#1 — At the Intersections — June 13, 2026

What factors determine the financial toll of a hailstorm? That question is part of our focus on At the Intersections, where we look at how modeling connects our environment, actions, and well-being. This week, we examine new research that uses models to understand complex systems. This week, we're looking at how modeling helps connect our environment and our actions to our well-being. Right. And we have a few examples. One projects the financial outcomes of weather, and another looks at the health consequences of how our workplaces are structured. We'll also get into work that refines medical knowledge, assessing treatment safety during pregnancy, clarifying prevention strategies for women, and tracing the effects of childhood habits. And we end with a system where a human expert and an artificial intelligence work together to process neural data. So, let's start with that first model, the one for predicting financial consequences specifically from severe weather. This study with member Gregory KOP modeled insured loss data from hailstorms to derive what are called vulnerability functions. It used data from three major hailstorms in Calgary, Canada. And what data did they look at? It can't just be the financial losses. No, they analyzed that loss data against environmental variables like hail size from radar and precipitation from satellites, but also against nine socioeconomic variables from the Canadian census. So, it's not just about the storm itself, but about the community it hits. And what did they find? They found that including cumulative precipitation data improved the model's performance. Adding those socioeconomic variables also enhanced the predictive power. And in terms of method, one type of model performed better than another. Correct. Random forest models consistently produced better results, reducing the mean squared error by 35 to 50% compared to lasso regression. The takeaway seems to be that for accurate hail loss modeling, you have to account for both the environmental hazard and the socioeconomic dimensions of the area. From the financial effects of weather, we then turn to the health consequences of workplace structure. This research with member Anders Holm estimates how much sickness absence could be reduced by improving an employee's influence at work. So, how do you measure something like influence at work? In this case, they used register data from a large Danish cohort and defined influence as the ability to affect how and when tasks are performed, which they assessed using a job exposure matrix based on job titles. And they found a connection, that higher levels of influence were associated with fewer days of sickness absence. They did, and they ran a statistical simulation to project the impact. It showed that a standard deviation increase in influence would correspond to a reduction of 0.1 days of sickness absence per person per year. That might not sound like a lot per person, but across the entire study population of over 300,000 people, it adds up. It amounts to an estimated reduction of over 126,000 sick days. The largest reductions were seen in care work and education. Next, we examine HIV prevention strategies for women. A study with member Mia Biondi analyzed post-exposure prophylaxis or PEP, and a related strategy, PEP in Pocket or PIP. The context here is that HIV prevalence among cis and trans women is increasing, but the uptake of prevention medication remains low. So, this study used what's called an evolutionary concept analysis to get a better understanding of PEP and PIP specifically for women. What's the distinction between the two? PEP is used reactively after a potential exposure. It's initiated by a clinician and requires time-sensitive access to care. Whereas PIP is different. It's a self-initiated option for someone who might have infrequent high-risk exposures. It offers more autonomy and can be accessed outside of a healthcare setting. The analysis concludes that both are promising women-centered strategies, but a key finding is that they remain underutilized. Right. It identifies gaps in both research and implementation, pointing to a need for more investigation to advance equitable HIV prevention. Staying on the theme of treatment decisions, another piece of research with Mia Biondi looks at the safety of treating hepatitis C during pregnancy. And there's a specific reason for this. Offering direct-acting antivirals or DAAs is the only potential way to prevent vertical transmission to the child. Pregnancy might also be the only time a person with HCV is engaged in the healthcare system. So, this work analyzed a large data set on pregnancy outcomes for people exposed to a specific DAA combination. What did it find? The analysis found that reported rates of spontaneous abortion were in line with rates in the general population, and it looked at congenital anomalies. And the details there are notable. Of 10 anomalies reported, six occurred with drug exposure in the second or third trimester. The fetal organs were already formed, meaning the anomalies were present before the drug was even taken. And the other reported anomalies showed no unifying pattern that would suggest a single drug-related cause. So, the data to date has not demonstrated major safety signals for using these drugs in pregnancy. From treatment during pregnancy, we shift to the long-term effects of childhood habits. Research with member Paul Tremblay explores whether early exposure to video games is a predictor of problematic gaming in adulthood. The question was whether internet gaming disorder or IGD shares a pattern seen in substance use disorders, where the timing of first exposure is linked to adult symptom severity. Using growth mixture modeling, the study identified four different patterns of gaming onset and frequency. One group stood out, the consistently high group. This group had high levels of gaming during both childhood and adolescence, and they displayed higher adult IGD symptoms than the groups that escalated their gaming later in life. The regression analysis confirmed it. Gaming during preschool and high school predicted current IGD symptoms, and preschool gaming emerged as the strongest predictor. So, the study does highlight a parallel between IGD and substance-related disorders with an early age of onset predicting adult symptom severity. Finally, we look at a system where human expertise and artificial intelligence collaborate to process neural data. A project with member Paul Fruin presents a system called EEG AI. And it's designed to solve a known problem with electroencephalography or EEG data. The raw signal is used to measure brain activity, but it has a low signal-to-noise ratio. It's mixed with artifacts from things like muscle activity or eye blinks. Traditionally, cleaning that data is a laborious process, requiring extensive manual inspection by an expert. So, this new framework integrates a large language model-driven AI agent that assists a human expert. The agent calls on analysis tools and interprets outputs to decide which signal components to retain or reject. And it's a collaborative process. It happens in an iterative closed loop, which allows for continuous feedback from both the model and the human expert. In evaluations, the AI agent system performed as well as or better than conventional methods at cleaning these artifactual components from the signal. The framework is designed to streamline EEG preprocessing while keeping that expert oversight, which ensures the process is both reproducible and auditable. That concludes this week's issue. We'll return next week with new research from across the disciplines. From At the Intersections, thanks for listening.