Disease Trajectories, Danish Style
Lifelong patterns for 6.2 million people
In the first large-scale study of its kind, researchers in Denmark computed the disease trajectories for the country’s entire population—approximately 6.2 million people—over the course of 15 years, revealing some surprising and interesting patterns.
“The advantage in Denmark is that we can follow people with this lifelong perspective,” says Soren Brunak, professor of systems biology at the Technical University of Denmark, and lead researcher on the paper published in Nature Communications in January 2014.
Since 1968, every Danish medical record has been linked to the individual’s social security number. Even when people move or visit different physicians or hospitals, the data tracks them. The medical records also document, in chronological order for each patient, every diagnosis. And the dataset covers thousands of diseases.
To determine disease trajectories, Soren Brunak and his colleagues relied on the diagnostic codes from the Danish dataset and looked for pairs of diagnoses with positive relative risk—the probability that one diagnosis would follow another. They then went through two levels of condensation to get at the patterns in the data. First, they condensed the 6.2 million patient trajectories into the 1200 most common pairs of positive relative risk. Next, in a non-hypothesis driven way, the researchers condensed the 1200 trajectories into five trajectory networks. “We took the data, clustered it, and saw what popped up,” Brunak says. The five largest networks covered diabetes, cardiovascular disease, cerebrovascular diseases, prostate disease and chronic obstructive pulmonary disease (COPD).
Naturally, many of the strongest trajectories were known to physicians, Brunak says, but there were a few surprises. “Gout, for example, showed up in a very convincing way in the cardiovascular disease landscape,” he says, a connection that wasn’t previously confirmed.
Brunak is now working on making the networks predictive. By digging into the details of a specific network and layering it with biomarker data (which is also available for many patients) or lifestyle data (such as income or school performance, which are also tied to the social security number), Brunak and his colleagues hope to discover ways to predict individual disease paths. For example, the researchers might look at whether the data predict which diabetic patients will develop renal failure or blindness or some other complication. “What is there in the trajectory or genetic makeup that leads them down a different route?” Brunak says.
The team is also interested in understanding inverse comorbidities—where patients with one disease are less likely to get another disease. Having schizophrenia, for example, is associated with reduced risk of certain types of cancer. Is that the result of medications having a protective effect or do the cellular networks driving the disease provide protection? “It’s important to view these trajectories as ways to discover what you don’t see as well as what you do,” Brunak says.