Mathematical models are fundamental tools for understanding the transmission mechanisms of emerging infectious diseases, forecasting epidemic trends, and assessing transmission risks. Yet, a critical flaw hinders their accuracy: to date, many epidemiological models of disease spread have assumed that at-risk populations are homogenous. In reality, populations are diverse and complex, hence rendering predictions from such models inaccurate.
Scientists from two Chinese universities in the city of Xi'an collaborated with NRI's Professor Robert Cheke to develop a novel technique that incorporates heterogeneity in populations into epidemiological models, making them more realistic and enhancing prediction accuracy. This innovative approach simulates how population heterogeneity changes during an epidemic, leading to more reliable disease risk assessments.
In a recent study published in the journal PLOS Computational biology, the scientists showed that better estimates of outcomes based on real world data can be achieved by accounting for heterogeneity in disease models.
To accomplish this, the authors used a combination of models and deep learning technology to ‘train’ one of the models using epidemic data on an outbreak of the Omicron variant of COVID-19 that occurred in Shanghai after China abandoned its dynamic zero-COVID case policy. Deep learning is a method in artificial intelligence that teaches computers to process data in a way that mimics the human brain.
Projections from the model showed an attack rate (the proportion of a population that becomes ill with a disease during a specific period) of 80.1% and a peak of new infections of 3.2% of the population. When compared with actual data from the outbreak, these figures represent an 18.6% improvement in prediction accuracy over calculations based on assuming that at-risk populations were homogenous. This new modelling framework also estimated that the epidemic lasted 70 days rather than the 30 days based on the conventional estimates, and that herd immunity was decreased for about a third of the population, which is much closer to the real events.
Emerging infectious diseases like COVID-19 and SARS have posed significant threats to global public health and controlling them has become a top priority. Epidemic models are crucial in supporting decision-making to mitigate these threats and optimizing strategies to protect public health and minimize economic costs. By accurately predicting epidemics, these models help to determine medical resource allocation, vaccination coverage, and other critical factors, ensuring effective and cost-efficient outbreak management.
The novel modelling framework by Professor Cheke and his colleagues adopts a simple method to incorporate heterogeneity into mathematical models and presents a practical and effective method for addressing the above considerations. This new approach also facilitates the extension of the framework to consider different aspects of infectious disease transmission with heterogeneity in susceptible populations. Professor Cheke said: ‘The straightforward approach that we used for dealing with spatial heterogeneity, deliberately concentrating on disease spread around epicentres rather than assuming a uniform population at risk, is practical and allows for further improvements to our understanding of infectious disease dynamics and how to model them accurately.’