SISMID 2026: Forecasting and evaluation

Forecasting & evaluation of infectious disease dynamics

Introductions

Introduce yourself with:

Nowcasts make statements about current trends based on partial data
Forecasts are unconditional statements about the future: what will happen
Scenarios state what would happen under certain conditions

Collected as part of routine activities by state & local PH departments.
Not exhaustive list of all cases, ideally a representative sample.
Often looked to as real-time indicators of “what is going on”.
E.g., Influenza-like illness (ILI) = fever AND additional “flu-like” symptom.

How can we use surveillance data to answer real-time questions like:

Mechanistic or phenomenological? — do model parameters have epidemiological meaning?
Generative or data-led? — can we write the model as a data-generating process we can simulate and get a likelihood from?
Frequentist or Bayesian? — how do we estimate and propagate uncertainty in the model?
Simple or complex? — Model complexity ≠ skill. How do we encode epidemiological complexity/realism without sacrificing predictive accuracy?

We can classify models by the level of mechanism they include
All of the model types we will introduce in the next few slides have been used for COVID-19 forecasting (the US and/or European COVID-19 forecast hub)

NOTE: level of mechanism \(\neq\) model complexity

Complex agent-based models

Conceptually probably the closest to meteorological forecasting, if with much less real-time data.

Compartmental models

Aim to capture relevant mechanisms but without going to the individual level.

Semi-mechanistic models

Include some epidemiological mechanism (e.g. SIR or the renewal equation)
Add a nonmechanistic time component inspired by statistical models (e.g. random walk)
For those familiar with the nowcasting course, this should be familiar

Statistical and Machine Learning models

Models that don’t include any epidemiological background e.g. ARIMA; also called time-series models
“Black-box” machine learning approaches
Various regression-based approaches (could include some epi-relevant covariates)

Other models

Throughout the course we will

work with real epidemiological surveillance data in R
fit time-series forecasting models and make predictions
evaluate forecasts using proper scoring rules
combine models into ensembles and contribute to collaborative modelling hubs