JMIR Publications

ABSTRACT

Surveillance data are essential for public health initiatives; however, missing data is often a challenge, impacting the ability to assess accurate vaccine coverage by introducing bias, particularly when addressing disparities. While methods like multiple imputation using chain equations (MICE) are robust, they can be computationally expensive when applied to large datasets. We explored the use of the machine learning techniques Iterative-Imputer and miceforest and cloud-based computing to reconcile missing demographic data. 2021-2022 flu vaccination and demographic data came from the WV Immunization Information System (N=2,302,036) where race (15%) and ethnicity (34%) were missing. We utilized MICE, Iterative-Imputer, and miceforest, where we jointly imputed missing variables and created 15 datasets each. After imputations, we obtained an additional 780,339 observations compared to the complete case. MICE and miceforest best preserved the proportional distribution of demographics relative to the complete case. MICE required 14 hours to complete 15 imputations, while Iterative-Imputer took 2 minutes and miceforest took in 10 minutes to complete the same number of imputations. After applying post-imputation estimates to flu data, vaccination coverage rates dropped between 0.87-18%. Utilizing miceforest to reconcile missing demographic data poses as a potential solution, offering a flexible, fast, and iterative approach that can improve data completeness while preserving underlying distributions and mitigating potential bias. By offloading resource-intensive tasks to a cloud-based server, public health officials can mitigate processing constraints, enabling parallel execution of multiple tasks while minimizing downtime. This enhanced efficiency accelerates analyses, facilitates culturally responsive decision-making, and optimizes organizational performance, ultimately improving communication and productivity by eliminating prolonged processing delays.