Amenazas

The results represented in this thesis were obtained from the organised Finnish National Breast Cancer Screening Programme (FNBCSP), which covers nearly 100% of the target age population (50-69 years). The unique personal identifier allows accurate linkage of data available at the comprehensive national registers. The collection of information on screening invitations, mammography tests, further assessments by the mass screening registry, and cancer outcomes for the whole population by the cancer registry provides several major advantages to conducting research in the Finnish programme. The collection of information on breast symptoms reported during the mammography screening visits allows for studying the impact of symptoms not yet available or published in any other programme.

No prior studies have estimated the risk of breast cancer, false-positive test findings and cancer mortality in relation to breast symptoms in a prospective manner. Follow-ups were done for a maximum of 24 years and a maximum of 10 screening visits after screening visits with or without symptoms. The study used screening history and symptom history information of all screening-age women in Finland who had biennial screening mammography, and followed up for further assessments until cancer diagnosis and death. The main benefit of using data on breast symptoms reported within the screening programme is that one can study cancer incidence and deaths from breast cancer in symptomatic women as compared to non-symptomatic. Additionally, harms in terms of false-positive findings were assessed for the first time in this study.

One key strength of this study is that follow-ups were performed until the date of the last screening visit or the diagnosis of cancer or death, whichever came earliest. The individual-level data allowed us to extract screening history information of every woman who presented with symptoms or who did not have symptoms. This is unique in the Finnish programme.

In studies III and IV, we made the symptomatic and asymptomatic group more comparable by matching with the background variables at any given period of the screening visits. This minimized the bias in the risk estimates by the confounders. As we used the full screening cohort, the study power was sufficient to compute the difference in risk and rates for symptomatic and asymptomatic visits. Because not all symptoms are equally sensitive, each symptom was analysed independently thus it was possible to analyse the risk of cancer and mortality for every individual symptom. It is important to ensure that women with symptoms have equal opportunity to participate in the screening programme as those without symptoms. This means that

the screening effectiveness is to be evaluated based on all eligible women invited, regardless of any personal or risk history. Such evaluation reflects the actual scenario in the existing screening-age population. However, this study also highlights the importance of differential follow-up assessment of women who have a differential risk.

The definitions of the programme process and outcome indicators used in these studies adhere to the WHO-IARC and EU guidelines, as well as other international guidelines on breast cancer screening and diagnosis (Perry et al., 2006; IARC, 2016). Thus, future studies conducted in other screening programmes could utilize our study methods and findings. Furthermore, the findings of our studies could be applied to other existing and new programmes to improve the performance and effectiveness of screening mammography.

The studies contained several limitations.

One of the potential limitations is the collection of symptoms information. The collection was based on the women’s self-reporting and by radiographers at the screening visits. The radiographers’ physical inspection of the breast is likely to be less comprehensive than a full clinical examination. Information on symptoms is mainly collected in order to support the interpretation of the mammograms. However, in most of the cases, the presence of symptoms is confirmed by having the radiographer examine the breasts before the mammography is performed. This supports that the collected symptom information is valid, albeit not perfect.

Women with symptoms were possibly more likely to attend mammography screening than asymptomatic women leading to a self-selection bias. However, about 84% of all invited women participate in mammography screening in the Finnish programme, which is the highest among any existing mammography screening programme. In addition, most of the women who attended were asymptomatic (about 97%), and thus the attendance bias caused by symptoms is likely to be small. Another issue is the recall bias. We do not know about any possible delay in presentation of symptoms: whether women waited for their first invitation (i.e., at age 50 years) or a subsequent screening invitation. The symptom information is based on the women’s reporting of symptoms in the past two to six months and the examination by the radiographer at the screening visit. Women were more likely to remember any recent abnormalities in their breasts, and thus the recall bias is not likely to be of higher importance.

An important limitation is the inability to address the potential confounding effect of important risk factors that do not exist in the database. Thus, effect adjustment by factors such as family history, breast density, hormone use and socio-

78

economic status, which are known to influence breast cancer risk, was not possible in this study (Anderson et al., 2014; Barnard et al., 2015; Collaborative Group on Hormonal Factors in Breast Cancer, 2001; IARC, 2012a; IARC, 2012b).

The estimate of sensitivity, specificity and positive predictive value (PPV) of diagnosing breast cancer cases with symptoms in study I was limited to those women who attended mammography screening. Cancer cases detected outside of screening were excluded. The low sensitivity of symptoms in this study indicate that a mammography screening programme is still justified.

Study III was sensitive to lead-time bias and overdiagnosis because we used detection method instead of the use of background incidence of breast cancer to estimate the interval cancer rate in the absence of screening. However, it was not possible to find a comparable non-screened group as the Finnish screening programme has a high coverage (almost 100%) and attendance rate (about 84%). As well, there was no possibility to estimate the background incidence of breast cancer in women with symptoms. In addition, the positive predictive value (PPV) of mammography did not differ between the symptomatic and asymptomatic groups between the first and subsequent screening round, and thus the lead time bias due to prevalent screenings is negligible. Furthermore, our estimates of incidence rates and hazard ratios were based on the analysis of invasive breast cancers and on advanced and fatal breast cancers, both of which are less affected by overdiagnosis. Nonetheless, the proportion of in situ carcinomas was only 5% in those with symptoms.

In study IV, we analysed the difference in socio-economic statuses between symptomatic and asymptomatic women who were diagnosed or died of cancer. In women who reported a lump, we observed extra all-cause death cases in lower socio- economic class than in upper classes and also an extra number of deaths as compared to those without a lump. However, the socio-economic status of the whole cohort was unknown; thus, the proportion of breast cancer deaths or all-cause deaths in those with or without symptoms might differ from the study estimates. The higher all-cause mortality rates than breast cancer mortality rates, and also the difference in the rates between symptomatic and asymptomatic group throughout the follow-up period might be explained partially by the difference in socio- economic status.