For the aforementioned approaches, the clinical cancer registries are a source of supplementary data on treatment adherence and therapeutic care. In Main Study I, the Association of German Tumour Centres (ADT) was tasked with designing and implementing a corresponding procedure.

In the approach based on health insurance funds, the two subcontractors (the Leibniz Institute for Prevention Research and Epidemiology – BIPS GmbH and the SOCIUM Research Centre on Inequality and Social Policy of the University of Bremen) will analyse billing data from a total of six funds, some of which operate on a nationwide basis. To this end, the health insurance fund data must be comprehensively processed and supplemented with the causes of death. For each fund, this requires authorisation from the relevant authority and the fund itself on the basis of a data protection concept and a study protocol. These documents and the technical implementation were finalised and the authorisation processes were initiated in Main Study I. For some of the health insurance funds, work on securing the data has already begun.

In the population-based approach, data from the state cancer registry of North Rhine-Westphalia (LKR NRW) and the local mammography screening units will be supplemented with data from the two associations of statutory health insurance physicians of North Rhine and Westphalia-Lippe. As the data is amalgamated, maximum security will be ensured using the “blind anonymisation” method, whereby the data is not present in uncoded form at any point as it moves from the data owner to the evaluating body at the BfS. In Main Study I, the technical IT concept was drawn up for the evaluating body, and the software components were completed, implemented and tested. For the evaluations, work was carried out on the methodological further development of the concept of correction factor-based analysis.

In this phase of the study, the fund-based approach succeeded in making most of the raw data from the total of six health insurance funds involved in the project available and in incorporating this data and checking it for consistency. Here, in the case of the health insurance fund collaborating with SOCIUM, the database was fully incorporated for the period from 2006 to 2018 (inclusive). For the health insurance funds collaborating with BIPS, most of the database was also fed in from the year 2004 onwards. Initial descriptive analyses show 39 % coverage of the female population aged 50 to 69 as a Germany-wide average.

Where possible, the causes of death are to be imported directly from the epidemiological cancer registries. This direct comparison of causes of death was delayed at both BIPS and SOCIUM for various reasons beyond the research contractor’s control and could ultimately only be successfully completed between LKR NRW and the health insurance fund collaborating with SOCIUM. Nevertheless, it was shown that the causes of death could be successfully compared with the designed procedure under the given technical circumstances.

Where it is not possible to compare them with the cancer registries, the causes of death are to be determined from the information stored in the health insurance fund data using a Gagnon algorithm that is adapted to the specific stock of data of each individual health insurance fund. The corresponding programming instructions have been adapted. This process was successfully carried out up to the data year 2018 (inclusive) for the health insurance fund collaborating with SOCIUM and up to 2015 for the health insurance funds collaborating with BIPS. In addition, the potential of various other methodological approaches (random forest methods, inspection of insured persons’ profiles) was also systematically investigated in order to optimise the algorithm. The current form of the algorithm was validated using data from the direct comparison and exhibits a high level of sensitivity and specificity.

The study design follows an approach involving emulated target trials. On the one hand, this allows adequate consideration of the distortion effects (including with regard to time-dependent factors) that are possible in observational studies – as opposed to in randomised controlled trials (RCTs). On the other hand, it also allows efficient use of the available data for the analyses. To this end, a study protocol was prepared and submitted to a journal for publication. Within the framework of the strategy of converging evidence, the analyses specified in the study protocol form a key part of the implementation and analysis concept.

This approach suffered from a considerable time lag due to delays in development and implementation, a lack of preselection of duplicates in the provision of data from the two associations of statutory health insurance physicians, and challenges in terms of record linkage. The time lag meant that only a partial dataset – consisting of data from the screening units and LKR NRW as well as the North Rhine association of statutory health insurance physicians – could be made available at the evaluating body.

However, the transfer to the evaluating body, the issuing of this partial dataset to the general contractor, and the consistency and plausibility checks carried out by the general contractor allowed researchers to demonstrate that the designed data flow concept can produce valid data within the framework of productive operation and is therefore suitable for further use in the evaluation of mortality. Moreover, the retrieved “best of” partial dataset allowed empirical determination of how the selected parameters for anonymisation influence suppression rates. This allows an initial estimation to be made of the actual suppression rates in the final evaluation dataset, which is still pending. It was shown that the anonymity threshold, k, needed to ensure the anonymity of data leads to low suppression rates with acceptable simplification of characteristics.

To compensate for the delays, work was carried out on correcting the structural differences between women from later phases of the project who were and were not participating in the MSP. On the one hand, this work included further analyses in relation to the correction factor for adapting the classical correction factor approach. On the other hand, further approaches to correcting the distortions in this approach were also discussed and analysed, taking account of current research findings.

In this phase of the study, LKR NRW also worked with the University of Münster to determine the incidence rates of advanced stages of breast cancer in accordance with EU directives. With a relative decrease among regular participants in the mammography screening programme of 16.5% for the first 24-month period, or 21.3% following the first follow-up screening examination, the results of the study indicated a reduction in the incidence of advanced stages of breast cancer.

In addition, a concept was developed for the cross-model data amalgamation body that provides for record linkage of the minimal datasets from the fund-based and population-based approach in order to enrich the population-based approach with individual information on distorting factors. Despite reduced temporal resolution in this dataset, it was therefore possible to carry out the desired evaluation within the framework of the emulated target trial design in accordance with the methodology of the fund-based approach.

The approach using clinical cancer registry data examined the possibility of retrospective amalgamation and enrichment with information from screening units. In the process, it was found that routine data comparison is so far only established in North Rhine-Westphalia and Lower Saxony and is yet to be implemented in other federal states. Nevertheless, the availability of clinical cancer registry data for the relevant study period is limited to some regions of Bavaria and Baden-Württemberg as well as Brandenburg, Mecklenburg‐Vorpommern, Saxony, Saxony‐Anhalt and Thuringia.

After a technical concept was developed for the data’s amalgamation, pilot testing was planned in the model region of Mecklenburg‐Vorpommern before the procedure’s nationwide rollout in the event of a positive evaluation. Following an initial negative evaluation by the local state representative for data protection, the University of Münster and ADT adapted the jointly produced research concept to the model region of Saxony-Anhalt and set it out in specific terms.

Despite considerable efforts, it was not possible to obtain the necessary authorisations – or, therefore, to carry out the pilot study – by the end of this phase of the study. As these authorisations are unlikely to be issued in good time for Main Study II either, this approach is not being pursued further. However, the successfully completed test data flow did at least demonstrate a possible transfer of the research concept into practice and the fundamental suitability of the procedure for joining up the information.