Defining and Estimating Outcomes Directly Averted by a Vaccination Program when Rollout Occurs Over Time

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Accurate estimation of vaccine effectiveness during pandemics like COVID-19 is crucial for public health.
• Previous studies often estimated "directly averted" deaths, potentially underestimating the total impact by not fully accounting for indirect effects.

Purpose of the Study:

• To define a causal estimand for "directly averted" outcomes by vaccination, representing the impact on vaccinated individuals while holding coverage constant.
• To develop an unbiased estimator for this causal estimand in randomized controlled trials (RCTs).
• To analyze the bias of the commonly used "hazard difference" estimator.

Main Methods:

• Causal inference framework to define "directly averted" outcomes.
• Development of an unbiased estimator for a one-stage RCT.
• Simulation studies to compare the proposed estimator with the "hazard difference" estimator.

Main Results:

• The "hazard difference" estimator is biased in RCTs if vaccination has a non-null effect, due to failure to account for "depletion of susceptibles" among the unvaccinated.
• Overestimation by the "hazard difference" estimator is minimal for deaths with low infection-fatality rates but can be substantial for infections with high vaccine efficacy and high transmissibility.
• Defined and compared "avertible" outcomes (outcomes that could have been averted but were not due to lack of vaccination).

Conclusions:

• The proposed causal estimand provides a lower bound for total vaccine-averted outcomes.
• The "hazard difference" estimator can lead to significant overestimation, particularly for averted infections.
• Further research is needed to identify causal estimands in observational studies.