Data discovery and data harmonization<\/li>\n<\/ul>\nMidterm Goals<\/strong>
\nFollowing action points had been identified:<\/p>\n\n- Running trains within Germany between MII partners.\n
\n- Existing consortial use cases will not serve the purpose.<\/li>\n
- Interconsortial use cases should be set. Rare Diseases would be a good demonstrator.<\/li>\n<\/ol>\n<\/li>\n
- Running trains cross borders\n
\n- Train rails between the Netherlands and Germany:\n
\n- Advantage: Joint development within the GO FAIR initiative<\/li>\n
- Disadvantage: No national initiative exists yet.<\/li>\n<\/ol>\n<\/li>\n<\/ol>\n<\/li>\n
- Train rails between Switzerland and Germany:\n
\n- Advantage: The Swiss Personalized Health Network (SPHN) is well aligned with MII.<\/li>\n
- Disadvantage: Switzerland is not part of the EU.<\/li>\n<\/ol>\n<\/li>\n<\/ol>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
<\/p>\n
Technical Infrastructure:<\/strong><\/p>\nThe afternoon session on the Technical Implementation of the PHT identified several key challenges and suggested directions for future implementations.<\/p>\n
\n\n\nKey Points from the Technical Infrastructure Working Group Discussion<\/strong><\/em><\/td>\n<\/tr>\n\nOvercome the Trusted Third Party Restriction<\/strong>
\nCurrent implementation efforts include at least one central service, the Docker Registry, that is known and addressable by several other services via HTTP. The operator of such an instance can access all images within the registry, hence the operator needs to be trusted. Avoidance of such trusted third parties leads to a preference for decentralized approaches. A potential Peer-to-Peer approach technology might avoid centralization and constitute a platform for developments efforts.Rich Station Metadata Publishing<\/strong>
\nA train is associated with a set of stations that it visits, potentially with a certain order. The construction of such routes needs to be assisted by metadata, which is provided by the stations. Station metadata should provide information on data quality and completeness. A metadata server may keep an index of available data resources at stations<\/p>\nTrusted Broker<\/strong>
\nKnowledge of the stations that a train decided to visit leaks information on the station’s data to the train issuer. Hence, the train issuer is generally not allowed to know the stations being visited by the train. This leads to an anonymization component, which hides the set of stations visited by the train from the issuer.<\/p>\nData Standards<\/strong>
\nThe diversity of data resources at each station poses significant challenges for implementing algorithms that operate on the data. Hence, the wide adoption of the FHIR communication protocol and common schemata directly enhances the potential of the PHT. Exposition of data as FHIR poses challenges for clinical data warehouses, such as tranSMART. The interoperability between tranSMART warehouses and FHIR resources should be explored.<\/p>\nImprove Data Findability<\/strong>
\nThe development of machine learning algorithms requires data exploration and potential pre-analyses. Pre-analysis trains that can cache results for subsequent trains. The PHT infrastructure can benefit from the ability to store results from such pre-analyses at stations for later consumptions by main algorithms.<\/p>\nStation – Train Communication<\/strong>
\nIn the current implementation, the stations do not have any assumptions on the content of trains. Consequently, stations cannot assist trains in their execution, for instance by adaption computational resources to the need of the station. A semantic communication between train and station should be facilitated ideally before routing. The station should know who runs the analytics task and its intention. There could be multiple trains, each focus on another language, format, technology. The right train for the station should be identified.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n <\/p>\n Use Cases<\/strong><\/p>\nThe use cases working group identified the possible application areas of PHT in healthcare and life science domain.<\/p>\n \n\n\nKey Points from the Use Cases Working Group Discussion<\/strong><\/em><\/td>\n<\/tr>\n\nThis Group had identified a set of Use Cases that could benefit from the PHT approach, such as:<\/p>\n\n- Text Mining (application): access to medical plain text documents which cannot be shared otherwise, run algorithms or models on corpus, could be useful for training and productive phase (for training, human interaction\/acccess is necessary)<\/li>\n
- Feasibility requests: getting the number of eligible patients, popular task with low data protection problems (only aggregated data)<\/li>\n
- Support for recruitment: contacting and including a patient as a subject, important for clinical trials and registers, more problematic with regard to data protection due to personal identity and mostly lack of consent to data processing or contacting<\/li>\n
- Record Linkage (PPRL) to find and count e.g. rare disease patients<\/li>\n<\/ol>\n
The importance of providing FAIR data at the creation had been emphasized:<\/p>\n \n- Instruments should be developed to support FAIR data at creation<\/li>\n
- Success stories: provide plausible scenarios for utilizing FAIR data and show ways for a data protection-compliant implementation<\/li>\n
- Simple, easy-to-deploy-environment for PHT to demonstrate the advantages (project \u201cFeldbahn\u201d)<\/li>\n<\/ol>\n
The role of ontologies and vocabularies are underlined:<\/p>\n \n- Phenotyping: is everything but HPO<\/li>\n
- A mapping between European countries f.i. medical terminologies (e.g. BRIDGE DB)<\/li>\n<\/ol>\n
Next Steps:<\/p>\n \n- Prioritization of use cases and goals<\/li>\n
- Identify a mapping between goals and existing projects<\/li>\n<\/ol>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
<\/p>\n Session 4: Discussions, Summary and Outlook<\/strong><\/p>\nWorking groups presented and discussed their results.<\/p>\n The alignment between ongoing projects was discussed. Suggested collaborations:<\/p>\n | | | |