TRISTAN CHIAPPISI · Field Report · No. 3 · June 2026
I · The Claim Is Real II · Control, Not Just Hours III · The Dose Is the Hours IV · The Mechanism V · Can We Call It Cause? VI · The Living Data What This Means
Field Report · No. 3 · 2026 Published 2026-06-07 · Columbus, Ohio

Is Work Stress Fatal?

A field report on the epidemiology of work stress. The honest answer is an association, not a coroner's verdict: more than 120,000 American deaths a year are linked to it, the risk climbs with the dose, and it has held for decades. The surprise is what the data blames. Not effort. Control. Every claim sourced, every estimate flagged.

By Tristan Chiappisi Columbus 12 charts · ~29 min read Connect on LinkedIn →
Not a metaphor · A hazard ratio
Field Report No. 3 · Cinematic Edition

Prefer to watch? Watch the cinematic version.

The same data and the same sources, rebuilt as a particle documentary. About nine minutes. Or keep reading the field report below.

Play the cinematic version →
From the Editor

Work stress is not a feeling. It is an exposure with a body count, and the effect sizes are large enough to put on a chart.

For most of a decade I have worked at a pace I chose, and then could not stop choosing. The work is good. I am a builder, and builders do not punch out. There is always one more thing that would be better if I stayed with it another hour, and for a long time I treated rest as something to get back to later, after the next ship. I am also a father. The two facts sit in the same calendar, and the calendar does not have enough hours in it for both to win.

I used to read "work is killing you" as a figure of speech. A thing people say at the end of a bad quarter, the way they say a meeting could have been an email. Then I went and read the studies. Not the LinkedIn version. The actual cohort papers, the WHO estimates, the Whitehall data that has been running since the 1960s. The figure of speech turned out to be a measurement. There is a hazard ratio attached to it. There is a death count, and it has a confidence interval.

Here is the part that surprised me, and it is the reason this report exists. The data does not blame hard work the way the wellness industry wants it to. It blames something more specific. The single most replicated finding in this literature is not "long hours hurt you," though they do. It is that the people with the least control over their work die younger, on average, than the people with the most, and not because they work harder. They often work less. They just work with less say. The danger is not effort. It is effort you cannot steer.

When I read that, I recognized my own working life in it. I have been lucky. The leadership I work under now gives me real autonomy, the room to choose the right things to work on and how to do them. And when I look back at the stretches that actually wore me down, they were not the longest or the busiest ones. They were the ones where I had no choice, or a hard deadline I had not set and could not move. I got to keep the one thing the data says protects you most. That was luck, and it was good management. Not everyone gets either, and the charts that follow are partly about who does not.

This is a field report, not a wellness essay, so it follows the same rule as the last two: every load-bearing number is sourced to a primary or near-primary study, and where a number is a modeled estimate or an attributable fraction rather than a clean count, that is flagged on the chart, not hidden under it. One caveat belongs up front, because the whole piece rests on it. Almost all of this evidence is observational. It shows association, not a coroner writing a cause on a certificate. The papers say work stress is associated with dying sooner. They stop short of cause. What earns the title is that the associations are large, they step up with the dose, and they replicate across decades and continents. That is as close to the word kill as epidemiology is allowed to get, and it is closer than I expected.

If you ship for a living, or set the pace for people who do, or you are a parent trying to do both, this is the field report I went looking for and could not find. So I read the papers and drew the charts. Here they are.

Tristan Chiappisi, ed. · Field Report No. 3 · It is not a mood · It is a measurement
Seven Numbers, One Thesis
120k
Excess US deaths a year the model links to ten workplace stressors. Its actual range runs 122,000 to 142,000, with wide error bars. The paper's word is "associated with," not "caused."
Goh, Pfeffer & Zenios · Mgmt Science 2016
745k
Global deaths in 2016 from long working hours alone: 398,000 from stroke, 347,000 from heart disease. Up 29% since 2000.
WHO / ILO · Environment International 2021
23.3M
Healthy years of life lost to death and disability (DALYs) from long hours in 2016, the same WHO/ILO accounting behind the 745,000 deaths. The death toll is only the tip.
WHO / ILO · DALYs · 2016
+35%
Stroke risk for people working 55+ hours a week versus 35–40, in the WHO/ILO joint estimates. Heart-disease death risk: +17%.
WHO / ILO · 2021
1.9×
Heart-disease risk for civil servants with persistent low job control versus high (odds ratio 1.93). A companion paper showed personality and coping did not explain it away.
Whitehall II · Bosma et al. · BMJ 1997
$180B
US health spending the same model ties to workplace stress: 5 to 8 percent of the national total. The model's range runs $117B to $190B.
Goh, Pfeffer & Zenios · 2016
41%
Workers who report feeling "a lot of" stress the previous day. Worldwide, 23% are engaged at work. One in five felt lonely yesterday.
Gallup · State of the Global Workplace 2024

I.

The number is real, and rounder than it looks

The Number Has a Citation. And it is bigger, and blurrier, than the round figure suggests.

Start with the hardest version of the claim, the one that sounds like hyperbole until you source it. A 2016 study estimated that workplace conditions are associated with more than 120,000 excess American deaths a year, a toll its authors call comparable to the country's leading causes of death. The number comes from a model, not a morgue, and that suspiciously round 120,000 deserves to be opened up. So the honest move is to show exactly what kind of number it is.

A.

Set the estimate beside the leading causes of death, for scale.

Goh, Pfeffer, and Zenios built a model from public data: ten workplace stressors crossed against the mortality risk each one carries. The figure everyone quotes, 120,000, is what the paper actually rounds down from. Its own phrase is "more than 120,000," and its point estimates run from 122,000 to 142,000, with confidence intervals as wide as 74,000 to 225,000. So the honest figure is not a clean 120,000. It is a broad band that starts around there. Set that band beside the CDC's 2023 causes of death and it keeps company with the top five or six. The authors are careful with the comparison: a toll "comparable to" the largest causes, one that exceeds diabetes and Alzheimer's. Pfeffer's talks sharpen that into "the fifth leading cause of death." The paper does not. Neither will I.

SOURCES: Death counts, CDC/NCHS Data Brief No. 521, "Mortality in the United States, 2023" (final data, Dec 2024), cdc.gov/nchs/databriefs/db521. Workplace-stress estimate: Goh, Pfeffer & Zenios, "The Relationship Between Workplace Stressors and Mortality and Health Costs in the United States," Management Science 62(2):608–628 (2016), doi:10.1287/mnsc.2014.2115. The estimate is "more than 120,000" (model point estimates 122,000–142,000; 95% CI as wide as 74,000–225,000), described by the authors as associated with, not proven to cause, these deaths, and comparable in scale to the leading causes rather than a discrete cause itself.
This is not 120,000 deaths stacked neatly beside the others. It is an attributable estimate. Most of it is already counted inside the top bars of this very chart, heart disease and stroke, because that is mostly how the exposure acts: through the cardiovascular system, over years. The chart is not claiming a new disease, and it is not ranking work as the fifth or sixth thing that kills us. It is showing that the modeled toll is the same order of magnitude as the causes we already fear. Read it as scale, not as a separate column in the ledger. One caveat the American figure earns: the model leans on relative risks whose US replication is the weakest arm of the literature (Nordic cohorts run positive, several US cohorts null), so treat 120,000 as the middle of a wide, US-uncertain band rather than a hard count.
B.

The 120,000 is not one exposure. It is ten.

The model behind Chart A bundles ten distinct workplace conditions, and they are not interchangeable. Pull each one out and stand it next to its best meta-analysis, and the field comes into focus. Losing your job tracks the highest mortality. Long hours, low control, insecurity, and shift work sit in a modest middle band. Effort-reward imbalance brings up the rear. The outcomes differ too, so the chart keeps them in separate colors: a risk ratio for stroke is not the same animal as one for all-cause death.

SOURCES: Unemployment, Roelfs et al. 2011, Soc Sci Med (mortality HR 1.63, 42 studies). Organizational justice, Kivimäki et al. 2006, Scand J Work Environ Health (2 cohorts; thin). Health insurance, Wilper et al. 2009, AJPH (US; single cohort, contested). Long hours, Kivimäki et al. 2015, Lancet. Job insecurity, Virtanen et al. 2013, BMJ (age-adjusted; attenuates toward 1.0 with full adjustment). Shift work, Vyas et al. 2012, BMJ. Job strain, Kivimäki et al. 2012, Lancet. Effort-reward imbalance, Dragano et al. 2017, Epidemiology. All observational. Three stressors in the model (work-family conflict, high job demands, low social support) lack a clean hard-endpoint meta-analysis and are not plotted.
Read the colors and a pattern surfaces. The two largest risks here, unemployment and going uninsured, are not about the work itself. They are about losing it, or carrying it without a net. The work-content stressors, low control, long hours, insecurity, shift work, cluster in the 1.2 to 1.35 band: modest one person at a time. That is the whole point. The body count in Part I does not come from any single heroic hazard ratio. It comes from how many people are standing under the modest ones at once.
Stanford GSB · on Pfeffer's Dying for a Paycheck · 2018
The workplace is killing people and nobody cares.
Headline of the Stanford Graduate School of Business interview · the study's estimate is "more than 120,000" deaths a year

II.

The most replicated finding is also the least intuitive

Control, Not Just Hours. The danger is effort you cannot steer.

If the story were only "hard work hurts," the executives would die first. They do not. The longest-running data on work and the heart says the risk tracks control, the amount of say you have over how the work gets done. Two pictures: the gradient that surprised everyone, and the model that named it.

C.

The Whitehall gradient runs the wrong way for the hustle story.

The Whitehall II study has followed thousands of British civil servants since the 1980s, a workforce sorted into clean grades from clerk to permanent secretary. People with persistent low job control developed coronary heart disease at close to twice the rate of those with high control: an odds ratio of 1.93 over a 5.3-year window. The people at the top, with the most demanding jobs, had the most control and the lowest risk. A companion paper found the link survived adjustment for hostility, coping style, negative affectivity, and minor psychiatric symptoms. Job demands and social support, tellingly, were not associated with the risk at all. Control was.

SOURCES: Bosma, Marmot, Hemingway, Nicholson, Brunner & Stansfeld, "Low job control and risk of coronary heart disease in Whitehall II (prospective cohort) study," BMJ 314(7080):558–565 (1997), PMID 9055714: odds ratio 1.93 (95% CI 1.34–2.77) for persistent low job control; n=6,895 men and 3,413 women, age 35–55 at baseline. The finding that the link held after adjustment for hostility, coping, negative affectivity, and minor psychiatric morbidity (a 1.5–1.8× range) is the companion paper, Bosma, Stansfeld & Marmot, J Occup Health Psychol 3(4):402–409 (1998), PMID 9805284. Observational cohort: association, not proven cause.
The gradient is the finding. Marmot's civil servants were not coal miners. They were office workers in a single organization, with similar commutes and the same national health service. What separated the bottom from the top was status and say, not danger or hours. The clerk with no control over her day carried more cardiac risk than the permanent secretary running the department. That is not a story the wellness app can sell you, because the fix is not on the employee's side of the desk.
D.

The model that named it: high demand plus low control equals strain.

In 1979 Robert Karasek drew the two-by-two that organizes the whole field. One axis is how much a job demands of you. The other is how much control, or decision latitude, you have over how to meet those demands. High demand is not the problem on its own. High demand crossed with low control is. He called that quadrant job strain, and three decades of cardiology have been filling it in. The largest pooled analysis, 197,473 people across thirteen European cohorts, put the heart-disease hazard of job strain at 1.23, adjusted for age and sex (95% CI 1.10 to 1.37). Adjust further for socioeconomic status and it eases to about 1.17, still elevated.

SOURCES: Model, Karasek, "Job Demands, Job Decision Latitude, and Mental Strain: Implications for Job Redesign," Administrative Science Quarterly 24(2):285–308 (1979), jstor.org/stable/2392498. Empirical anchor, Kivimäki et al., "Job strain as a risk factor for coronary heart disease," Lancet 380:1491–1497 (2012), IPD-Work Consortium, n=197,473, 2,358 CHD events; age- and sex-adjusted hazard ratio 1.23 (95% CI 1.10–1.37), attenuating to ~1.17 after adjustment for socioeconomic status. The quadrant diagram is a conceptual model, not a data plot.
The two-by-two explains why the wellness perks miss. A meditation app raises nothing on the control axis. Neither does a fruit basket, a resilience workshop, or a reminder to take your PTO. They land in the demand half of the model, asking the worker to absorb the load better, when the replicated risk lives in the control half. The 1.23 is modest per person. Multiply a modest hazard across a whole working population and you get the population-scale estimates in Part I. The leading honest alternative is residual confounding by social class: job control tracks socioeconomic position, which is why the 1.23 eases toward 1.17 once status is adjusted. That attenuation is the confounding showing itself; what survives it is the 1.17.
E.

Three different theories of work stress. The same heart disease.

If only one research group, using one questionnaire, found this, you should not believe it. But the control story shows up through three lenses that were built separately. The demand-control model (how much say you have), Siegrist's effort-reward imbalance (high effort, thin payoff), and organizational justice (whether you are treated fairly) each have been independently linked to coronary heart disease, sometimes in the very same cohort, though the literature is mixed and job strain is the least consistent of the three. And they stack. In 90,164 people, carrying both job strain and effort-reward imbalance raised the risk well past either one alone.

SOURCES: Additive risk, Dragano et al., "Effort-Reward Imbalance at Work and Incident Coronary Heart Disease," Epidemiology 28(4):619–626 (2017), IPD-Work, n=90,164: either exposure HR 1.16 (1.01–1.34), both HR 1.41 (1.12–1.76). Convergence across models, Sara et al., J Am Heart Assoc 7(9):e008073 (2018). Effort-reward model, Siegrist, J Occup Health Psychol 1(1):27–41 (1996). Organizational justice, Kivimäki et al., Arch Intern Med 165 (2005), HR ~1.5 inverted. Observational; the lower bound on the "either" estimate touches 1.0.
Convergent validity is the quiet workhorse of a causal argument. A single finding can be a fluke of one scale or one sample. Three theories, built by different people to measure different things, that keep landing on the same disease, are much harder to wave off as measurement artifact. They share some conceptual overlap, and the reviewers say so plainly. But the reward model catches something the control model misses, which is exactly why having both is worse than having either.

III.

Where the dose-response is clean enough to draw

The Dose Is the Hours. And the curve only bends one way.

Control is the subtle finding. Hours are the blunt one. Here the data behaves like a textbook exposure: the more you work past full time, the higher the risk climbs, in steps you can see. This is the part of the literature solid enough that the WHO and the ILO put their names on a global death toll.

F.

Stroke risk climbs with the hours, in steps.

The largest meta-analysis of working hours, 603,838 people, found a dose-response for stroke. Against a standard 35-to-40-hour week, working 41 to 48 hours carried 10 percent more stroke risk, though that first step was not statistically significant. 49 to 54 hours carried 27 percent more, and 55 or more hours carried 33 percent more, both significant, with a trend so consistent its p-value runs below 0.0001. Coronary heart disease moved too, but more weakly: only the 55-plus band reached significance, at 13 percent, and the paper reports no clean step-ladder for it.

SOURCE: Kivimäki et al., "Long working hours and risk of coronary heart disease and stroke," Lancet 386:1739–1746 (2015), stroke analysis n=528,908, thelancet.com. Relative risks vs 35–40 h: 41–48 h RR 1.10 (95% CI 0.94–1.28, not significant); 49–54 h RR 1.27 (1.03–1.56); ≥55 h RR 1.33 (1.11–1.61); p for trend <0.0001. CHD at ≥55 h RR 1.13 (1.02–1.26). These are relative risks, distinct from the WHO/ILO joint-estimate model (+35% stroke, +17% IHD death) and from the job-strain hazard ratio in Chart C. Observational: association, not proven cause.
A dose-response is what an epidemiologist looks for before saying "cause." Random associations do not usually climb in steps. This one does, for stroke, though the first step is drawn hollow because it did not reach significance on its own. The coronary signal is real but softer, which is itself honest data: not every outcome moves the same way, and a chart that pretended both lines were equally steep would be lying to make a cleaner picture. The stroke line is the one to trust here.
G.

Scaled to the planet: 745,000 deaths in a single year.

In 2021 the WHO and the ILO ran the first global accounting of long working hours. For 2016 alone they attributed 745,000 deaths to weeks of 55 hours or more: 398,000 from stroke and 347,000 from ischemic heart disease. The toll had risen 29 percent since 2000, with heart-disease deaths up 42 percent and stroke deaths up 19 percent over that stretch. Nearly three in four of the dead were men. About 9 percent of the world works these hours.

SOURCE: WHO/ILO Joint Estimates; Pega et al., "Global, regional, and national burdens of ischemic heart disease and stroke attributable to exposure to long working hours," Environment International 154 (2021), sciencedirect.com; summary at who.int. The ~577,000 figure for 2000 is implied by WHO's stated 29% rise to 745,000 and is shown as a derived comparison, flagged.
This is the most conservative possible version of the claim. It counts one exposure, long hours, and two outcomes, stroke and heart disease. It does not count job strain, low control, job insecurity, shift work, or the work-family conflict that the American model in Part I includes. Add those back and the number grows. The 745,000 is the floor, built from the one exposure solid enough for two UN agencies to co-sign.
H.

The deaths are the tip. The disability is the iceberg.

A death is the part of this that gets counted. It is also the smaller part. The same WHO and ILO accounting that found 745,000 deaths from long hours in 2016 also found 23.3 million DALYs, disability-adjusted life-years: the healthy years of life erased by those deaths and by the strokes and heart attacks people survive but never fully recover from. Most cardiovascular events are survived. A stroke is far likelier to disable than to kill. So the death toll is the floor, not the ceiling.

SOURCE: WHO/ILO Joint Estimates; Pega et al., Environment International 154 (2021), summarized in the WHO/ILO Q&A: long working hours were responsible for about 745,000 deaths and 23.3 million DALYs in 2016, who.int. DALYs combine years of life lost to early death with years lived with disability, so the iceberg shows visible-versus-hidden, not a literal numeric multiple. Stroke as a leading cause of long-term disability: CDC. The US 120,000 model (Goh, Pfeffer & Zenios) estimates deaths and dollars, not a non-fatal event count, so no equivalent US figure is shown.
This is the answer to "120,000 is a lot, but it feels like there is more." There is. The death count is the one number a model like this can put a clean estimate on, so it is the number that gets quoted. Underneath every death sits a far larger body of non-fatal strokes and heart attacks, the ones that do not end a life but reshape it, and they rarely make it onto a chart at all. For the US specifically, nobody has cleanly counted them, which is its own kind of telling. The toll we can count is the smallest version of the toll there is.

IV.

The objection is "it is only correlation." Here is the biology.

How It Reaches the Heart. Stress is not a metaphor in the arteries.

The strongest move against "work stress kills" is that stressed people differ in a hundred ways, and one of those, not the stress, does the killing. The reply is a mechanism. If you can trace a path from a stressful decade to a narrower coronary artery, the bare association stops looking like coincidence. Three decades of physiology have drawn that path, and it is not the tidy single switch the headlines want.

I.

Stressor to artery, in three routes and one syndrome.

Chronic job strain reaches the heart along two direct biological routes and one behavioral one. Directly, it keeps the body's alarm systems switched on: the sympathetic axis raises working blood pressure and flattens heart-rate variability, and the HPA axis lifts morning cortisol. Indirectly, it nudges people toward smoking, inactivity, and bad sleep. Both kinds of route converge on the metabolic syndrome, the cluster of belly fat, high blood pressure, high blood sugar, and bad lipids that runs ahead of heart disease, and the more stress exposures a person carries, the higher the odds, in clean steps.

SOURCES: Pathway reviews, Chandola et al., Eur Heart J 29 (2008); Steptoe & Kivimäki, Nat Rev Cardiol 9 (2012). Metabolic-syndrome dose-response, Chandola, Brunner & Marmot, "Chronic stress at work and the metabolic syndrome," BMJ 332:521–525 (2006), Whitehall II: OR 1.13 → 1.55 → 2.25 (95% CI for 3+ exposures, 1.31–3.85). Ambulatory blood pressure, Landsbergis et al. meta-analysis, PLoS ONE (2013): +3.4/+2.1 mmHg at work. Behavioral odds ratios, Nyberg et al., PLoS ONE (2013), IPD-Work. Behavioral mediation ≈32%, Chandola et al. 2008. Allostatic-load frame, McEwen, NEJM 338 (1998).
The behavioral route explains only about a third of it. Chandola's group put the share of the work-stress effect that runs through health behaviors and the metabolic syndrome at roughly 32 percent. The rest is consistent with the direct routes, the cortisol and the blood pressure. One honest caveat sits on the chart in gray. Inflammation, the most fashionable mechanism, is the weak link here: for work stress specifically, the association with markers like CRP and IL-6 is inconsistent. The biology is real. It is just not a single clean wire from the inbox to the artery.

V.

The question the whole report has been circling

Can We Call It Cause? Score it in the open, and see.

The title says kill. Almost all of the data says associated with. The grown-up way to close that gap is the checklist epidemiologists used to pin smoking to lung cancer: Austin Bradford Hill's nine considerations for moving from correlation toward cause. So score the work-stress-to-heart-disease link against all nine, in public, including the one box it cannot check.

J.

The scorecard, including the test it fails.

Work stress does well on the criteria hardest to fake. Temporality is strong: the stress is measured years before the heart attack, and dropping the early years of follow-up barely moves the result, which argues against the sick simply reporting more stress. Consistency is strong: dozens of cohorts on three continents. Dose-response, plausibility, and coherence are moderate, each propped up by the earlier charts. Strength is honestly weak, the risk ratios modest. And then there is the box that stays empty.

SOURCE: criteria framework, Hill, "The Environment and Disease: Association or Causation?" Proc R Soc Med 58:295–300 (1965). Scoring synthesized from Kivimäki & Kawachi, "Work Stress as a Risk Factor for Cardiovascular Disease," Curr Cardiol Rep 17:74 (2015); Eller et al., Cardiol Rev 17 (2009); and the meta-analyses cited throughout this report. The strong/moderate/weak ratings are the author's reading of that literature, not a published index.
Experiment is the one weak leg, and it carries the most weight. No large randomized trial has ever lowered people's job strain and then counted their heart attacks. That gap is exactly why the careful experts, Kivimäki and Kawachi, stop at "a moderately elevated risk" instead of "proven cause." It is also why this report's title earns a comma and a caveat rather than a coroner's certainty. The evidence is strong nearly everywhere it is allowed to be. It is thin in the one place that ethics and money have kept the trial from running.
K.

The closest thing we have to that experiment.

If you cannot run the trial that lowers stress, look at the two natural experiments that change the work instead. The cleanest one runs the wrong way for comfort. When people lose their jobs involuntarily, a real shock you cannot pin on their personality, heart attacks and strokes roughly double in the years that follow. The other direction, deliberately improving the work, has been tried once at scale in a randomized design, and the result is both encouraging and humbling.

SOURCES: Job loss and cardiovascular events, Gallo et al., Occup Environ Med 63 (2006), Health and Retirement Study: heart attack HR 2.48 (1.49–4.14), stroke HR 2.43 (1.18–4.98). Job displacement and mortality, Sullivan & von Wachter, Quarterly Journal of Economics 124 (2009). Intervention, Work, Family & Health Network: Berkman et al., AJPH 113 (2023, cardiometabolic risk); Olson et al., Sleep Health 1 (2015, sleep); Moen, Kelly et al., Am Sociol Rev 81 (2016, distress). The cardiometabolic benefit was a subgroup effect among higher-risk workers, with no significant main effect.
Read the two directions together and the honest verdict falls out. Removing work damages the heart, clearly and measurably. Improving work helps, but so far only for the workers already at risk, and only on a risk score rather than counted heart attacks. That asymmetry is not a weakness to bury. It is the current state of the evidence, and saying it plainly is what lets the rest of the argument keep its footing. The lever looks real. We have not yet pulled it hard enough, in a clean enough trial, to measure how many lives it moves.

VI.

How many people are standing in the exposure right now

The Living Data. The population the risk is acting on.

Epidemiology tells you the hazard. Gallup tells you the size of the crowd standing in it. Every year they survey workers in more than 140 countries, and the 2024 read is not a story about a stressed minority. It is the working condition of the majority.

L.

Stress is the base rate, not the exception.

In Gallup's 2024 global survey, 41 percent of workers said they felt a lot of stress the previous day. Only 23 percent are engaged at work, the share who feel involved and committed; in the US and Canada that rises to 33 percent. One in five workers felt lonely the day before. Just 34 percent say they are thriving in life overall. The exposure the studies measure is not rare. It is the water most of the workforce swims in.

SOURCE: Gallup, State of the Global Workplace: 2024 Report (figures are the 2024 edition; Gallup refreshes the same landing URL each year, so later editions report different numbers), gallup.com/workplace/349484. Self-reported survey across 140+ countries. Engagement figure is the global rate; the US/Canada regional rate (33%) is marked separately.
Gallup's other 2024 finding is the lever. Managers account for an estimated 70 percent of the variance in team engagement. Engagement is not the same thing as job control, but it rhymes with it, and it points to the same place the cardiology pointed: the conditions are set above the worker, not inside them. The crowd standing in the exposure is most of the people you work with. Some of them are setting the dose for the others.
What this means if you build, lead, or parent

The data does not say stop working. It says stop working this way.

Read the literature end to end and a single instruction falls out of it, and it is not the one the wellness industry sells. The harm is not the effort. It is the combination the studies keep naming: high demand, low control, no end to the hours, thin support. The literature says changing the combination should lower the hazard, though the clean trial that would prove it has not been run. The hazard was never the work itself. It was the shape of the work. That shape is a design choice, and design choices have owners.

1. If you set the pace, control is the cheapest intervention you have.

The most replicated protective factor in this entire literature is decision latitude, and it costs nothing. Give people real say over how and when the work gets done. A perk asks the worker to absorb the load better. Autonomy changes the load. One lands in Karasek's demand half, where the risk is not. The other lands in the control half, where it is. When the US Surgeon General published a workplace mental-health framework in 2022, autonomy, control over when, where, and how the work happens, was one of its five essentials. The cardiology had a name for the same thing forty years earlier: decision latitude. If you only do one thing with this report, move something off your desk and onto theirs.

2. If you are in it, watch the helplessness, not just the hours.

The hours have a clean dose-response and you can count them. The harder exposure to notice is the one Whitehall measured: the slow accumulation of doing work you cannot steer. The 2 a.m. message you cannot ignore. The deadline you did not set and cannot move. Track that the way you track your calendar, because the body is tracking it whether or not you are. The stretches I remember as the worst were never the busiest. They were the ones I could not steer.

3. If you are a parent, the cost is not only yours.

Work-family conflict is one of the ten stressors in the American model, sitting in the same column as the ones that move mortality. The pace you keep is not a private arrangement between you and your job. When researchers followed families across five waves, about two-thirds of the effect of a parent's work-family conflict on a child's mental health ran through the parent it wore down first: the distress, the strained marriage, the shorter fuse at home. The work does not reach the child directly. It reaches the child through you. The studies measure the worker. The household feels it too.

Conclusion.

I started reading this literature to find out whether "work is killing you" was a real claim or a thing people say. The honest answer is narrower than the title, and harder to shake. The evidence is association, not a coroner's verdict. But the associations are large, they climb with the dose, and they replicate across decades and continents. Work without control, without limit, without support is reliably linked to dying sooner, and each of those three has a name, an effect size, and an owner. That is the finding under all the charts: what tracks dying sooner is not the work, it is the shape of it, and a shape is something someone designs. And it is measured in one life. About seventy-eight years, a third of the waking middle spent at work: yours, the people you will see tomorrow, and the one who wrote this.

Sources

Every claim, where to check it.

Goh, Pfeffer & Zenios: The Relationship Between Workplace Stressors and Mortality and Health Costs in the United States · Management Science 62(2):608–628 (2016). "More than 120,000" excess US deaths/yr (model point estimates 122,000–142,000; 95% CI as wide as 74,000–225,000) associated with ten stressors; 5–8% of US health spending (~$180B; range $117–190B); modeled from GSS + MEPS + a 228-study meta-analysis. Observational; the authors flag association, not causation. doi:10.1287/mnsc.2014.2115
Pfeffer: Dying for a Paycheck · HarperBusiness (2018); Stanford GSB, "The Workplace Is Killing People and Nobody Cares" (Mar 2018). Source of the popular "fifth leading cause of death" framing. The peer-reviewed paper says only that the toll is "comparable to" the leading causes; this piece uses that careful wording, not the popular one. gsb.stanford.edu
CDC / NCHS: Mortality in the United States, 2023 · Data Brief No. 521 (Dec 2024). Final data. Heart 680,981 · cancer 613,352 · accidents 222,698 · stroke 162,639 · CLRD 145,357 · Alzheimer's 114,034 · diabetes 95,190. cdc.gov/nchs/databriefs/db521
WHO / ILO: Long working hours increasing deaths from heart disease and stroke · May 17, 2021. 745,000 deaths in 2016 (398k stroke + 347k IHD); +29% since 2000; +35% stroke / +17% IHD-death at 55+ hrs. The 29% total rise and the 72%-male figure appear in this news release rather than the journal paper. who.int
Pega et al.: Global, regional, and national burdens... attributable to long working hours · Environment International 154 (2021), art. 106595. The WHO/ILO joint-estimates primary paper: 745,194 deaths; stroke RR 1.35 (1.13–1.61), IHD RR 1.17 (1.05–1.31) at ≥55 h. Open access: PMC8204267. sciencedirect.com
WHO/ILO Joint Estimates: Q&A (work-related burden, 2000–2016) · Long working hours: about 745,000 deaths and 23.3 million DALYs from heart disease and stroke in 2016 (the disability burden behind the death toll). who.int Q&A
Kivimäki et al.: Long working hours and risk of coronary heart disease and stroke · Lancet 386:1739–1746 (2015). Stroke n=528,908. RR vs 35–40 h: 41–48 h 1.10 (0.94–1.28, n.s.) / 49–54 h 1.27 (1.03–1.56) / ≥55 h 1.33 (1.11–1.61); p-trend <0.0001. CHD 1.13 (1.02–1.26) at ≥55 h. PMID 26298822
Kivimäki et al.: Job strain as a risk factor for coronary heart disease · Lancet 380:1491–1497 (2012). IPD-Work Consortium, 13 European cohorts, n=197,473, 2,358 CHD events. Age/sex-adjusted HR 1.23 (95% CI 1.10–1.37); ~1.17 after adjustment for socioeconomic status. PMID 22981903
Bosma, Marmot et al.: Low job control and risk of CHD in Whitehall II · BMJ 314(7080):558–565 (1997). Odds ratio 1.93 (95% CI 1.34–2.77) for persistent low job control. Demands and social support were not associated; control was. PMID 9055714
Bosma, Stansfeld & Marmot: Job control, personal characteristics, and heart disease · J Occup Health Psychol 3(4):402–409 (1998). The companion paper: low control → 1.5–1.8× CHD, surviving adjustment for hostility, negative affectivity, minor psychiatric disorder, and coping. PMID 9805284
Karasek: Job Demands, Job Decision Latitude, and Mental Strain · Administrative Science Quarterly 24(2):285–308 (1979). The demand-control model; job strain = high demand + low decision latitude. jstor.org/stable/2392498
Gallup: State of the Global Workplace 2024 Report · 41% "a lot of" daily stress; 23% engaged globally (US/Canada 33%); 20% lonely; 34% thriving; managers ≈70% of engagement variance. Figures are the 2024 edition; Gallup updates the same URL yearly, so it now serves later numbers. gallup.com/workplace/349484
Roelfs et al.: Unemployment and all-cause mortality (meta-analysis) · Soc Sci Med 72:840–854 (2011). 42 studies, >20M people. Fully-adjusted mortality HR 1.63 (1.49–1.79). PMC3070776
Wilper et al.: Health insurance and mortality in US adults · Am J Public Health 99(12):2289–2295 (2009). Single cohort; HR 1.40 (1.06–1.84). Contested; later re-analyses weaker. PMID 19762659
Vyas et al.: Shift work and vascular events (meta-analysis) · BMJ 345:e4800 (2012). n≈2.0M. Myocardial infarction RR 1.23 (1.15–1.31). PMC3406223
Virtanen et al.: Job insecurity and incident CHD (IPD-Work) · BMJ 347:f4746 (2013). Age-adjusted RR 1.32 (1.09–1.59); attenuates to ~1.19 (CI to 1.00) fully adjusted. PMC3738256
Dragano et al.: Effort-reward imbalance and incident CHD (IPD-Work) · Epidemiology 28(4):619–626 (2017). n=90,164. Either HR 1.16 (1.01–1.34); both ERI + strain HR 1.41 (1.12–1.76). PMC5457838
Sara et al.: Work-related stress and CHD across three models (review) · J Am Heart Assoc 7(9):e008073 (2018). Demand-control, effort-reward, and justice each associated with CHD; the review calls the body mixed and heterogeneous (job strain positive in 8 of 19 studies). PMC6015274
Siegrist: Adverse health effects of high-effort/low-reward conditions · J Occup Health Psychol 1(1):27–41 (1996). The effort-reward imbalance model. PMID 9547031
Kivimäki et al.: Justice at work and incident CHD (Whitehall II) · Arch Intern Med 165(19):2245–2251 (2005). High justice HR 0.65 (0.47–0.89); survived adjustment for strain and effort-reward. PMID 16246990
Chandola, Brunner & Marmot: Chronic stress at work and the metabolic syndrome · BMJ 332:521–525 (2006). Dose-response OR 1.13 → 1.55 → 2.25 (1.31–3.85). PMC1388129
Chandola et al.: Work stress and CHD, what are the mechanisms? · Eur Heart J 29(5):640–648 (2008). Two routes; ≈32% behavioral mediation; lower HRV, higher cortisol. PMID 18216031
Steptoe & Kivimäki: Stress and cardiovascular disease (review) · Nat Rev Cardiol 9(6):360–370 (2012). HPA and sympatho-adrenomedullary pathways. PMID 22473079
Landsbergis et al.: Job strain and ambulatory blood pressure (meta-analysis) · PLoS ONE 8(6) (2013). Work systolic +3.43 mmHg (2.02–4.84), diastolic +2.07. PMC3673518
Nyberg et al.: Job strain and cardiovascular risk factors (IPD-Work) · PLoS ONE 8(6) (2013). Inactivity OR 1.34, diabetes 1.29, smoking 1.14, obesity 1.12; null for clinic BP/lipids. PMC3688665
Kaltenegger et al.: Working conditions and chronic low-grade inflammation (systematic review & meta-analysis) · Scand J Work Environ Health 47(8):565–581 (2021). Job demand-control vs CRP/IL-6: "weak, if any." The mechanistic weak link (ERI and long hours fared better). PMC9058622
McEwen: Protective and damaging effects of stress mediators · New England Journal of Medicine 338(3):171–179 (1998). The allostatic-load concept. PMC3181832
Kivimäki & Kawachi: Work Stress as a Risk Factor for Cardiovascular Disease (review) · Curr Cardiol Rep 17:74 (2015). The verdict: work stressors are "associated with a moderately elevated risk," not proven cause. PMC4523692
Eller et al.: Work-related psychosocial factors and ischemic heart disease (review) · Cardiol Rev 17(2):83–97 (2009). Nordic studies positive; US studies largely null. PMID 19367150
Hill: The Environment and Disease: Association or Causation? · Proc R Soc Med 58:295–300 (1965). The nine considerations scored in Chart I. PMC1898525
Gallo et al.: Involuntary job loss and acute cardiovascular events · Occup Environ Med 63(10):683–687 (2006). MI HR 2.48 (1.49–4.14); stroke HR 2.43 (1.18–4.98). PMC1839969
Sullivan & von Wachter: Job displacement and mortality · Quarterly Journal of Economics 124(3):1265–1306 (2009). Mortality +50–100% the year after displacement; +10–15% long-term. UCLA PDF
Berkman et al.: A workplace intervention and cardiometabolic risk (WFHN cluster-RCT) · Am J Public Health 113(12):1322–1331 (2023). Subgroup benefit in higher-risk workers; no significant main effect. Sleep: Olson et al., Sleep Health 1 (2015). Distress: Moen, Kelly et al., Am Sociol Rev 81 (2016). PMID 37939328
Amstad et al.: Work-family conflict and outcomes (meta-analysis) · J Occup Health Psychol 16(2):151–169 (2011). 427 effect sizes. Work-to-family conflict ↔ depression r=.23, somatic symptoms r=.29. PMID 21280939
Strazdins / Dinh et al.: Parental work-family conflict and child mental health · Soc Sci Med 70 (2010); Soc Psychiatry Psychiatr Epidemiol (2021). ~66% of the effect mediated via parent distress, marital strain, and parenting. PMID 32813025
U.S. Surgeon General: Framework for Workplace Mental Health & Well-Being · HHS (2022). Five Essentials; "Work-Life Harmony" centers autonomy and flexibility. hhs.gov (PDF)
Holt-Lunstad et al.: Social relationships and mortality risk · PLoS Medicine 7(7):e1000316 (2010), OR 1.50 survival, "comparable with quitting smoking"; and Perspect Psychol Sci 10(2):227–237 (2015), isolation/loneliness ~26–32%. PMC2910600

About this field report.

Field Report No. 3 in an irregular dispatch from Tristan Chiappisi, an engineer who works in data, builds in the AI space, gives talks about both, and writes things down when the data points somewhere it should not have to. The writing is set in Source Serif 4 and Inter Tight. There are no advertisements, sponsored sections, or affiliate links.

Every numerical claim in this issue is sourced to a primary or near-primary study, and triple-checked against it. Death tolls that are modeled estimates or attributable fractions are labeled as such on the chart, and the line between association and proven cause is kept visible throughout. The full citation manifest ships alongside the draft.

If you want to know more, or have me speak on this, reach out on LinkedIn.

Reach me on LinkedIn

What's next

Tristan elsewhere