# Lesson 4: Nutrition, Physical Activity, and Sleep

*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 four, Nutrition, Physical Activity, and Sleep.

**Sarah:** It's meant for after you've worked through the lesson. We'll add some perspective on where these ideas came from and some critique of the evidence behind them, work through a few questions students tend to find thorny, and do some 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. You'll hear a few seconds of quiet, and you can pause the audio if you'd like more time to think it through.

**Sarah:** Here's what we'll cover. First, why fortification worked so well against deficiency diseases, and why that success is hard to repeat for chronic disease. Second, why nutrition science is so contested, including a measurement trap that can flip a finding. Third, whether the London bus driver result could be explained by who ended up in which job. Fourth, what the U-shaped curve for sleep and mortality means. And fifth, a question Kiffer and I see a little differently, about how much weight to put on advice to individuals.

**Kiffer:** Question one. Deficiency diseases were among the great successes of twentieth-century public health, and folic acid in Canada is a good place to start. In November 1998, Canada made it mandatory to add folic acid to white flour, enriched pasta and cornmeal. A study by De Wals and colleagues, published in 2007, looked at seven provinces and found that neural tube defects fell from 1.58 per thousand births before fortification to 0.86 per thousand once fortification was fully in place. That's a drop of about forty-six percent.

**Sarah:** Let's make that concrete with a made-up province. Suppose it has forty thousand births a year, and the before and after rates from that study apply to it. About how many fewer neural tube defects would it see each year once fortification is fully in place? Take a few seconds.

*(Pause)*

**Kiffer:** Before fortification, 1.58 per thousand births, times forty thousand births, is forty times 1.58, which is about sixty-three. Afterward, forty times 0.86 is about thirty-four. So the difference is about twenty-nine fewer neural tube defects a year in a province of that size.

**Sarah:** And what I find most interesting is why advice hadn't already done this. Women had been told to take folic acid supplements for years before 1998.

**Kiffer:** They had, and the same paper reported that the recommendations were not routinely followed. In Quebec, roughly three in ten pregnant women reported taking vitamins before conception, both before and after fortification began. The timing makes it harder still. The neural tube closes about four weeks after conception, often before a woman knows she's pregnant, so a supplement started after a positive test usually comes too late. Fortified flour reaches everyone who eats bread or pasta, whether the pregnancy was planned or unplanned.

**Sarah:** So fortification worked because one missing nutrient caused one disease, and one change to the food supply could replace it. Why doesn't that model carry over to heart disease or type 2 diabetes?

**Kiffer:** Those diseases have no single missing ingredient. Scurvy has a necessary cause, so someone with enough vitamin C can't develop it. Heart disease has many contributing causes, each adding a modest amount of risk over decades, and removing one shifts the risk only a little. That's why Lind could settle a question with twelve sailors and six days, while research on diet and chronic disease needs large cohorts, long follow-up and trials that are very hard to run. Diets are hard to blind, people drift away from them, and the effects are small.

**Sarah:** Question two. Why is nutrition science so contested? Students often come away from the Seven Countries Study and the sugar story thinking the whole field is unreliable.

**Kiffer:** I'd put it more carefully. The field has three recurring weak points. The first is design. Keys compared whole populations, and population averages can hide what's happening inside them. His earlier 1953 graph showed six countries when data for twenty-two were available, and in 1957 two critics, Yerushalmy and Hilleboe, argued that the choice of countries greatly exaggerated the association. The second is funding. In 2016, Cristin Kearns, Laura Schmidt and Stanton Glantz showed that the Sugar Research Foundation had paid for a 1967 review in the New England Journal of Medicine that played down sugar and blamed fat, and the funding was never disclosed.

**Sarah:** And that wasn't an isolated case. A 2013 analysis by Maira Bes-Rastrollo and colleagues looked at seventeen systematic reviews on sugary drinks and weight gain. Reviews whose authors had financial ties to industry were about five times as likely to conclude that there was no association.

**Kiffer:** The third weak point is measurement, and I'd like to spend some time on it, because it can quietly reverse a finding. Most diet studies ask people what they ate. Doubly labelled water measures how much energy a person burns, and in someone whose weight is stable, energy in has to equal energy out. When researchers compare the two, people under-report what they eat, and heavier people tend to under-report more. A 2015 study of five populations of African descent found that under-reporting rose steadily with body mass index.

**Sarah:** Activity has a similar problem in the other direction. In the Canadian Health Measures Survey from 2007 to 2009, accelerometers showed that only about fifteen percent of adults reached a hundred and fifty minutes of moderate to vigorous activity a week, while more than half of Canadians described themselves as at least moderately active.

**Kiffer:** Right. Here's the harder example, and it's made up. A hypothetical study has two groups. The leaner group truly eats twenty-three hundred calories a day and under-reports by ten percent. The heavier group truly eats twenty-six hundred calories a day and under-reports by twenty-five percent. What will each group report, and what would a researcher who trusted the reports conclude? Take a few seconds.

*(Pause)*

**Sarah:** Ten percent of twenty-three hundred is two hundred and thirty, so the leaner group reports two thousand and seventy calories. Twenty-five percent of twenty-six hundred is six hundred and fifty, so the heavier group reports nineteen hundred and fifty. That means the heavier group appears to eat a hundred and twenty calories a day less than the leaner group, when in truth it eats three hundred more. A researcher who trusted the reports might conclude that eating more has nothing to do with body weight, or even that it protects against weight gain.

**Kiffer:** That's the trap. Error that's the same in every group usually blurs a real difference. Error that differs between groups can create a false difference or reverse a true one. Now a second step. Suppose a validation substudy tells you the heavier group under-reports by twenty-five percent. How would you correct their reported nineteen hundred and fifty?

**Sarah:** I'd add back twenty-five percent. Nineteen fifty plus a quarter of nineteen fifty is about twenty-four hundred and thirty-eight. Hmm, that doesn't match the twenty-six hundred we started with.

**Kiffer:** Good, you checked it against the starting value. So which amount was the twenty-five percent taken from in the first place?

**Sarah:** From the true intake. The reports are seventy-five percent of the truth, so I divide by 0.75. Nineteen fifty divided by 0.75 is twenty-six hundred. Now it matches.

**Kiffer:** That's the logic behind pairing questionnaires with doubly labelled water or other biomarkers in a subsample of participants. And Keys's central claim, that population diets shape heart disease, has held up through decades of later work. The practical habit is to ask how diet was measured, whether the error could differ between groups, and who paid for the study.

**Sarah:** Question three. Jerry Morris found that London bus drivers had more coronary heart disease than conductors, with the biggest difference in sudden deaths. The obvious worry is selection, and I'd like to put some numbers on it.

**Kiffer:** Here's a made-up version. Picture a thousand drivers and a thousand conductors, and suppose the job has no effect on the heart at all, so each group would go on to have thirty heart attacks during the study. Now suppose ten conductors develop early chest pain, find the stairs too hard, and transfer to driving, and all ten later have heart attacks. The study counts each man by the job he holds when the heart attack happens. What's the risk in each group, and how do they compare? Take a few seconds.

*(Pause)*

**Sarah:** Okay. The conductors lose ten men, so they're down to nine hundred and ninety, and the drivers go up to a thousand and ten. Each group still has thirty heart attacks, so it's thirty out of nine hundred and ninety compared with thirty out of a thousand and ten. That's about three percent in both, so no real difference.

**Kiffer:** Check that against what we assumed about those ten men and when their heart attacks happen.

**Sarah:** Oh, the ten heart attacks move with them. They happen to men who are drivers by then. So the conductors have twenty heart attacks among nine hundred and ninety, which is about two percent, and the drivers have forty among a thousand and ten, which is about four percent.

**Kiffer:** And when you divide the drivers' risk by the conductors' risk, what ratio do you get?

**Sarah:** Four percent divided by two percent is about two. So the drivers look twice as likely to have a heart attack, and in this made-up example the job did nothing at all.

**Kiffer:** Ten men out of two thousand were enough to do it. Your first answer kept those heart attacks with the conductors, and that points to the fix. If you classify every man by the job he was hired into, the transfers and their heart attacks stay with the conductors, each group has thirty in a thousand, and the difference disappears. Trials use the same idea when they analyse participants in the groups they were assigned to.

**Sarah:** Did Morris check for this kind of selection himself?

**Kiffer:** He did, in a clever way. In a 1956 paper, Morris and colleagues used the measurements taken for the busmen's uniforms. Drivers measured more around the waist than conductors, even among the youngest men, which suggested the difference was there before they started the job. That's evidence of selection. Yet in follow-up, sudden death as the first sign of coronary disease was still more than twice as common in drivers, whatever their physique.

**Sarah:** So selection was real, and it still didn't explain the whole difference.

**Kiffer:** That's how I read it. The bus study generated the hypothesis, and later trials and other designs supported the idea that activity protects the heart. It's a good model for reading any natural experiment. Ask who moved between the groups, and why they moved.

**Sarah:** Question four is about sleep. Both short and long sleep are linked with higher death rates. A 2010 meta-analysis by Francesco Cappuccio and colleagues pooled about 1.4 million participants and found a relative risk of about 1.12 for short sleep and 1.30 for long sleep, compared with seven to eight hours a night. Students sometimes read that as saying long sleep is worse for you than short sleep.

**Kiffer:** Let's test that with a made-up cohort. A thousand people who sleep seven to eight hours each have a two percent chance of dying during follow-up. A thousand long sleepers are made up of eight hundred and fifty healthy people with that same two percent risk, and a hundred and fifty people with an illness nobody has diagnosed yet, which makes them sleep longer and gives them a ten percent risk. Sleep itself does nothing here. What's the risk among the long sleepers, and the relative risk compared with normal sleepers? Take a few seconds.

*(Pause)*

**Sarah:** The healthy long sleepers give two percent of eight hundred and fifty, which is seventeen deaths. The ill ones give ten percent of a hundred and fifty, which is fifteen. That's thirty-two deaths in a thousand, or 3.2 percent. Divided by two percent, that's a relative risk of 1.6.

**Kiffer:** So we get a relative risk larger than the one in that meta-analysis, produced entirely by hidden illness. This is reverse causation, because the illness causes the long sleep. Cappuccio and colleagues made a similar point themselves. They suggested that long sleep may be better read as a marker of ill health, and possibly a useful signal of illness that hasn't been diagnosed.

**Sarah:** Then why does the short-sleep side hold up better, when both sides come from the same kind of observational study?

**Kiffer:** Because experiments back it up. Sleep-restriction studies cut people's sleep for a few days under controlled conditions and measure what happens, and they find effects on metabolism, the heart and blood vessels, immune function and thinking within days. So the two arms of the U look alike on a graph and rest on very different evidence.

**Sarah:** Shift work raises a related question about evidence. The International Agency for Research on Cancer, or IARC, classified night shift work as probably carcinogenic in 2007, and kept that classification when it looked again in 2019.

**Kiffer:** Yes, and the 2019 review rated the evidence in humans as limited, with sufficient evidence in animals for disrupted light and dark schedules, and strong mechanistic evidence. It's worth knowing what a classification like Group 2A tells you. It describes how strong the evidence is that something can cause cancer. It doesn't tell you how much risk a given amount of night work adds.

**Sarah:** Question five, and this is where Kiffer and I disagree a little. The lesson has a line comparing telling a shift-working nurse to get more sleep with telling a nineteenth-century slum dweller to drink cleaner water. It ends by calling behaviour change without structural change a moral position. I think that goes too far.

**Kiffer:** Go ahead and make your case, Sarah, and then I'll say where I think the line should sit.

**Sarah:** Structural change is slow. Limits on consecutive night shifts and longer rest between shifts may take years to win, and the nurse working nights now still needs help. Advice about darkening her bedroom, timing her light exposure and planning naps may help her today. If clinicians hear that giving advice is a moral position, some of them may stop giving it, and that doesn't serve anyone.

**Kiffer:** I agree the advice is worth giving, and the line grants that the advice is correct. My concern is what happens when advice is the whole response. It puts the burden on the person with the least control over the schedule, because the employer sets the rotation. Forward-rotating schedules, enough rest between shifts and limits on consecutive nights reach every nurse on the ward, including the ones who never hear the advice.

**Sarah:** Individual tools are changing, though. GLP-1 drugs act on one person at a time, and the weight loss in the trials is large.

**Kiffer:** They do, and they show the limits of a strict line between individual and structural approaches. They also bring back the equity question. The drugs are expensive and need long-term use, so who gets them depends on drug coverage and income, which are structural conditions in their own right.

**Sarah:** So where do you land on it?

**Kiffer:** I think advice to individuals is part of good practice, and I'd give it. I'd judge a public health response by whether it changes the conditions that make the healthy choice possible for the people with the least room to choose. When a response stops at advice, I think we should say so plainly.

**Sarah:** I still think the moral position wording is too strong, but we agree on the practical rule. When you give advice, name the structural condition that would make it easier to follow, and push for that condition as well.

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

**Sarah:** First, deficiency diseases yielded to fortification because one missing nutrient caused one disease, and one change to the food supply reached people that advice never reached. Chronic diseases have many small causes, so the evidence about them builds slowly.

**Kiffer:** Second, check how an exposure was measured. Error that differs between groups, such as heavier people under-reporting more, can reverse a finding, which is why researchers check self-report against doubly labelled water and accelerometers.

**Sarah:** And third, before you read an association as a cause, ask who moved between the groups and which way the arrow could run. Sick conductors moving to driving and hidden illness lengthening sleep can each produce a convincing ratio with no causal effect behind it.

**Kiffer:** If you'd like more practice, rework the bus driver example with twenty transfers in place of ten, and then see what happens when you classify each man by the job he was hired into.

**Sarah:** Next time, it's Lesson five, Sexual and Reproductive Health, where we look at how contraception, maternal health and sexual health became public health concerns.

**Kiffer:** Take care, everyone.

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