German mammography screening program: program sensitivity between 2010 and 2016 estimated based on German health claims data

To the best of our knowledge, this is the first study investigating the feasibility and plausibility of estimating the rate of screen-detected and interval cancers as well as program sensitivity of the German MSP based on health claims data. Using data until 2016 covering 12% of the German population, we estimated a program sensitivity of about 70–72% for the German MSP, which is well in line with program sensitivity determined for North Rhine-Westphalia by linking data from screening units and cancer registry data [3]. While such a data linkage is resource-intensive and poses several other challenges in Germany, claims data are readily available, i.e. the analyses described here are a useful and pragmatic approach to monitoring quality parameters of the German MSP.

While the interval cancer rate and program sensitivity were reported for North Rhine-Westphalia from the years 2005 to year 2011 and for Lower Saxony from the years 2006 to 2011, there are no other estimates of these parameters for Germany [3]. Accordingly, there is no external data source to which we could compare our more recent estimates. Our results do not indicate that there have been major changes in these parameters over time, but it has to be considered that the precision of our estimates was limited, so detecting minor trends would require an even larger sample size. The Cooperative Association of the German MSP does not have information on interval cancer rates nor on program sensitivity but it annually reports other nationwide (quality) parameters such as the rate of screen-detected breast cancers [10‐16]. These rates are slightly higher than the rates of screen-detected cancers observed in our study. A likely explanation for this is that we used rather loose criteria to classify breast cancers as prevalent cases. For example, one outpatient code of breast cancer in the pre-observation period already led to classification as a prevalent case (i.e. this case could no longer become an incident case) and miscoding in the outpatient setting is not very rare. In other words, we may partly have misclassified cancers detected at screening or during follow-up as prevalent cancer and thus may have underestimated the absolute number of screen-detected and interval cancers. As this potential misclassification concerns both screen-detected and interval cancers, we do not think that it has a relevant impact on program sensitivity, which is supported by the good agreement between program sensitivity observed in our study and external data sources.

According to the European guidelines for quality assurance in breast cancer screening and diagnosis, interval cancer rates should not exceed 30% of the background incidence rate in the target population in the months 0–11 after the screening mammography and they should not exceed 50% in the months 12–23 after the screening mammography. The Cooperative Association of the German MSP estimated the background incidence rates of breast cancer using data from the pre-introduction period of the MSP (1999–2005) and cancer registries [17]. For the years 2010 to 2016, the nationwide background incidence was estimated to be 2.6–2.8 cases per 1,000 women. In 2016, for example, the nationwide background incidence was 2.5714 per 1,000 and the rate of IC in 2016 was 0.6956 per 1,000 for the first year in our study, yielding a proportion of 27.05%, i.e. it was below the threshold of 30%; the same applies to the years 2010–2015. Also in the second year after screening the proportion was either below or (with a maximum of 52%) very close to the threshold of 50% (see Supplemental Table 1). However, it has to be taken into account that background incidence rates show regional variation (e.g. 2.98 per 1,000 in Western Germany and 2.18 per 1,000 in Eastern Germany) and should thus ideally be applied on a regional level.

Comparison of our approach to algorithms developed for databases in other countries is difficult given that the type of data, the health care and reimbursement systems (affecting the type of information available) as well as the characteristics of the MSPs differ between countries. Fenton et al. developed and validated an algorithm focusing on the identification of screen-detected breast cancers in Medicare data, so the age range of women differed (≥ 68 as opposed to 50 to 69 in our study) [18]. Furthermore, the algorithm considered only diagnoses and treatments occurring up to one year after a mammogram, which also hampers comparability.

In the interpretation of our study, the following should be taken into account. First, our algorithm identifies screen-detected cases by the occurrence of a pre-operative case conference in combination with an incident diagnosis of breast cancer as defined based on claims data. Prior analyses linking the data from screening units with cancer registries had information on pre-operative histological findings and post-operative cancer verification, so the type of information is not identical. While case conferences are supposed to be held as the last step of diagnostic verification, we cannot rule out that for some MSP participants no case conference is held, which would result in an underestimation of screen-detected cancers. Second, when trying to distinguish incident and prevalent cases in claims data, one has to make compromises [19]. In our study, we prioritized making sure that a case is incident and therefore may have excluded a too large number of prevalent cases. As mentioned before, this may have slightly underestimated the absolute number of screen-detected and interval cancers (and thus also the respective rates) but we do not expect that it had an impact on program sensitivity. Also, the definition of incident breast cancers could be done based on a more sophisticated algorithm (i.e. searching for inpatient diagnosis later during follow-up), but our results do not suggest that misclassification was a major issue. Third, as the underlying populations of different statutory health insurance providers in Germany partly differ e.g. with respect to socioeconomic characteristics and as we did not consider patients with private health insurance (about 10% of the German population), caution is needed when extrapolating the rates observed in our study to the whole of Germany. Our data cover between 12.4 and 14.2% of all German non-initial screenings between 2010 and 2016 [10‐16]. To account for differences in the age distribution and facilitate comparison with Bokhof et al. we used age standardization with the old standard population of Europe. The age distribution of this standard population is rather different from the current age distribution in Germany, which has to be considered when interpreting absolute rates. Fourth, incidence rates vary between initial and non-initial MSP participants [20], so it is reasonable to distinguish between both types of participants. Given that identification of initial MSP participants in claims data is subject to uncertainty due to left truncation, we decided to focus on non-initial MSP participations, i.e. women who already had an MSP screening mammography during the pre-observation period. Finally, it should be taken into account that a transfer of individual information on interval cancers from claims data to the screening units is not possible due to data protection regulations. This also means that an evaluation to distinguish interval cancers into overseen breast cancers and de novo breast cancers cannot be done based on our approach.