Follow a national cohort of 2,307 veterans for seven years, and one number rearranges everything you thought you knew about who attempts suicide.
Of the four strongest predictors, baseline loneliness came out on top—accounting for 41.3% of the explained variance, ahead of self-harm thoughts, low dispositional gratitude, and post-baseline trauma.
Not PTSD. Not a psychiatric diagnosis. Loneliness.
And the risk stacked. Veterans with none of those four factors had a 2.0% probability of attempting suicide. With one, 5.3%. With two, 13.5%. With three, 30.4%. With all four, 55.0%.
Now hold that finding against the arithmetic of the last decade.
Between 2019–2020 and 2025–2026, after years of expanded VA mental-health spending and near-universal depression and suicide screening, the share of veterans reporting frequent loneliness did not fall. It rose—from 19.7% to 22.4%.
That is roughly 460,000 additional veterans who now feel isolated, left out, or lacking companionship most of the time, even as the clinic's toolkit grew larger and better funded.
The money went where the model could measure. Loneliness sat just outside its field of view.
Here is the reflexive assumption that keeps it there. As one VA psychiatrist, Alan Teo, put it: "We typically think of loneliness as a consequence of depression, rather than a cause."
If that were true, the strategy would be sound. Treat the depression, and the loneliness lifts as a side effect—no separate intervention required.
But the longitudinal data cuts the other way. In a seven-year prospective cohort where participants had no suicidal thoughts at enrollment, loneliness didn't trail behind depression as downstream wreckage. It led.
Earlier work from the same research group found the same shape: over four years, post-traumatic stress and related psychiatric problems predicted suicidal thinking, but ranked behind loneliness as a risk factor for the group as a whole.
This matters because it inverts the order of operations. If loneliness is merely a symptom, you can leave it to the antidepressant. If it's a driver, treating the depression while ignoring the isolation is like managing cholesterol and calling it a cure for the artery.
And loneliness has a dose-response signature that behaves like a vital sign. As loneliness climbs from hardly ever to sometimes to often, suicidal ideation rises in lockstep—3.8%, 14.8%, 32.3%—even after adjusting for demographics, military history, trauma, and health.
Frequently lonely veterans carried up to 32-fold greater odds on mental-health outcomes, and 15.9-fold greater odds of future suicidal intent relative to their moderately lonely peers.
So why does the clinic keep missing it?
Start with who does the diagnosing. Most depression, Teo notes, isn't caught by psychiatrists or psychologists—it's caught by primary care doctors, who "see many more patients with depression."
And loneliness, he says plainly, is "probably not at the top of their mind." Not on the radar screen. Providers are "sort of aware" of it when they probe, but it isn't a metric, a diagnosis, or a billable code.
A fifteen-minute med check is built to adjust a dose and screen a symptom list. It is not built to surface the question of whether anyone would notice if you didn't come home.
Then there's the structural cruelty of the risk profile itself. The frequently lonely are disproportionately the unmarried and the unpartnered—precisely the veterans with no one at the appointment reminding them to go, no one flagging the withdrawal, no one to be the second set of eyes.
Among veterans with problematic substance use, loneliness tracked independently with being unmarried, smaller social networks, and—tellingly—lower purpose in life and less optimism. Nearly half of that group reported clinically significant loneliness.
The same VA research found loneliness was tied not only to the highest levels of depression, but to the lowest levels of patients' own efforts to manage their health and seek help.
Read that carefully. The condition that most predicts a suicide attempt is the same condition that erodes the motivation to walk through the clinic door.
The condition that most predicts a suicide attempt is the same condition that erodes the motivation to walk through the clinic door.
The people at highest risk are, by the mechanism of the risk itself, the least likely to be sitting in front of a doctor. A treatment-centric model can only help the veterans who show up. Loneliness selects against showing up.
This is where the VA's own suicide data compounds the problem: veterans not engaged with VA healthcare are at greater risk. The system's field of view narrows exactly where the danger is densest.
None of this makes the clinic useless. It makes the clinic the wrong primary instrument for this particular risk factor.
Which suggests a different opening question. For years, veteran suicide prevention has asked, "Why are veterans killing themselves?"—a question that pulls attention toward individual pathology, trauma, and crisis points.
The reframe some advocates now urge is smaller and more actionable: "How do we prevent loneliness and isolation among veterans?" That shift moves the target from what's broken inside a person to what's missing around them.
And what's missing is often structural, not clinical. Leaving military life removes the camaraderie, the structure, and the shared purpose that had defined a person's sense of belonging—and the absence leaves a void no medication is formulated to fill.
This is why loneliness earns its label as a transdiagnostic risk factor. It doesn't belong to one disorder; it sits beneath many, and it moves before them.
The levers that reach it are non-clinical by design. Peer support networks. Social prescribing—the practice of connecting people to community activities rather than only to pharmacies. Group-based and community engagement that rebuilds the belonging that discharge dismantled.
There's even a hint of what protects. In the substance-use cohort, higher purpose in life buffered the link between depressive symptoms and loneliness—veterans with depression but strong purpose were less likely to be lonely.
An earlier study found the same protective logic: veterans with greater social support, more curiosity, and more confidence in their ability to bounce back were less likely to develop suicidal thoughts. As Pietrzak framed it, prevention "may not only be about fixing what is wrong, but also building what is strong."
Honesty requires a caveat, though. These are targets, not proven cures.
The prevalence data is cross-sectional; it establishes that loneliness travels with suicidality, not that a peer program will reliably reduce attempts. The intervention evidence for social prescribing and community engagement in this population is thin and still accumulating.
What the evidence does establish is harder to dismiss. The single best predictor of who actually attempts—not who thinks about it, but who acts—lives outside the space a diagnosis and a prescription can see.
You cannot medicate a missing community back into existence. And you cannot screen for a risk factor that isn't on the radar, in a patient who never arrives.
The through-line, as with moral injury, is connection—that what wounds in silence heals in company. The number climbed to 22.4% because we kept funding the answer to a question loneliness was never going to hear.