Frequently asked questions

  • Chevron Down

    What is the PHT?

    The term Personal Health Train is used for technologies and agreements that enable decentralised analysis of health-related data. Modern health care and research generates and requires vast amounts of data captured and stored at different locations. Managing these data in a centralized database is neither desirable nor feasible. Central data processing gives rise to issues on agreements for data-accessibility, data standardization and privacy. Researchers found ways to use health data from various sources in decentralized databases. They coined this approach, the Personal Health Train (PHT). The PHT is designed to enable health care innovators and researchers to work with health data from various sources. It gives controlled access to data, while ensuring privacy protection and optimal engagement of individual patients and citizens. The personal health train is a vehicle that moves the analysis to the data and by that avoiding that data must be moved or centrally collected. The PHT stands for a collection of technologies and agreements that enable decentral data analysis.

  • Chevron Down

    What is the promise of the PHT?

    The goal of the PHT is to improve quality of health care and health support efficiently, fast and affordable, while retaining privacy and safety. It enables reuse of data from different sources that previously was not findable or accessible. It will become possible to combine data that previously had not been possible such as lifestyle and health care data and eventually will enable personalised medicine.

  • Chevron Down

    How does the PHT approach work?

    With the PHT-approach, the analysis visits the data while the (privacy) sensitive health data remain where they are. Only the outcome of the analysis is exported. So, with the PHT-approach, less to no data will need to be moved.  This is beneficial for the quality of data, sometimes data-volumes are too large to move within a realistic timeframe and effort. Moreover, moving data is often done by copying privacy sensitive data. This is not desirable. The PHT approach preserves privacy yet allows to gain vital insights from health-related data.

  • Chevron Down

    Who can make use of the PHT?

    Up to now, the PHT-approach is mainly used by health care innovators and researchers. The aim of the PHT-program is to make the approach scale-able. In the future, health applications such as decision support aids might be supported by the PHT-approach and therefore the PHT might potentially be used (knowingly or not) by anybody. Other sectors such as the farming sector use similar approaches. 

  • Chevron Down

    Is there an instruction manual available?

    There is no instruction manual. The implementation of decentral data-analysis requires different (technical) agreements that depend on the purpose and context. In a research-setting the implementation may be more experimental by nature than an implementation in health care or management, which needs to be more robust. The Dutch PHT-program aims to define a set of agreements that enables interoperability between these different implementations and forms a base on which PHT-implementation can be accomplished.

    The PHT is a principle that requires context-specific choices. However, to achieve the intrinsic interoperability of the PHT-principle, a general set of technical and non-technical agreements is necessary. This set of agreements will be developed by the PHT-program. 

    The Dutch PHT-program aims to publish a set of (technical) agreements that will enable various stakeholders to implement the PHT-approach. The international PHT-network is working on international convergence.

  • Chevron Down

    Where can I find examples of the PHT?

    There have been several practical approaches to the PHT.

  • Chevron Down

    Is FAIR data a prerequisite for the PHT?

  • Chevron Down

    How does the PHT contribute to the AI–agenda?

    The PHT can provide a framework for the responsible application of AI in health care. The PHT-network is overlapping with the AI-network and is working in close synergy for example with the Dutch National AI coalition.

  • Chevron Down

    Are there other data-trains?

    Yes. Many other sectors, for instance the farming sector, have acknowledged the need and promises of working with federated data. The Personal Health Train keeps close contact and cooperates with multiple stakeholders from other sectors to learn from each other.

  • Chevron Down

    Does the PHT provide any kind of data analysis?

    The PHT-program focuses on describing the prerequisites for federated data analysis. It does not provide analyses. Yet, it aims to facilitate analyses to federated data. Within the PHT-network there are several projects that aim on sharing, re-using and improving data-analysis approaches.

What are the components of the Personal Health Train-metaphor? 

The train metaphor is useful to explain the principles of federated data analysis. The following explanation illustrates the different components of the PHT  metaphor. 

  • Chevron Down

    FAIR Data

    The datasets must be described according to the FAIR principle (Findable, Accessible, Interoperable, Reusable) by means of a common ontology. An ontology is a computer-interpreted description (model) of reality, also known as a knowledge representation. 

  • Chevron Down

    FAIR Data Point

    The location where the owners of the data records their data is called a FAIR Data Point. In general, organisations make a selection of their (operational) data available at their data point and provide the data according to the FAIR Data principles of an ontology. Placing data in a FAIR Data Point does not mean that this data is automatically available to everyone. 

  • Chevron Down

    FAIR Data Station

    The FAIR Data Station facilitates that the train can access the data. Depending on the application, the service must be more or less secured; and depending on the complexity of the analysis, the service must offer more or fewer possibilities. This can range from an open SPARQL endpoint to a virtual machine. 

  • Chevron Down


    The train is the logic of the analysis. The form of the train varies, depending on the type of service. The train can take the form of a query, an interface call or a fully programmed and self-learning algorithm. 

  • Chevron Down


    The rails are the agreements, guarantees and interfaces of the PHT.  

  • Chevron Down


    The analysis is sent from a dashboard. This location has a monitor, which can track the progress of the train and show the final results. 


Artificial intelligence 

Artificial Intelligence is a collective term that refers to systems that automize specific human stategies for reasoning and decision-making. 

Artifical Intelligence may for amongst others refer to Natural Language Processing, Knowledge Representation, Machine Learning and Robotics.In the context of medical data, Artificial Intelligence mostly refers to machine learning. 

Deep learning 

Deep learning is part of a larger family of methods of mechanical learning. It makes use of artificial neural networks that can analyse larger amounts of data based on examples. 


Encryption is the encoding of information using a complex mathematical formula. Encryption uses a key to make information unavailable in its original form. Your encrypted data can only be restored if you use the key. This is called decryption. 

Federated/distributed learning 

Federated or distributed learning are techniques which being used to develop artificial intelligence (AI) - algorithms in an automated way - in other words ‘to train’ these algorithms - based data that are stored at different locations. This Machine learning is thus performed in a distributed or decentral manner. This technique offers the possibility to work with data without it having to be centralised. 

FAIR-data principles 

FAIR-data principles are guidelines for the way data is described, stored and published. The letters ‘FAIR’ are an acronym for: Findable, Accessible, Interoperable and Reusable 

Partitioned data 

Vertically partitioned data refers to data concerning different aspects of the same person. Horizontally partitioned refers to the same type of data of different peoples. 

Informed consent 

Describes the agreement of a person – who has received/ access to adequate information – to e.g. participate in medical studies. 

(Natural)Language processing 

(Natural) Language processing is a combination of statistical techniques and machine-learning techniques. This enables to find key words from unstructured texts, in order to understand a human language with a computer program. 

Machine learning 

Machine learning refers to the development of an algorithm by letting a computer perform data-analyses. In the medical context, machine learning is applied for example to let algorithms perform specific tasks based on pattern-recognition often large and historical data-sets. Machine learning may for example be applied to enhance the speed of certain tasks or to automize them. For example to perform a prediction whether a person has a certain disease. 

Multi-centric learning 

When multiple centres work together on a study. 

Multi-party computing 

Multi-party computing is a cryptographic protocol that allows different parties to jointly compute data without centralising that data. The outcome of this activity is the result of the calculation, without the parties having seen the data of the other parties. 

Neural network 

Neural networks are based on the functioning of the brain. They model the data using models of neurones. These will be applied to try to solve complex tasks that cannot be solved by the current computer science and AI methods. 

Knowledge representation 

Knowledge representation is a disciplin that focusses on presenting information in a way that computing systems can make use of this information. 

Open source software 

Software that has its source code published and is freely available to everyone. 

Personalised medicine 

Personalised medicine is a medical model with the aim to tailor medical decisions, interventions and/or products to an individual patient. Based on a person’s predicted response or risk of disease.  


When designing an information system, privacy is taken into account. This remains a focal point throughout the lifecycle of the system. 

Self Sovereign Identity(SSI) 

SSI is a collective term for cryptographic technologies to give users control over which personal data is shared with whom, while the recipient of the personal data can quickly verify them electronically. The aim of SSI is to enable safe and efficient exchange of digital information. 

Shared decision-making 

Shared decision-making refers to the joint decision by a patient together with a health care professional. This requires a proper exchange of information between the two. 

Text mining 

Text mining is a method to recognize patterns in unstructured text. It is a form of natural language processing. It provides way to manage the growing amount of available information. 

Share this page…