Using regression tree analysis to examine demographic and geographic characteristics of COVID-19 vaccination trends over time, United States, May 2021–April 2022, National Immunization Survey Adult COVID Module

Abstract

Using data from the nationally representative National Immunization Survey (NIS), we applied conditional linear regression tree methodology to examine relationships between demographic and geographic factors and propensity of receiving various doses of COVID-19 vaccine over time; these analyses identified temporal changes in these relationships that heretofore had not been identified using conventional logistical regression methodologies.

Three regression tree models were built using an R package, Recursive Partitioning for Modeling Survey (rpms), to examine propensities over time of receiving a (1) first dose of a two-dose COVID-19 mRNA primary vaccination series or single dose of the Janssen vaccine (vaccine initiation), (2) primary series completion, and (3) monovalent booster dose, using a conditional linear effect model. Persons ≥50 years were more likely to complete a primary series and receive a first booster dose; persons reporting having received non-COVID-19 vaccines recently were more likely to initiate vaccination, complete the primary series, and get a first booster dose; persons reporting having work or school requirements were more likely to complete the primary series. Persons not reporting having received non-COVID-19 vaccines in 2 years but reporting having work or school vaccination requirements were more likely to initiate vaccination than those without work/school requirements. Among persons not reporting having received non-COVID-19 vaccines in 2 years and not reporting having work or school vaccination requirements, those aged ≥50 years were more likely to initiate vaccination than were younger adults. Propensity of receiving various doses was correlated with age, having recently received non-COVID 19 vaccines, and having vaccination requirements at work or school.

Regression tree methodology enabled modeling of different COVID-19 vaccination dose propensities as a linear effect of time, revealed changes in relationships over time between demographic factors and propensity of receipt of different doses, and identified populations that may benefit from vaccination outreach efforts.