The Swiss Personalised Health Network developed a flexible, pragmatic framework that promotes interoperability between diverse biomedical datasets. The proposed three-pillar strategy synergises data and permits its use according to the specific needs of researchers and communities, thereby stimulating progress in data-driven medicine.
The challenge of interoperability in data-driven medicine
The advent of next-generation sequencing technology has represented a huge stride towards realising the potential of personalised medicine. By shifting away from the traditional one-size-fits-all approach, personalised medicine aims to achieve the best possible outcome for treating or preventing disease based on a patient’s unique genetic makeup and response to environmental factors.
Large volumes of data must be cross-referenced and collated to realise the potential of personalised medicine, including genomic, epidemiological and clinical data. However, data from multiple diverse sources is often incompatible, which poses a major challenge for data-driven personalised medicine.
Interoperability denotes the ability of different computer systems to connect, share, understand and use information. Medical knowledge and data are represented and organised in a way that suits a specific purpose and community. Different scales, scores and classifications are used to partition medical knowledge into finite classes that represent the meaning of data (semantics). Depending on its purpose, each classification is characterised by a specific way of partitioning.
Meanwhile, data acquired from research and healthcare systems are structured in purpose-specific data models. As such, there is no single data classification or model that can serve every purpose.
Swiss Personalised Health Network
Data sharing across systems is the key to unlock the potential of data-driven personalised medicine. However, before the development of personalised medicines, interoperability must first be achieved. For this purpose, numerous organisations were established, including data sharing initiatives such as Integrating the Healthcare Enterprise and SNOMED International. Their attempts to promote interoperability mostly involved mapping data onto a single common standard. However, this meant that these standards are only applicable to specific cases.
In this work, the Swiss Personalised Health Network (SPHN) developed a different approach to solve this problem. SPHN aims to develop a national infrastructure for using health data to promote research in personalised medicine. Specifically, their goal is to facilitate data exchange between the Swiss healthcare system, research communities and regulatory agencies in a safe, interoperable and meaningful fashion.
Informed by the shortcomings of previous attempts to address interoperability, the SPHN adopted a three-pillar strategy (Figure 1). This leverages existing semantics to unify databases in a purpose-specific and complementary manner, without relying on any particular standard. Consequently, the framework has the flexibility to be translated into any existing or future data model.
Figure 1: The three-pillar strategy for interoperability. (ICD-10: 10th International Classification of Diseases; SNOMED CT: Systematized Nomenclature of Medicine Clinical Terms, LOINC: Logical Observation Identifiers Names and Codes; OMOP: Observational Medical Outcomes Partnership; CDISC: Clinical Data Interchange Standards Consortium)
The first pillar involves the multidimensional encoding of concepts using data from diverse sources. It develops a semantic framework encoding a set of concept definitions that rely on existing knowledge representations. As a result, it allows the meaning of data to be expressed without imposing a particular standard.
The second pillar then uses descriptive formalisms to describe and integrate data from the first pillar. These include the resource description framework (RDF), Arden syntax and Web Ontology Language. These languages intuitively describe data and its semantics as they are collected at the source. Since no specific data model is enforced, there is flexibility in the storage and transport of information. This pillar also permits the use of different formalisms depending on the type of information and purpose.
Finally, the third pillar translates the formal descriptive languages into any target data model suitable for the needs of any user. These can be existing or ad hoc models, owing to the flexibility of the first and second pillars. Such one-to-many mapping permits data to be efficiently shared across communities. Critically, the data in this pillar is structured as different data models that conform to the requirements of specific research projects and communities.
Since mid-2019, the strategy has been implemented in all 5 Swiss university hospitals and high research organisations related to personalised medicine, including eHealth Suisse, the University of Geneva, the University Hospitals of Geneva, SIB Swiss Institute of Bioinformatics and Lausanne University Hospital. These efforts were coordinated by the SPHN Data Coordination Centre, part of the SIB Swiss Institute of Bioinformatics. The proposed strategy holds promise to stimulate research and innovation for personalised health in Switzerland.
Overall, the SPHN has developed a nationwide framework to enable interoperability of health data. The success of the three-pillars strategy lies in the development of a pragmatic semantic-based framework that synergises existing standards instead of building new ones.
Consequently, data from diverse sources can be integrated and applied to meet the specific needs of the research and communities involved. This work is important as solving the challenge of interoperability may finally enable the promise of data-driven medicine to be fulfilled.
Image credit: creativeart – Freepik