Abstract
Register
Abstract
Year 2021
September 2021

SHBC1655

Abstract Title
Forecasting Intensive Care Unit (ICU) Bed Demand for COVID-19 Patients at Tan Tock Seng Hospital (TTSH) in Singapore
Authors

W.LIAN1, H.P.PHUA1, A.S.H.TAN1, W.Y.LIM1, A.CHOW1

Institutions

Tan Tock Seng Hospital1

Background & Hypothesis

In many countries, the COVID-19 pandemic has imposed immense strain on healthcare systems, especially on Intensive Care Units (ICU). To better anticipate ICU demand for capacity planning, we aimed to forecast ICU demand due to COVID-19 infections.

Methods

Linear, quadratic, logistic, and exponential models were used to fit daily COVID-19 admissions to TTSH/NCID, by age group, with weighting on more recent observations. Daily admission forecasts were extrapolated for the next 2 months and updated weekly. From April 2021, an exponential smoothing (ETS) model was used instead. ICU in-flights were then calculated using in-hospital data for the proportion of admissions requiring ICU, time lag between admission and ICU entry, and mean length of stay in ICU. We evaluated model performance by calculating the root mean square of the forecast errors (RMSE) for the subsequent 7 days.

 

Results

The logistic model performed best in forecasting admissions with the lowest RMSE among all models (Logistic:6.56; Quadratic:7.27; Linear:8.70; Exponential:15.7) before April 2021. The ETS model also performed relatively well (RMSE≈6) after April 2021. Both logistic and ETS models tend to over-predict admissions. With these models, 85% of predicted daily ICU in-flights were overestimated with mean +2 beds/day prior to April 2021 whereas approximately 70% of that were underestimated with mean -1 beds/day after April 2021.

Discussion & Conclusion

The forecasting models provided good guidance for manpower and resource planning throughout the COVID-19 pandemic, despite limitations in available parameters and the dynamic nature of the evolving pandemic with changing transmissibility of novel strains and effectiveness of control measures.

Top