Scholarlink

<p>Forecasting the demand for diphtheria, tetanus, and pertussis (DTaP) vaccines is crucial to efficiently allocating resources, reducing vaccine waste, and improving health outcomes, particularly in rural and underprivileged areas. Challenges specific to these areas include limited access to healthcare facilities, inconsistent supply chains, and a lack of data on vaccination rates and patient demographics. This study presents a predictive modeling framework that is capable of improving resource planning at a federally qualified health center (FQHC) in upstate New York by better predicting demand for the DTaP vaccine from the rural population it serves. Using a large dataset of 160,289 patient visits from 2021 to 2023, patient demographic, clinical, insurance, visit-specific, and temporal features&#xa0;were extracted and incorporated into multiple machine learning (ML) models. Multiple ML algorithms—such as XGBoost, CatBoost, and histogram-based gradient boosting—&#xa0;were benchmarked&#xa0;against widely used baseline models including random forests, logistic regression, and decision trees. CatBoost was the top-performing model, with a receiver operating characteristic–area under the curve (ROC-AUC) score of 99.8% and an F1 score of 89.0% at a decision threshold of 67.2%. Seasonal decomposition revealed peaks during back-to-school periods, highlighting temporal dynamics. In addition, SHapley Additive exPlanations (SHAP) analysis improved model interpretability, allowing the identification of age, insurance type, visit type, and previous immunization history as the key predictors that enhance the models’ performance. The predictive modeling framework delivers very accurate forecasts and actionable insights while opening black-box models to guide immunization strategies and optimize resource planning by the existing healthcare system.</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

Integrating machine learning models and explainable AI to predict DTaP vaccine demand in rural primary care

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月29日

查看全文
  • Osamah Yaeesh,
  • Shoog Nimri,
  • Yong Wang,
  • Shaw-Ree Chen,
  • Karen Kinter,
  • Danielle Renodin-Mead,
  • Mohammad T. Khasawneh

摘要

Forecasting the demand for diphtheria, tetanus, and pertussis (DTaP) vaccines is crucial to efficiently allocating resources, reducing vaccine waste, and improving health outcomes, particularly in rural and underprivileged areas. Challenges specific to these areas include limited access to healthcare facilities, inconsistent supply chains, and a lack of data on vaccination rates and patient demographics. This study presents a predictive modeling framework that is capable of improving resource planning at a federally qualified health center (FQHC) in upstate New York by better predicting demand for the DTaP vaccine from the rural population it serves. Using a large dataset of 160,289 patient visits from 2021 to 2023, patient demographic, clinical, insurance, visit-specific, and temporal features were extracted and incorporated into multiple machine learning (ML) models. Multiple ML algorithms—such as XGBoost, CatBoost, and histogram-based gradient boosting— were benchmarked against widely used baseline models including random forests, logistic regression, and decision trees. CatBoost was the top-performing model, with a receiver operating characteristic–area under the curve (ROC-AUC) score of 99.8% and an F1 score of 89.0% at a decision threshold of 67.2%. Seasonal decomposition revealed peaks during back-to-school periods, highlighting temporal dynamics. In addition, SHapley Additive exPlanations (SHAP) analysis improved model interpretability, allowing the identification of age, insurance type, visit type, and previous immunization history as the key predictors that enhance the models’ performance. The predictive modeling framework delivers very accurate forecasts and actionable insights while opening black-box models to guide immunization strategies and optimize resource planning by the existing healthcare system.