Addressing Issues and Challenges in the Real-World Evaluation of Immunisation Strategies

Estimating the effectiveness of vaccines and other immunisation interventions, and quantifying the public health impact of their implementation, are central tasks in modern epidemiology. While randomised trials provide crucial evidence regarding the efficacy and safety of immunisation prior to implementation, public health decisions also necessitate evidence obtained under routine conditions, in heterogeneous populations and within evolving epidemiological contexts. This dissertation explores the real-world evaluation of immunisation strategies, with vaccine effectiveness as its primary focus, paying particular attention to the methodological issues that arise when routine surveillance, vaccination registries, hospital records, administrative archives, and primary care databases are used to inform public health decisions. The document presents a practice-driven methodological narrative based on five applied studies conducted in operational public health settings. While these studies address different viral infection settings, populations, endpoints and data sources, they all share the same inferential problem: how to estimate the effectiveness and impact of preventive strategies in the absence of random allocation. The methodological focus is therefore on aligning epidemiological questions with study design, exposure definition, outcome ascertainment, confounding control, temporal structure and statistical modelling. Rather than presenting the included studies as isolated applications, this dissertation interprets them as successive contributions to the more general issue of generating credible and interpretable real-world evidence. The first study used national surveillance data and vaccination coverage information to estimate the number of SARS-CoV-2 infections, hospitalisations, intensive care admissions and deaths averted by the Italian vaccination programme in 2021. The second study evaluated the impact of rotavirus vaccination on paediatric gastroenteritis hospitalisations in Italy, taking into account long-term trends, seasonality, and heterogeneous regional implementation of the programme. The third study estimated the relative effectiveness of the bivalent Original/Omicron BA.4-5 booster against severe cases of COVID-19 during the circulation of Omicron XBB sublineages, paying particular attention to the time elapsed since vaccination and the resulting waning protection (i.e., the gradual decline in protection over time). The fourth study compared multivariable regression and propensity score-based approaches for estimating relative vaccine effectiveness, examining how alternative adjustment strategies affect covariate balance, overlap, and interpretation. The fifth study extended this framework to passive immunisation by evaluating the effectiveness of the Nirsevimab programme against bronchiolitis in primary care. Unlike vaccines, which induce an active immune response, Nirsevimab is a long-acting monoclonal antibody that provides passive protection against respiratory syncytial virus. This study used regional real-world data and analytical strategies designed to address confounding factors, time-dependent exposure, clustering, and incomplete covariate information. These applications demonstrate that real-world evaluation of vaccines and other immunisation strategies requires more than applying a statistical model to routinely collected data. Credible estimates depend on constructing meaningful comparison groups, correctly aligning eligibility, exposure and follow-up, distinguishing betweencalendar time and time since administration, and defining outcomes operationally. The studies also demonstrate the distinction between individual-level effectiveness and population-level impact. Individual-level analyses estimate protection among comparable subjects, whereas population-level analyses quantify the burden prevented under actual implementation conditions. Both perspectives are necessary for public health decision-making, but they address different issues and require different assumptions. The dissertation makes both an applied and a methodological contribution. In terms of substance, it provides evidence of the value of vaccination and passive immunisation strategies in reducing the burden of viral infections in Italy. In terms of methodology, it clarifies how design choices, data structure, and modelling assumptions influence the interpretation of real-world estimates. Particular emphasis is placed on confounding, time-dependent exposure, waning immunity, implementation delays, outcome definition, use of routine data, heterogeneity, missing information and causal interpretation. The dissertation therefore supports the role of real-world evidence as an important resource for public health decision-making. Routine data can inform decisions regarding programme effectiveness, impact and implementation, provided they are analysed within a coherent and transparent framework, as resulting estimates may be highly sensitive to the underlying assumptions.