VPB0141 - A near real-time surveillance algorithm to identify in-hospital VITT cases

Background: In spring 2021, various groups described Vaccine-induced immune thrombotic thrombocytopenia (VITT) a rare thrombotic syndrome associated with COVID-19 vaccine ChAdOx1-S (incidence 1:100,000 exposures).
In the UK, reporting at a national level depended on specialists reporting cases through a variety of mechanisms, as national datafeeds were insufficiently granular to identify the VITT phenotype. Real-world data analytics of EHRs to identify vaccine complications could strengthen national reporting.

Aims: We developed an algorithm to define VITT using a regional hospital and primary care data system. Secondary aims included a reporting dashboard for specialists and generalisable methods for identifying rare events in admitted inpatients.

Methods: A linked primary-secondary care, near real-time datafeed of patients admitted to Barts Health NHS Trust covering a 2.2 M population. An algorithm was developed to identify clinical features of VITT and categorise patients by likelihood of VITT, using SNOMED-coded admission, vaccine and laboratory data (Table 1). Clinical validation of the algorithm’s output was completed.

Results: The algorithm identified 698 cases in routine data within Barts Health [definite (n=1) probable (n=10) and possible (n=687)]. 91 patients were validated. The algorithm had low precision for defining separate VITT categories (9.1% ‘definite and probable’ and 32% ‘definite, probable and possible’) but identified the one definite case of VITT. Inter-validator agreement was moderate (Randolph’s kappa 0.45). All cases identified clinically and included in the full dataset were also found by the algorithm (high recall).

Dashboard design for care teams to flag future VITT cases for hospital in-patients will be presented.

Conclusion(s): We have successfully demonstrated identification of thrombotic complications from COVID-19 vaccines in EHR data. Alternative diagnosis trends could improve precision of the algorithm. Barriers to progress are lack of pathology data standards and complex approvals for data sharing. This study was supported by HDR-UK.