Background: Different definitions and estimates of overdiagnosis in mammographic screening reflect a substantial need to investigate and understand the complexity of the issue. This modeling study aims to estimate the number of overdiagnosed women, defined as those diagnosed with breast cancer who die from any cause within the lead-time period.

Methods: We used numbers from incidence and death statistics available online and published estimates of lead-time. Postulated cohorts of screened and not screened women ages 50 to 51 were followed for a period corresponding to 10 biennial screening exams during 20 years, and a further 10 years, to ages 78 to 79. The increase in breast cancer incidence because of screening was estimated based on lead-time. The proportion of women diagnosed with breast cancer who died within the lead-time period was assessed based on the differences in the cumulative number of breast cancer diagnosed in a nonscreened and screened cohort.

Results: The proportion of inevitable overdiagnosed women in a screened versus nonscreened cohort was 1.9% for England and Wales and 1.8% for Norway. Sensitivity analyses using various assumptions increased the estimates up to a maximum of 4%.

Conclusion: The proportion of women with breast cancer diagnosed after participation in a screening program who died within the estimated lead-time period was less than 4%. This inevitable proportion of overdiagnosis should be emphasized in the definition and communication of the issue.

Impact: The issue of overdiagnosis is complex and estimates should be interpreted with substantial care. Cancer Epidemiol Biomarkers Prev; 25(5); 759–65. ©2016 AACR.

The reported estimates of overdiagnosis in mammographic screening vary from 0% to 75% (1, 2). Different study designs, age groups included, follow-up time, comparison groups, inclusion of ductal carcinoma in situ (DCIS), time span, choice of estimator, and adjustments for lead-time are examples of reasons for the varying estimates (1, 3–10).

As researchers in mammographic screening we have struggled with the issue of overdiagnosis for many years. However, we have realized that the vast majority of the definitions of this dispute include a time aspect related to death. Overdiagnosis can be divided into two components based on the time at which death occurs among women diagnosed with breast cancer; women dying from all causes within the period corresponding to the lead-time, and women dying after the estimated lead-time. Lead-time is defined as the time from detection of preclinical breast cancer by screening to detection of clinical cancer in the absence of screening. The first component is related to potentially progressive preclinical breast cancer. Such cancers are destined to cause symptoms and thus defined to have finite lead-time. Woman diagnosed with first scenario breast cancer is expected to die of other causes before the cancer would have given rise to clinical symptoms, within the lead-time period. The second component assumingly relates to slow- or nonprogressive preclinical cancer that never would have resulted in symptoms or death. The latter is an extreme form of length bias where the lead-time, in theory, is infinitive.

One of the main challenges when estimating overdiagnosis is how to take the lead-time into account. Most studies have used estimates of mean lead-time based on models that assume all preclinical cancers to be progressive. This assumption could potentially yield a large bias (11). Separating the two components is thus essential in the estimation of overdiagnosis. As far as we know, only one study including such a separation has been published (12). That study used the sojourn time to separate the estimated incidence of overdiagnosis into potentially progressive and less or no progressive tumors by applying Markov Chain Monte Carlo estimation on data from two randomized controlled trials. Sojourn time is defined as the time between the point at which a lesion can be found by screening until it is clinically detectable (13). The study concluded that progressive tumors dominated the overdiagnosed cases and only a small fraction consisted of tumors with infinite length bias and thus less or no potential to progress (12).

In principle, all screening programs will detect breast cancer among women who die of other causes in the near future since there exist competing risk of death among women targeted by screening. Although the all-cause mortality rates are low, it is inevitable. In this study, we focus on the first component of overdiagnosis in mammographic screening by using a modeling approach. We aimed to estimate the number of women diagnosed with breast cancer who died within the estimated lead-time period caused by screening.

We postulated a cohort of 100,000 women ages 50 until age 79 according to death statistics. We used numbers from national statistics from England and Wales (14) and Norway (15). From the interim life tables, we found that the yearly decrement of death in the period 2008 to 2010 increased from 2.3 per 1,000 at age 50 to 38.2 per 1,000 at age 79 in England and Wales (14) and from 1.9 per 1,000 to 34.0 per 1,000 in 2010 to 2012 among Norwegian women (15). Yearly decrement was defined as the all-cause mortality rate between age x and (x + 1), which is the probability that a woman at age x will die before reaching age (x + 1). The expected remaining lifetime, the number of years the woman will live thereafter, decreased from 34 years at age 50 to 17 years at age 69 in both England and Wales (14) and in Norway (15). Based on these numbers, women 50 to 69 years are expected to live until age 85. The majority would thus live beyond any realistic lead-time period (9, 16).

Nonscreened cohort

The incidence rates of invasive breast cancer in England and Wales (late 1980s) were adapted from Duffy and Parmar (3). The rate increased from 158.4 per 100,000 women years at age 50 to 272.7 per 100,000 at age 79. Data from Norway was adapted from the Cancer Registry of Norway. The smoothed incidence rate (three-year moving average) of invasive breast cancer increased from 114.1 per 100,000 at age 50 to 234.5 per 100,000 women years at age 79 in 1980 to 1984. The reference incidence was consciously chosen from periods not influenced by opportunistic or organized screening.

Screened cohort

We assumed biennial mammographic screening for women ages 50 to 69 years, following the recommendation of 10 screening rounds. Furthermore, we calculated the proportion of breast cancer diagnoses advanced by screening according to the formulae given by Duffy and Parmar (3), which assume an exponentially distributed lead-time. We applied the rate of progressive cancers as exponential distributed with λ equal to 0.34 as given by Duffy and colleagues (12). These figures correspond to an average lead-time of 3.2 years, ranging from 1 to 10 years (Table 1).

Analysis

We followed two cohorts of women, 50 and 51 years old, for a period corresponding to 10 biennial screening exams during 20 years, and a further 10 years after screening, to 78 and 79 years old. We compared the cumulative number of women diagnosed with breast cancer in the nonscreened and screened cohort and applied the decrement mortality to estimate the proportion of women diagnosed with breast cancer who died within the estimated lead-time period (first component). We called this measure the inevitable proportion of overdiagnosed women. Calculations were made for England and Wales and for Norway, separately. Sensitivity analyses were performed to study the influence of the yearly decrement in mortality, the lead-time distribution, and the incidence of DCIS and/or breast cancer in the prescreening period.

This is a modeling study and thus no ethical approval or consent was required. Microsoft Excel was used for calculations.

The estimated number of breast cancers diagnosed in a postulated cohort of 100,000 women 50 years old, screened prevalently at age 50 to 51 and screened subsequently at age 52 to 68 and 53 to 69, and followed for a further 10 years in England and Wales is given in Table 2 and Supplementary Table S1. In the absence of screening, the number of women diagnosed with breast cancer was calculated as the breast cancer incidence multiplied by the decremented population. In the screened 50-year old cohort (upper diagonal in Supplementary Table S1), the number of women with breast cancer was affected by lead-time, that is, the number of diseased women aged 50 was calculated as 158 + 0.85 × 162 + 0.60 × 165 + … + 0.05 × 196 =650; and for women aged 51 the number was calculated as (1 − 0.85) × 162 = 25. For the subsequent screening exams, we multiplied the lead-time coefficient with the remaining number of breast cancer cases. For women aged 52, the number was 448 [(1 − 0.60) × 165 + (1 − 0.43) × 169 × 0.85 + … + 0.05 × 206]. For those aged 53, the number was 15 [(1 − 0.85) × (1 − 0.43) × 169]. Similar calculation was performed for the screened 51-year-old cohort (lower diagonal in Supplementary Table S1). The effect of the biennial screening interval is clearly visible in the fluctuating number of breast cancers diagnosed during the age range 50 to 69 in the screened cohort. The cumulative number of women with breast cancer observed during the 30 years period was equal in the nonscreened and screened cohort, but the internal distribution differed. Because of lead-time, there are more women at risk of dying after a diagnosis of breast cancer in the screened compared with the nonscreened cohort. Applying the mortality rates yields an excess of 223 deaths among women with breast cancer in the screened cohort (Table 2). The inevitable proportion of overdiagnosed women in the screened versus nonscreened cohort was thus 1.9% (223/11,474) for England and Wales.

We used the same approach for Norway (Table 3 and Supplementary Table S2). The prescreening incidence of breast cancer was somewhat lower for Norway compared to England and Wales. The age slope was about the same for both populations. The mortality decrement had the same curve in Norway as in England and Wales, but on a slightly lower level in Norway. The inevitable proportion of over-diagnosed women because of implementation of organized mammographic screening in a Norwegian setting was estimated to be 1.8% (160/8,993).

Sensitivity analyses

We applied mortality statistics from the prescreening period to study the influence of the yearly decrement in mortality. For England and Wales, the mortality rates increased from 3.3 to 58.5 per 1,000 for age 50 to 79 in the prescreening period 1985 to 87. Using these numbers resulted in a 3.2% inevitable overdiagnosis (Supplementary Table S3). For Norway, the mortality rates increased from 3.0 to 55.6 per 1,000 for women 50 to 79 years in the prescreening period 1980 to 1984, which resulted in 2.6% inevitable overdiagnosis (Supplementary Table S4).

To study the influence of the lead-time distribution we applied an additional distribution with more years affected (range 1–15; Table 1). This has an average lead-time of 4.7 years (λ = 0.22), instead of an average of 3.2 years (range 1—10). We were thus able to follow the women until age 84 (within the expected life-time for women aged 50 years). Then the proportion of cancers advanced was 90% for year 1, 71% for year 2, …, 6% for year 15 (Table 1). This approach increased the inevitable overdiagnosis estimates to 4.0% for England and Wales (Supplementary Table S5) and to 3.7% for Norway (Supplementary Table S6).

Three different settings were used to study the influence of a rise in the prescreening incidence of DCIS and/or invasive breast cancer. The settings were (1) a fixed increase in incidence of 20% for women aged 50–79 years; (2) an 1% absolute increase for each year of age, starting with a 1% increase in women aged 50, ending up at 30% increase among women aged 79; and (3) 1% absolute decrease for each year of age, starting with a 30% increase in women aged 50 and ending up at 1% increase among women aged 79. The proportion of inevitable overdiagnosis was similar for all of the three settings and the same as the prescreening incidence in England and Wales (1.9%; Supplementary Table S7), and in Norway (data not shown).

In this modeling study, 2% to 4% of women diagnosed with breast cancer in a population offered mammographic screening are dying from any cause within the estimated lead-time.

There are at least three factors to consider when estimating the inevitable proportion of overdiagnosed women in this study: the death rates, the lead-time, and the incidence of breast cancer.

First, we used the observed rates of all-cause mortality in England and Wales and in Norway to estimate the competing risk of death. If the rate increases, the inevitable proportion of overdiagnosed women increases and vice versa. The death rates are relatively stable in developed countries, but small differences, as shown for England and Wales and for Norway are of substantial influence for the estimate. The shape of the death curve by age is also of influence for the estimates. In this study, a steep curve indicates a higher mortality among the elderly and thus a higher inevitable proportion of overdiagnosed women among the elderly. The curve is steeper for England and Wales compared to Norway, which partly explain the different outcome for the two places.

Second, the assumption of the lead-time distribution highly influences the results. We based our estimations on the exponential distribution of lead-time given by Duffy and Parmar (3), which presents mathematical details in the calculation of probabilities of cancers shifted by screening. We assumed that lead-time is only related to progressive cancers, thus the instantaneous rate of transition to symptomatic disease was changed to λ = 0.34 instead of 0.30 (12). However, considerable development of screening equipment, from screen film to full-field digital mammography and further to tomosynthesis has taken place during the last decades, resulting in an increased rate of screen-detected cancer (17–19). The increase is assumingly because of longer lead-time. To quantitatively investigate this bias, we applied an average lead-time of 4.7 years (λ = 0.22), yielding a two-fold increase of first component overdiagnosed women.

Third, a constant increase in the age-specific prescreening incidence did not influence our estimate of inevitable overdiagnosis. Adding a fixed proportion of DCIS to the incidence did not influence the estimate of the first component of overdiagnosis. Similarly, the inevitable overdiagnosis estimate changed marginally for an age dependent increase. The yearly decrement in death and the lead-time distribution are thus the main contributors in this estimate.

Several assumptions were included in our estimations. Biennial mammographic screening in women ages 50 to 69 years was the basis, which is shown to reduce mortality from breast cancer (20). Furthermore, we assumed an attendance rate of 100%, which probably overestimate the outcome of interest. Compliance in the Norwegian screening program is 84%, which gives a 16% reduction in the estimate (1.5% instead of 1.8%). In addition, the estimates are valid for England and Wales and for Norway. No generalization is thus possible. In particular, the breast-screening program in England and Wales invites the women every 3 years and has now expanded the target age range to cover 47 to 73 years (21).

The strength of the study was using a fixed lead-time distribution to study the impact of organized screening exclusively. The impact of opportunistic screening was evaded by using reference incidences from the period before opportunistic screening was available (1980s). Furthermore, the estimates are based on easy available numbers and numbers from previous published papers. The estimations are explained in detail in the text, which makes the results transparent and easily approvable.

Our estimate is in contrast with the result given by Duffy and colleagues (12). They separated the estimate into potentially progressive and less or no progressive tumors using the sojourn time and concluded that progressive tumors dominated. However, both studies state that the absolute risk of death from other causes in the near future after a diagnosis of breast cancer is a minor phenomenon.

Focusing only on the competing risk of death during the estimated lead-time period (first component of overdiagnosis) represents a limitation of this modeling study. The second component is related to women diagnosed with breast cancer and die from any cause after the estimated lead-time period has passed (i.e., maximum 10 years). These women might be diagnosed with slow- or nonprogressive breast cancer and might thus went through treatment as a result of a breast cancer diagnosis that did not prolong their life. However, we are not able to separate women with an incessant slow- or nonprogressive tumor from tumors with potential to change their growing pattern. The screening might thus save lives by detecting and treating these women for breast cancer in an early stage and it is well known that early diagnosis allows for effective and less toxic treatment with reduced mortality compared with symptomatic breast cancer (22). Hypothetically, a woman with a screen-detected early stage slow growing tumor is treated at age 50 and die at age 85 of other causes. She might be overtreated (i.e., second component), but she might also be saved from breast cancer death because of early diagnosis and treatment. Another woman had the same tumor, but do not attend screening. When she turns 62, she is diagnosed with a late-stage symptomatic cancer and die 3 years later. She got her breast cancer diagnosis 12 years later compared with the first women, but the breast cancer killed her at age 65. Continuous diagnostics and research on these early breast cancer cases is thus of utmost importance to get knowledge and making us able to tailor the treatment. Furthermore, follow-up time after the estimated lead-time period has passed is important to get knowledge about the progression rate and to estimate the percentage of breast cancers that never will cause symptoms or death. Optimally, a cohort of screened women along with an accompanied unscreened cohort should be followed until death, that is use of observational data. Such data are not available of today. To estimate a precise extent of the second component is impossible, because of reasons described above. To get a view about the extent of the second component, estimates of the overall rate of overdiagnosis could be considered. The Independent UK Panel on Breast Cancer Screening (23) estimated the proportion of overdiagnosis to be 10.7% in randomized controlled trials. Observational studies using a compensatory drop represent an appropriate method for estimating the proportion of overdiagnosis (1). Two of the studies included in the review by Puliti and colleagues (1) were from England and Wales, showing a proportion of overdiagnosed invasive cancers among invited women of 10% (24) and 2.3% (25). The latter was recalculated to 3.3% (1). A recent study from Norway showed a proportion of 10% to 11% for the same group of women (5). Using these numbers yield an average of 8% [(10 + 3.3 + 10.5)/3] overdiagnosis. This number represents both the first and second component. Our estimate for the first component was approximately 2%. Subtracting our estimate for the first component leave 6% (8–2%) for the second. It means that 75% (6%/8%) of the overdiagnosed women have a lead-time longer than 3 years on average. Applying a lead-time of 4.7 years on average yields 50% [(8–4%)/8%] of the estimate in the second component. These rough estimates indicate that the major contribution of overdiagnosis is from slow- and nonprogressing tumors. However, our estimates are based on lead-time estimated from tumors detected in the 1970s to 1990s (3). More recent studies and newer technologies indicate a longer lead-time (16). The proportion of women overdiagnosis within the first component might be underestimated in our study. However, we need information about prognostic and predictive characteristics and other biological markers of the tumor to understand this complex issue of overdiagnosis. Such knowledge could help to differentiate the non-progressive from the progressive tumors, and thus make us able to identify the most efficient treatment.

In conclusion, a small proportion of women diagnosed with breast cancer died within a time-period corresponding to the lead-time caused by mammographic screening. The inevitable proportion of overdiagnosis ranged from 2% to 4%. We consider the disadvantage of this component of overdiagnosis negligible when weighted against the mortality reduction because of mammographic screening.

No potential conflicts of interest were disclosed.

The funders played no role in the design or conduction of the study, collection, management, analysis or interpretation of the data, preparation, review, or approval of the manuscript, or the decision to submit the manuscript for publication.

Conception and design: R.S. Falk, S. Hofvind

Development of methodology: R.S. Falk, S. Hofvind

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.S. Falk

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.S. Falk, S. Hofvind

Writing, review, and/or revision of the manuscript: R.S. Falk, S. Hofvind

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.S. Falk, S. Hofvind

Study supervision: S. Hofvind

The study was performed as a part of the regular jobs of the two authors. The South-Eastern Norway Regional Health Authority employs R.S. Falk, whereas S. Hofvind is the head of the governmentally funded Norwegian Breast Cancer Screening Program.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Puliti
D
,
Duffy
SW
,
Miccinesi
G
,
de Koning
H
,
Lynge
E
,
Zappa
M
, et al
The EUROSCREEN Working Group. Overdiagnosis in mammographic screening for breast cancer in Europe: a literature review
.
J Med Screen
2012
;
19
:
Suppl 1
:
42
56
.
2.
Zahl
PH
,
Mæhlen
J
. 
Overdiagnosis of breast cancer after 14 years of mammography screening
.
Tidsskr Nor Laegeforen
2012
;
132
:
414
7
.
3.
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
.
4.
de Gelder
R
,
Heijnsdijk
EA
,
van Ravesteyn
NT
,
Fracheboud
J
,
Draisma
G
,
de Koning
HJ
. 
Interpreting overdiagnosis estimates in population-based mammography screening
.
Epidemiol Rev
2011
;
33
:
111
21
.
5.
Falk
RS
,
Hofvind
S
,
Skaane
P
,
Haldorsen
T
. 
Overdiagnosis among women attending a population-based mammography screening program
.
Int J Cancer
2013
;
133
:
705
12
.
6.
Welch
HG
,
Black
WC
. 
Overdiagnosis in cancer
.
J Natl Cancer Inst
2010
;
102
:
605
13
.
7.
Biesheuvel
C
,
Barratt
A
,
Howard
K
,
Houssami
N
,
Irwig
L
. 
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
.
8.
Ciatto
S
. 
The overdiagnosis nightmare: a time for caution
.
BMC Womens Health
2009
;
9
:
34
.
9.
Duffy
SW
,
Lynge
E
,
Jonsson
H
,
Ayyaz
S
,
Olsen
AH
. 
Complexities in the estimation of overdiagnosis in breast cancer screening
.
Br J Cancer
2008
;
99
:
1176
8
.
10.
Gunsoy
NB
,
Garcia-Closas
M
,
Moss
SM
. 
Estimating breast cancer mortality reduction and overdiagnosis due to screening for different strategies in the United Kingdom
.
Br J Cancer
2014
;
110
:
2412
9
.
11.
Baker
SG
,
Prorok
PC
,
Kramer
BS
. 
Lead-time and overdiagnosis
.
J Natl Cancer Inst
2014
;
31
:
106
.
12.
Duffy
SW
,
Agbaje
O
,
Tabar
L
,
Vitak
B
,
Bjurstam
N
,
Björneld
L
, et al
Overdiagnosis and overtreatment of breast cancer: estimates of overdiagnosis from two trials of mammographic screening for breast cancer
.
Breast Cancer Res
2005
;
7
:
258
65
.
13.
Feinleib
M
,
Zelen
M
. 
Some pitfalls in the evaluation of screening programs
.
Arch Environ Health
1969
;
3
:
412
5
.
14.
Office for National Statistics [Internet] UK: England and Wales, Interim Life Tables, 1980–1982 to 2008–10 (Excel Sheet) [updated 2011 Oct 11, cited 2015 Jul 14]
.
Available from
: http://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/lifeexpectancies/datasets/interimlifetablesenglandandwales
15.
SSB.NO [Internet] NO: Statistics Norway. Stat Bank. Table 07902 [cited 2015 Jul 14]. Available from
: https://www.ssb.no/en/statistikkbanken/
16.
Weedon-Fekjaer
H
,
Vatten
LJ
,
Aalen
OO
,
Lindqvist
B
,
Tretli
S
. 
Estimating mean sojourn time and screening test sensitivity in breast cancer mammography screening: new results
.
J Med Screen
2005
;
12
:
172
8
.
17.
Skaane
P
. 
Studies comparing screen-film mammography and full-field digital mammography in breast cancer screening: updated review
.
Acta Radiol
2009
;
50
:
3
14
.
18.
Skaane
P
,
Bandos
AI
,
Gullien
R
,
Eben
EB
,
Ekseth
U
,
Haakenaasen
U
, et al
Comparison of digital mammography alone and digital mammography plus tomosynthesis in a population-based screening program
.
Radiology
2013
;
267
:
47
56
.
19.
Pisano
ED
,
Gatsonis
C
,
Hendrick
E
,
Yaffe
M
,
Baum
JK
,
Acharyya
S
, et al
Digital Mammographic Imaging Screening Trial (DMIST) Investigators Group. Diagnostic performance of digital versus film mammography for breast-cancer screening
.
N Engl J Med
2005
;
353
:
1773
83
.
20.
Lauby-Secretan
B
,
Scoccianti
C
,
Loomis
D
,
Benbrahim-Tallaa
L
,
Bouvard
V
,
Bianchini
F
, et al
Breast-cancer screening—viewpoint of the IARC Working Group
.
N Engl J Med
2015
;
372
:
2353
8
.
21.
National Health Services
. 
UK: National Health Services Breast Screening Programme [cited 2015 Oct 23]
.
Available from
: http://www.cancerscreening.nhs.uk/breastscreen/
22.
Hofvind
S
,
Ursin
G
,
Tretli
S
,
Sebuødegård
S
,
Møller
B
. 
Breast cancer mortality in participants of the Norwegian Breast Cancer Screening Program
.
Cancer
2013
;
119
:
3106
12
.
23.
Independent UK Panel on Breast Cancer Screening
. 
The benefits and harms of breast cancer screening: an independent review
.
Lancet
2012
;
380
:
1778
86
.
24.
Duffy
SW
,
Tabar
L
,
Olsen
AH
,
Vitak
B
,
Allgood
PC
,
Chen
TH
, et al
Absolute numbers of lives saved and overdiagnosis in breast cancer screening, from a randomized trial and from the Breast Screening Programme in England
.
J Med Screen
2010
;
17
:
25
30
.
25.
Waller
M
,
Moss
S
,
Watson
J
,
Møller
H
. 
The effect of mammographic screening and hormone replacement therapy use on breast cancer incidence in England and Wales
.
Cancer Epidemiol Biomarkers Prev
2007
;
16
:
2257
61
.