Toward accurate prediction of pediatric epidemic disease patient volume in the Chaoshan region: A deep learning framework

TL;DR

A deep learning model called LSTM-BEATS accurately predicts pediatric disease patient volume in Chaoshan, helping hospitals prepare for outbreaks.

Contribution

LSTM-BEATS improves pediatric epidemic prediction by capturing local dynamics and outperforms existing methods in accuracy.

Findings

LSTM-BEATS achieved a 6.12% improvement in average correlation coefficient over other models.

The model achieved the lowest RMSE (0.130) and MASE (0.610) in predictions.

It supports proactive planning for staffing and medication during influenza surges.

Abstract

Accurate prediction of pediatric epidemic infectious diseases is critical for effective prevention and personalized treatment. Herein, we developed a deep learning framework for the epidemiological characteristics of the Chaoshan region, using electronic health records data from 278,506 pediatric outpatient and emergency visits at the Second Affiliated Hospital of Shantou University Medical College between 2017 and 2023. Our framework is designed to learn pediatric representations that capture local epidemic dynamics and to meet regional clinical prediction needs. Results demonstrate that the framework achieves strong predictive performance on the regional dataset. Our framework yields at least a 6.12% improvement over its counterparts in terms of average correlation coefficient; it achieves the lowest errors in both root-mean-square error (RMSE = 0.130) and mean absolute scaled error (MASE = 0.610). Our framework provided targeted decision support for local healthcare institutions in workforce allocation, medication, and supply planning, thereby contributing to improved prevention strategies.

•LSTM-BEATS learns region-specific pediatric epidemic dynamics from 2017 to 2023 Chaoshan EHR•LSTM-BEATS with trend seasonality modules and self-attention improves multivariate prediction•Achieved lowest errors, RMSE 0.130 and MASE 0.610 with higher correlations•Predictions support proactive staffing and medication planning during influenza-season surges

LSTM-BEATS learns region-specific pediatric epidemic dynamics from 2017 to 2023 Chaoshan EHR

LSTM-BEATS with trend seasonality modules and self-attention improves multivariate prediction

Achieved lowest errors, RMSE 0.130 and MASE 0.610 with higher correlations

Predictions support proactive staffing and medication planning during influenza-season surges

Wang et al. develop a deep learning framework (LSTM-BEATS) that accurately predicts pediatric epidemic patient volume in the Chaoshan region by capturing local epidemiological dynamics. This model outperforms traditional methods and enables proactive resource planning—helping hospitals optimize staffing and medical supply allocation ahead of seasonal outbreaks

Figures