|Home | About | Journals | Submit | Contact Us | Français|
Several reviews have estimated the balance of benefits and harms of mammographic screening in the general population. The balance may, however, differ between individuals with and without family history. Therefore, our aim is to assess the cumulative risk of screening outcomes; screen-detected breast cancer, interval cancer, and false-positive results, in women screenees aged 50–75 and 40–75, with and without a first-degree relative with a history of breast cancer at the start of screening. Data on screening attendance, recall and breast cancer detection were collected for each woman living in Nijmegen (the Netherlands) since 1975. We used a discrete time survival model to calculate the cumulative probability of each major screening outcome over 19 screening rounds. Women with a family history of breast cancer had a higher risk of all screening outcomes. For women screened from age 50–75, the cumulative risk of screen-detected breast cancer, interval cancer and false-positive results were 9.0%, 4.4% and 11.1% for women with a family history and 6.3%, 2.7% and 7.3% for women without a family history, respectively. The results for women 40–75 followed the same pattern for women screened 50–75 for cancer outcomes, but were almost doubled for false-positive results. To conclude, women with a first-degree relative with a history of breast cancer are more likely to experience benefits and harms of screening than women without a family history. To complete the balance and provide risk-based screening recommendations, the breast cancer mortality reduction and overdiagnosis should be estimated for family history subgroups.
Early detection of breast cancer by mammographic screening reduces breast cancer mortality1, 2. However, mammographic screening has a number of harms including overdiagnosis, false-positive recall and false-negative results. Therefore, population-based mammographic screening programmes are only justified when the benefits outweigh the harms.
Currently, all population-based mammographic screening programmes have a one-size-fits-all regimen; age is the only risk factor used to select the target population and all women within the targeted age range are screened with the same modality and frequency. There is continuing debate about whether the benefits outweigh the harms in the current one-size-fits-all screening programmes for breast cancer2–5. Meanwhile, the focus is shifting from this one-size-fits-all approach towards personalized breast cancer screening6. The aim of personalized breast cancer screening is to improve the balance of benefits and harms for each woman by tailoring her screening regimen based on breast cancer risk. Key to the potential effectiveness of personalized screening is that the balance between the benefits and harms differs between subgroups of women with differing breast cancer risk.
Until now, the effect of risk-based screening on the balance between the benefits and harms of breast cancer screening has only been estimated using modeling studies which showed favourable outcomes for risk-based screening7–9. Modeling studies rely on data and assumptions that may be incorrect9, and therefore, these findings should be supported by observational studies. Although several large cohort studies have been initiated to predict the benefits and harms of screening in individual women, results of these studies are still limited and are mainly focused on breast cancer risk prediction.
The aim of this study is to assess the cumulative risk of both favourable (screen-detected breast cancer) and unfavourable (interval cancer and false-positive results) screening outcomes in women screened from aged 50–75 and 50–69, with and without a first-degree relative with a history of breast cancer at the start of screening. Because women aged 40 to 49 years with a two-fold increased risk of breast cancer, i.e. those with a first-degree relative with breast cancer7, 10, have a similar benefit-harm ratio as average-risk women aged 50 to 75 years8, we also assessed the cumulative risk of screening outcomes in women invitees aged 40–75 and 40–69 with and without a first-degree relative with a history of breast cancer at the start of screening.
In 1975, a biennial mammographic screening programme was implemented in Nijmegen, the Netherlands, inviting women aged 35 years and older (born before 1939). In 1989, a national biennial screening programme for women aged 50–69 years was implemented throughout the Netherlands. Subsequently, the Nijmegen screening programme invited the same target group as the national programme, and women aged 70 and older could undergo screening only if they made an appointment for the examination themselves11. In 1997, the upper age limit of both the national and Nijmegen mammographic screening programme was raised to 74 years.
Until 2013, the screening examination consisted of two views, a mediolateral oblique and craniocaudal view, in initial screens and a single view in subsequent screens. Initially, the subsequent screens consisted of a lateral view, which was replaced by a mediolateral oblique from the fourth round onwards. An additional craniocaudal view was taken during the screening examination if abnormalities were suspected or if the quality was not good enough for evaluation11. From 2013 onwards, subsequent screening examinations standardly consisted of a mediolateral oblique and a craniocaudal view. Mammograms are read by two independent radiologists, who must reach consensus on recall12.
All women consented to the use of their anonymous data for scientific research from 1990 onwards. Because there was no law on the registration of personal data before 1989, women participating in the screening programme during the first eight rounds (1975–1990) were not asked to give informed consent. After the introduction of the law, women were asked to give explicit permission for the use of their personal data to evaluate the mammographic screening programme (1990–2004). From 2004 onwards, women received a leaflet with their invitation stating that their personal data is used to improve the screening programme. In case women do not agree, they can object by signing a form, i.e. an opt-out procedure.
During the first round, women invited to participate in the breast cancer screening programme were asked to fill in a questionnaire before undergoing the screening examination. In the questionnaire, participants were asked about, among other things, the history of breast cancer in their family (grandmothers, mother, aunt(s), sister(s), and daughter(s)). We defined a family history of breast cancer as having at least one first-degree relative with a history of breast cancer, i.e. a history of breast cancer in mother, sister(s) and/or daughter(s). Recall was defined as the recall of a woman by screening radiologists for further diagnostic work-up to assess whether the suspicious finding on the screening mammogram was malignant. If histological and/or cytological diagnostic work-up confirmed the presence of breast cancer within 6 months following recall, the woman had a so-called ‘screen-detected breast cancer’, i.e. true-positive. The nature of the screen-detected breast cancer could be invasive or non-invasive, i.e. ductal carcinoma in situ (DCIS). Invasive screen-detected breast cancers were further subdivided in tumours <15 mm and ≥ 15 mm based on tumour diameter at pathological examination. In a few cases we used tumour diameter at the diagnostic mammogram because it was not reported at pathological examination (e.g. because chemotherapy preceded surgery). A recall was considered to be false-positive when diagnostic work-up did not confirm the presence of breast cancer during the first year after screening, even if breast cancer was subsequently detected before the next screening examination. Invasive diagnostic work-up included the procedures fine needle aspiration cytology, core needle biopsy and surgical biopsy. We defined interval cancer (false-negative results) as a breast cancer that was not detected at screening but was diagnosed before the next screening examination.
The cumulative risks of screen-detected breast cancer, invasive screen-detected breast cancer stratified by size, interval cancer, false-positives and false-positives with invasive work-up were estimated by discrete-time survival models. Separate models were used to estimate the risks for each screening outcome. Women were censored after the event of interest, a competing event (see Table 1), loss to follow-up or death – whichever came first.
The discrete-time survival model was fitted using the LOGISTIC procedure in SAS statistical software, version 9.2 (SAS Institute Inc, Cary, NC)13, 14. All logistic regression models included number of prior screening rounds attended, family history and age. Prior number of screening rounds attended served as the discrete time-scale and ranged from 1 to 19, i.e. covering 36 years of follow-up. Family history was classified into three groups; a first-degree family history, no family history and an unknown family history. Age was grouped in two-year age groups between 40 and 75, i.e. 40–41, 42–43 up till 74–75. Women below 40 were grouped together and above 75 were also grouped together, because of small numbers.
Hubbard et al15 found that women participating in more screening rounds had a different risk of false-positives than women participating in fewer screening rounds, i.e. dependent censoring. Because the number of screening rounds appeared to depend on the result of previous screening rounds for both cancer and false-positives, we could not test formally whether dependent censoring occurred in our population. Because previous studies have shown dependent censoring in the risk of false-positives, we calculated the risk of false-positives and false-positives with invasive work-up under the assumption of independent censoring (standard model) and adjusted for dependent censoring. The models adjusted for dependent censoring included five categories for the proportion of screening rounds attended (≤0.25, >0.25–≤0.5, >0.5–≤0.75, >0.75–<1.0, 1.0) in addition to the standard model. Because there is no evidence yet that cumulative risks of cancer are affected by dependent censoring, we did not adjust the cumulative risk of screen-detected breast cancer and interval cancer.
Additionally, we adjusted the risk of each screening outcome for competing events (see Table 1). Without adjustment for competing events, the models estimate the latent risk of an event in the absence of competing events, i.e. risk of an event had a woman continued screening. This is not relevant for women who stop screening because of breast cancer diagnosis or death. Therefore, we adjusted the risk of an event in each round for competing events by multiplying the risk by the proportion of women who had no competing event16.
Elveback’s formula17 was used to calculate the cumulative risk of each screening outcome. In this formula, the risk of the event of interest for each screening round is multiplied by the proportion of women without the event of interest until that screening round. The 95% confidence intervals (CI) were obtained by taking the 2.5 and 97.5 percentiles of 10,000 samples from the multivariate sampling distribution of all model parameter estimates.
During the first screening round, 19,703 of the 23,210 women invited to participate underwent a screening examination (84.9%) and filled in a questionnaire including questions about family history of breast cancer. Of the women who filled in the questionnaire, 90.8% (n=17,882) had no family history of breast cancer, 5.6% (n=1,101) had one or more first-degree relatives with a history of breast cancer and 3.7% (n=720) did not know whether her mother, sister(s) and daughter(s) had a history of breast cancer. Because women can quit screening after one or more rounds, the number of women in follow-up decreased with an increasing number of screening rounds (Table 2).
Tables 3 and and44 show the cumulative risk of screen-detected breast cancer, interval cancer and false-positives for women with and without a family history in women screened from age 50 to 75 and age 40 to 75, respectively. Estimates for screening from age 50–69 and 40–69 are provided in Supplement 1. The tables 3 and and44 show that women with a family history had a higher cumulative risk of a screen-detected breast cancer, interval cancer and a false-positive than women without or with unknown family history. Women screened from 50 to 75 with a family history of breast cancer had a 9.0% (95% CI, 6.7–12.1) risk of screen-detected breast cancer, 4.4% (95% CI, 3.1–6.9) risk of interval cancer, and 7.6% (95% CI 6.2–10.7) risk of a false-positive. Women screened from age 40 to 75 with a family history had a 9.1% (95% CI, 7.0–12.7), 5.0% (95% CI, 3.6–8.5) and 13.1% (95% CI, 10.2–17.5) risk of screen-detected breast cancer, interval cancer and false-positives, respectively. For women without a family history screened from age 50–75 and 40–75, the cumulative risks were 6.3% (95% CI, 5.8–7.3) and 6.4 (95% CI, 5.8–7.7) for screen-detected breast cancer, 2.7% (95% CI, 2.4–3.4) and 3.1% (95% CI, 2.8–4.3) for interval cancer, 5.0% (95% CI, 4.8–6.2) and 8.8% (95% CI, 8.1–10.3) for false-positives, respectively.
Overall, women with a family history had an increased risk of 1.4 (95% CI, 1.1–1.9), 1.7 (95% CI, 1.1–2.4) and 1.5 (95% CI, 1.1–1.9) times the risk of women without a family history for screen-detected breast cancer, interval cancer and false-positives, respectively. Women with a family history of breast cancer also had higher cumulative risks of an invasive breast cancer, invasive breast cancer smaller than 15 mm, and false-positives with invasive work-up than women without a family history of breast cancer.
Comparison of the cumulative risk in women screened from 50–75 and 40–75 shows that inclusion of five more biennial screening examinations in the age range 40–49 especially increases the cumulative risk of interval cancers and false-positives. The cumulative risk of screen-detected breast cancers increases by only 0.1% in both family history groups when adding five extra screens while false-positives increase by 3.8–5.5% in the model assuming no dependent censoring and 8.1–11.6% in the model adjusted for dependent censoring.
This is the first study estimating the cumulative risk of screen-detected breast cancer, false-positives and interval cancers over the course of a screening programme for women with and without a first-degree relative with a history of breast cancer. We found that risks of both favourable and unfavourable screening outcomes were larger for women with than for women without a first-degree relative with a history of breast cancer. This gives some first insights into the relative balance of benefits and harms of screening for women with and without a family history.
It is well-known that women with a first-degree family history of breast cancer have a higher risk of breast cancer than women without a family history18, 19. However, the relative and absolute effect of mammographic screening on breast cancer mortality for women with a family history compared to those without remains largely uninvestigated. We found that women with a family history have a higher cumulative risk of screen-detected breast cancer, but also a higher cumulative risk of small invasive screen-detected breast cancer. Otten et al showed that women with invasive breast cancer smaller than 15 mm detected at screening have the same life expectancy as women in the target population invited to screening, i.e. women without breast cancer20. Therefore, the detection of small invasive breast cancers is regarded as favourable. However, detection of these favourable breast cancers cannot directly be translated to a breast cancer mortality reduction and some small invasive breast cancers detected by screening may represent overdiagnosed cancers. To our knowledge, no studies have been undertaken investigating breast cancer mortality reduction or overdiagnosis by presence or absence of family history. Only modelling studies have investigated the benefits and harms of risk-based screening, including a first-degree family history of breast cancer, and concluded that risk based screening could reduce harms and costs7, 9.
This study also indicates that women with a family history have a greater risk of a false-positive and a false-positive with invasive work-up compared to women with no family history. Most other studies have also found higher risks of false-positive recalls16, 21–23 or false-positives with invasive work-up22, 24 for women with a family history. However, the effect of family history on a false-positive result is larger in this study than in other studies from the US. A possible explanation might be that radiologists had access to the family history of screened women from 1991 until 2003 and recalled women with a family history more readily than women without a family history. However, in the US, radiologists also have often access to family history. Therefore, a more likely explanation is that the high absolute risk of false-positives in the US minimizes the differences between family history groups leading to the observed difference between the US and the Netherlands.
False positive results, in particular those requiring invasive follow-up procedures25, 26, have a negative impact on psychological state25–27 and screening re-attendance27, 28. According to a review29, the psychological impact of false-positives for women with a family history is similar to that in the general population, although some studies show a greater negative impact in women with a family history30. In the Netherlands28 and other European countries27, false-positive results negatively affect re-attendance. It is, however, unknown whether this effect differs between women with and without a family history of breast cancer.
We used different models to estimate the cumulative risk of false-positives and false-positives with invasive work-up: unadjusted and adjusted for dependent censoring. Dependent censoring occurs when women who quit screening have a different risk of a screening outcome than women who continue screening16. Previous research indicated that this is likely for false-positives and provided some solutions for statistical analyses31. However, these solutions are not appropriate when women quit because of their false-positive, which is the case in the Netherlands28. Therefore, we adjusted for dependent censoring by marginalizing the cumulative risk over the attendance probability. Because half of the cumulative risk of false-positives was adjusted for dependent censoring and the cumulative risk of screen-detected breast cancer was not, we did not estimate the cumulative risk of recall. However, the results indicate that recall is likely to be higher in women with than without a family history, because the risk of screen-detected breast cancer and false-positives were both higher in women with a family history.
The results of our study cannot be generalized directly to the current era of digital mammography, to different screening intervals and/or to other countries such as the US. Our results are based on women who started screening in 1975. Over time, recall rates changed32, and it is expected that the cumulative risk of screen-detected breast cancer and false-positives, with and without invasive work-up, are higher and the cumulative risk of interval cancers is lower for women who recently started screening31. However, we expect that the relative risk difference between women with and without family history will remain more or less the same. The cumulative risk of screening outcomes should also not directly be translated to other screening intervals. For example, the small number of extra screen-detected breast cancers found when starting at age 40 instead of 50 are likely the result of a 2-year screening interval. A 1-year screening interval in the age group 40–50 would probably yield a higher number of extra screen-detected breast cancers33. Furthermore, our results should only be generalized to countries with similar recall rates, including Europe and Australia14.
Furthermore, our analysis has focused on women with and without a first-degree family history at the start of screening, which has several implications for the study results. Firstly, family history of breast cancer was self-reported, which may cause misclassification. Previous research has shown that self-reported family history of breast cancer is accurate34 and therefore it is unlikely that this affected our results. Secondly, we did not exclude women with a strong family history who are screened outside the mammographic screening programme nowadays, i.e. women with BRCA mutations and women below 60 with a relative risk of breast cancer of 3 or more35. The effect of this inclusion on the cumulative risk of both screen-detected and interval cancer is expected to be small, because BRCA mutations are rare, even among women with a first-degree family history36, and women with a relative risk of 3 or more are invited to participate in the national mammographic screening programme after the age of 60. Thirdly, the presented cumulative risks are based on assessment of first-degree family history at screening initiation and therefore the cumulative risk are only valid for women who are making a decision about starting screening. Fourthly, the subgroups of women with a first-degree and with unknown family history were small, which may have resulted in wide confidence intervals and some unexpected values (i.e. small risk of interval cancer in women with unknown family history and small difference in screen-detected cancer between women screened from age 40–75 and age 50–75 in all family history subgroups). However, in general, we believe that the point estimates are informative despite the broad confidence intervals. Finally, we did not investigate the effect of possible confounders, such as breast density, on the cumulative risks, which might be able to explain some part of the differences found between women with and without family history.
Our study has two major strengths: the use of a population-based approach and many years of follow-up. All women who were invited to participate in the mammographic screening programme and attended the first screening exam were included in this study, thereby minimizing selection bias. Furthermore, the Nijmegen screening database followed women for more than 35 years and has an accurate registration of attendance and screening outcomes. As a result, we could calculate the cumulative risk of screen-detected breast cancer, interval cancer and false-positives over the entire period of screening eligibility without extrapolation.
To conclude, our study shows that women with at least one first-degree relative with a history of breast cancer have a higher risk of favourable (small screen-detected breast cancers) and unfavourable (interval cancers and false-positives) screening outcomes than women without such family history. Before giving risk-based screening recommendations to women with and without family history, further research should complete the benefit-harm balance, i.e. by estimating the breast cancer mortality reduction and overdiagnosis in family history subgroups, and weight the benefits and harms.
Research and practice are shifting from a ‘one-size-fits-all’ approach to personalized medicine. Key to personalized breast cancer screening is that the balance between screening benefits and harms differs between subgroups. This article is the first observational study showing that screening benefits and harms are higher for women with a first-degree family history than for women without such family history, screened from age 40–75 and 50–75, a first step towards personalized breast cancer screening.
RAH was supported in part by a grant from the National Cancer Institute of the National Institutes of Health under award number R03CA182986. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
The authors declare that they have no competing interests