Scholarlink

COVID-19 disrupted time series trends in Emergency Departments (ED), causing a notable decrease in patient visits. Many studies focus on pre-pandemic or pandemic-specific data, raising concerns about models trained on altered reality. This paper proposes original strategies for patient visit post-pandemic forecasts. The proposed study couples deep learning models, such as LSTM or CNN, to pandemic-included preprocessing data to reduce accuracy loss. The first approach incorporates explicit COVID-19 features labeling the data in the deep model used for forcecasting. In the second approach, the COVID anomaly period is manually excluded for the data and aggregated pre-COVID and post-COVID data are used. To illustrate the study, we assess the models using a dataset from a public French hospital, acknowledging challenges in time series forecasting. Results reveal that the proposed strategies achieve high performance, showcasing forecasting accuracies with MAPE values of 7.26% and 7.92% for the first strategy, and 8.55% and 8.24% for the second strategy.

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

Patient Visits Forecasting in the Post-pandemic Era at Emergency Departments

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Nicolas Haxaire,
  • Farah Mourad-Chehade,
  • Alice Yalaoui

摘要

COVID-19 disrupted time series trends in Emergency Departments (ED), causing a notable decrease in patient visits. Many studies focus on pre-pandemic or pandemic-specific data, raising concerns about models trained on altered reality. This paper proposes original strategies for patient visit post-pandemic forecasts. The proposed study couples deep learning models, such as LSTM or CNN, to pandemic-included preprocessing data to reduce accuracy loss. The first approach incorporates explicit COVID-19 features labeling the data in the deep model used for forcecasting. In the second approach, the COVID anomaly period is manually excluded for the data and aggregated pre-COVID and post-COVID data are used. To illustrate the study, we assess the models using a dataset from a public French hospital, acknowledging challenges in time series forecasting. Results reveal that the proposed strategies achieve high performance, showcasing forecasting accuracies with MAPE values of 7.26% and 7.92% for the first strategy, and 8.55% and 8.24% for the second strategy.