Scholarlink

Background <p>COVID-19 vaccination has prevented substantial morbidity and mortality in the United States, yet its uptake remains an issue, with similar challenges persisting for seasonal flu vaccination. Public trust in federal statistics and federal institutions may influence vaccination behavior, but its role in predicting COVID-19 vaccine uptake has not been extensively evaluated.</p> Methods <p>We analyzed data from three waves of the United States Census Bureau Household Pulse Survey. The outcome was COVID-19 vaccine uptake for the 2024 to 2025 season. Predictors included sociodemographic characteristics, health insurance, flu vaccine uptake, and three measures of trust in federal statistics. Missing data were addressed using multiple imputations. Data from October to December 2024 were used for training, and data from February to March 2025 served as the test set. Five supervised learning models were tuned using cross validation to maximize the area under the curve. Model performance and calibration were evaluated on the test set, and feature importance was assessed using SHAP (SHapley Additive exPlanations) values.</p> Results <p>Models performed overall with area under the curve values between 0.867 and 0.895. The model achieved discrimination. SHAP analyses identified flu vaccination uptake, age, number of children, and trust in federal statistics as the strongest predictors of COVID-19 vaccine uptake.</p> Conclusion <p>Machine learning models predicted COVID-19 vaccine uptake in a national sample. Flu vaccination behavior, age, number of children, and trust in federal statistics emerged as key predictors. These findings may help inform population-level outreach strategies, while recognizing that the results reflect associations rather than causal effects.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative evaluation of machine learning models for predicting COVID-19 vaccine uptake in U.S. adults

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Nathaniel J. Maxey,
  • Taliyah S. Griffin,
  • Plamena P. Powla,
  • Felix M. Pabon-Rodriguez

摘要

Background

COVID-19 vaccination has prevented substantial morbidity and mortality in the United States, yet its uptake remains an issue, with similar challenges persisting for seasonal flu vaccination. Public trust in federal statistics and federal institutions may influence vaccination behavior, but its role in predicting COVID-19 vaccine uptake has not been extensively evaluated.

Methods

We analyzed data from three waves of the United States Census Bureau Household Pulse Survey. The outcome was COVID-19 vaccine uptake for the 2024 to 2025 season. Predictors included sociodemographic characteristics, health insurance, flu vaccine uptake, and three measures of trust in federal statistics. Missing data were addressed using multiple imputations. Data from October to December 2024 were used for training, and data from February to March 2025 served as the test set. Five supervised learning models were tuned using cross validation to maximize the area under the curve. Model performance and calibration were evaluated on the test set, and feature importance was assessed using SHAP (SHapley Additive exPlanations) values.

Results

Models performed overall with area under the curve values between 0.867 and 0.895. The model achieved discrimination. SHAP analyses identified flu vaccination uptake, age, number of children, and trust in federal statistics as the strongest predictors of COVID-19 vaccine uptake.

Conclusion

Machine learning models predicted COVID-19 vaccine uptake in a national sample. Flu vaccination behavior, age, number of children, and trust in federal statistics emerged as key predictors. These findings may help inform population-level outreach strategies, while recognizing that the results reflect associations rather than causal effects.