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera