This study was approved by the Hospital Clínic de Barcelona Ethics Committee for Investigation with Drugs (HCB/2018/0573).

We included in this project patients with COVID-19 admitted to the emergency room (ER) or hospitalized between February 17, 2020 (beginning of the first wave in Spain), and February 15, 2022 (end of the sixth wave in Spain).

The following methodology comprises a series of steps in order to achieve the study’s objectives.

The main deliverable of this project is the methodology described in the previous section. By following the aforementioned steps, any health care institution can go from local raw data to standardized, semantically interoperable EN/ISO 13606 and OMOP repositories. This methodology also led to the creation of 12 EN/ISO 13606-standardized archetypes that model important clinical variables in the ER and hospitalization settings, allowing the reuse of clinical information by using it in accordance with the Creative Commons terms.

Another interesting result of this study is the development of the ontologies that represent OMOP CDM, as well as their mappings to EN/ISO 13606 AM and RM. This process was carried out by members of the MIU at HCB after carefully reading the pertinent documentation of these standards and designing the optimal way of using them to represent clinical concepts.

Furthermore, representing clinical variables by means of ontologies is another way of reusing clinical information. With the creation of new ontologies for each project at HCB, where we have developed the ontology-based clinical repository OntoCR, we continue to extend our clinical knowledge representation.

Table 1 shows the correspondence between EHR archetypes, OMOP CDM tables, number of extracts created throughout the study, and the number of COVID-19 patients they pertain to. We have included the diagnoses recorded in the episodes of the study period as well as the historical ones. Health problem entries are part of the aforementioned project with BSC to extract clinical entities from unstructured texts through natural language processing, so they will not be included in this table.

We still do not have the final number of records in our OMOP tables, since the initial approach of adapting the preexisting API had to be replaced by the creation of the application that inserts data from the extracts into the ontologies. However, an OMOP database for a random small subset of patients was successfully created to test the queries and validate the methodology. This was performed using a locally developed Protégé plugin (“OntoLoad”) that imports a set of data from a relational database into the ontologies [17]. Table 2 describes the OMOP tables that were created.

This study proposes a methodology for standardizing clinical data, thus allowing its reuse without any change in the meaning of the modeled concepts. Although the focus of this paper is health research, our methodology suggests that the data be initially standardized according to EN/ISO 13606 to obtain EHR extracts with a high level of granularity that can be used for any purpose, as previous studies have suggested [20]. Afterward, its transformation to OMOP CDM–compliant tables allows its consolidation in joint repositories for research purposes.

Although EN/ISO 13606 was chosen because of the operational mechanisms it offers for data exchange, due to the flexibility and standard-agnostic nature of our methodology, there is complete independence regarding any specific standard. Thus, by modeling ontologies of other standards and mapping them to local variables, we may, for example, carry out transformations between EN/ISO 13606, OpenEHR [21], Fast Healthcare Interoperability Resources (FHIR) [22], OMOP CDM, and Informatics for Integrating Biology and the Bedside (i2b2) [23] with the minimum use of resources and without the need for changes in the database structure. Health information standards such as EN/ISO 13606 and OpenEHR allow the modeling and formalization of clinical knowledge through their RMs and archetypes [24], and ontologies are precisely a tool for carrying out such tasks. This is what makes them ideal in the context of an implementation of a dual-model strategy, allowing the representation of concepts in the health domain, its standardization, and the storage of instantiated patient data.

Furthermore, ontologies provide several advantages for the conceptualization of entities in a domain. It explicitly represents domain knowledge, allows the application of inference processes, enables the reuse of domain knowledge, allows data aggregation, and detects new associations between concepts [17].

It is clear for us that loading normalized data onto clinical repositories (instead of ad hoc data loading) provides many benefits. It is possible to reuse the same interoperability standards used in health care, adapting them to this new paradigm [25]. This approach allows the availability of clinical data for further single- or multicenter research.

We would like to highlight the vital importance of continuous collaborative research. This study is framed within a continued line of research since 2009 between HCB, ISCIII, and H12O. In this line of collaborative research, a standardized and transparent process has been designed and implemented for obtaining standardized data models for research from EHR raw data. Hence, in the first stage, the basis for a semantically interoperable clinical information management system based on EN/ISO 13606 was defined, proving that clinical information residing in heterogeneous systems could be normalized, combined, and communicated without loss of meaning. In the second stage, a common information model that reflects the clinical process and the relationships between the clinical records components was developed. In the third stage, a normalized information model based on EN/ISO 13606 archetypes was implemented and applied to local information systems for specific clinical use cases. With this model, it is possible to construct and order information recovered from these complex systems for the exchange of integral health and social information of patients and to use it for secondary purposes.

Many of the requisites of clinical data repositories for primary use are common to those for secondary use, such as normalized clinical information models, controlled terminologies, identification of actors, and contextual information. Developments carried out for primary use repositories are also profitable for secondary uses, and the progresses derived from secondary uses accelerate the advances in shared clinical records. A lot of work has been reported in this field throughout the globe in the last years, which has led to developing policies, repository models and its application in the form of competitive projects [2,26,27].

It is very usual for researchers to resort to the generation of their own data for research and its manual introduction into data management systems. It is also quite common for them to use general purpose tools, particularly spreadsheets, as data management systems [28], while there is perception of a high need of additional support for the analysis of high volumes of data. This represents a significant problem, since these applications cannot guarantee the consistency of data, and they present difficulties for sharing and consolidating data and a limited capability of data exploitation.

Different methodologies have been proposed to create OMOP repositories from raw data. Some approaches are based on a simple mapping of local variables to their OMOP CDM counterparts, an alignment of vocabularies using the Athena tool provided by OHDSI and an ETL process through SQL scripts [29]. Other authors have proposed transforming source data to RDF, carrying out a semantic mapping (in some cases, using an ontological representation of OMOP CDM), and loading it to a data store [30,31].

Likewise, other standard-agnostic approaches have been reported in the literature. The ongoing INFOBANCO project of the Madrid Region [32] seeks to create a platform for the management, persistence, exchange, and reuse of health data focused on applying each health information standard for the purpose it was intended to, offering multiple interoperability and exploitation services suited for specific use cases [24]. Furthermore, the 3-pillar strategy of the Swiss Personalized Health Network [33] pursues a semantically interoperable clinical data landscape based on a multidimensional encoding of concepts, an RDF-based storage and transport of the instances of these concepts and a conversion of RDF to any target data model.

This study has many strengths that are worth mentioning. On the one hand, it describes a real-world collaborative effort between 3 health care institutions in Spain to model, share, and consolidate standardized patient data. Furthermore, the standard-agnostic nature of the proposed methodology leads to a significant scalability, allowing transformation between different health information standards and common data models. The software used in our methodology (LinkEHR, Protégé, and Liferay) either have a free version or are open source, which make them accessible to low-income areas and institutions with limited funding for interoperability projects.

We must also mention the limitations of this study. First of all, the ontology-based clinical repository used in our institution was developed throughout many years, and it might not be a suitable approach for institutions that seek a rapid implementation of a methodology. This can limit the external validity of the study. Moreover, since the tool to insert data from standardized extracts into the ontologies is not ready yet, we still have not completed the creation of OMOP CDM tables. However, an OMOP CDM database for a small subset of patients was successfully created to test the queries and validate the methodology.

The MIU team at HCB is working on creating the ontological representation of different health information standards (FHIR and OpenEHR) and CDMs (i2b2, International Cancer Genome Consortium Accelerating Research in Genomic Oncology (ICGC Argo) [34], and Clinical Data Interchange Standards Consortium (CDISC) [35]). This will extend the current metamodel and allow us to carry out multistandard transformations, which will also help us compare the performance of such standards for different scenarios.

Semantic interoperability plays a very important role within HISs, providing meaning and clinical context to the clinical information and allowing for better clinical decision-making and research. This study has demonstrated that ontologies constitute a valuable approach for knowledge representation and standardization of health information in a standard-agnostic manner. With the proposed methodology, institutions can go from local raw data to standardized, semantically interoperable EN/ISO 13606 and OMOP repositories.