# Lesson 10: Occupational Health and Worker Safety

*Companion-podcast transcript • Sarah & Kiffer*  
*Office Hours episode to listen to after working through the lesson*

---

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. This episode goes with Lesson ten, Occupational Health and Worker Safety.

**Sarah:** It's meant for after you've worked through the lesson. We'll bring in some perspective on how occupational health works in practice here in Canada, some critique of the evidence and the regulation built on it, a run at the questions students tend to find thorny, and a few extra worked examples, including a couple that are harder than the ones on the lesson page.

**Kiffer:** There are a few places where we'll ask you to work something out before we give the answer. When that happens, you'll hear a few seconds of quiet. Pause the audio if you'd like more time to think it through.

**Sarah:** We have four questions. First, why does the gap between evidence and regulation last for decades, and why is asbestos still killing Canadian workers years after the ban? Second, what does it mean when a group of workers has a standardized mortality ratio below one? Third, how can anyone tell whether a safety law or rule actually worked? And fourth, is burnout a problem with the worker or a problem with the job?

**Kiffer:** Question one, then. Sarah, Pott described the chimney sweeps' cancer in 1775, Britain banned child sweeps in 1875, and Canada banned new uses of asbestos only in 2018. Why do students find gaps like that so hard to accept?

**Sarah:** Because it feels like the science should settle it. Once a cohort shows that insulation workers are dying of mesothelioma at far higher rates than other people, you'd expect the rules to change within a few years. A century, or even forty years, sounds like somebody simply failed to act.

**Kiffer:** Sometimes somebody did, and the asbestos industry's campaign against the evidence is well documented. There's also a second reason, and it's built into the disease itself. It's the latency. Mesothelioma usually appears thirty to forty years after the first exposure. So the people whose deaths prove that a hazard kills are almost always people who were exposed a generation earlier.

**Sarah:** Let me put some made-up numbers on that. Imagine a pipefitter who started work in a Canadian shipyard in 1978 and breathed asbestos dust from the insulation. With a latency of thirty to forty years, his mesothelioma, if it comes, would appear somewhere between 2008 and 2018. So by the time the ban arrived, his disease had already happened or was already on its way.

**Kiffer:** Now run it forward. A hypothetical worker first exposed in 2017, the year before the ban, would develop the disease, if at all, somewhere between 2047 and 2057. And renovation and demolition workers who disturb asbestos in older buildings are still being exposed today. So a ban changes future cases, and the counts we see this year mostly reflect exposures from the 1970s and 1980s.

**Sarah:** You can see that in British Columbia. WorkSafeBC reported a hundred and thirty-eight work-related deaths in 2025. Seventy-nine of them were from occupational diseases, and thirty-six of those were from asbestos exposure.

**Kiffer:** Thirty-six divided by a hundred and thirty-eight is about 0.26, so roughly one work-related death in four in the province that year traced back to asbestos. And thirty-six out of seventy-nine is a bit under half of the disease deaths. That's seven years after the national ban, and given the latency, we should expect asbestos to stay near the top of that list for years to come.

**Sarah:** So latency is part of why regulation lags. If a hazard takes thirty years to show itself, the first clear evidence arrives thirty years after the exposure, and only then does the political fight start.

**Kiffer:** And latency makes that political fight harder. The people harmed are retired or dead, they're spread across many employers, and the company that exposed them may no longer exist. That's the same dispersed harm that kept children in chimneys for a century. A fire that kills a hundred and forty-six people in one afternoon moves a legislature. Several hundred mesothelioma diagnoses a year, scattered across the country, rarely do.

**Sarah:** Canada's own position made it worse. Canadian workers were exposed here at home, and with the big mines at Thetford Mines and the town of Asbestos in Quebec, we were one of the largest producers and exporters in the world, and the federal government defended those exports long after peer countries had banned the material.

**Kiffer:** That's an uncomfortable part of the history, and it matters for how we read the lag. When a hazard supports jobs in a particular region, the people who would bear the cost of regulation are organized and visible, and the people who bear the cost of delay are scattered and often decades away from getting sick.

**Sarah:** Engineered stone seems like the test of whether anything has been learned. Accelerated silicosis can develop within about five to ten years of heavy exposure, so the evidence arrives much faster than it did for asbestos. Fabricators in their thirties are getting sick, and Australia banned the product in July 2024.

**Kiffer:** Which makes it a useful case to watch. The latency excuse is much weaker here, because the harm shows up within a decade. If Canadian provinces still take decades to act, that would suggest the lag owes more to politics and economics than to the biology of the disease.

**Sarah:** Question two. Occupational cohorts are usually summarized with the standardized mortality ratio, or SMR. That's the number of deaths observed among the workers, divided by the number you'd expect if they had died at general population rates for their age and sex.

**Kiffer:** Here's a made-up cohort to try. Four thousand workers at a hypothetical foundry, where silica dust is the main hazard, are followed for twenty years. Based on provincial death rates, you'd expect two hundred and fifty deaths. You observe two hundred. What's the SMR, and what does it tell you about working at this foundry? Take a few seconds.

*(Pause)*

**Sarah:** Let me try this one. Two hundred observed divided by two hundred and fifty expected is 0.8. So the workers had twenty percent fewer deaths than expected. That would mean the foundry isn't harming them, and if anything, working there looks mildly protective.

**Kiffer:** The arithmetic is right. Before you accept the conclusion, though, think about who gets hired at a foundry in the first place, and who stays there for twenty years.

**Sarah:** Oh, of course. People with serious heart or lung disease don't take heavy industrial jobs, and people who get sick tend to leave. The comparison group is the whole population, which includes people too ill to work at all. That's the healthy worker effect. An all-cause SMR of 0.8 is right in the usual range for working cohorts, so it tells us almost nothing about the foundry.

**Kiffer:** Exactly. A working population would be expected to come in below one even if the job were harmless, so a low all-cause SMR can hide real harm. Where would you look next if you suspected the dust was doing damage?

**Sarah:** At the specific causes the hazard should produce. Suppose, still with made-up numbers, we expected twelve deaths from non-cancer lung disease and observed thirty. Thirty divided by twelve is an SMR of 2.5, so two and a half times the expected number, and that's with the healthy worker effect pulling the ratio down.

**Kiffer:** That's the same logic that made the Selikoff insulator cohort so convincing. The insulators were a working population too, with all the selection that implies, and their deaths from mesothelioma, lung cancer and asbestosis still came in far above what was expected. Disease-specific excesses that large are hard to explain away.

**Sarah:** Now a harder version. Split the same foundry cohort by dust exposure. In the low-dust jobs, we expected a hundred and fifty deaths and observed ninety. In the high-dust jobs, we expected a hundred and observed a hundred and ten. A company report says the high-dust workers' mortality is only ten percent above the general population, so the dust risk is small. What's the better comparison, and how large does the dust effect look? Take a few seconds.

*(Pause)*

**Kiffer:** First, check that the pieces add up. Ninety plus a hundred and ten is two hundred observed, and a hundred and fifty plus a hundred is two hundred and fifty expected, so this is the same cohort. The low-dust SMR is ninety divided by a hundred and fifty, which is 0.6. The high-dust SMR is a hundred and ten divided by a hundred, which is 1.1.

**Sarah:** And the better comparison is inside the cohort. Both groups went through the same hiring, so both carry the healthy worker effect. 1.1 divided by 0.6 is about 1.83. So the high-dust workers died at roughly eighty-three percent above the rate of comparable workers in low-dust jobs. That's a much larger effect than the ten percent in the report.

**Kiffer:** That's the internal comparison approach from the lesson. It depends on an assumption, that the two groups would have had similar mortality without the dust, and that they're similar enough in age for the ratio of their SMRs to be a fair comparison. There's also a trap that pushes the estimate in the other direction.

**Sarah:** Are you thinking of workers moving from one group to the other during follow-up?

**Kiffer:** Yes. A worker whose breathing gets worse in a dusty job is often moved to a cleaner job or leaves the work altogether. That's the survivor side of the healthy worker effect. If workers are grouped by the last job they held, the sick ones end up counted in the low-dust group, so the low-dust group looks worse than it should and the high-dust group looks better. In this situation, the eighty-three percent is more likely to understate the dust effect than to overstate it.

**Sarah:** So the company's ten percent figure was misleading twice over. It used the general population as the comparison, and it ignored who moved between jobs.

**Kiffer:** That's a good way to put it. Whenever you read an occupational study, ask who the workers are being compared with, and whether the sickest workers could have drifted out of the exposed group before the counting was done.

**Sarah:** Question three. How can anyone tell whether a safety law or a new rule actually worked? I think this is harder than it sounds, mainly because injury rates are often falling anyway. In Ontario, for example, lost-time claim rates fell between 2000 and 2007 in every age group, by somewhere between about twenty and fifty percent depending on age and sex, and an earlier analysis of Ontario claims from 1990 to 2003 found that fewer people working in manual jobs may explain part of the decline. So if a province brings in a rule and injuries fall afterward, the rule can get credit for a trend that was already under way.

**Kiffer:** Here's a made-up example. Province A brings in a rule requiring fall-protection training on residential construction sites. In the year before, its construction injury claim rate was 3.0 per hundred workers, and two years later it's 2.4. Province B had no new rule, and its rate went from 2.8 to 2.5 over the same period. How much of Province A's drop would you credit to the rule? Take a few seconds.

*(Pause)*

**Sarah:** Province A fell by 0.6 per hundred workers. Province B, with no rule, fell by 0.3. If Province A would have followed the same trend as Province B, then 0.3 of its drop would have happened anyway. So the rule gets credit for the other 0.3 per hundred workers, which is half of the drop we saw.

**Kiffer:** That's a difference-in-differences estimate, one of the designs used to evaluate regulation. Compared with Province A's starting rate of 3.0, a fall of 0.3 is about a ten percent reduction. The key assumption is that the two provinces would have moved in parallel without the rule, so it helps to check whether their rates were moving together in the years before.

**Sarah:** And claims data add a second problem. If the rule also changed how willing workers were to report injuries, the claim rate could move for reasons that have nothing to do with how many people were actually hurt. Underreporting is already a known weakness of compensation data, especially in non-union workplaces.

**Kiffer:** There is at least one randomized test. In California, the state safety agency, Cal OSHA, selected some workplaces for inspection at random between 1996 and 2006. Levine, Toffel and Johnson compared the inspected workplaces with similar ones that weren't picked. Over the four years after inspection, the inspected workplaces had 9.4 percent fewer injury claims and twenty-six percent lower workers' compensation costs, and the researchers found no detectable effect on jobs.

**Sarah:** That's unusually clean evidence, and most laws can't be tested that way. The disaster-driven ones are the hardest of all. Canada's best-known example is probably Westray. On May 9, 1992, an explosion at the Westray coal mine in Plymouth, Nova Scotia, killed twenty-six miners. The federal Westray law, Bill C-45, came into force on March 31, 2004, and it set out how organizations and the people who direct work can be held criminally responsible.

**Kiffer:** Twelve years from the explosion to the law. It follows the pattern of disaster, outrage, inquiry and legislation very closely. Now think about how you'd measure its effect on worker safety. There's no Province B to compare with, because it's a federal criminal law that applies everywhere at once, and part of its purpose is accountability after deaths have happened.

**Sarah:** So the laws that come from disasters are hard to evaluate, and the chronic, everyday harms, like hearing loss or back injuries, are the least likely to get laws of their own.

**Kiffer:** That's a fair summary of the critique. For reading this kind of research, the practical habit is to ask what the comparison group is, and whether something else changed at the same time as the rule.

**Sarah:** Question four. Burnout. In 2019, the World Health Organization included burn-out in ICD-11 as an occupational phenomenon. It describes a syndrome resulting from chronic workplace stress that hasn't been successfully managed, and the WHO says it is not classified as a medical condition.

**Kiffer:** Before we argue about that framing, let's practise telling burnout apart from moral injury, because the two get mixed up all the time. Here's a made-up case. Priya is an emergency nurse. During a winter surge, her hospital is so short of beds that she's told to discharge an older patient she believes isn't safe to go home. She does it, and she can't stop thinking about it. She isn't especially exhausted, and she still feels that her work matters. Burnout, moral injury, or both? Take a few seconds.

*(Pause)*

**Sarah:** This one is mostly moral injury. The harm comes from being made to act against her own judgment about what good care requires. Burnout's three dimensions are exhaustion, depersonalization or cynicism, and a reduced sense of accomplishment, and the case tells us she doesn't have the first or the third. If the same thing kept happening every shift for a year, and she started to feel drained and detached, you'd expect burnout to develop alongside it.

**Kiffer:** Right, and the distinction matters for the response. A mindfulness app does nothing about a shortage of beds. Moral injury points straight at the decisions that the system forces onto individual workers, and that's where the fix has to happen.

**Sarah:** Which brings me to where we disagree. I think the ICD-11 framing, useful as it is, has a cost. Calling burnout an occupational phenomenon, and saying it isn't a medical condition, can leave the person who is struggling in a gap. Their employer may treat it as a health issue, while their doctor may feel there's no diagnosis to treat.

**Kiffer:** I see that risk, and I still think the framing is mostly right. The intervention research points one way. Individual programs like resilience training have modest effects, and structural changes like adequate staffing and manageable workloads do better. If we label burnout as an individual illness, the easy response is to send people to a workshop and leave the schedule alone.

**Sarah:** But people need help in the meantime. A nurse who is exhausted today can't wait three years for a staffing plan. And some people with burnout also meet the criteria for depression, which does need treatment. If the workplace framing makes anyone hesitate to assess the person in front of them, that's a real harm.

**Kiffer:** That's fair, and it's the strongest version of your argument. The overlap with depression is one of the measurement problems researchers keep raising. Where I land is that individual support should be available and easy to reach, and it should never be the whole response. When an organization offers wellness programs and leaves workloads exactly as they were, it's treating the symptom and keeping the cause.

**Sarah:** I'd put more weight on individual care than Kiffer does, but we agree on a practical rule. When you see a burnout program, ask what it changes about the job, and make sure the people affected can still be assessed for depression and other conditions.

**Kiffer:** Gig work raises the same question one step further out. A delivery driver classified as an independent contractor often has no employer whose workplace could be changed, and in most provinces has no workers' compensation coverage, although some provinces have begun to extend it. It's the evidence-to-regulation lag again, with a new kind of work arriving faster than the rules.

**Sarah:** So the question of who counts as an employee turns out to be an occupational health question as much as a labour law one, because it decides whether there's a workplace to fix at all.

**Kiffer:** Let's pull it together with three things to take away.

**Sarah:** First, latency shapes almost everything in occupational disease. A hazard that takes thirty or forty years to cause illness produces late evidence, scattered victims and a long tail of cases after any ban, as the asbestos deaths in British Columbia still show.

**Kiffer:** Second, an SMR below one in a working population is expected because of the healthy worker effect. Look at the specific causes the hazard should produce, and compare more exposed and less exposed workers within the same cohort.

**Sarah:** And third, to judge whether a safety rule worked, ask what would have happened without it. A comparison group, such as a similar province, or better still a randomized design, is what turns a before-and-after story into evidence.

**Kiffer:** If you'd like more practice, rework the foundry example with your own made-up numbers, and see how the dust effect changes when sick workers move into the low-dust jobs.

**Sarah:** Next time, it's Lesson eleven, Social, Economic, and Political Determinants of Health, where we look at how income, policy and power shape who gets sick and who stays well.

**Kiffer:** Thanks for staying with us through some hard history and some careful comparisons. Take care, everyone.

**Sarah:** See you in Lesson eleven.
