US COVID-19 VACCINE HESITANCY: A MACHINE LEARNING APPROACH TO POLICY SOLUTIONS

Abstract

Vaccines are one of the greatest triumphs of biomedical science and public health. However, hesitancy to vaccines hinders their benefits, both historically and especially during the COVID-19 pandemic. Drivers of individual decisions leading to COVID-19 vaccine hesitancy are complex, but recent survey data sets from the Census regarding pandemic experiences open a new window to identifying groups of vaccine hesitant individuals to target with policy initiatives. This study employs machine learning to explore whether US COVID-19 vaccine hesitancy can be predicted well with survey data, discover which characteristics and pandemic behaviors predict vaccine hesitancy best, and propose policy informed by the results of the models. The analysis finds that random forest classifiers using imputed survey data predict best for COVID-19 vaccine hesitancy, producing AUC values from 0.79 to 0.81. The results suggest using survey data allows for good predictions, as the performances are in line with and even exceed models using patient level indicators to predict vaccine hesitancy. The analysis finds that among the features included in the survey, age, previous contraction of COVID-19, education level, and food insecurity are the most powerful predictors of COVID-19 vaccine hesitancy. The analysis proposes targeted policy interventions for Americans who have recently tested positive for COVID-19 and for Americans who are food insecure. The model and policy interventions aim to reduce vaccine hesitancy across the US and hopefully contribute to an end to the pandemic, which has altered so many lives.