There are different ways to model transmission and each model has its strengths and weaknesses. There is no such thing as "the perfect model," and it is always useful to run different approaches in parallel and compare the results side by side to adjust and validate conclusions and recommendations each time.

The compartmental transmission or SEIR model, the individual-based model and the meta-population model are three different mathematical approaches we have used intensively to monitor the COVID-19 epidemic in Belgium, among other models and variations used both locally and internationally.

The stochastic [compartmentalized transmission](https://covid-project2022.vercel.app/article/how-can-we-model-viral-transmissions-compartmental-model) model allows detailed age-specific case-fatality data to be integrated with results from blood analyses, social contact patterns, the emergence of new variants, vaccination coverage and so on.

The [individual-based model](https://covid-project2022.vercel.app/article/how-can-we-model-viral-transmissions-individual-based-model) is ideally suited to measure the effectiveness of contact tracing and of isolation and quarantine measures.

The [meta-population model](https://covid-project2022.vercel.app/article/how-can-we-model-viral-transmissions-meta-population-model) is best for looking at how mobility affects virus spread.

The calculations of both the meta-population and the compartmental transmission model can be done faster than for the individual-based model, an important factor in times of crisis, especially when you want to compare multiple scenarios and have to recalculate the model each time.

In addition to these three models, there is a range of other ways to look at viral transmission.

Growth models follow trends

Growth models chart the number of cases or hospitalizations to express the evolution of these numbers over time. They help to pick up trends in a very simple way: are we seeing an upward trend, are we approaching a peak or plateau, or have we passed the peak?

Growth models primarily let the data speak for itself without making assumptions about the underlying diffusion process. More advanced and realistic growth curves take into account that at a certain point a peak will be reached and that the rate of decline may be higher or lower than the upward stage of the curve.

Growth models were an important tool during the first wave, when little was known about the parameters that determine viral transmission. These types of models allow us to look up to 7 days ahead and help estimate how the growth factor changes over time. Although they are often misused as such, these types of models are not suitable for making long-term projections, or for estimating the effects of interventions on the course of the curve.

A short-term model to follow hospital capacity during the first wave

In the early weeks, when we had very little data, we developed a practical model to track hospital capacity in our country in real time. We the number of new hospitalizations, and the total number of people in the hospital and in intensive care to estimate how likely a complete flood of our healthcare system would be on the short term.

Implementing Bayesian statistics, where probabilities are constantly revised based on new available information, this model allows us to make estimates about how long someone will stay in the hospital or what proportion of admitted patients will need intensive care, in the absence of actual data on these parameters.

While this model helps us to make valuable short-term projections, e.g., worst-case and best-case scenarios, it also has some important limitations. The model follows the course of a single wave, and does not handle erratic changes to the course of that wave quite well, for example, due to a sudden and drastic change in measures. The model also assumes that the proportion of hospitalized patients that end up in intensive care is constant over time, which is unrealistic over the course of an epidemic or pandemic, but plausible in the short term. Finally, it also makes the assumption that there are no behavioral changes that affect the course of the epidemic. For all of these reasons, this type of model is not suitable to make long-term projections.

Who infects whom? Transmission Models

Transmission models or trees try to map successive infections between individuals, much like new branches growing on a tree. To find out how someone got infected, you can either look at the genetic similarity of the viruses of infected individuals or can trace their contacts.

Especially in the early days of the pandemic, where the testing capacity was limited, contact tracing helped to monitor infections. The actual purpose of contact tracing is of course to be able to design more targeted interventions so that only those with a large risk of being infected are quarantined. For that reason, contact tracing can play an important role as part of an exit strategy, where lockdown measures are being relaxed. Contact tracing can prevent larger outbreaks from happening.

Transmission trees are very useful to estimate the serial and generational interval (respectively, the time between the onset of symptoms and the time between infection of the infector and infected person) and thus gain a view of how quickly an infectious disease like COVID-19 spreads.

To be able to accurately account for the complex interplay of viral transmission, the effect of measurs, changing behavior and, eventually, vaccination, we use different types of models, the most important ones are metapopulation, individual-based and stochastic compartment models.