"It takes effort, but you’re investing for impact."
The ‘Personal Health Train’ (PHT) concept allows you to share and analyse large amounts of data without compromising privacy. Together with the NCDC research team, Dr Pedro Da Costa Mateus first applied the PHT to dementia data (at the time whilst still at Maastro-Maastricht University). Da Costa Mateus, now a clinical data scientist at Radboudumc, explains how the PHT works, what benefits it offers, and shares some tips.
Analysing data from other cohort studies, or sharing your own data with others, is difficult to reconcile with protecting the privacy of research participants (see also: Making research data FAIR – how?). Yet exchanging data is crucial for making an impact. It enables you to carry out data analyses on a much larger scale and thus answer more research questions more quickly.
A look at data
For this reason, it is good that the Personal Health Train (PHT) has been developed. This is a digital infrastructure designed to analyse data without sharing privacy-sensitive information. It works like this: instead of copying all the data to a single central location, the PHT makes it possible to bring an analysis algorithm or script (the train) to the data. In other words: to data owners such as research institutions, hospitals or clinics (the stations). They then specify what a passing train is or isn’t permitted to do with the data. Finally, the railway lines ensure that the relevant algorithms can access the data securely and that interaction can take place. Ultimately, the algorithm only brings back the results. The researcher wishing to analyse the data cannot, therefore, view the privacy-sensitive information.
The PHT was jointly developed by Maastro (Maastricht University), the Dutch Techcentre for Life Sciences and the LUMC. It has been further developed into an open-source application by the Netherlands Comprehensive Cancer Organisation.
PHT applied to dementia data for the first time
Da Costa Mateus and the team at the NCDC (Netherlands Consortium of Dementia Cohorts) were the first to apply the PHT to dementia data. This took place as part of their research into blood and MRI biomarkers related to Alzheimer’s disease.
The main challenge for the team was harmonising data from nine Dutch cohorts – in other words, ensuring that all variables were named and coded in the same way.
When designing an analysis algorithm, it is important to know exactly how the data is organised. When using the PHT, you cannot view the data, making it difficult to ‘debug’ it. “That is why data harmonisation beforehand is crucial. That was also our biggest challenge,” says Mateus. “Particularly because no suitable data model yet existed for longitudinal cohort data.” Mateus created scripts and mappings that enabled the harmonisation of the nine cohorts. DEMPACT established the Dutch Dementia Data Community on zenodo.org. There you will find a publication about these scripts and mappings.
What has the NCDC achieved thanks to the PHT?
Mateus: “The PHT increased the volume of data we were able to analyse so significantly that we were able to answer research questions that we would not have been able to answer by analysing each cohort separately.”
So, all in all, was the use of the PHT a success?
“Yes. Ultimately, it worked for the overarching use case: training an AI model to predict brain age.”
Mateus does emphasise, however, that it was certainly not a smooth process. “It proved difficult to reach legal agreements. There was quite a bit of miscommunication and differences in interpretation.” This cost the team a great deal of time. As did coordinating the implementation of data harmonisation. “And practical matters, such as installing certain software at the data owner’s premises. Fortunately, the agreements are now in place, and they provide a solid foundation for the future.”
Do you have any practical tips for fellow researchers?
“Make sure you bring the right expertise to the table: people with legal knowledge, but also people with the right technical expertise. We spent a lot of time and energy explaining to hospitals, clinics and other parties what the PHT entails and how we wanted to use it. We really had to convince them of the major benefits. If you don’t do this properly, parties are often unwilling to cooperate. They’re afraid of handing over data and then accidentally breaching privacy legislation.”
“And yes, this does indeed take quite a bit of effort. But you’re investing for long-term impact. Thanks to the results we’ve achieved, we can now carry out the analysis for another research question, once again using the PHT.”
What was your biggest insight regarding the PHT?
"A lot of the technical knowledge was new to me. But the most important insight was just how crucial collaboration is to success. You can’t carry out this kind of research on your own; you have to bring together the right expertise to make it happen."
Ready to get started or have any questions? Get in touch!
Would you like to use the PHT for your cohort data? Or do you have any other questions about data management? Please feel free to contact Data Management Coordinator Dr Didi Lamers.
