Infection study, breath timing and AI shame: what this week's health research actually shows

Three new health studies landed within 36 hours of each other, on 28 and 29 September. They come from a World Health Organization agency counting cancer cases, a Northwestern University psychology lab timing people's breathing, and a survey of 1,138 researchers about their own AI habits.

Start with the largest number. Infections caused an estimated 2.3 million new cancer cases worldwide in 2024, or 12 per cent of all new cases, according to a paper published on Monday in The Lancet Oncology and reported by CBC News on 29 September. The work was carried out by the International Agency for Research on Cancer (IARC), an arm of the WHO. The bacterium Helicobacter pylori accounted for 760,000 of those cases, mostly in eastern Asia. Human papillomavirus came close behind at nearly 750,000, with high rates in sub-Saharan Africa. Hepatitis B and hepatitis C were linked to a significant share of liver cancer. For the first time, the IARC team expanded its list to 16 additional cancers. It added HIV as a cause of cervical cancer and Merkel cell polyomavirus as a driver of an aggressive skin cancer.

Epstein-Barr moves up the list

The single biggest revision is Epstein-Barr virus. It was linked to about 260,000 cases in 2024, including nasopharyngeal cancer, some gastric cancers and Hodgkin's lymphoma. "That's quite a big burden of cancer caused by Epstein-Barr virus that was not counted in the past, but is now recognized," said Gary Clifford, a cancer epidemiologist and study co-author based in Lyon.

That matters because EBV has no vaccine and no antibiotic treatment. Clifford told CBC that the virus is "increasingly becoming a target that people need to focus on for prevention." The study also flags the size of the gap between rich and poor countries. The researchers attribute it to screening and vaccination programmes built up over decades in high-income settings.

The regional contrast is stark enough to make the global average misleading. In North America, H. pylori and HPV remain the most common infectious causes, but the overall burden is much lower. In Canada, a 2019 study estimated that four per cent of cancers were attributable to infections, according to Karena Volesky, an epidemiologist in Montreal who led that Canadian research while at McGill University. "It tells us how important each of the 12 infections are," Volesky said of the new global figures. That is roughly one third of the global rate. The gap reflects vaccination and screening rather than biology alone.

A 41 millisecond difference

While the IARC team was counting cases, neuroscientists at Northwestern University were measuring something much smaller. Response speed on an exhalation is faster than on an inhalation by an average of 41 milliseconds, about 1/25th of a second, according to a study published on 29 September in the journal iScience. Responses during the pauses between breaths were 21 milliseconds faster than on inhalation.

"By using a tangible and easy to understand task, we were able to show the relationship between respiration and cognition, which I hope people in a range of fields will find application for," said Erika M. Yamazaki, the study's lead author and a former member of the lab of senior author Ken Paller.

The experiment is small and specific. Thirty-five adults aged 18 to 33 pressed a space bar as quickly as possible when a red square turned yellow, a standard task called the Psychomotor Vigilance Task. They wore a device under the nostrils to measure airflow and repeated the task twice, once before and once after a nap or an eight-hour sleep in the lab. It is the first experiment to measure response speed across all phases of respiration, including the pauses.

Paller was careful about what the result means. "This study documents an important link between respiration and the brain systems for responding to environmental events," he said. "We don't yet know exactly how they are linked, but we suspect neurophysiological efficiency, because other studies have shown that various brain oscillations are synchronized with the rhythms of one's breathing." The lab is now working on sleep apnea research funded by the National Institutes of Health, led by Yamazaki.

Read together, the two studies show the range of what counts as a health finding this week. One estimates millions of cases with a model built on global cancer registries. The other measures a difference of four hundredths of a second in a lab of 35 people. Neither is a wearable study in the consumer sense, but both feed the same pipeline: cheap sensing, large datasets and a willingness to attach numbers to things that were previously left to intuition.

The researchers who use AI and won't say so

The third study turns the lens on the people doing the research. A survey of 1,138 communication scholars in 77 countries, published in Information, Communication and Society and reported by Nieman Journalism Lab on 28 September, found that nearly 70 per cent already used generative AI for at least one research-related task. Most of the same respondents also considered AI use inappropriate for most research-related tasks.

Just over half said using AI for writing texts (55 per cent) and generating code (50.4 per cent) were appropriate. The authors report no consensus on where the ethical line sits. The result, they argue, is a stigma they call "AI shame" and a "transparency paradox": researchers feel obliged to disclose AI assistance while fearing that disclosure makes their work look less credible.

"While such hesitation is understandable given that reporting genAI use might reduce the chances of publication, it ultimately hampers open discussion about how these tools can be used ethically and responsibly in research," the study says. The authors, from National Yang Ming Chiao Tung University in Taiwan, the University of Amsterdam and Utrecht University, recommend that journals and institutions build a "disclosure-positive" culture instead of banning the tools outright.

Hannes Cools, one of the researchers, posted a short thread on Bluesky on 28 September describing the work as the first systematic, field-wide, international empirical investigation of generative AI practices in communication research. The study combined the survey with four focus group discussions.

What the week does not settle

None of these three papers closes a question. The IARC estimate is a model, not a count, and it depends on assumptions about what fraction of each cancer type is attributable to a given infection. The Northwestern result is a timing difference in a small sample, with the mechanism explicitly unknown. The AI survey describes what researchers say they do, in a field where saying it carries a cost.

There is also a limit to how much the three connect. The IARC paper is about prevention at population scale. The Northwestern paper is about milliseconds in a laboratory. The AI survey is about professional norms. What they share is a common method problem: each one has to decide how much of an observed effect to attribute to a cause it cannot directly see.

For readers tracking digital health, the practical takeaway is narrower than any headline. The cancer numbers point to vaccines and antibiotics that already exist and are unevenly distributed. The breathing study points to sleep and respiration as an underused measurement channel, which is exactly the kind of claim wearable makers like to borrow. The AI survey points to a disclosure problem that will follow every algorithm that ends up in a clinic.

Clifford's summary of the cancer work applies more widely than he intended it to. "We really still have a lot of work to do implementing things that we know work and we've known for a long time but haven't really made it into all populations in the world," he said.