Skip to content

Latest commit

 

History

History
36 lines (26 loc) · 1.57 KB

README.md

File metadata and controls

36 lines (26 loc) · 1.57 KB

COVID-19 hospital admissions forecasts for the ECDC Forecast Hub

Current forecasts

The most recent forecasts (from 27 November 2021) cover the following locations: Belgium (BE), Czechia (CZ), Denmark (DK), Estonia (EE), France (FR), Greece (GR), Iceland (IS), Ireland (IE), Latvia (LV), Malta (MT), Netherlands (NL), Norway (NO), Slovenia (SI), United Kingdom (GB).

Methods

Autoregressive time series ensemble

A mean ensemble of three autoregressive time series models: ARIMA, ETS and a “naive” model (where future admissions are equal to the most-recently observed admissions). Each of the three time series models are fit independently to weekly incident admissions data. The mean-ensemble quantile forecast is made by taking the mean of the three quantile forecasts.

ARIMA regression

A regression model with ARIMA errors, with 1-week lagged COVID-19 cases as a predictor. The model is fit with weekly admissions/cases data. To make forecasts of future admissions for a forecast horizon of two weeks or more, we use the ECDC Hub ensemble case forecast.

Case convolution

A convolution of COVID-19 cases and a report-to-admission delay distribution. The model is fit with weekly admissions/cases data. To make forecasts of future admissions, we use the ECDC Hub ensemble case forecast.