Can natural experiments unlock causes of disease?

Beyond measuring the effects of medication or surgery, real-world data can help us understand the underlying mechanisms of disease.

Map of the 1854 Broad Street cholera outbreak in London. Credit: Wikimedia Commons/Wellcome Images, a website operated by Wellcome Trust, a global charitable foundation based in the United Kingdom.

No Epidemiology 101 course is complete without the story of the 19th century physician John Snow (not to be confused with Jon Snow, the character from Game of Thrones—though we imagine many public health students are familiar with both).

In 1854, there was an outbreak of cholera in London. At the time, it was not known how patients became infected with the potentially fatal diarrheal disease. In investigating the outbreak, Snow noticed that not all Londoners living in the same neighborhood were getting sick—some were dying from cholera while neighbors in the next building were completely healthy.

The difference? Some homes happened to be served by one water company, while neighboring homes were served by another water company. He found that homes served by one company had dramatically higher rates of cholera infection compared to the other, suggesting that the disease was being spread through the water, since there were no other appreciable differences between neighbors that might explain why only some were getting sick. As it turned out, the water from one company was contaminated with sewage.

Snow’s study—a natural experiment that took advantage of the as-good-as-random assignment of neighbors to one water company or another—would support the “germ theory” of infectious disease that we know to be true today. Cholera is caused by the bacteria Vibrio cholerae, a microbe that Snow could not see but that his data correctly showed was being transmitted via contaminated water.

Still good a century and a half later

Snow’s story showed how powerful it can be to look at health data through the right lens. Although randomized controlled trials are a mainstay of modern medical research, natural experiments remain a powerful tool when randomized trials aren’t feasible. Between the Random Acts of Medicine book and newsletter, we’ve looked at a bunch of natural experiments that tell us something about the effects of treatment, whether it be medications, procedures, or practices in the health care system.

But as Snow’s story illustrates, natural experiments could be used for something more: to help us understand the underlying biological mechanisms of disease.

Let’s take the case of peptic ulcer disease—ulcers in the stomach or small intestine that can cause life-threatening internal bleeding and dangerous perforations in the GI tract. Examining biopsies of these ulcers under a microscope, the pathologist Robin Warren noted that when there was more inflammation in the tissue, there tended to be more bacteria, specifically one bacterium called Helicobacter pylori. What wasn’t clear was whether H. pylori was causing infection and subsequent inflammation, or whether the tissue was inflamed for another reason and the bacteria were simply taking advantage of the inflamed environment to flourish.

So Warren teamed up with physician Barry Marshall in 1981 to try to establish a causal link between H. pylori and peptic ulcer disease, suspecting that H. pylori was causing infection, inflammation, and ulceration. In a move that can only be described as “old school,” Marshall swallowed a culture of H. pylori (OK, we suppose we could describe this as “gross,” too). He developed a stomach infection, and then had an endocscopy and a biopsy which showed inflammation full of H. pylori. The bacteria was causing the ulcer. Warren and Marshall would be awarded the Nobel prize in 2005 for their work, which opened opportunities for doctors to treat ulcers caused by H. pylori with antibiotics to prevent serious complications.

But had the right data been available to them, Marshall and Warren might have been able to make this discovery without swallowing any bacteria.

By 1981, antibiotics were being used regularly to treat all kinds of infections. If H. pylori was causing ulcers, then it stands to reason that if patients were receiving antibiotics for other reasons that happened to also kill H. pylori, then those patients should be getting fewer ulcers, right?

Today, we could look at a database with millions of patients from across the country. We could find the many thousands that come in complaining of a sore throat or an ear infection; some of them will happen, by chance, to be treated with antibiotics for a suspected bacterial infection, while others won’t receive antibiotics if the infection is suspected to be viral (note, doctors vary in their likelihood of prescribing antibiotics, which could be leveraged to estimate the effect of receiving antibiotics if people are as-good-as-randomly assigned to high- vs. low-prescribing doctors).

If we followed patients out for the next year or two, we could see if there was any difference in the rate of peptic ulcers between those who, by chance, got antibiotics for their sore throat or ear infection—and thus might have also killed off ulcer-causing H. pylori in the process—and those who didn’t get antibiotics. If the antibiotic group had fewer ulcers, it would suggest that bacteria were causing ulcers.

What could be next?

While it is difficult to grasp just how much more we know about the causes of disease today than back in the days of John Snow, there is still plenty we don’t know about why some people get certain diseases and others don’t. A new study demonstrates how natural experiments may still help elucidate underlying biological pathways (thanks to

Adam Cifu, MD

for tweeting this onto our radar).

Researchers in Europe were curious to see if there was evidence of a factor in the blood that could lead to spontaneous intracerebral hemorrhage (bleeding in the brain without a clear cause). They collected data from hundreds of thousands of Swedish and Danish blood donors and the transfusion recipients who received the donor blood—which is essentially randomly assigned to patients as long as it’s a compatible match. They then looked at both donors and transfusion recipients to see who developed a spontaneous brain bleed in the years following the donation/transfusion.

Photo by Nguyễn Hiệp on Unsplash

They found that if a donor would go on to develop a spontaneous brain bleed, the patients who received a transfusion of their blood were about 2-3 times more likely to develop a spontaneous brain bleed themselves than if they had received a transfusion from a donor who didn’t develop a brain bleed. The findings suggest there could be some yet-unknown factor in the blood that can cause brain bleeding (it’s theorized that it may possibly be amyloid proteins).

Fortunately, spontaneous brain bleeding is a rare outcome for both donors and recipients, and these results shouldn’t stop us from transfusing blood to patients who really need it. But much like John Snow’s study, this natural experiment may reveal something new about how spontaneous brain bleeding can develop—elucidating the mechanism of a disease without even having to see it.