A Weather Forecast for Disease

In a world where pandemics and lockdowns are still fresh in people’s minds, there has never been more of an appetite for real-time tracking of infectious disease. With new advances in genomics, as well as epidemiological and clinical data, modern surveillance techniques allow for unprecedented insight into the current status of disease. But although the technology may exist, the infrastructure for disease forecasting is a still fledgling science. One group of researchers, in a bid to support the creation of such a system, has outlined their perspective on the steps needed to design a successful disease forecast in the future.

First, it’s vital to address the looming threat of antimicrobial resistance. To date, disease forecasts have been unable to incorporate predictions on pathogen diversity – but, to ensure that the models are useful to practitioners and policymakers, they must be able to describe current infectious agents and their risk of resistance diversification. Pathogens evolve fast – and we need to keep up if we want to effectively monitor our antimicrobials’ ability to keep us safe.

So what can we do? The authors propose a marriage between disease forecasting and genomic data. Sequencing technology is faster and cheaper than ever, and our ability to handle large volumes of data is only increasing. We’re also expanding our understanding of resistance mechanisms, causative mutations, and predictive parameters. As turnaround times decrease and access to sequencing technology increases, we can track the evolution of the most pressing pathogens and the effectiveness of our antibiotic treatments against them. Embedding this data into prediction models and refining them over time in light of real-time pathogen evolution could significantly improve the accuracy and utility of infectious disease forecasting.

Despite the availability of extensive public pathogen sequence databases and the range of projects underway to compare sequences and combat resistance, the authors highlight that differences in sampling strategies and lack of context can impact the data’s forecasting utility. To remedy this, they recommend continual sampling in the context of long-term surveillance – but standardized approaches to sampling, sequencing, and reporting (including metadata) could also help.

Although mathematical modeling for epidemiology has grown significantly more accurate in recent years, there are still improvements to be made – and real-world observations, particularly in genomics, don’t always match up with the math. In light of the expanding opportunities, the authors call for the incorporation of molecular data – genetics, genomics, and ultimately phylodynamics – into disease forecasting to ensure that our predictions, and the actions we take as a result, are as accurate and well-considered as possible.