End of course summary

Nowcasting and forecasting of infectious disease dynamics

Aim of this course:

How can we use data typically collected in an outbreak to answer questions like

• what is the number of cases now? (nowcasting)
• is it rising/falling and by how much? (\(R_t\) estimation)
• what does this mean for the near future (forecasting)

in real time.

Timeline

• delay distributions and how to estimate them (day 1)
• \(R_t\) estimation and nowcasting (day 2)
• forecasting, evaluation, and ensemble methods (day 2)
• forecasting hubs, applications, and linking nowcasting and forecasting (day 3)

Key takeaways

Day 1: Delay distributions

• delays play a fundamental role in nowcasting/forecasting
• we characterise them with probability distributions
• estimating delays requires correction for biases due to
  • double interval censoring (daily data)
  • right truncation (real-time analysis)

Day 2: \(R_t\) estimation

• \(R_t\) estimation using the renewal equation is a convolution problem
• improving the generative model leads to improvements in estimation (geometric random walk vs. independent priors)
• generation time is a key delay distribution for transmission dynamics
• understanding the role of the generative model in the estimation of \(R_t\)

Day 2: Nowcasting

• nowcasting is the task of predicting what data will look once delays have resolved
• it is a right truncation problem (same as discussed before)
• a joint generative model can combine delay estimation, nowcasting and \(R_t\) estimation

Day 2 (cont.): Forecasting and evaluation

• forecasting is the task of making unconditional statements about the future
• meaningful forecasts are probabilistic
• we can assess forecasts using proper scoring rules (MAE, RMSE, CRPS)
• methods include ARIMA models, ensemble approaches, and time-series cross-validation
• we can use visualisation and scoring to understand the predictive performance of different models

Day 3: Collaborative modelling and applications

• forecasting hubs enable collaborative modelling efforts across institutions
• hubverse tools provide standardised formats for forecast submission and evaluation
• linking nowcasting and forecasting through joint or pipeline approaches
• the methods introduced here have wide applications in infectious disease epidemiology
• open-source tools are available to make this task easier in practice

Outlook: Contributing to forecast hubs

• apply these methods in practice to learn about typical nowcast/forecast performance
• contribute to collaborative forecast hubs to compare approaches
• use hubverse standards for forecast formatting and submission
• example: European Respiratory Forecasting Hub

https://respicast.ecdc.europa.eu/

Feedback

• please tell us if you enjoyed the course, what worked / didn’t work etc.
• we will send out a survey for feedback

Thank you for attending!

Return to the session