Pitfalls and best practices in extracting data from multiple sources

Make sure to have a clear idea of what you need early on. Think carefully about the level of detail you actually require and make use of the support structures available.

Jonas Björk’s research focuses on public health questions, and he also has a strong interest in methodological aspects of epidemiological research. Over the years, he has gained substantial experience in managing data from multiple registers. Many of his studies are based on data from various health data registers that have been linked to data from population registers.

“There are many steps in the process when data from several different sources is to be requested and linked. So it is important to be aware that the initial phase can be long,” he says.

In many cases, approval from the the Swedish Ethical Review Authority is required, for example if medical records, health-related data, or other sensitive personal data are to be processed.

“The ethical review is the foundation for conducting the project and then serves as input to the confidentiality assessments carried out by data controllers. So already at the ethical review stage, you need to be able to describe the benefit of the study and specify the data required,” he says.

Clear justification is essential

Jonas Björk emphasises that deciding at an early stage – and being able to justify – which data are required to address the research question is crucial to ensuring a smooth process. It is important to think carefully about the level of detail required in the requested data, otherwise there may be unnecessary delays along the way.

“As a researcher, it is important to be pragmatic and to consider when detailed data are necessary and when more aggregated data will do. If your request is too general, without clearly stating why this is necessary to address the research question, you are likely to end up in discussions with the register holder. After all, the register holder needs to take into account both confidentiality requirements and data minimisation, meaning they should not disclose data ‘unnecessarily’ unless it is justified by the research question“ he says.

Jonas Björk recommends contacting the register holders at an early stage in the planning process. Especially if there is uncertainty regarding what data is available and can be requested.

“There are substantial support available to help researchers understand the registers. Sometimes register holders have designated contact persons who can describe the content, and most major register holders, such as Statistics Sweden or the National Board of Health and Welfare, have websites providing information about the registers and the variables they contain. The register holder can also often advise on how to formulate requests to increase the likelihood that your request will be approved and that the data will be granted,“ he says.

Collaboration and support can enhance efficiency

Jonas also recommends seeking knowledge and support within your own university. Many universities and higher education institutions offer advice and support to researchers on data extraction from registers and the management of research data. It can also be beneficial to contact register centres, as well as research communities and networks at the institutions, which have experience of working with register data in the relevant field of research. Identifying productive collaborations can make the work smoother and more time-efficient.

“Managing an ethical review and combining data from several different sources is complex and requires a certain level of experience. If you are not familiar with using registers, it may be advisable to seek out potential collaborations. This can make the research process more efficient and help you avoid errors and pitfalls in data management,“ he says.

This applies particularly if you are obtaining data from registers other than those you normally use, Jonas Björk points out.

“For example, if you want to identify suitable data on prior morbidity among individuals participating in a population-based study, and you have no previous experience of using registers to summarise levels of morbidity it is easy to make basic mistakes. Fortunately, there is often a wealth of expertise available from other researchers who have experience in identifying relevant variables and definitions in register data for the information you wish to obtain.“

Complex data linkage requires expertise and resources

Linked very large datasets requires considerable time, as well as a certain level of data expertise. Sometimes it can simply be too complex and demanding for researchers to handle the linkage themselves.

“Data is usually provided as a set of separate files, and the linking process is complex and requires a high degree of precision. In the case of larger register studies, this expertise may therefore need to be secured. Some projects engage a dedicated data manager − someone with specialist expertise in linking data into a research database,” he says.

Another important consideration is how large datasets should be stored and managed.

“When data are disclosed electronically, the university must have robust systems and secure servers in place. This generally works well, but it is important to follow the university’s data management procedures. For example, it is necessary to specify which researchers in the project will have access to the data, “ says Jonas, and continues:

“Another solution is for all data to remain on a server hosted by the register holder, with the researcher being granted an account to log in to that specific environment. Statistics Sweden offers the MONA (Microdata Online Access) service, which is a platform for providing access to microdata. If a project uses data from other register holders, it is possible to consult Statistics Sweden to determine whether these data can be stored and made available via MONA.“

Explore the potential of cohort studies

Jonas points out that, in some cases, it may be possible to answer a research question using data collected from a study that has already been conducted.

“Most major universities have ongoing cohort studies, which follow groups of individuals who have voluntarily enrolled over time. These can be a valuable source for addressing many research questions. At the same time, it is important to be aware that the legal aspects surrounding data extraction from an existing study can be quite complex.“

The reuse of previously collected data in new research must also be carried out in accordance with current ethical review legislation. It is also advisable to contact those responsible for the relevant cohort study at an early stage of the process to discuss what data are available and the possibilities for access.

Positive developments in register-based research

Although large-scale register requests can be both time-consuming and complex, Jonas believes that the preconditions for conducting research using health and register data in Sweden have improved somewhat in recent years.

“During the pandemic, the limitations became clear, but now a great deal is happening both in Sweden and within the EU to improve the conditions for conducting research and development using register data. This is a welcome development that is beneficial both for us as researchers and for society,” he says.

Text: Ulrika Ernström