>> After a very detailed presentation of the data availability and use available in twelve carefully selected countries, Busse postulates that a good data basis for research purposes is particularly available “when all information about the interaction of individual patients with the health system is kept in a central location “. In this way, care processes across service providers and, if necessary, across sectors could be mapped, ideally also over time. According to the report submitted on behalf of the Zi, this condition is generally met in those health systems that have anchored a “gatekeeping” system in primary care; A typical example of this is the family doctors in Great Britain. According to the Zi report, the most typical example of a database that brings together data collected by primary care physicians is the Clinical Practice Research Datalink (CPRD), which brings together data from practices that have enrolled in the program (“opt-in”). Patients of these practices can in turn opt out of data transfer (“opt-out”), in line with the NHS national opt-out policy. The data held by CPRD can be applied for observational studies or used to support experimental studies. They can also be linked to other NHS Digital datasets to provide richer, cross-sector insights. More than 2,000 scientific studies have already used CPRD data, explains the report.
Another positive example is Scotland, where the Scottish Primary Care Information Resource (SPIRE) was founded in 2016 to bring together data from Scottish practices in order to give general practitioners working there the opportunity to find out more about their patient population in comparison to better plan care and support the Scottish National Health Service (NHS) to monitor and improve the quality of care.
The initiative emphasizes that SPIRE is not about a central database of all information on Scottish patients, but rather the targeted aggregation of parameters from GP data for specific, approved purposes. Data would therefore not be stored longer than is necessary to process the respective applications; the information would then be destroyed.
The primary care database of the Dutch research institute NIVEL also uses data from primary care, although the data from the electronic practice management systems also comes from service providers beyond general practitioners. In the Netherlands, this model is also used by other institutions and initiatives.
According to an interim conclusion, what all of these initiatives have in common is that, in addition to research opportunities, the data is also used explicitly for feedback to the participating practices, including an opportunity to compare your own results with those of other practices. This also exists in countries such as England, Canada and the USA, where a “robust technical infrastructure and strong scientific and political support were essential” for successful implementation.
There is definitely something comparable in the German healthcare system: the Zi practice panel. The aim of this panel is to monitor quality and costs, but also to use the data for health services research. However, the structure of specialist medical care in Germany means that, in contrast to the examples described above, specialists also take part in this practice panel.
Data from independently
created stocks
In the course of the research for the report presented, various data sources with different content were identified, which can be linked via one or more link variables. This is not data that has been collected in an integrated manner for the primary purpose of use or that is merged into a new data set, but rather must first be linked for the secondary purpose of use (in this case for research). On the one hand, this could be necessary because the data for the primary purpose only covers a very limited area (for Germany, the report cites the §21 KHEntgG data as an example); or on the other hand, because an additional expansion of an already extensive data set may be necessary in order to be able to answer certain research questions accordingly. This refers, for example, to links between health care data and employment/unemployment data or between clinical data and billing data.
The overarching goal of these data initiatives is to link data from different sources, especially from different public institutions. This could, for example, link data on health care with data from the education sector and crime statistics.
In the opinion of the report authors, central data storage locations for merging independently created data sets open up opportunities for health services research. Examples of this are initiatives such as eDRIS in Scotland or PHARMO in the Netherlands and the Finnish FinData campaign.
The extent of the consolidation of the different data sets is different and relevant with regard to the considerations for potential idea generation for the German region. For example, the Australian Population Health Research Network (PHRN) and the Canadian Social Data Linkage Environment (SDLE) link data together on a project-specific basis, while the source data sets remain generally distributed. In contrast, New Zealand's 'integrated' Integrated Data Infrastructure (IDI) stitches together incoming data using a 'spine'. What is also interesting about the PHRN approach is that the regional authorities at the state level have their own linking offices that cooperate in the PHRN network, but also carry out their own linking. According to the authors of the report, this idea could also be considered for the German healthcare system. For example, you are thinking of a combination with a design of the German Record Linkage Center (GRLC) of the University of Duisburg-Essen, which was founded in 2011 in cooperation with the Research Data Center (FDZ) of the Federal Employment Agency (BA).
Differences in the linking process
In the opinion of the report authors, successful linking requires the possibility of a clear assignment of information from separate data sets. This requires suitable key variables that must be present in all data sets to be linked. For a personal link, it is often necessary for data protection reasons to use pseudonyms instead of clear personal identifying characteristics from the original data (such as an insurance or ID number). However, the same pseudonymization process must be used for all data sources to be linked.
According to the authors, the responsibilities for the pseudonymization of the data as well as the process of de-identification of the data sets can contribute significantly to maintaining anonymity and protecting privacy.
In Germany, it would previously have been challenging in terms of data protection to establish widespread use of linking health-related data. Nevertheless, successful case studies would show that the corresponding opportunities can be better exploited. Methodological standards for linking personal data in the context of research projects have also been formulated with the “Good Practice Data Linkage”.
Although, in the authors' opinion, a detailed technical description of possible de-identification and linking options goes beyond the scope of the report presented, it nevertheless seems useful to them to highlight individual approaches as examples in order to allow initial ideas for future discussions.
In the sample of case studies included in the report, there is a cluster that uses a procedure based on an underlying, “anchoring” list of people to use identifying characteristics to identify these people in the data sets to be linked and thus to link data. The “IDI Spine”, which is used in the New Zealand IDI, is mentioned as an example. According to the explanation in the report, this spine represents a list of individuals that combines information from the tax offices with birth and visa details. The aim of the spine is to identify all possible individuals who have ever lived in New Zealand (target population), but to record them only once.
All microdata of the sectors mapped in the IDI are then linked to the spine and could therefore also be linked to each other, but without using a unique identifying feature across the different sectors. The “Population Spine” (PS) in Scotland, which is used for the links via eDRIS, also works in a similar way. PS maps everyone in Scotland who has ever been in contact with NHS Scotland, along with their identifiers.
The linking initiatives mentioned here would demonstrate that even without the presence of a unique identifier, linking based on other identifying characteristics is possible and has been used successfully for several years. <<
by: MVF editor-in-chief Peter Stegmaier