Scholarlink

Objective <p>To develop and validate a Seasonal Auto-Regressive Integrated Moving Average (SARIMA) model for forecasting daily patient admissions to the Post-Anesthesia Care Unit (PACU), and to evaluate its potential for optimizing nursing staff allocation.</p> Method <p>Daily admission data from 16,637 patients between November 2, 2020, and January 2, 2022, were analyzed. The SARIMA model was developed on a training set (Nov 2020 - Dec 2021) and its forecasting accuracy was rigorously assessed on a test set (Dec 2021 - Jan 2022) using five-fold rolling cross-validation. Model selection was based on Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC), with residual diagnostics conducted to ensure validity. The model’s performance was compared against a Long Short-Term Memory (LSTM) neural network. An operational simulation for nurse staffing was conducted based on the forecasts.</p> Results <p>The SARIMA(1,0,2)(0,1,2)7 model was identified as optimal. It demonstrated strong forecasting performance with a mean RMSE of 14.53, MAE of 11.14, and R<sup>2</sup> of 0.75 on cross-validation. Performance was superior during stable periods (e.g. Fold 4: RMSE = 9.27, R<sup>2</sup> = 0.88) but declined during periods of potential COVID-19 disruption. The model significantly outperformed the LSTM benchmark (LSTM RMSE = 15.51, R<sup>2</sup> = 0.483). A staffing simulation showed the model’s recommendations could potentially reduce overstaffing on 32.1% of days while maintaining safe coverage.</p> Conclusion <p>The SARIMA model provides accurate and reliable short-term forecasts for PACU patient admissions under normal operational conditions. It serves as a valuable decision-support tool for optimizing nursing staff scheduling and improving resource allocation efficiency, demonstrating superior performance and practicality compared to a more complex LSTM model in this clinical setting.</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

Optimizing PACU nursing resource allocation through SARIMA-based patient volume forecasting: a case study from a tertiary hospital in China (2020–2021)

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月31日

查看全文
  • Juan Xiong,
  • Ping Tu,
  • Zhi Hao Li,
  • Na Li,
  • Liang Fang

摘要

Objective

To develop and validate a Seasonal Auto-Regressive Integrated Moving Average (SARIMA) model for forecasting daily patient admissions to the Post-Anesthesia Care Unit (PACU), and to evaluate its potential for optimizing nursing staff allocation.

Method

Daily admission data from 16,637 patients between November 2, 2020, and January 2, 2022, were analyzed. The SARIMA model was developed on a training set (Nov 2020 - Dec 2021) and its forecasting accuracy was rigorously assessed on a test set (Dec 2021 - Jan 2022) using five-fold rolling cross-validation. Model selection was based on Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC), with residual diagnostics conducted to ensure validity. The model’s performance was compared against a Long Short-Term Memory (LSTM) neural network. An operational simulation for nurse staffing was conducted based on the forecasts.

Results

The SARIMA(1,0,2)(0,1,2)7 model was identified as optimal. It demonstrated strong forecasting performance with a mean RMSE of 14.53, MAE of 11.14, and R2 of 0.75 on cross-validation. Performance was superior during stable periods (e.g. Fold 4: RMSE = 9.27, R2 = 0.88) but declined during periods of potential COVID-19 disruption. The model significantly outperformed the LSTM benchmark (LSTM RMSE = 15.51, R2 = 0.483). A staffing simulation showed the model’s recommendations could potentially reduce overstaffing on 32.1% of days while maintaining safe coverage.

Conclusion

The SARIMA model provides accurate and reliable short-term forecasts for PACU patient admissions under normal operational conditions. It serves as a valuable decision-support tool for optimizing nursing staff scheduling and improving resource allocation efficiency, demonstrating superior performance and practicality compared to a more complex LSTM model in this clinical setting.