Article
Models for forecasting the occupancy of intensive care beds with COVID-19 patients. What aspects are important for patient management?
Search Medline for
Authors
Published: | September 27, 2021 |
---|
Outline
Text
Background/Research question/Problem: For better planning of intensive care bed occupancy during the COVID-19 pandemic, forecasting models for the expected number of COVID-19 patients requiring intensive care can be a helpful tool. Over the past year, several forecasting models have been developed that differ in terms of the characteristics to be predicted, the data basis, and thus the prediction results. The task of the authors is to assess the SARS-CoV-2 infection incidence on state level. For this purpose, we are in close exchange with clinicians. Based on our previous experience with three selected forecasting models, we would like to provide suggestions for further development so that these models can be used in practice in a targeted manner.
Solution and suggestion: Forecasting models often use a wide prognosis interval (e.g., 95% confidence interval) or different, partly opposing scenarios. From a practitioner's point of view, a forecasting model that represents a trend as precisely as possible without presenting different scenarios would be sufficient. The forecasting models included here consider a period between 7 days and 8 weeks. For occupancy management, a weekly forecast over a 7-day period might be sufficient for clinicians, as it seems possible to manage patients within this time horizon. Also, forecasts over a shorter time period seem to be more accurate, as relevant changes (e.g., changing contagiosity) can be reacted to. Due to age-related differences in resource requirements and likelihood of severe disease progression, either a prognostic value for younger (<60 years) and older patients requiring intensive care or integration of the age distribution into the model would be helpful. Not all models predict the number of COVID-19 suspected cases. Because these are extremely resource intensive due to the need to separate them from both non-COVID patients and confirmed COVID-19 patients, forecasts should ideally also predict suspected cases.
In a supplementary analysis, the validity of two forecasting models was examined using retrospective data on occupancy rates from the IVENA and DIVI hospital reporting systems.
Conclusion/Discussion/Lessons learned: Forecasting models can make an important contribution to resource management during a pandemic. Ideally, future models would consider the suggestions listed above for further development. In addition, it would be important to place the forecasting models on a uniform and comparable data basis regarding the actual occupancy rates.