Fig. 4: Extent of bias in symptomatic vaccine effectiveness (symptomatic VE) estimates and the epidemic size by testing scenarios and levels of vaccine efficacy against susceptibility (VEsusc).
The extent of bias—how much symptomatic VE estimates diverged from the vaccine’s “true” protection against symptomatic infection—is shown for estimates derived from the retrospective cohort design (VERR; (a)) and the retrospective test-negative design (VEOR; (c)). VERR and VEOR were calculated from studies conducted at the point of highest positive epidemic growth for each epidemic (i.e., the first inflection point), across three testing scenarios (green): equal testing by vaccination status, moderately unequal testing (vaccinated with 1.76 times higher testing), and highly unequal testing (vaccinated with 2.36 times higher testing). The epidemic size—the cumulative proportions of individuals with symptomatic SARS-CoV-2 who were unvaccinated (yellow) and vaccinated (dark gold) (b)—was also recorded at that time point. Simulated epidemics with fewer than 0.1% cumulative infections were removed. Each boxplot shows the median (centre line), the interquartile range (IQR; the box spans the 25th and 75th percentiles), and whiskers that extend to the most extreme data points within 1.5 times the IQR. The violin plot outlines illustrate the kernel probability density. Boxplots and violin plots are based on n = 100 epidemic realisations per testing scenario and level of VEsusc, except when VEsusc = 0.9 (n = 97). The median time to the point of highest positive epidemic growth by the level of VEsusc was t = 50 for 0.1, t = 58 for 0.3, t = 68 for 0.5, t = 92 for 0.7, and t = 185 for 0.9.
