Repeated Low-Level Blasts in Military Training Are Linked to Later Anger and Aggression

The fuse, in this story, was never lit by a single explosion. That is what made it so easy to miss. The veterans whose records Eamonn Kennedy began reading had not all lived through a roadside bomb or a battlefield concussion. Many had simply done their jobs.

They fired howitzers. They stood a few feet from breaching charges, day after day, through training exercises that never made the news and never knocked anyone out.

And yet something had accumulated in those people, and Kennedy, a research assistant professor of epidemiology at University of Utah Health and the VA Salt Lake City Health Care System, wanted to know what. When his team pulled the numbers together, veterans from high-blast careers, infantry, artillery, weapons instructors, were turning up in their own medical files with a particular kind of trouble. Anger. Difficulties with self-control.

Seventeen per cent of them carried some documented sign of it, against 12% of comparable veterans whose jobs had kept them clear of the shockwaves. Not a chasm, then, but not nothing either.

Teaching a Machine to Read Three Million Notes

To find that signal, the team had to read more clinical notes than any human group could manage in a working lifetime: roughly 3.6 million of them, scattered across 10,000 veterans, an average of 364 separate entries per person. So they handed the reading to a machine. Anger, after all, rarely announces itself plainly in a medical record.

It hides in phrasing. A clinician might note that a patient discussed anger-provoking situations, or remark, drily, that someone had been a bit more than put out, and the meaning just sits there in the syntax, waiting for someone, or something, to catch it. That last point matters more than it sounds. Some of the words most specific to anger turned out not to be the obvious ones like outburst, but unremarkable scraps of everyday speech that only a reader attuned to implication would ever flag.

So the researchers built a stack of language models, running entirely offline on a single VA server so that no patient data ever left the building, to catch exactly these shadings. One model measured how closely each note’s meaning resembled the idea of anger; another, a version of Google’s Gemma, read the notes much as a person might; a third simply counted telltale word stems, and their combined verdicts were checked, painstakingly, against a thousand notes that humans had labelled by hand. The final system agreed with the human raters 96 per cent of the time. Out of every hundred notes, fewer than three mentioned anger at all. The thing they were hunting was rare, which is precisely why it needed a machine to find it.

A Small Effect That Refuses to Vanish

What it found is unsettling precisely because it is so modest. After the researchers stripped out every confounder they could name, age, combat exposure, traumatic brain injury, the blunt weight of having been to war, low-level blast still nudged the odds of anger upward. “Although the effect was moderate, our findings do suggest that long-term occupational blast exposure is a risk factor for anger, even independently of other military exposures,” Kennedy says.

The honest difficulty is that almost nothing in a soldier’s life arrives on its own. The men who fired the heavy weapons were also likelier to have deployed, to have seen combat, to carry PTSD, of which anger is itself a defining symptom; add PTSD into the statistics and the blast signal nearly washed out, surviving only at the margins.

You can read that two ways, and Kennedy holds both. Either blast is a faint contributor riding alongside far bigger forces, or a high-blast job is really just shorthand for a whole accumulated burden, what he describes as a network of stress, trauma, physical injury and psychiatric illness, that no single number can ever properly untangle.

Still, he keeps circling a word that, here, sounds almost hopeful: modifiable. A roadside bomb cannot be undone; a training drill can be redesigned. Most of this exposure happens not in combat but on the controlled, repeatable world of the range, where distance from the muzzle and rounds fired in a day are things someone could, in principle, decide differently. Blast exposure is largely modifiable, Kennedy notes, because it occurs in very controlled situations that give researchers access for reducing harm, and there is a balance to be struck where troops stay trained and ready for the mission with rather less risk to their long-term health.

There is something faintly vertiginous about the whole enterprise. A machine, reading millions of small human notations, surfaces a pattern too diffuse for any one doctor to have caught, and points back toward an ordinary act, a man firing an artillery piece on a quiet morning, repeated until it leaves a mark on his temperament he may never trace to its source.

https://doi.org/10.1093/milmed/usag217

Frequently Asked Questions

If the blasts are too weak to cause a concussion, how could they affect someone’s mood years later?

That is the puzzle at the heart of this work. Individually these training blasts are subconcussive, meaning they pass without any obvious injury, but the damage researchers suspect is cumulative rather than dramatic: small repeated insults to the brain’s wiring, blood vessels and possibly even the gut, building up over a career. The effect on anger was modest but persisted even after accounting for combat and head injury, which is exactly why it has proved so hard to spot until now.

Why use an AI to read the medical records instead of researchers?

The sheer volume made human reading impossible: 3.6 million clinical notes across 10,000 veterans, roughly 364 per person. Anger also rarely appears as a tidy diagnosis; it surfaces in offhand clinical phrasing that a keyword search would miss entirely. The team trained a combination of language models, ran them offline so no patient data left the building, and checked the results against a thousand human-labelled notes, reaching 96% agreement.

Does this mean military training itself needs to change?

The researchers frame blast exposure as largely modifiable, which is the hopeful part of an otherwise sobering finding. Because most of it happens on controlled training ranges rather than in combat, factors like distance from heavy weapons and the number of rounds fired per session can in principle be adjusted to lower the dose. Some safety measures, such as minimum stand-off distances and overpressure limits, already exist, though the study suggests cumulative exposure deserves closer tracking.

Is anger actually common among these veterans?

No, and that is worth stressing. Fewer than three per cent of all the clinical notes mentioned anger, aggression or violence at all, and the condition was rare in both groups. The finding is about a relative difference in risk between high and low blast exposure jobs, not a portrait of widespread rage among former service members.