We used the OMOP ATHENA online browser to identify mappings to SNOMED CT, Read v2 terminology, and the International Classification of Disease Tenth Revision (ICD-10). We reported vaccine exposure using SNOMED CT for vaccine administration and Dictionary of Medicines and Devices (dm+d; National Health Service [NHS]) codes for vaccine prescriptions; dm+d is a dictionary of descriptions and codes which represent medicines and devices in use across the NHS. Each national team validated these mappings and any differences between the terms they would use to represent each concept discussed with a decision made by consensus. We then used to create monthly reports of vaccine coverage and AEIs. We elected to use CVST and anaphylaxis as demonstration AEIs [16].

The data were drawn from the data sources of the 4 DaC-VaP partners.

We performed repeated cross-sectional reports of the incidence of the AEIs in the vaccinated population in a single time interval postvaccination.

DaC-VaP partners ran cross-sectional studies for the previous 28 days.

The first search started on December 8, 2020 (first dose of the Pfizer vaccine given in the United Kingdom), and the second search on January 5, 2021. These ran in 28-day intervals (February 2, March 2, March 30, April 27-August 17, 2021).

The cross sections included all individuals registered with general practices on the date of vaccination and remained registered for 28 days. The outputs were reported in the following age bands: <16 years old, 16-39 years, 40-64 years, and 65 years and older. Mortality in the postvaccination period were also reported for those with AEIs.

We reported by vaccine brand, including reporting unknown vaccines. We presumed that the unknown vaccine brand for December 2020 was Pfizer-BioNTech, as Oxford-AstraZeneca was unavailable until January 2021 and other vaccine types later.

We aimed to include statistical reporting and disproportional analysis metrics, a proportional reporting ratio (PRR), and a reporting odds ratio (ROR). We used a Bayesian method that provided a framework to combine prior information/knowledge and data to account for conceptual transparency. Our aim was to use IC = log₂ (observed + 1/2)/(expected + 1/2).

The DaC-VaP collaborators had individual ethical control of their data. No data were reported that might risk identifying individuals. Where less than 5 individuals were in a group, this was reported as <5. This study aims to demonstrate the potential of the DaC-VaP collaboration to report outcomes of interest.

We initially reported whether the data items or clinical concepts required for the study existed within the OMOP CDM and whether there were mappings to SNOMED CT, Read v2, or dm+d (Table 1).

We then reported the components by terminology (Table 2). Age did not map to the Read v2 terminology, but this is of no practical significance. We noted that the SES only exists as a generic concept in OMOP and that a custom mapping would be required. There was no mapping of vaccine dose (first or second) to Read v2. However, like age, this would not be practically important as these data are well ordered in the DaC-VaP collaborators’ data sources.

Of most importance are the AEIs. For CVST, the specific concept exists within OMOP, and it also exists in SNOMED CT: SNOMED concept IDs 95455008 and19522900 from CVST (concept ID 195229008). For anaphylaxis, SNOMED CT and the ATHENA online browser showed 130 and 161 items, respectively. SNOMED CT and OMOP had 16 and 15 items, respectively. Read v2 codes were generic for CVST and anaphylaxis. Of these, those relevant to vaccination are shown in Tables 2, 3, 4, and 5 for England, Wales, Scotland, and Nothern Ireland, respectively.

COVID-19 vaccine exposure was well recorded with dm+d and SNOMED CT. The vaccine unsurprisingly was listed as a VTM at the top of the drug dictionary hierarchy, with VMPs created for each vaccine type. There were virtual and actual packs and products to match the vaccines available. Additional administration and vaccine-type clinical terms were also within SNOMED CT (Table 6). Finally, we found a small number of vaccine administration codes within dm+d that were not mapped to OMOP.

This study shows a mapping method to identify codes relevant to CVST and anaphylaxis using the OMOP CDM to link common concepts required for COVID-19 vaccine pharmacovigilance to different terminologies relevant to the United Kingdom. All our predefined concepts were represented in the OMOP CDM. However, some, such as SES, did not have specific mappings and, thus, custom mappings would need development. We noted that local codes and curation of variables may be used to enable specificity where the concepts are less granular, especially for CVST.

The OMOP CDM may be suboptimal to overcome the limitation in the granularity of the coding systems used for AEIs. As well as being less granular, the Read v2 terminology has not been updated formally since April 2016, so local adaptions have been undertaken in the developed UK nations to enable new conditions and treatments, such as COVID-19 and vaccination, to be recorded.

Conventionally, CDMs, such as OMOP, are used by each database, mapping data and querying them using the script created by 1 of the teams. The cataloguing is carried out using applications such as White Rabbit and Rabbit in a Hat (Figure 1).

The OMOP CDM provides a framework for capturing patient demographic and socioeconomic characteristics, varying vaccine exposure [23], and AEI data. Although others could replicate our approach, avoiding the need to map whole databases, initial findings reflect the relative size of the terminologies we use, and their granularity needs improvement. Therefore, to conduct future research, concepts will require localization to evaluate COVID-19 vaccination associated with CVST and anaphylaxis. We selected dm+d rather than the better-known British National Formulary (BNF), although the latter is mapped to SNOMED CT. Its limitations are that it only lists prescribable drugs. Its chapter headings change from time to time, and it is not mapped to the ATHENA OMOP hierarchy. We considered MedDRA with clincically validated medical terminologies for clinical conditions, medical devices, and medicines that is commonly used to share AEIs mainly with regulators. MeDRA is primaily used to report pharmacovigilance using acute care and clinical trial data. This approach cannot be directly used within primary care but will be considered for future studies.

Concept mapping to a large number of terminologies, such as within OMOP and its ATHENA online browser, are usable and valuable for those conducting studies that draw together heterogeneous data to perform pooled analyses. Comprehensive mappings have to set a level of granularity that may be more or less specific than the terminologies they map to. Clinical variable curation at a local database level would prove useful to address issues around granularity. This would allow local expert refinement of the mappings that could be used by others looking to do a limited pooled analysis of a small number of clinical concepts. The interconnectivity of the pooled analysis may also support the MHRA’s Sponstaneous Report System used for optimizing patient safety.