Georgios Varotsis: Vaccination, Pharmacovigilance and Surveillance in Times of Pandemic: Epidemiological Evidence from Routine and Participatory Health Data

Abstract

Pandemics require rapid public health decisions, often before purpose-built research data is available. Decisions may therefore rely on routine healthcare, administrative, surveillance and public-participation data. The aim of this thesis was to examine how routine and participatory health data can generate epidemiological evidence for vaccination delivery, vaccine safety monitoring and infectious disease surveillance during the COVID-19 pandemic. 

Study I used Swedish population registers and vaccination records to evaluate whether mailed letters with pre-booked COVID-19 vaccination appointments increased first-dose uptake compared with SMS prompts to self-book in Uppsala County. A regression discontinuity analysis based on an administrative birth-year threshold included 96,194 adults. Pre-booked appointments increased uptake by 2 percentage points, from approximately 92% to 94% (odds ratio 1.30). 

Study II evaluated free-text symptom narratives from the UK ZOE COVID Study app for vaccine pharmacovigilance. Overall, 1,615,139 narratives from 452,144 vaccinated participants were embedded with a sentence-transformer model and clustered with a hierarchical clustering algorithm. The analysis identified 46 clusters reported more frequently after vaccination and 15 that differed between participants vaccinated with two different vaccines. Most findings corresponded to known adverse reactions; others captured experiences less routinely represented in standard symptom categories, including menstrual changes and sensory symptoms. 

Study III evaluated Swedish 1177 medical helpline data for population-level COVID-19 surveillance by linking calls to population registers and SARS-CoV-2 PCR test data. In 2020, 18.8% of residents aged ≥15 years had at least one answered call registered. A prediction model trained among callers with linked PCR results was applied to estimate COVID-19 probability over time. Aggregated and calibrated estimates tracked major epidemic waves through much of 2021, but performance declined during the Omicron wave, demonstrating the need for repeated evaluation as epidemic conditions change. 

Together, the studies show that routine and participatory health data can support pandemic evidence generation. They highlight the value of reducing administrative friction in vaccination delivery, using participant narratives to identify candidate safety signals, and using selected healthcare-contact data for surveillance, while recognising limitations arising from selection, measurement, reporting and transportability.