Using machine learning to identify COVID-19 vaccine-hesitancy predictors in the USA.

Introduction: Vaccine hesitancy is complex, multicausative phenomenon that undermines public health efforts to contain the spread of infectious diseases. Improving our understanding of the drivers of vaccine hesitancy might improve our capacity to address it. Methods: We used the results of the May 2021 Assistant Secretary for Planning and Evaluation's survey on COVID-19 vaccine hesitancy, which estimated the proportion of adults for every US county that felt either hesitant or unsure and strongly hesitant towards taking the COVID-19 vaccine when it becomes available. We developed a prediction model to identify the most important predictors of vaccine hesitancy. The potential predictors included demographic characteristics, the Centers for Disease Control and Prevention's Social Vulnerability Index and the Republican Party's voting share in the 2020 presidential election as a proxy of political affiliation, all at the county level. Results: We found that the main drivers of vaccine hesitancy are income level, marital status, poverty, income, schooling, race/ethnicity, age, health insurance status and political affiliation. While the drivers are shared by both levels of hesitancy, the order changes. Particularly, political affiliation is a more important predictor for strong hesitancy than for hesitancy or unsure. Conclusions: These results deepen our understanding of the phenomenon and could help design more targeted interventions to reduce hesitancy in specific subgroups of the population.