Increased incidence of invasive breast cancer after the introduction of service

screening with mammography in Sweden

Hakan Jonsson*, Robert Johansson and Per Lenner

Department of Radiation Sciences, Oncology, Umea University, Umea, Sweden

Screening with mammography has been shown to substantiallyreduce mortality from breast cancer. The incidence of invasivecancer will increase as screening starts, and it is desirable that itgradually returns to the same level as before screening. Age-spe-cific incidence of invasive breast cancer in 11 Swedish counties,including 463,000 women aged 40–74 years, was analysed beforeand after the start of service screening with mammography. Inci-dence, as observed on average during 12.8 years from screeningstart, was compared to expected incidence based on the incidenceduring a 15-year period preceding screening start. The height ofthe incidence peak during the first screening round was increasingwith increasing age, compatible with the accumulation in the pop-ulation of slowly growing tumours by age. All analysed age groupsshowed an increased ratio between observed stabilised incidence7–14 years after screening start and expected incidence. When rel-ative risks were adjusted for lead time, the estimates were 1.54(95% confidence interval [CI] 1.33–1.79) and 1.21 (95% CI 1.04–1.41) for the age groups 50–59 and 60–69 years, respectively. Inthe age groups 40–49 and 70–74, no change was observed. Thefindings were further confirmed by the observation of a disappear-ance in the screened population of the notch in the increasingtrend of age-specific breast cancer incidence for the ages aftermenopause. This notch could indicate hormone-related retarda-tion in tumour growth around menopause. It appears that manyof these clinically insignificant, retarded tumours are detectedwith screening mammography.' 2005 Wiley-Liss, Inc.

Key words: breast cancer; incidence; mammography; screening;overdiagnosis

Many studies have demonstrated a decline in breast cancermortality after the introduction of screening for breast cancerwith mammography, both in trials and service screening.1–6

There are, however, also drawbacks connected with screeningsuch as anxiousness among women who are recalled and popula-tion radiation load. Another drawback is the effect of lead time,which for several individuals causes an increased time to livewith a diagnosis of breast cancer. Another potential disadvantageis related to length bias. The screen-detected cases are not a ran-dom sample of cases in the preclinical detectable phase. Instead,the cases with long such phases, i.e., slowly growing tumours,are overrepresented. For a certain number of subjects, a latentbreast cancer may be disclosed with mammography, although thetumours would otherwise never have been detected. A corre-sponding phenomenon is well known in prostate cancer fromstudies on autopsy material.7 The presence of latent cancers hasbeen debated in breast cancer, but studies addressing this issuehave been few.8,9 Particularly, the follow-up after start of screen-ing has hitherto been too short in many materials to allow validconclusions.

Because of lead time, an increased incidence should beexpected in the first screening round, the prevalence round. This isalso a well-known fact.10–12 What level of incidence that will fol-low in the next rounds is, however, uncertain.

The objective of our study was to estimate the incidence ofinvasive breast cancer after introduction of service screening withmammography in Sweden and to compare this rate with theexpected incidence. A higher observed than expected incidencemay indicate an increased detection of clinically insignificantcancer.

Material and methods

After recommendations, in 1985, by the National Board ofHealth and Welfare, service screening with mammography wasintroduced in Sweden. The screening is population based and isgenerally ruled on a county level. The first counties started servicescreening in 1986, followed by other counties up to 1997 when thewhole country was covered. In some of the counties, 2 or moredifferent programs were performed. The programs can thereforediffer between counties and sometimes within counties concerningage limits for invitation and start time. The average attendancerate in Sweden was 81% in 1995–96.13 However, the attendancewas lower (around 70%) in the 3 large cities (covering 19% of theSwedish population). Opportunistic screening exists in the largecities but is rare elsewhere, which might explain some of the dif-ferences in attendance rates. However, in our study none of thelarge cities was included.

Individual data on breast cancer incidence 1971–2000 wereretrieved from the Swedish Cancer Registry and aggregated popu-lation data from Statistics Sweden. Some characteristics for thedifferent screening programs, such as start times, age limits andscreening intervals, were collected by a questionnaire that wassent to the screening centres.

Only invasive breast cancer was analysed in our study. Caseswere defined by site code 170 (International Classification of Dis-eases, revision 7) and histopathologic code 096 (WHO C24.1).

We used data from 11 counties where service screening startedbetween 1986 and 1990 (Table I). To obtain as distinct start timeas possible, only counties with one screening program were con-sidered. There was, however, one exception, namely the county ofJ€onk€oping. In this county, 2 programs were used but with the sameage limits for invitation and only 8 months between the startpoints. Counties involved in randomised trials were excluded, aswell as the county of G€avleborg, where screening started in themid 1970s.

The counties were followed 15 years before the screening startand for 11–15 years during screening depending on the year ofscreening start, up to year 2000 for all. However, one county,Skaraborg, was in 1998 merged with 2 other counties into onelarger county and was therefore only followed to 1997 (9 years ofscreening). The average follow-up after screening start, weightedby the size of the population, was 12.8 years.

For each county, we identified the year of screening start (TableI). The years of follow-up were numbered consecutively with theyear of screening start as 0. All counties were followed from 15years before screening start and until calendar year 2000 (year215,214, . . . , 0, 1, . . . , k), where k differed between 8 and 14.

We defined a ‘‘stabilised phase’’ (SP) for the incidence as theperiod from year 7 and forward. It is well known that the inci-dence shows an increase during the first screening round (the prev-alence peak). However, the level of incidence thereafter has neverbeen clearly demonstrated. To define a SP, we had to choose the

Grant sponsor: V€asterbotten county council.*Correspondence to: Oncological Centre, Umea University Hospital,

S-901 85 Umea, Sweden. E-mail: hakan.jonsson@oc.umu.seReceived 22 November 2004; Accepted after revision 21 March 2005DOI 10.1002/ijc.21228Published online 14 June 2005 in Wiley InterScience (www.interscience.

wiley.com).

Int. J. Cancer: 117, 842–847 (2005)' 2005 Wiley-Liss, Inc.

Publication of the International Union Against Cancer

start of the phase not too early to avoid involvement from theprevalence peak and the decreased incidence some years there-after. The choice of year 7 as start of SP was based mainly on thiscriterion. Still, it was early enough to give a sufficient number ofyears of follow-up for valid estimation.

The crude age-specific (5-year classes) breast cancer incidencebefore the screening start and during the SP was plotted. The pre-screening incidence was estimated as an average over year 25 to21. The average incidence during the SP of screening was esti-mated from year 7 and forward. Only age groups that were invitedto screening were included in the analysis from each county. Forthe county of Skaraborg, the 70–74-year group was not usedbecause the upper age limit was changed during the studiedperiod.

Using a log-linear Poisson model based on year 215 to 21, anestimation of the expected incidence in absence of screening wascalculated. Similar models were used to estimate the RR betweenobserved incidence and expected incidence in absence of screen-ing in year 1 (prevalence phase, PP) and year 7 and forward (stabi-lised phase, SP), respectively. As covariates in the model, year(continuous) and county were used. We estimated RR in relationto expected incidence in the PP and SP. For this estimation, weintroduced dummy variables as covariates defined as zero for allyears except year 1 (PP) and year 7 and forward (SP), where itwas set to one. Year 0 and 2–6 were excluded from all Poissonestimation. Each of the 4 age groups (40–49, 50–59, 60–69 and70–74 years) was analysed separately. Only data from countiesthat invited women to screening were used in the respective agegroup.

Observed age-specific incidence was plotted for the age groups40–49, 50–59, 60–69 and 70–74 years. Also, the expected inci-dence in absence of screening as well as the mean incidence dur-ing the SP, estimated in the Poisson model above, were plotted foreach year. At the end of the observation period, it was not possibleto use all counties for estimation of the incidence in the SP due toa later screening start in some counties. Thus, the summation wasmade over the remaining counties.

Due to lead time, incident cancer cases may be moved to alower age class if they were detected at screening compared tothose not screened. For example, a breast cancer can be screendetected at age 49, whereas in absence of screening, hypotheti-cally, for the same women the cancer had been detected at age 51.Thus, the comparison between prescreening and screening periodsfor specific age groups can be biased. We, therefore, also used anadjustment for lead time. Mean sojourn times have been estimatedat 2.4, 3.7 and 4.2 years for the age groups 40–49, 50–59 and 60–69, respectively.14 Assuming that lead time is exponentially dis-tributed and that the screening intervals are all of fixed equallength, the corresponding expected lead times among screen-

detected cases are equal to the mean sojourn times. However, onlyscreen-detected cases will have lead time. To find the average leadtime for all cases, we had to estimate the proportion of screen-detected cancer in the material. Based on registered data from 2counties in Northern Sweden (Norrbotten and V€asternorrland), wefound that the proportion of screen-detected cases at age 40–74years was 65%. Using this figure, the resulting average lead timesfor all breast cancer cases were 1.6, 2.4 and 2.7 years for the agegroups 40–49, 50–59 and 60–69 years, respectively. For the agegroup 70–74 years, we had no estimate of the mean sojourn time.Based on the increase over age, we approximated lead time to 3.0years for this age group. These average lead times were added toage at diagnosis for all cases year 1 and forward. The age-adjustedRR was then calculated using the Poisson models above.

For the majority of the estimated Poisson models, a slight over-dispersion was observed. We, therefore, extended the model toaccount for overdispersion.15 Confidence intervals then becamesomewhat wider, whereas the point estimates remained unchanged.The models were estimated using the S-PLUS software.16

Results

The observed incidence in the 5-year period preceding screen-ing start and in the stabilised phase year 7 and forward was plottedby age in Figure 1. The lower curve can be seen as the expectedincidence only if there had been no trend. The change of age-spe-cific breast cancer incidence in a screened population after stabili-sation, compared to absence of screening, can clearly be seen.

For all age groups, a substantial increase in incidence was seenat the start of screening (Fig. 2a–d). For the age groups 50–59 and60–69 years, a decrease followed by an increase toward the stabi-lised phase was seen after the prevalence peak (Fig. 2b,c). In theage group 70–74 years, a stabilisation was also seen but after asteep decrease (Fig. 2d). In Table II, the variation between thecounties is shown. However, due to the relatively small populationsize in individual counties, some of the observed differences aredue to chance.

For all age groups, the incidence at prevalence screening (meas-ured at the year after the start year) was clearly higher thanexpected without screening. It increased with age from 51% forages 40–49 to 100% for ages 70–74 years (Table III). When agewas adjusted for lead time, the figures became lower and increasedover age from 22% (40–49 years) to 84% (70–74 years).

For the age group 70–74 years, the incidence level in the stabi-lised phase seemed to be almost equal to the expected incidencewithout screening, whereas the incidences in the other age groupswere clearly higher (Fig. 2a–d). The estimated unadjusted RRswere 1.31, 1.69, 1.27 and 1.06 for the age groups 40–49, 50–59

TABLE I – TIME FOR START, AGE GROUPS INVITED TO SCREENING, NUMBER OF YEARS OF SCREENING,MEAN POPULATION OF THE INVITED AGE GROUP AT FIRST YEAR OF SCREENING AND SCREENING

INTERVAL IN THE STUDIED COUNTIES

CountryStart ofscreening

(month/year)

Invited agegroup (years)

No. ofyears ofscreening

No. ofwomen in target

population

Screeninginterval(months)

Norrbotten 03/1989 40–74 12 52,071 20V€asternorrland 01/1990 40–74 11 55,153 20V€astmanland 10/1986 40–69 15 45,483 20–24Skaraborg 04/1989 50–741 92 35,316 24Halland 01/1989 50–74 12 31,564 21Blekinge 03/1988 45–69 13 16,6273 24Kalmar 10/1986 40–74 15 48,210 24Kronoberg 08/1990 50–69 11 18,475 18J€onk€oping4 08/1986 40–74 15 60,649 22–24S€odermanland 11/1989 40–74 12 51,886 18–24Uppsala 02/1988 40–74 13 47,971 21–27

1Changed to 50–69 10/1991.–2Followed to 1997.–3Age 50–69.–4Two screening programs. The start ofthe later program was 04/1987.

843INCIDENCE DURING MAMMOGRAPHY SCREENING

and 60–69 and 70–74 years, respectively (Table III). After ageadjustment for lead time, the figures became 0.96 (95% confidenceinterval [CI] 0.77–1.21), 1.54 (95% CI 1.33–1.79), 1.21 (95% CI1.04–1.41) and 1.03 (95% CI 0.82–1.30) for age groups 40–49,50–59, 60–69 and 70–74 years, respectively (Table III).

Discussion

Our study demonstrates a substantial increase of breast cancerincidence after the introduction of service screening with mam-mography. The observed incidence during screening may be div-ided into 2 or more phases. The first phase is related to anincreased cancer detection rate during the first, or ‘‘prevalence,’’screening round. After the first round, a decrease apparently takesplace, and thereafter the incidence tends to stabilise, for most agegroups, at a higher level than before screening.

In a Dutch study, a contemporaneous comparison of incidenceof breast cancer in birth cohorts between two cities, one witha screening program (Nijmegen) and the other without (Arnhem),was made.9 After excluding DCIS, the rate ratio for ages 54–66was 1.10 measured 8–11 years after screening start. However, in a5-year period before screening start, rate ratio was 0.71 for thisage group. In another study, the incidence of invasive breast can-cer in women 50–69 years in Norway and Sweden before and afterintroduction of nationwide screening was analysed.8 An increaseby 45% in Sweden and 54% in the part of Norway that first startedscreening was found. The introduction of screening on the countylevel is a gradual process, implying that in Sweden the follow-uptime varied a lot between different counties. However, due to thelong follow-up period, the influence of prevalence rounds wasprobably negligible. In a study in Italy, breast cancer incidencewas followed for 10 years after introduction of screening. Theresults, adjusted for lead time, showed no evidence for overdiag-nosis of invasive cancer.17 In Denmark, a slightly increased inci-dence of invasive breast cancer in women aged 50–69 years wasseen after the prevalence peak in 3 districts when compared to therest of the country.12 However, the follow-up was short, 3–6 yearsafter screening start.

The peak in incidence connected with the first screening roundbecomes larger with increasing age. This finding is well compati-ble with previous studies18 and with a longer preclinical phase inolder women.14 The observation is not surprising because it couldbe regarded as natural that slowly growing tumours tend to accu-mulate as the population grows older.

The increase of breast cancer incidence compared to theexpected level that could be seen in the later screening rounds wasobservable in all age groups except the oldest (70–74 years). Oneexplanation could be lead time. Due to lead times, the patients willon average have their diagnoses at a lower age. To make postscre-ening and prescreening populations comparable concerning agegroups, we added the average lead time to the age of the post-screen cases. It did not substantially change the results in 50–69age group.

Another explanation for these findings is the presence of smalllatent tumours that during lifetime do not grow sufficiently largeto give symptoms and thereby they never become clinicallydetected. This view is comparable with the frequent presence ofsubclinical tumours in prostate cancer, which has been demon-strated on autopsy material.7 The implications of a high presenceof small latent tumours in the population are considerable, bothfor the individual and for society. When such tumours aredetected, a substantial number of subjects will be unnecessarilydiagnosed with cancer. The strains on the individual in this situa-tion may be large, psychologically by being aware of having thedisease and physically by arduous treatment. Therefore, this find-ing is not a trivial one.

It is a well-known phenomenon that the age-specific incidencecurve for breast cancer increases up to around the age of 50, thenlevels off to increase again at 55–60. This phenomenon has beencalled ‘‘Clemmesens hook.’’19 In this study, it was notable thatthis ‘‘hook’’ disappeared in the population of women who hadbeen exposed to several years of screening (Fig. 1). Attempts toexplain this finding must be highly hypothetical. One possibility isthat tumour growth is retarded and levels off at a subclinical levelat menopause due to a diminished exposure to estrogen. Withmammography, however, these small and clinically insignificantlesions are detected. This explanation is also compatible with ourfinding of the highest increase of incidence in age group 50–59(Table III, Fig. 2b).

A possibility that cannot be ruled out is that the observedincreased incidence may be caused by the increased use of hor-mone-replacement therapy (HRT) alone or in combination withscreening. HRT is used at ages around and after menopause. Thismeans that if HRT increases the risk for breast cancer in age 50–59years, the latency time has to be short. However, if tumours thatwould be retarded at menopause at a clinically nondetectable stageare stimulated by HRT and thereby become detectable by mammo-graphy, the ‘‘latency time’’ might be short. An increased incidencemay then immediately be observed. Further studies are needed toexplore the significance of these hypothetical mechanisms.

We had no possibility to estimate the expected incidence inabsence of screening otherwise than by using historical databecause screening was introduced nationwide. The results, there-fore, must be interpreted carefully. In Sweden, breast cancer inci-dence has increased almost linearly from 1960 to the mid-1980swhen service screening started. We have estimated a log-lineartrend during a 15-year period before screening start and assumedthis trend would have continued in absence of screening. A trend-break after screening start in the counties could hypothetically becaused by changes in known risk factors for breast cancer such ashormone replacement or childbirth. We are, however, not awareof any indications of such abrupt changes in risk factors for breastcancer. It is also unlikely that such changes have been parallelwith the introduction of service screening. Ideally, the incidenceafter introduction of screening should be studied in randomisedcontrolled trials where contemporaneous comparison is possible.Unfortunately, the trial time, i.e., the time before the control group

FIGURE 1 – Age specific mean incidence of invasive breast cancerbefore screening (dashed line) and in the stabilised phase (solid line,not adjusted for lead time or incidence trend) in the studied counties,ages 40 to 74 years. For the period before screening start the last5 years before screening start (25,24, . . . ,21) were analysed and forthe stabilised phase 7 years after start and forward.

844 JONSSON ET AL.

was invited to screening, generally was too short in the trials thathave been undertaken in Sweden and elsewhere.

We used an overall estimate for women 40–74 years of the propor-tion of breast cancer detected in screening based on 2 counties inNorthern Sweden. We are aware that this proportion generallyincreases with age and is higher for the first screening round andmightalso vary between different screening programs. However, we did notuse age and round specific estimates of the proportion of screendetected. The expected consequences should be a too big adjustmentin the stabilised phase but also a bias toward a too big adjustment inages 40–49 and a too conservative adjustment in ages 70–74.

The estimation of mean sojourn time14 is made under theassumption of exponential distribution. Mean lead time amongscreen-detected cases based on the estimated sojourn time wastherefore also based on this assumption. However, other distribu-tions of sojourn time may give other estimates of mean lead time.We, therefore, also made a sensitivity analysis for the age group50–59 by assuming a 50% higher and 50% lower mean lead timethan the one we presented in the results. This resulted in an

adjusted RR at 1.43 and 1.60 for the increased and decreased leadtime, respectively.

From round 2 and later, approximately one-fifth of the womenin age group 40–49 were annually invited for the first time, andtheir contribution to the incidence can be regarded as equal to thatfor prevalence screening. This will also be the case in age group50–59 years for the 3 counties (26% of the women) with the lowerage limit of 50 years. This could cause a bias toward a too highRR for the age group 40–49, but for age group 50–59 the adjustedRR in the prevalence phase was the same as for the stabilisedphase. Thus, there is no indication for a bias in this age group dueto prevalence screening.

During the first year of screening, only a limited proportion ofthe population was screened because in most counties the screen-ing did not start in the beginning of the year. However, during thefollowing year (year 1), there was generally close to 100% preva-lence screening.

The length of the prevalence round, which can differ betweencounties, will affect the intensity of the screening and thereby the

FIGURE 2 – (a–d) Incidence of invasive cancer summarised over the counties (solid line). For the period 7 years after screening start and after-wards (SP), mean incidence is indicated by a dashed line. Expected incidence (lower dashed line) in absence of screening estimated in the modelbased on the annual incidence 15 years before the screening start. (a) 40–49 years; (b) 50–59 years; (c) 60–69 years; (d) 70–74 years.

845INCIDENCE DURING MAMMOGRAPHY SCREENING

height of the observed incidence peak. If, hypothetically, the firstround was accomplished in 1 year instead of 2 years, the observedincidence during the prevalence round had been considerably higher.

In conclusion, mammography screening has in previous studiesbeen shown to fulfill its primary goal, namely to reduce mortalityfrom breast cancer. However, our study indicates an increasedbreast cancer incidence with mammography screening. The rea-son appears to be a high presence of small subclinical breastcancers in the female population, especially at ages 50–59years. Many of these tumours would never have been disclosedwithout mammography. It is, therefore, a high priority task forfuture research to attempt to find methods to sort out these

clinically irrelevant tumours from those that require adequatetreatment.

Acknowledgements

We thankDr. P. Olsson, whowas one of the promoters of this study.The authors are indebted to the responsible radiologists in the respec-tive screening centres who gave us valuable and necessary informationby kindly answering the questionnaire: Dr. P. Bord�as, Dr. J.-O.Englund, Dr. B. Epstein, Dr. E. Frodis, Dr. H. Hakansson, Dr. M.Kubista, Dr. E.-M. Raupach, Dr. H. Svensson, Dr. P.-A. Svensson,Dr. Z. von Palffej, Dr. E. Thurfjell andDr. K. Yde.

References

1. Shapiro S, Venet W, Strax P, Venet L, Roeser R. Ten- to fourteen-year effect of screening on breast cancer mortality. J Natl Cancer Inst1982;69:349–55.

2. Tab�ar L, Fagerberg CJG, Gad A, Baldetorp L, Holmberg LH,Gr€ontoft O, Ljungquist U, Lundstr€om B, Manson JC, Eklund G, DayNE, Pettersson F. Reduction in mortality from breast cancer aftermass screening with mammography. Randomised trial from the BreastCancer Screening Working Group of the Swedish National Board ofHealth and Welfare. Lancet 1985;1:829–32.

3. Nystr€om L, Rutqvist L, Wall S, Lindgren A, Lindqvist M, Ryd�en S,Andersson I, Bjurstam N, Fagerberg G, Frisell J, Tabr L, Larsson L-E.Breast cancer screening with mammography: overview of Swedishrandomised trials. Lancet 1993;341:973–8.

4. Duffy SW, Tab�ar L, Chen HH, Holmqvist M, Yen MF, Abdsalah S,Epstein B, Frodis E, Ljungberg E, Hedborg-Melander C, Sundbom A,Tholin M, et al. The impact of organized mammography servicescreening on breast carcinoma mortality in seven Swedish counties.Cancer 2002;95:458–69.

5. Hakama M, Pukkala E, Heikkila M, Kallio M. Effectiveness of thepublic health policy for breast cancer screening in Finland: populationbased cohort study. BMJ 1997;314:864–7.

6. Jonsson H, Nystr€om L, T€ornberg S, Lenner P. Service screening withmammography of women aged 50–69 years in Sweden: effects onmortality from breast cancer. J Med Screen 2001;8:152–60.

7. Sakr WA, Haas GP, Cassin BF, Pontes JE, Crissman JD. The fre-quency of carcinoma and intraepithelial neoplasia of the prostate inyoung male patients. J Urol 1993;150:379–85.

8. Zahl PH, Strand BH, Maehlen J. Incidence of breast cancer in Norwayand Sweden during introduction of nationwide screening: prospectivecohort study. BMJ 2004;328:921–4.

9. Peeters PH, Verbeek AL, Straatman H, Holland R, Hendriks JH,Mravunac M, Rothengatter C, Van Dijk-Milatz A, Werre JM. Evalu-ation of overdiagnosis of breast cancer in screening with mammo-graphy: results of the Nijmegen programme. Int J Epidemiol 1989;18:295–9.

10. Shapiro S, Strax P, Venet L. Periodic breast cancer screening inreducing mortality from breast cancer. JAMA 1971;215:1777–85.

11. Tab�ar L, Akerlund E, Gad A. Five-year experience with single-viewmammography randomized controlled screening in Sweden. RecentResults Cancer Res 1984;90:105–13.

TABLE III – INCIDENCE OF BREAST CANCER, RELATIVE RISKS (RR) FOR THE OBSERVED INCIDENCECOMPARED TO THE EXPECTED INCIDENCE ESTIMATED IN THE MODELS AND 95% CONFIDENCE INTERVALS

FOR RR IN THE PREVALENCE ROUND (THE YEAR AFTER SCREENING START) AND THE STABILISEDPHASE (7 YEARS AFTER SCREENING START AND FORWARD) IN THE STUDIED AGE GROUPS

Age Incidence per100,000

RR estimatedin model

95% confidenceinterval (RR)

RR age adjustedfor lead time

95% confidenceinterval (RR)

Prevalence phase (year 1)40–491 152 1.51 1.24–1.84 1.22 0.99–1.5150–59 260 1.84 1.61–2.10 1.54 1.34–1.7860–69 388 1.86 1.65–2.09 1.62 1.43–1.8470–742 532 2.00 1.64–2.45 1.84 1.50–2.24

Stabilised phase (year �7)40–491 134 1.31 1.06–1.63 0.96 0.77–1.2150–59 246 1.69 1.45–1.96 1.54 1.33–1.7960–69 293 1.27 1.10–1.47 1.21 1.04–1.4170–742 306 1.06 0.83–1.34 1.03 0.82–1.30

1The counties Skaraborg, Blekinge, Halland and Kronoberg were excluded because the age group wasnot invited.–2The counties V€astmanland, Skaraborg, Blekinge and Kronoberg were excluded because theage group was not invited.

TABLE II – INCIDENCE OF INVASIVE BREAST CANCER DURING THE 5 YEARS PRECEDING SCREENING START(25, 24, . . . , 21) AND THE YEAR AFTER SCREENING START (PREVALENCE PHASE) IN THE AGE GROUP 50–69

YEARS FOR THE STUDIED COUNTIES

CountyPrescreening Prevalence screening

Relative incidenceNo. of cases Incidence per 100,000 No. of cases Incidence per 100,000

Norrbotten 189 135 94 329 2.4V€asternorrland 259 169 89 302 1.8V€astmanland 254 176 76 271 1.5Skaraborg 224 155 94 329 2.1Halland 226 178 91 351 2.0Blekinge 169 201 46 276 1.4Kalmar 233 168 99 372 2.2Kronoberg 173 187 67 365 2.0J€onk€oping 303 173 94 279 1.6S€odermanland 254 184 97 357 1.9Uppsala 223 188 87 363 1.9

846 JONSSON ET AL.

12. Olsen AH, Jensen A, Njor SH, Villadsen E, Schwartz W, Vejborg I,Lynge E. Breast cancer incidence after the start of mammographyscreening in Denmark. Br J Cancer 2003;88:362–5.

13. Olsson S, Andersson I, Karlberg I, Bjurstam N, Frodis E, Hakansson S.Implementation of service screeningwithmammography in Sweden: frompilot study to nationwide programme. JMed Screen 2000; 7:14–8.

14. Duffy SW, Day NE, Tabar L, Chen HH, Smith TC. Markov models ofbreast tumor progression: some age-specific results. J Natl Cancer InstMonogr 1997;22:93–7.

15. McCullagh P, Nelder JA. Generalized linear models, 2nd ed. London:Chapman and Hall/CRC, 1989.

16. Venables WN, Ripley BD. Modern applied statistics with S, 4th ed.New York: Springer, 2002.

17. Paci E, Warwick J, Falini P, Duffy SW. Overdiagnosis in screening:is the increase in breast cancer incidence rates a cause for concern? JMed Screen 2004;11:23–7.

18. Jonsson H, Larsson LG, Lenner P. Detection of breast cancer withmammography in the first screening round in relation to expectedincidence in different age groups. Acta Oncol 2003;42:22–9.

19. Clemmesen J. Results from statistical research (Symposium). Brit JRadiol 1948;21:853–90.

847INCIDENCE DURING MAMMOGRAPHY SCREENING