Thanks for subscribing to the free version of RMN. Here are this week's stories for pain subscribers.
OECD Report Measures Trade-Offs Between Catastrophe Bonds vs. Taxation
While CAT bonds are seen as preferable policy, the report says there are real-world barriers to their adoption.
Oregon Drops Wildfire Map Effort
The political fallout exposed a deeper tension between centralized risk modeling and local implementation.
Ports Face a Multibillion-Dollar Physical Risk Threat: JP Morgan
$63 billion in trade is at risk each year due to physical risk disruptions.
Modeling the Next Outbreak: How AI Is Rewriting Pandemic Forecasting
The COVID-19 pandemic exposed major flaws in infectious disease forecasting, particularly the inability of traditional models to adapt quickly to dynamic conditions like new variants and shifting public health policies.
Now, artificial intelligence is changing that.
A team of researchers from Johns Hopkins and Duke University has developed an advanced AI tool, PandemicLLM, that uses large language models (LLMs) to forecast the spread and severity of infectious diseases.
The tool consistently outperforms the best existing models used by the Centers for Disease Control and Prevention (CDC), especially during periods of high volatility in outbreaks, according to a recent article in Nature Computational Science.
Unlike conventional models that treat forecasting as a statistical exercise based primarily on numerical data like case counts and hospitalizations, PandemicLLM takes a fundamentally different approach. It interprets the pandemic landscape using AI-driven language reasoning—processing both numerical and non-numerical data through text-based prompts
“COVID-19 elucidated the challenge of predicting disease spread due to the interplay of complex factors that were constantly changing,” said Lauren Gardner, a modeling expert at Johns Hopkins said. “When conditions were stable, the models were fine. However, when new variants emerged or policies changed, we were terrible at predicting the outcomes. The new tool fills this gap.”
PandemicLLM integrates four types of data:
- Demographic and spatial data on healthcare systems and political affiliations
- Epidemiological time series, including hospitalizations and vaccination rates
- Public health policies, such as lockdown measures and school closures
- Genomic surveillance data on emerging variants and their characteristics
These diverse data streams are converted into approximately 300-word prompts through a human-AI collaborative process. For instance, time-series data might be translated into natural language summaries, which are then combined with qualitative policy or demographic descriptions. The AI uses these prompts to predict short-term hospitalization trends across all 50 U.S. states, classifying outcomes as substantial increase, moderate increase, stable, moderate decrease, or substantial decrease.
The model was tested retrospectively on COVID-19 data from 2021 to 2023. Compared with the CDC’s ensemble forecasting model, PandemicLLM showed significantly improved accuracy, particularly in detecting turning points in the trajectory of the disease, such as those driven by the emergence of new variants .
“Traditionally we use the past to predict the future,” said Hao “Frank” Yang, a Johns Hopkins engineer who specializes in AI reliability. “But that doesn’t give the model sufficient information to understand and predict what’s happening. This framework uses new types of real-time information.”.
Beyond forecasting, researchers are now exploring whether LLMs can simulate individual decision-making under public health risk—potentially aiding in the design of more effective and responsive interventions.
“We know from COVID-19 that we need better tools so that we can inform more effective policies,” Gardner added. “There will be another pandemic, and these types of frameworks will be crucial for supporting public health response.”
Here is a paragraph you can add to your article on pandemic modeling, detailing the potential impact of another global pandemic on insurers and reinsurers, based on the Lloyd’s report:
For insurers and reinsurers, the resurgence of a severe global pandemic would present a complex and far-reaching risk scenario.
In a 2024 report from Lloyd’s of London, pandemics disrupt not only public health systems but also global economic activity through lockdowns, workforce absenteeism, supply chain breakdowns, and shifts in consumer behavior.
The economic consequences are particularly acute for sectors reliant on in-person interaction—such as travel, hospitality, and live events—many of which were heavily exposed during COVID-19. In an “extreme” pandemic scenario, with high virulence and vaccine delays, insurance losses could be compounded by extended recessions, inflation, and civil unrest.
Lloyd’s emphasizes that while insurance plays a role in resilience and recovery, the evolving nature of infectious disease—driven by urbanization, global mobility, and climate change—demands that the industry model more severe and novel outbreak dynamics.
