Estimating the population-level impact of pneumococcal conjugate vaccines

Introducing pneumococcal conjugate vaccines (PCVs) into childhood immunization schedules has substantially reduced pneumococcal diseases globally. PCVs offer direct protection against colonization with vaccine-targeted serotypes (VT), reducing transmission and leading to indirect protection. This dissertation aimed to analyze the population-level impact of PCVs and the influencing factors through statistical and mathematical modeling. Estimating the population-level impact of PCVs faces challenges in finding appropriate control groups due to confounding. A potential solution involves comparing the current disease burden with the counterfactual disease burden without PCVs. Various statistical methods, including interrupted time series (ITS), synthetic control with Bayesian variable selection (SC), and seasonal-trend decomposition using LOESS plus principal component analysis (STL + PCA), have been employed to model the counterfactual. In my dissertation's first project, I compared a novel approach using LASSO regression in a synthetic control model with established methods through simulations, finding both the LASSO approach and SC to be accurate and reliable in estimating vaccine impact. When applied to real-world data from Chile, Ecuador, Mexico, and the US, these models demonstrated variability in PCV impact on pneumonia hospitalization across age groups and countries. The population-level impact of PCVs is influenced by factors driving vaccine direct effect (like vaccine valency and host age) and population factors (like vaccine coverage and social contact structure). In my dissertation's second project, I developed an age-structured Susceptible–Colonized transmission model, verified against colonization data from children in France, England and Wales, Alaska, and Massachusetts. With this model, I simulated the time needed to eliminate VT colonization after introducing PCVs and found a range of 3.8–6 years using contact matrices from 34 countries while holding all other parameters constant. Through regression analysis, I discovered that contact rate and assortativity in children under 5, who had the highest colonization prevalence, significantly predicted time-to-elimination. While most of the contact matrices used in our study represent high-income, low-transmission settings and our findings require further research for broader applicability, the results nevertheless show that social contact structure can drive variability in the population-level vaccine impact. The first project added to the toolkit of vaccine impact evaluation and uncovered the variability in PCV impact across populations while the second explored social contact structure as a potential driver of this variability. Together, these projects highlight the variability in the population-level impact of PCVs and the need to consider population factors, such as social contact structure, when assessing vaccine impact.