The AI ‘nudge’ taking aim at pulmonary embolism

Feature Stories | 25 Mar 2025 | Read time: 5+ min

It’s a story about a running club, an AI algorithm, and a programmer misdiagnosed with a heart attack who sat across from an emergency physician over coffee and said: I have an idea.

But let’s start at the beginning.

When retired Apple iOS and macOS programmer Dennis Christopher first developed a pain in his knee, he assumed that it was the typical aches and pains familiar to any avid runner.

“I thought maybe it had to do with a long run I’d recently completed,” Dennis recalls. “You know with runners; the pains come and go.”

The pain eventually disappeared, but persistent fatigue and a growing shortness of breath led him to a walk-in clinic near his home in Burlington, Ontario. Clinicians there weren’t too concerned, and he was sent home—where his symptoms only worsened.

“Finally, it came to where I was walking up some stairs and I had trouble. That’s when I thought…something is really wrong here.”

At the emergency department, doctors initially suspected a heart attack. It was luck that a sharp-eyed heart specialist who’d treated Dennis in the past recognized his name and suggested investigating for pulmonary embolism (PE). The original pain in his knee? It was a warning sign of deep vein thrombosis, a serious condition that can lead to pulmonary embolism (PE), when blood clots in the lungs can cause a life-threatening blockage. Testing confirmed PE, and Dennis quickly received treatment. Once diagnosed, the condition is easy to address, usually with blood thinner medication.

Dennis Christopher, running a race in 2022.

The bogeyman of the emergency department

Even though he was back to running within a month or two, the experience left Dennis with the distinct impression that PE wasn’t an easy diagnosis to make—even for the most seasoned emergency care doctor. After all, his own diagnosis was made thanks to a physician who was already familiar with his health history. This certainly wouldn’t be the typical emergency department experience for most patients.

Pulmonary embolism is a notorious diagnostic challenge, often described as “the bogeyman” of the emergency department. Most of its symptoms—chest pain, fatigue, shortness of breath, dizziness—mimic those of heart attacks, pneumonia, or asthma. It’s not common either; only two in every thousand patients in the emergency department have pulmonary embolism. Testing everyone with these symptoms can overwhelm an already busy emergency department’s resources, making the decision a critical judgment call with potentially fatal consequences for patients.

A ‘double double’ shot of innovation.

With over 25 years of programming experience and an interest in machine learning, Dennis saw an opportunity: could big data and artificial intelligence (AI) better predict pulmonary embolism in patients? He just needed someone with the right medical expertise to tell him whether it was possible.

Enter Dr. Kerstin de Wit, an emergency physician and Queen’s researcher specializing in bleeding and clotting disorders...and at the time, a member of the same running club as Dennis. They’d crossed paths before but never exchanged more than a few words.

That changed when a mutual friend suggested Dennis share his idea with her.

“We met at Tim Hortons,” Dennis says, with a smile. “I laid out the idea and said let's put it into machine learning and see what it comes up with. What's unique about Kerstin and my project is we're not looking at the issue of what to do after you suspect pulmonary embolism. We're looking at ‘do you suspect pulmonary embolism in the first place?’”

The birth of PE Nudge

Assembling a research team and with the help of AI tools, Dennis and Kerstin analyzed over 14 million emergency department visits in Ontario. They found that more than 1,500 cases of pulmonary embolism are misdiagnosed in the province every year.

That’s where PE Nudge comes in. It’s a predictive AI model designed by the team to assist health care providers in recognizing patterns and symptoms that could indicate a possible PE diagnosis. Once it does this, the program offers a data-driven prompt to the practitioner to consider testing for PE, giving them a potentially lifesaving ‘nudge’.

So far, the model has performed well using the de-identified provincial health care data it’s been trained with. Recently, the research team also obtained data from the Ontario Health Data Platform and uncovered an opportunity to refine the model further by incorporating data trends seen during the COVID years.

To ensure PE Nudge’s effectiveness, the final step will be real-world testing in emergency departments, where split-second decisions can mean life or death. This requires health care organizations who are willing to pilot the model.

“The data we've accessed so far doesn’t include real-time data available when [patients] are checked into the emergency department,” explains Dennis. For example, their blood pressure or their oxygen saturation. So, the next step is to find a platform to get that data and incorporate it into the model and see if it helps improve it even further.”

“We created a basic starter model, which can tell if any emergency patient has a higher risk of having blood clots or pulmonary embolism,” Kerstin adds. “And we have to validate that. I'm always looking for partners who might have good, high-performing electronic medical records where we could start running on their data.”

Kerstin de Wit standing on a dark stage... — Kerstin de Wit discussing the PE Nudge research project at her recent 5à7 research talk.

It’s a bold, high-impact project. Unique, not just for its unconventional origin, but also for its ambition. Few have tried to use machine learning in health research at this scale, with this scope, or for this level of impact. The path forward is long, but clear: more data, more testing, more refinement. All to give emergency care providers a tool to catch PE cases that might otherwise be missed.

Dennis continues to lead the project alongside Kerstin, contributing research design, statistical analysis, and programming expertise. What began as an idea over coffee has grown into something much larger—but for Kerstin, Dennis’s greatest strength isn’t just technical ability, it’s perspective.

“Sometimes when you work in research, you get a little bit jaded,” she says. “It just gets so overwhelming when you can’t always stretch your wings and make an impact in places that you think would be helpful because there's so many constraints. But with Dennis, you know he’ll try. He'll really try to get to the next step, to do better, to get further.”

Read time

5+ min

The AI ‘nudge’ taking aim at pulmonary embolism

It’s a story about a running club, an AI algorithm, and a programmer misdiagnosed with a heart attack who sat across from an emergency physician over coffee and said: I have an idea.

The bogeyman of the emergency department

A ‘double double’ shot of innovation.

The birth of PE Nudge

Related topics

Explore Our Stories