Overdiagnosis in the population-based service screening programme with mammography for women aged 40 to 49 years in Sweden

Abstract

Objectives

To estimate the level of overdiagnosis of all breast cancers and of invasive breast cancers in women aged 40–49 invited to the subsequent screening rounds in the Swedish service-screening programme 1986–2005.

Methods

To estimate the level of overdiagnosis in subsequent screening, the rate ratios (RR) of the breast cancer incidence in the study group (women in areas with screening in ages 40–49) and the control group (women in areas with no screening in ages 40–49) were calculated for all breast cancers and for invasive breast cancers. The RR estimates were adjusted for the prescreening difference in incidence between study and control group and for lead time.

Results

The prescreening incidence rate ratio was estimated at 0.92 (95% confidence interval [CI]: 0.88–0.97). The number of breast cancer cases and person-years were 6047 and 3.8 million, and 7790 and 5.2 million, in the study group and control group respectively during the study period. The RR estimate for all cancers was 1.01 (95% CI: 0.94–1.08) when adjusted for prescreening difference and a lead time of 1.2 years. The corresponding estimate for invasive breast cancers was 0.95 (95% CI: 0.88–1.02).

Conclusions

We found no significant overdiagnosis for women aged 40–49 in the Swedish service screening programme with mammography.

INTRODUCTION

In 1986 the National Board of Health and Welfare (Sweden) issued guidelines recommending that county councils invite women aged 40–54 to screening every 18 months and women aged 55–74 every 24 months. The guidelines were modified in 1987 and 1988 recommending that, in case of limited resources, the county councils should focus on ages 50–69. Consequently, about half of the Swedish counties invited women aged 40–69 or 40–74 and the remaining counties invited women aged 50–69 or 50–74 (Table 1).

Table 1

Lower age limit for invitation to screening and period of inclusion into the study and control group in the areas for screening in Sweden for each year of the follow-up 1986 to 2005. n = screening had not yet started at beginning of year. Study group (screening of women age 40–49) = shaded cells with lower invitation age in italics. Control group (no screening of women age 40–49) = shaded cells with lower invitation age in normal or no lower invitation age.

County \ Year	1986	1987	1988	1989	1990	1991	1992	1993	1994	1995	1996	1997	1998	1999	2000	2001	2002	2003	2004	2005
Screening from age 40
Östergötland	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Dalarna	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Gävleborg	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Jönköping (Höglandet)	n/40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Västmanland	n/40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Jönköping (Jönköping)		n/40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Uppsala			n/40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Södermanland				40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Västernorrland					40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Norrbotten					40	40	40	40	40	40	40	40	40	40	40	40	40	40	40	40
Gotland												40	40	40	40	40	40	40	40	40
Screening from age 40 and 45
Örebro (SoE)		n/40	40	40	40	40	40	40	40	40	40/45	45	45	45	45	45	45	45	45	45
Screening from age 40 and 50
Kalmar	n/40	40	40	40	40	40	40	50	50/40	40	40	40	40	40	40	40	40	40	40	40
Skåne (SoW)		40	40	40	40	40	40	40	40	40	40/50	50	50	50	50	45	45	45	45	45
Skåne (SoE)		n/40	40	40	40	40	40	40	40	40	40	40	50	50	50	50	50	50	50	50
Skåne (middle)				n/40	40	40	40	40	40	50	50	50	50	50	50	45	45	45	45	45
Skåne (NW)				n/50	50	50/40	40	40	40	40	40	50	50	50	50	50	50	45	45	45
Jönköping (Habo, Mullsjö)				50	50	50	50	50	50	50	50	50	50	40	40	40	40	40	40	40
Kronoberg					50	50	50	50	50	50	50	50	50	50/40	40	40	40	40	40	40
Screening from age 50 with a short period of screening from age 40 and 45
Västra Götaland (N.Bohuslän)	n/50	50	50	50	45	45	45	50	50	50	50	50	50	50	50	50	50	48	48	48
Västra Götaland (S.Bohuslän)	n/50	50	50	50	45	45	45	50	50	50	50	50	50	50	50	50	48	48	45	45
Västra Götaland (S Älvsborg)			n/40	40	40	50	50	50	50	50	not	50	50	50	not	not	not	50	50	50
Stockholm				50	50	50	50	50	50	50	50	50	50	50	50	50	50	50	50	50/40
Västra Götaland (Skaraborg)				50	50	50	50	50	50	50	50	50	50	50	50	50	50/48	48	48	48/45
Halland				50	50	50	50	50	50	50	50	50	50	50	50	49/48	45	45/42	42	42
Västra Götaland (N Älvsborg)								50	50	50	50	50	50	50	50	50	50	48	48	48
Västerbotten										n/50	50	50	50	50	50/40	40	50	50	50	50
Screening from age 50
Skåne (NE, kristianstad)				n/50	50	50	50	50	50	50	50	50	50	50	50	50	50	50	50	50
Skåne (Malmö)					50	50	50	50	50	50	50	50	50	50	50	50	50	50	50	50
Värmland								n/50	50	50	50	50	50	50	50	50	50	50	50	50
Jämtland											n/50	50	50	50	50	50	50	50	50	50
Västra Götaland (Göteborg)										50	50	50	50	50	50	50	50	50	50	50

N = Northern, So = Southern, E = Eastern, W = Western

The goal of mammography screening is to detect breast cancers at an early stage, which can save lives. In our previous cohort study, SCRY (SCReening of Young women), we showed that the breast cancer mortality was 26% lower among those invited to screening and 29% lower among those attending screening in the ages 40–49 compared with those not invited in the ages 40–49.¹

The potential harms of mammography screening are radiation exposure, discomfort and anxiety due, for example, to false-positives, false-negatives and unnecessary treatment, for example due to overdiagnosis. Overdiagnosis is the excess of cancers diagnosed with screening compared with without screening that is not due to earlier diagnosis. In other words, overdiagnosis is the cancers detected with screening that would otherwise have remained undiagnosed throughout the women's lifetimes.

Several attempts have been made to estimate the level of overdiagnosis in inviting women aged 40–49 years to screening. Moss concluded that for women aged 40–74 there is a possible shift from invasive to in situ breast cancer but no evidence of overdiagnosis,² while Jørgensen and Gøtzsche for the same age interval concluded that one in three breast cancers was overdiagnosed.³ Biesheuvel concluded that most estimates of overdiagnosis for ages 40–49 were biased, but ranged from −4% to 7% in the studies with the best precision in their estimates.⁴ A cohort study of 11 out of 24 Swedish counties followed for on average 12.8 years from screening start to the year 2000 found no excess incidence for ages 40–49.⁵

Three methods have been applied in the estimation of overdiagnosis: modelling, comparison of incidence rates, and comparison of cumulative incidences between screened and unscreened.⁴ The aim of this study was to estimate the level of overdiagnosis of all breast cancers and of invasive breast cancers in women 40–49 years invited to subsequent screening in the Swedish service screening programme 1986–2005 using incidence rates.

MATERIALS AND METHODS

The SCRY cohort

During the study period 1986–2005 about half of the Swedish counties invited women aged 40–49 to service screening with mammography (Table 1). The SCRY cohort consists of all 40–49-year-old women in Sweden, split into a control group and a study group based on whether the women were invited to screening or not. For the control group (without screening in ages 40–49) the follow-up time period was chosen so that the averages (weighted using population sizes) for follow-up time and mid-calendar year of follow-up corresponded to those of the study group (Table 1). For example, women in the counties of Halland and Jämtland were included in the control group from 1986–2001 and 1990–2005, respectively, to achieve similar follow-up in study and control groups. Halland was only followed through 2001 due to a gradual inclusion of younger women into the screening programme starting in 2001, disqualifying later years from the control group. Later years were not included in the study group due to short follow-up. Only first primary breast cancers were included in the study. Participation rates varied between 80% and 90%. In 1990, there were 620,620 women aged 40–49 years in Sweden.

Information on the service screening programmes, including initiation of the programme, invited age groups and changes during the study period, was collected through a questionnaire to the screening centres. Data on breast cancer cases were retrieved from the Swedish Cancer Registry. Breast cancers were defined as breast cancer according to the International Classification of Diseases for Oncology, Revision Seven [ICD7], code 170. Population data that were used in the calculation of person-years were supplied by Statistics Sweden. The SCRY cohort has previously been described in more detail.¹

In the present study Blekinge county and parts of Örebro county, comprising 10% of the person-years in the original study group, were excluded as only women aged 45–49 were invited to screening, which might bias the overdiagnosis estimates. The control group was identical to the original SCRY cohort control group. Thus, the study includes 23 out of 24 counties in Sweden (Table 1).

Lead time

Breast cancers may be detected earlier with screening, prolonging the elapsed time between diagnosis and death even if the time of death remains the same. The time intervals from when a cancer is detectable with mammography and from when a cancer actually is detected to when it would surface clinically is called sojourn time and lead time, respectively.

Lead time will cause a temporary increase in the incidence during the first, so-called prevalence screening round, due to prevalent ‘future’ cancers being detected and added to the incidence that would have been observed without screening. A similar increase takes place in the youngest ages invited to screening (i.e. ages 40–41). When comparing a screened and an unscreened population this will cause a bias due to lead time. We refer to this bias as prevalence peak bias.

Breast cancer incidence increases with age⁶ and may change over time, e.g. an increasing trend. Because a screened population corresponds to the future incidence in an unscreened population, another bias due to lead time may occur. Thus, there may be a bias when comparing the study group with the control group throughout follow-up, even if prevalence screening is excluded. We refer to this bias as trend bias.

The study period of the present study equals that of the original SCRY cohort study but was divided into prevalence screening, i.e. the first three years of each area's follow-up and ages 40 and 41, and subsequent screening. To adjust for prevalence peak bias, prevalence screening was excluded. It should be noted that the first round of screening is assumed to be included in the first three years of screening but does not necessarily take three years. In a similar way, some women may be invited to screening twice in ages 40–41. The weighted average follow-up years and average mid-calendar year of follow-up for subsequent screening were 12.4 and 1999 and 12.7 and 1999 in the study group and control group, respectively.

Overdiagnosis

To estimate the level of overdiagnosis in women aged 40–49 the crude rate ratio (RR) was calculated as the ratio of the breast cancer incidence (all breast cancer, i.e. in situ and invasive) in the study group and the control group. A crude RR between the study and control group was also calculated for a reference period with equal follow-up in all areas, 1970–1985, when screening had not yet begun in most counties (RR_ref). The corresponding RR and RR_ref were also calculated for invasive breast cancers. To adjust for the potential difference in baseline incidence between the study and control group the RR estimates of the study period were divided by the RR_ref estimates.

Five areas, covering 18% of the study population, were excluded from the calculation of the RR_ref due to ongoing screening activities (randomized controlled trial and pilot project) that took place before the start of their service-screening programmes. To estimate the impact of this exclusion a RR for the study period excluding the same five areas was calculated.

RRs were also estimated for the first three years of screening and the youngest ages invited to screening (40–41 years). These RRs are not estimates of overdiagnosis.

The software programme R was used for statistical analyses (R Foundation for Statistical Computing, Vienna, Austria).⁷

Adjustment for trend bias

To adjust for trend bias the RR estimates were divided by an adjustment for the annual relative incidence difference (RR_LT).

where

LT = average population lead time for all breast cancer, CD = relative change in incidence in the control group per calendar year, and AD = relative change in incidence in the control group per year of age.

The calculations of annual change in incidence in the control group (CD and AD) were based on incidence data for women aged 40–49 in the areas included in the control group in 1986–2005. AD is the average relative change in incidence per year of age, i.e. from age 40 to 41, from 41 to 42 and so on. CD is the corresponding average relative change in incidence per calendar year, i.e. from 1986 to 1987, from 1987 to 1988 and so on. The assumed average lead time for the screened population in the study group, including cancers not detected through screening can be calculated using estimates of mean sojourn time (MST) and sensitivity (S) for 40–49-year-olds. The population lead time LT can be calculated based on the lead time and the screening participation rate (PR) in the study group.^8–15

Due to the variation in the estimates of MST and S in the literature a sensitivity analysis was made by calculating RR estimates based on differing LTs.

RESULTS

Baseline incidence difference and trend bias

During the pre-screening reference period 1970–1985 there were 2486 breast cancer cases during 2.5 million person-years in the study group and 3879 breast cancer cases during 3.5 million person-years in the control group resulting in an RR of 0.92 (95% confidence interval [CI]: 0.88–0.97). The estimate was similar for invasive breast cancer (Table 2).

Table 2

Summary of results for reference period (prescreening period) by diagnosis and age and for prevalence screening, i.e. years and ages excluded in overdiagnosis estimates adjusted for prevalence peak. Numbers of breast cancer cases and person-years in the study group (SG) and the control group (CG), respectively, relative risks (RR) and 95% confidence intervals (CI). preD = prescreening difference

		Reference period (pre-screening)		Prevalence screening (follow-up excluded when adjusting for prevalence peak bias)
		All cancers	Invasive cancers only	Youngest women invited	3 first years of screening
	Group\Age	40–49 years	40–49 years	40–41 years	40–49 years
No. of breast cancer cases	SG	2486	2381	633	1704
	CG	3879	3705	658	1955
No. of person-years	SG	2,453,373	2,453,373	750,739	1,053,885
	CG	3,527,275	3,527,275	1,051,420	1,351,461
Crude estimate	RR	0.92	0.92
	95% CI	0.88–0.97	0.88–0.97

The mean relative increase in incidence in the control group (CD) was 2.45% per calendar year 1985–2005 and 5.07% per one year age group (AD). The participation rate in the study group was 0.8. Assuming a sojourn time of 2.4 years and a sensitivity of 0.6 resulted in a population lead time of 1.2 years and a lead time adjustment (RR_LT) of 1.09. Population lead times based on a range of possible sojourn and sensitivity estimates were made resulting in population lead times of up to 1.5 years. The lead time adjustments for a population lead time of 1.0 and 1.5 years were 1.08 and 1.12 respectively.

Overdiagnosis, main results

During the study period 1986–2005 there were 6047 breast cancer cases during 3.8 million person-years in the study group and 7790 breast cancer cases during 5.2 million person-years, resulting in an estimated crude RR of 1.07 (95% CI: 1.03–1.10) (Table 3).

Table 3

Summary of main results by diagnosis, type of adjustment and ages included. Numbers of breast cancer cases and person-years in the study group (SG) and the control group (CG), respectively, relative risks (RR) and 95% confidence intervals (CI). preD = prescreening difference

		All cancers		Invasive cancers only
	Group\Age	40–49 years	42–49 years	40–49 years	42–49 years
No. of breast cancer cases	SG	6047	4090	5081	3458
	CG	7790	5598	6994	5030
No. of person-years	SG	3,814,314	2,443,684	3,814,314	2,443,684
	CG	5,238,115	3,391,969	5,238,115	3,391,969
Crude estimate	RR	1.07		1.00
	95% CI	1.03–1.10		0.96–1.03
Adjusted for PreD	RR	1.16		1.08
	95% CI	1.09–1.23		1.01–1.15
Adjusted for PreD and prevalence peak bias	RR		1.10		1.03
	95% CI		1.03–1.17		0.97–1.10
Adjusted for PreD, prevalence peak bias and trend bias	RR		1.01		0.95
	95% CI		0.94–1.08		0.88–1.01

When adjusted for the prescreening difference the RR estimate was 1.16 (95% CI: 1.09–1.23) and when also adjusted for prevalence peak bias by excluding prevalence screening and for trend bias the RR estimate was 1.01 (95% CI: 0.94–1.08). Based on a population lead time of 1 year and 1.5 years the adjusted RR estimate was 1.02 and 0.99 respectively. For invasive breast cancers the corresponding adjusted RR estimate, i.e. based on a population lead time of 1.2 years, was 0.95 (95% CI: 0.88–1.01). Five areas could not be included in the calculation of the RR for the reference period. When these areas were excluded from the study period the study period RR estimates were two percentage points higher.

Prevalence screening

The RR estimate for the initial three years of screening and for ages 40–41 were 1.11 (95% CI: 1.02–1.21) and 1.34 (95% CI: 1.19–1.51) respectively when adjusted for the prescreening difference and trend bias but not for prevalence peak bias.

DISCUSSION

During the time period 1986–2005 about half of Sweden's counties invited women aged 40–69 (74) years and half invited women aged 50–69 (74) years. This facilitated a study of the overdiagnosis in the Swedish service screening programme with mammography of women aged below 50 years.

For the subsequent screening rounds the estimated overdiagnosis for all cancers (in situ and invasive) was one percent and non-significant when adjusted for prescreening difference, bias due to lead time such as prevalence peak bias and trend. Thus there was no overdiagnosis in the 40–49 year age group in subsequent screening.

Strengths and limitations

The incidence might have differed between the study and control group independently of screening because the current study was a geographical comparison where exposure was determined by screening policy in each area. This difference was adjusted for by utilizing data from the reference period. A few areas could not be included in the reference period due to variation in invited age groups. When these areas were excluded in the study period this increased the RR estimates, but only by two percentage units. Screening continued in ages above 50, hence no compensatory drop in incidence could be observed in this study. This necessitated adjustment for the potential biases associated with lead time. Trend bias was adjusted for based on population lead time and prevalence peak bias by excluding prevalence screening. There may have been some prevalence screening also in subsequent screening due to women not participating in the first round, leading to an overestimation of overdiagnosis, though the high participation rate indicates that it was small. Population lead time was calculated to be 1.2 years but varying estimates of mean sojourn time and sensitivity have been made for the age group 40–49.^8–15 A sensitivity analysis was therefore conducted. For population lead times of 1.0 and 1.5 years the overdiagnosis estimates were 1.02 and 0.99 respectively and not statistically significant.

Comparison with earlier studies on ages 40–49

Moss estimated that the absolute excess of breast cancers in the NBSS I trial for women 40–49 was 0.25 (95% CI 0.04–0.46) and 0.12 (95% CI −0.08–0.32) corresponding to RR estimates of 1.14 and 1.08 for all and invasive cancers respectively.² Gøtzsche estimated it at 1.30 (95% CI: 1.13–1.50) for all cancers, however the Moss study had a longer follow-up. Gøtzsche also estimated the RR in the Göteborg trial for women 39–49 years (RR = 1.13; 95% CI: 0.90–1.41).¹⁶ Biesheuvel noted that Gøtzsches estimates were not adjusted for lead time and are therefore biased upwards.⁴ In the current study, which included 23 out of 24 counties, the number of breast cancer cases was 6047 in the study group. The number of breast cancer cases in the NBSS I and Gothenburg trial study groups were 663 and 144 respectively. A previous Swedish cohort study based on 11 out of 24 Swedish counties estimated the overdiagnosis in ages 40–49 to be 0.96 (95% CI: 0.77–1.21).⁵

Prevalence screening

Prevalence screening was present in the first three years of screening and the youngest ages invited to screening (40–41). Though the screening round was closer to two years than three it is likely that there were backlogs, i.e. the screening centres could not keep the two-year screening interval. The RR estimates for the first three years of screening are low in comparison with the estimates for the first ages of screening (ages 40 and 41). This may be due to screening starting a few months into the first calendar year of screening so that a short period of non-screening is included. Also, in the areas with screening activities before start of screening the first years of inclusion in the study group were not prevalence screening and therefore diluted the increase of the observed incidence. Notable is that the RR estimates for prevalence screening were not adjusted for prevalence peak bias and therefore not meant as estimates of overdiagnosis.

In situ and invasive cancers

Though the majority of breast cancers are invasive, excluding the in situ cancers reduces the overdiagnosis estimate by six percentage points, hence in situ cancers make out a disproportionately large part of the estimate. For invasive cancers the RR estimate was 0.95 and non-significant. This indicates that fewer invasive cancers were diagnosed in ages 40–49 with screening than without, possibly due to cancers being detected already at the in situ stage. Without screening at age 40–49 pre-clinical cancer can progress to clinical cancer, remain as pre-clinical cancer or regress. A majority of in situ cancers seem to progress to invasive cancers.¹⁷ Cancers that remain as pre-clinical would probably be detected in screening at ages 50 and older because the screening programme in Sweden covers all women from 50 years of age.

CONCLUSION

The current large study of the Swedish service screening programme finds no significant overdiagnosis from subsequent screening for women aged 40–49.

Footnotes

ACKNOWLEDGEMENTS

This study was funded by the Swedish Cancer Society.

References

Hellquist

, Duffy

, Abdsaleh

, Effectiveness of population-based service screening with mammography for women ages 40 to 49 years: evaluation of the Swedish Mammography Screening in Young Women (SCRY) cohort. Cancer 2010;117:714–22

Moss

. Overdiagnosis and overtreatment of breast cancer: overdiagnosis in randomised controlled trials of breast cancer screening. Breast Cancer Res 2005;7:230–4

Jorgensen

, Gotzsche

. Overdiagnosis in publicly organized mammography screening programmes: systematic review of incidence trends. BMJ 2009;339:2587

Biesheuvel

, Barratt

, Howard

, Houssami

, Irwig

. 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

Jonsson

, Johansson

, Lenner

. Increased incidence of invasive breast cancer after the introduction of service screening with mammography in Sweden. Int J Cancer 2005;117:842–7

Krtolica

, Campisi

. Cancer and aging: a model for the cancer promoting effects of the aging stroma. Int J Biochem Cell Biol 2002;34:1401–14

Tabar

, Fagerberg

, Chen

, Duffy

, Gad

. Tumour development, histology and grade of breast cancers: prognosis and progression. Int J Cancer 1996;66:413–9

Tabar

, Fagerberg

, Chen

Efficacy of breast cancer screening by age. New results from the Swedish Two-County

Trial. Cancer 1995;75:2507–17

10.

Tabar

, Vitak

, Chen

, The Swedish Two-County Trial twenty years later. Updated mortality results and new insights from long-term follow-up. Radiol Clin North Am 2000;38:625–51

11.

Chen

, Brock

, Wu

. Estimating key parameters in periodic breast cancer screening-application to the Canadian National Breast Screening Study data. Cancer Epidemiol 2010;34:429–33

12.

Duffy

, Chen

, Tabar

, Fagerberg

, Paci

. Sojourn time, sensitivity and positive predictive value of mammography screening for breast cancer in women aged 40–49. Int J Epidemiol 1996;25:1139–45

13.

Duffy

, Day

, Tabar

, Chen

, Smith

. Markov models of breast tumor progression: some age-specific results. J Natl Cancer Inst Monogr 1997;22:93–7

14.

Brekelmans

, Westers

, Faber

, Peeters

, Collette

. Age specific sensitivity and sojourn time in a breast cancer screening programme (DOM) in The Netherlands: a comparison of different methods. J Epidemiol Community Health 1996;50:68–71

15.

Paci

, Duffy

. Modelling the analysis of breast cancer screening programmes: sensitivity, lead time and predictive value in the Florence District Programme (1975–1986). Int J Epidemiol 1991;20:852–8

16.

Gotzsche

. On the benefits and harms of screening for breast cancer. Int J Epidemiol 2004;33:56–64; discussion 69–73

17.

Yen

, Tabar

, Vitak

, Smith

, Chen

, Duffy

. Quantifying the potential problem of overdiagnosis of ductal carcinoma in situ in breast cancer screening. Eur J Cancer 2003;39:1746–54