The German Corona Consensus Dataset (GECCO): a standardized dataset for COVID-19 research in university medicine and beyond

In December 2019, first reports of a cluster of 41 patients infected by a novel coronavirus emerged from Wuhan, China [1]. Within a few months, the new virus, subsequently named “severe acute respiratory syndrome coronavirus 2” (SARS‑CoV‑2), has spread around the world causing the global COVID-19 pandemic. Currently (as of November 12, 2020), SARS-CoV-2 has infected more than 50 million and killed more than a million patients worldwide [2].

The pandemic has spurred intensive scientific research, including numerous regional, national and international epidemiological surveys and studies [3‐7]. While this research provides important new insights, the multitude of studies threatens to generate a dangerous fragmentation of information. This could delay or even prevent urgently needed scientific knowledge about SARS-CoV-2 and COVID-19. To avoid this fragmentation of information and make COVID-19 data more comparable and exchangeable across studies and institutions, interoperable datasets are needed.

Various initiatives have started to define uniform datasets and Common Data Elements (CDEs) for the collection of information about COVID-19. For example, questionnaires and case report forms (CRFs) have been developed to collect data about COVID-19 patients in a standardized way [5, 8, 9]. While the CDEs defined in these projects are an important step, they are not enough to ensure interoperability. To make data syntactically and semantically interoperable, data elements have to be embedded in standard data structures that can be exchanged across IT systems; they also have to use common terminologies that unambiguously define the meaning of clinical concepts.

To improve interoperability of COVID-19 data, we developed the German Corona Consensus Dataset (GECCO), which uses international health IT standards and terminologies for interoperable data exchange. GECCO defines a compact set of data elements to be collected in COVID-19 studies and was developed within the German COVID-19 Research Network of University Medicine (“Netzwerk Universitätsmedizin”) funded by the German Federal Ministry of Education and Research (BMBF) [10]. This article provides an overview of the GECCO dataset and its development.

Combining the initial draft dataset and the additional proposals from the expert board, 702 potentially relevant data elements were collected. From these data elements and based on the prioritization of the expert board, the editorial team compiled a core dataset consisting of 81 elements with 281 response options. These data elements were grouped into the following categories: anamnesis/risk factors (n = 16); imaging (n = 2); demographics (n = 7); epidemiological factors (n = 1); complications (n = 1); onset of illness/admission (n = 1); laboratory values (n = 25); medication (n = 4); outcome at discharge (n = 3); study enrollment/inclusion criteria (n = 2); symptoms (n = 2); therapy (n = 6); vital signs (n = 11) (Fig. 2).

For all data elements and their corresponding response options, value sets were created using codes from SNOMED CT, LOINC, UCUM, ICD-10-GM and ATC. Data elements, response options and associated value sets of the GECCO dataset can be accessed on the ART-DECOR platform [28].

Subsequently, FHIR profiles were created for the data elements. The following FHIR resources were used to model the data elements: Patient, Consent, Observation, Condition, Procedure, Encounter, Medication and MedicationStatement. The FHIR profiles can be accessed on Simplifier [29].

During the consolidation process, it became clear that some data elements are important for certain disciplines but irrelevant for others. These elements were not included in the GECCO core dataset as they would have inflated the size of the dataset. The editorial team decided to include these data elements in domain-specific extension modules, which will be specified in more detail at later stages of the project.

In this report, we presented the GECCO dataset, a core collection of data elements for acquiring and exchanging information about COVID-19 patients. By using standardized data structures (HL7 FHIR profiles) and international terminologies, the GECCO dataset is an important step towards interoperability of COVID-19 research data. It can facilitate harmonized data collection and analysis across institutions and IT systems, for example in clinical studies, registries or digital health applications.

A key factor to the successful application of standard datasets like GECCO is a close collaboration with the scientific community. To ensure a high acceptance of the dataset, the development of GECCO therefore included clinicians from a wide variety of medical disciplines and professional associations as well as experts in digital health, standardization and clinical terminologies. GECCO also collaborates closely with standards developing organizations such as HL7 and Integrating the Healthcare Enterprise (IHE) as well as other initiatives aiming to improve health data interoperability, such as the Medical Informatics Initiative [12], NFDI4Health [30] and the Corona Component Standards (cocos) [31].

For the successful application of standard datasets like GECCO, it is also important that these datasets are embedded in larger infrastructures for secure and interoperable data sharing across institutions. Initiatives like, for example, the National COVID Cohort Collaborative (N3C) in the US [32], OpenSAFELY in the UK [33] or the international project Secure Collective Research (SCOR) [34] are developing platforms for a secure, cross-institutional analysis of COVID-19 data. Similarly, GECCO is part of the German COVID-19 Research Network of University Medicine [10], which aims to bundle the resources of German university hospitals to improve diagnostics and treatment of COVID-19 patients. The network also includes a research data infrastructure for the secure and interoperable data exchange across university hospitals [35], for which GECCO provides a standard data structure. For example, projects such as NAPKON, a national project for collecting research data during pandemics [36], will collect their data according to the specifications of the GECCO dataset.

Note that the GECCO dataset was originally developed for German university medicine. However, we used international standards and terminologies to ensure that the GECCO dataset can be extended to use cases beyond university medicine and also be applied in international contexts. When defining the FHIR profiles, we also took into account international work (for example the IPS [23]) to make sure that the GECCO dataset can provide interoperability when used internationally.

Although the GECCO dataset was designed to be as compact and manageable as possible, acquiring and recording the information for all data elements still requires time (for example, when entering the information in an electronic case report form). Moreover, manual documentation is prone to transcription errors. Conversely, manually abstracted and structured information from unstructured health records may provide relevant insights for care-providers and improve their understanding of risk and outcome. For some of the data items, it is therefore desirable to automatically exchange data between a GECCO-based study database and existing IT systems, such as hospital information systems or clinical trial software. This requires standard interfaces between these systems. The FHIR profiles of the GECCO dataset provide an interoperable, machine-readable data structure that can facilitate this data exchange across IT systems. For example, LOINC-coded information about patients’ laboratory values could be directly transferred from the hospital information system to the GECCO dataset. For electronic data capture (EDC) systems used in clinical studies, converters are currently being developed that transform the underlying software formats into the GECCO HL7 FHIR format for interoperable data exchange. Independently of the work presented here, the GECCO dataset has also been converted to the CDISC Operational Data Model (ODM) and is published on the Portal of Medical Data Models (MDM) of the University of Münster [37].

The aim of the GECCO dataset was to define a compact set of core data elements for which most COVID-19 studies (particularly the studies conducted at German university hospitals) can provide full information. However, if not explicitly required, studies that want to use the GECCO dataset may use subsets of the dataset if they are unable to provide full information for all data elements.

Scientific knowledge about COVID-19 and SARS-CoV-2 is changing fast, which may necessitate modifications to the GECCO dataset in the future. Furthermore, the use of the GECCO dataset in clinical research projects will provide practical experience that may also motivate changes to the dataset. To incorporate new knowledge into the dataset, the COVID-19 Research Network of University Medicine [10] will put a governance framework in place that will coordinate revisions and extensions to the dataset. Domain-specific extension modules are already in preparation, which include many of the data elements that were not considered essential for the core dataset. Extension modules currently planned are: laboratory, diagnostics, immunology, gynecology and pregnancy, epidemiology, pediatrics, intensive care, oncology, radiology, virology, psychiatry and neurology (these extension modules are also made accessible on the ART-DECOR platform [38]).

The GECCO dataset provides researchers and healthcare professionals with a compact, interoperable dataset for collecting, exchanging and analyzing COVID-19 data across institutions and software systems. Developed by a multidisciplinary group of experts, GECCO builds heavily on international terminologies and IT standards. GECCO can thus help to improve the harmonization and coordination of research efforts to successfully fight the COVID-19 pandemic. Future inclusion of domain-specific extension modules will further expand the use of the GECCO dataset.