(A-337) Dependence of the Data Source on the Positive Predictive Value in the Retrospective Case Validation of Thrombosis and Thrombocytopenia Syndrome cases using Real-World Data

Background: Adverse events following COVID-19 vaccination, included Thrombosis with Thrombocytopenia Syndrome (TTS). The Brighton Collaboration (BC) case definition for TTS was based on clinical presentation, imaging, investigatory findings, and laboratory results. This definition was intended for use in clinical trials and post-licensing settings to allow Level Of Certainty (LOC) of cases to be determined. The Positive Predicted Value (PPV) of LOCs can estimate the validity of cases identified in Real-World Data (RWD) sources.

Objectives: To determine the PPV of TTS cases using the BC case definition across various RWD sources, while also assessing the impact of data source type on PPV. Secondary objectives were to establish whether risk windows, or publication of the signal of TTS, affected the PPV of cases.

Methods: As part of a global project, co-ordinated by the Global Vaccine Data Network (GVDN) in collaboration with Vaccine Monitoring Collaboration for Europe (VAC4EU), the association of TTS with COVID-19 vaccination was assessed. Sites (across Spain and the United Kingdom) contributed to this study and validated potential TTS cases.

Cases were assessed based on diagnostic codes from electronic healthcare records between 1 December 2020 to 31 October 2022. Data extraction to support validation was collated via Research Electronic Data CAPture (REDCap), a web-based platform for data capture hosted by GVDN. LOC of diagnosis (based on the BC case definition) was determined by an algorithm created by GVDN, where LOC1 or 2 are definite cases and 5 is a non-case. To determine the accuracy of LOCs, PPVs were calculated, with Yates’ continuity correction. Sensitivity analyses were also performed. Sub-group analyses were conducted to fulfil the secondary objectives. Analysis was conducted in RStudio using R 4.4.1.

Results: 187 TTS cases were identified, of which LOC was assessed in 136 cases, CPRD cases were not validated due to lack of information in CPRD Aurum to fulfil the BC definition. The PPV ranged from 0-65% for LOC1-2 cases. Sites with access to hospital records achieved higher PPVs for LOC1-2 cases compared to sites with primary care data, where a maximum of LOC4 was achieved. PPVs from hospital records were highest outside of vaccination risk windows (70% [95%CI: 47.0-85.9]) and prior to publication of the TTS signal (100% [95% CI: 19.8-100.0]).

Conclusions: The PPV of TTS case validation was highly variable. Validation of cases using the BC case definition required the availability of highly granular data; from clinical observations, scans, laboratory findings as well as prior medical history and medications, the availability of which was highly variable between different healthcare settings and countries.