The aim of WP1 in the PanCareSurFup project was to amalgamate data of survivors after childhood cancer from European cancer registries and other databases which were available for the three outcomes relevant to PanCare-SurFup (cardiac events, second cancer, late mortality). Based on this, clinical epidemiological studies were carried out on a selected set of serious late effects.
Through the cooperation of 16 project partners and 13 DPs from 12 countries, the project succeeded in generating the largest cohort of children with cancer in Europe to date. The resulting cohort of 83,333 5-year survivors is unique due to its size and the collection of a selected set of late effects. Additionally 32,263 non-five year survivors were collected, resulting in a cohort of 115,596 individuals. It provides an excellent opportunity to compare each decade since the 1940 with respect to childhood cancer and allows for a good comparison of survival rates.
Compared to other population-based European data collections, like ACCIS [1, 20, 21], some diagnoses differ in numbers, but the overall distribution in the PCSF cohort corresponds with the ACCIS data. While considering that variety of diagnoses in different countries is not uncommon to a certain extent [22] only few deviations can be seen in Table 6, primarily caused by the two further classification groups we implemented (“other further classifiable” like Langerhans cell-histiocytosis and “unclassifiable” with respect to ICCC-3). Furthermore, we have to take into account that we cannot entirely compare those two resources as ACCIS collects data since diagnosis, and our cohort is based on 5-year survivors, i.e. starts 5 years after diagnosis. Diagnoses with poorer survival (e.g. CNS tumours) were underrepresented compared to incidence data at time of diagnosis. Additionally, due to the fact that France delivered a cohort without leukaemia patients, this group contributes a little bit less than about a third to the data. Further on we seem to have a slight underreporting regarding tumours of the central nervous system (CNS), which is a known phenomenon as this diagnostic group with its different histology and behaviour is heterogeneously collected in cancer registries [23]. Neuroblastoma are somewhat less and lymphomas are somewhat more frequent compared to ACCIS. Regarding quality indicators, almost all of the data sources included in PanCareSurFup contributed as well to ACCIS, where no substantial difference between quality indicators was seen for the different data providers [24].
The assembled PanCareSurFup cohort is characterised by inclusion of all malignant diseases occurring from 0 to 20 years of age, with the exceptions previously mentioned. It should particularly be pointed out that the three outcomes relevant to PanCareSurFup are being investigated in approximately the same basic population. While cancer registries routinely collect mortality and second cancer incidence, other outcomes, such as cardiac disease, is not routine. In PCSF a small number of DPs were able to collect cardiac morbidity.
The project includes all DPs which were identified by a preceding survey and fulfil relevant requirements (e.g. good quality of follow-up, availability of relevant information, legal and organisational prerequisites). Thus, data are often collected through a population-based cancer registry, through a body with close connection to a population-based cancer registry, or within a clinical registry. In the future, statements largely representative of the population will be possible based on these analyses. Some countries that would have participated could not provide data for a variety of reasons. First, in some countries information on these outcomes was not centrally available; in other countries retrieving therapy data from clinical sources was not possible, and finally some potential DPs were uncertain that the data could be provided within the project period. The Nordic countries could not provide cardiac events due to the ongoing parallel Nordic study ALiCCS [25].
The cohort is based on data sets which were collected in very different contexts. For example, the Nordic countries had already established population-based cancer registries with high data quality and high completeness in the middle of the last century. However they lack precise information on treatment. Other countries, e.g., in Eastern Europe, also have long-standing data collections, not previously contributed to bigger projects. The persons responsible had very diverse backgrounds (epidemiologists, clinicians, registry experts) with different technical equipment and experience. DPs who were less experienced in delivering data to huge consortia received assistance from WP1 to deliver data which met the characteristics of the PCSF baseline variable list. Additionally, differences in background and level of experience were ironed out through regular meetings and bi-weekly conference calls. The use of a common data structure reduced differences between data sources.
The homogeneity of the PCSF cohort data was ensured by the following procedures: The creation of a common baseline variable list, standardised data flow and uniform data sets. All WP leaders early on determined the extent and content of the characteristics, the naming of variables, and the coding. The technical procedure of the transfer of encrypted data and the schedule for data delivery were also fixed. The call-for-data, i.e., the starting point for data delivery by the DP, included all these specifications. The harmonisation contained technical validity checks, plausibility checks, and further consultations with the DPs if there were implausibility or technical problems. For bigger plausibility problems, single new transfers of “corrected” data packages were also scheduled. The use of self-generated codes which are not defined in international diagnostic classifications in some cancer registries is an example to show that it makes sense to carry out basic validity checks centrally.
The basic principle of this project, namely that the cohort data of the single DPs were sent to a data centre (WP1) instead of three outcome-related WPs, proved successful: WP1 was responsible for carrying out validity checks of all variables which did not refer to the outcome relevant characteristics. Otherwise, each WP would have needed to come up with and could have realised its own solution, and the data sets would not have been comparable. So all WPs profited from this procedure. The WPs with additional case–control designs had to set up further specific procedures for additional case–control-related treatment data, which were collected by WP3 and WP4 separately.
However, in general it was the responsibility of the respective WP leaders with their specific know-how to decide upon the outcome-specific variables (e.g., to decide which events were ultimately classified as cardiac events). While inquiries to the DPs were carried out solely via WP1 in the beginning, implausibility in outcome-specific variables were arranged to be clarified directly with the responsible WP leaders for the remainder of the project duration. Within the scope of the case–control study conducted, DPs had to be contacted on the part of WP3 and WP4 (e.g., for assigning controls to cases or for providing therapy data for cases and controls which had not been provided for in the superordinate data set). Due to the amalgamation of the data by a central office and the plausibility checks carried out by these two levels, we can assume high data quality.
Despite the basically unambiguous rules, a number of obstacles occurred, which required complex solutions. These solutions were necessary in order to generate a harmonised, large, and meaningful cohort. Basically, cancer registries are dynamic data sources, in which older data may be modified (subsequent changes, e.g., of diagnosis or age can be seen from time to time) and follow-up information becomes more current the longer the follow-up duration. Therefore, it is recommended that the DPs freeze their data on a specified day and provide them for the overall project. This was, however, hard to communicate, and some DPs kept transferring modified data sets to WP1. This is acceptable in some degree if this leads to a considerably improved data quality. However, marginal changes should not result in new update deliveries. It proved to be difficult to find the right balance.
Limitations of the assembling of this huge retrospective European cohort are that DPs were not always able to provide data as specified in the call-for-data; instead, individual arrangements concerning the data delivery and an adjustment of the central WP1 data base to individual import strategies became necessary. In the end, an individual handling for almost each DP was necessary. This caused temporal delays and the risk was real that some outcome-related WPs would fall behind; as a result, some DPs delivered their data prematurely and multiple times via WP1 to WP leaders, even though data entry and data processing had not been completed. For this reason, many more data updates than intended had to be accepted. The following example demonstrates the complexity: One DP provided 20 data updates altogether, and one WP received 24 data transfers from WP1. In principle, updates were planned only as an exception (step 5 in Fig. 1), and only one single data transfer from WP1 to the respective outcome-related WPs was planned (step 7). In addition, the progress of the work packages went in parallel. However, this could be balanced and compensated by WP1, while three independent, parallel work packages would have been hard to coordinate. Some DPs did not provide data for all three outcomes. In part, this was planned from the beginning (e.g., no cardiac events from the Nordic countries), in part, it became apparent only during the project duration that data could not be provided (e.g. mortality data from Norway are in general available, but could not be provided within the scope of this project). The duration of observation differed for the single events among the data sets of some DPs (e.g., longer duration for cardiac events than for the occurrence of second tumours).
In order to make the ultimate cohort centrally available after the end of the project, the data bases of WP3-5 will finally be transferred to WP1 again. WP1 will store the data and make them available for future projects, should the occasion arise. The cohorts finally analysed in the work packages (e.g. as basis for case–control studies) will differ from the cohort described here due to WP-specific eligibility criteria. Nevertheless, the PCSF cohort described is the basis for all analyses to be carried out in PanCare-SurFup as well as for projects going beyond the end of the project.
In a consortium like this one, progress largely depends on iron discipline and rigour with respect to the common rules for project management. All partners must follow the specifications of the consortium (deadlines, agreements, definitions). As a basic principle, a transparent, prompt, and problem-oriented communication is a necessary basis for the success of such a complex project. Within the course of the project, these processes proceeded more and more smoothly.
Limitations of the consortium are that assembling a huge cohort like this takes a lot of time and this took in the end much longer than anticipated from the beginning. PCSF applied for and was granted a 1-year no-cost extension. Data assembled many decades ago were difficult to collect in some countries. Data management, databases, and data differed from country to country mainly due to different ways of collecting the cancer data and the outcomes, requiring major efforts to make the data homogeneous and comparable.
There are some lessons learned and ways to overcome problems during the implementation of such a diverse cohort to be composed by bringing together very different data sets from different countries. It is strongly recommended that one central institution is installed for doing all the work regarding harmonization, standardization and communication. An iron discipline has to be conformed as well as rigour with respect to the common rules for project management. A transparent, prompt and problem-oriented communication is needed, too. The involved parties should find the right balance between being adamant about standardized procedures while on the other hand considering individual country-specific and data provider-specific framework conditions. Regarding the practical approach data providers should freeze their relevant data set on a specific day and avoid updates with only marginal modifications. The ultimate cohorts should be made centrally available at the end of the project by each work package leader and should have backups to enable sustainability and long-lasting data security.
Benefits of the consortium assembling late effects data is that rare late effects detected in more countries can be pooled and this might lead to new strategies for identifying ways to treat late effects and reach best clinical follow up. The assembled cohort is the largest cohort in Europe and under a handful others under the largest worldwide. Amalgamations of this kind enable analyses which would not have been possible because the diseases are so rare. The scientific legacy produced by PanCareSurFup is available for maintenance, update, and future use in accordance with the regulations set up after the official funding end of the project. Therefore, a PanCareSurFup Sustainability, Publication and Authorship Policy has been developed, which includes that requests from outside investigators for use of the PCSF data will be welcome at least 5 years from the end of the study. The final datasets from each work package of PanCareSurFup are stored at the original work package leader’s institution. Back-ups of all data are stored at defined other institutions.
PanCareSurFup succeeded in compiling the largest and in itself homogeneous cohort of children with cancer during childhood and adolescence through the close cooperation of many European countries and by establishing a work package solely for the harmonisation of heterogeneous data sources. We can expect high quality results analysing this large data set with respect to the three outcomes in PanCareSurFup. The resulting data set provides an excellent opportunity to compare outcomes of patients diagnosed over seven decades.
Depending on the national situation per data provider, informed consent was obtained from all individual participants included in the study, or the data collection was done under national law. All data providers obtained ethical approval or approval from the relevant national body, and PanCareSurFup was supervised by the PCSF Ethical and Scientific Advisory Board.