Although lockdown, contact tracing; quarantine and social distancing are measures widely exercised by many countries to prevent the spread of coronavirus during Covid-19 pandemic, it’s challenging to quantify their respective efficacy. This is especially so as some may have infected others before they were tested positive with the virus; while others are still allowed to travel under special circumstances and also those who are violating the interventions on purpose.

Enhancing SEIR model with a neural network

Besides, publicly available data such as infection rates and death counts, may not thoroughly reflect the situation since testing remains limited in certain countries and there may be the possibility of under-reporting. Most importantly, some of the models that are forecasting how the ongoing pandemic may progress were built using data from the SARs or MERs outbreak earlier on.

To overcome these challenge, a team of engineers from Massachusetts Institute of Technology (MIT) created a model mixing standard epidemiology principles, Covid-19 pandemic data and a neural network to better prognosticate if strategies intervening viral outbreak has successfully slow down infection in each country.

Fundamentally, the spread of epidemics is modelled after SEIR or dividing the flow of people into four states: “susceptible”, “exposed”, “infected”, and “recovered”. The MIT team strengthened the compartmental technique by training a neural network to include the number of infected individuals who are under quarantine and as such, will not be able to infect others.

The neural network was trained through 500 iterations so it can teach itself how to predict patterns of how an infection is being spread. Researchers performed analyses on four places – Wuhan, Italy, South Korea and the US. The model was used to compare the impacts of quarantine and isolation measures have on each location in controlling the spread of the coronavirus.

Predicting when an infection will plateau

The accuracy of the neural network was tested using data from Wuhan, the supposed origin of the Covid-19 pandemic. Using data between 24 January and 3 March, the MIT researchers found that predictions rendered by the model matches with the eventual data on 1 April.

Moreover, the model also found that in South Korea, where the government promptly responded to the pandemic with strict quarantine measures, the rate of infection plateaued quickly. On the other hand, in Italy and the US, where interventions took a longer time to be in place, saw a more rapid spread of the virus.

In general, as the number of newly infected cases begin to drop, the corresponding prediction model will transit from an exponential regime to a linear one. With that, it’s suggested that infection in Italy and the US will reach its plateau between 15 and 20 April. A similar prediction was also made by the US Institute for Health Metrics and Evaluation.

Researchers also warned that by relaxing or reversing some of the strict quarantine measure that was imposed earlier on, may result in a second wave of exponential regime and probable catastrophic consequences, as in the case of Singapore. Overall, the model does prove that rigid quarantine can indeed suppress the virus from spreading. All findings are now made available on a preprint server while waiting to be peer-reviewed.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.