# Lesson 7: Genetics, Genomics, and Health

*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 seven, Genetics, Genomics, and Health.

**Sarah:** It's meant for after you've finished the lesson. We'll add some perspective on how these ideas show up in practice, some critique of the tools and of the claims made for them, a look at the questions students tend to find thorny in genetics, and a few extra worked examples, including some that go further than the ones on the lesson page.

**Kiffer:** At a few points we'll ask you to work something out before we give the answer. When that happens, 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:** We'll take five questions. First, what does a heritability figure tell you, and what can't it tell you? Second, what can a polygenic risk score do for one person, and what can it do for a whole population? Third, what makes a condition right for newborn screening, and what does a positive screen mean for a family? Fourth, where does reproductive choice end and eugenics begin? Kiffer and I see that one differently. And fifth, whose privacy is involved in a direct-to-consumer genetic test, and what does Canadian law protect?

**Kiffer:** Question one, heritability. It's the proportion of the variation in a trait, in a particular population, that can be attributed to genetic variation. The word variation carries the whole definition, so let's build a made-up town and watch the number move. In this hypothetical town, adult height varies, and we measure that spread with the usual statistic, the variance. Say the total variance is fifty units. Forty of those units come from genetic differences between people, and ten come from differences in their environments.

**Sarah:** So heritability is forty divided by fifty, which is 0.8, or eighty percent. That means that for someone in this town, about eighty percent of their height comes from their genes and the other twenty percent comes from things like what they ate growing up.

**Kiffer:** Try that last sentence on one real person. Picture someone in the town who is a hundred and seventy centimetres tall. Which hundred and thirty-six of those centimetres came from her genes?

**Sarah:** Okay, that doesn't make sense. Genes and nutrition work together on every centimetre, so one person's height can't be cut into slices. The eighty percent describes the spread across the whole town. It says that most of the differences in height between people in this town line up with genetic differences between them.

**Kiffer:** That's the right reading. Now a harder version. Suppose a bad harvest hits the town, and every family's diet gets worse by the same amount, so every child grows up about four centimetres shorter than they otherwise would have. Nobody's genes have changed. What happens to the heritability? Take a few seconds.

*(Pause)*

**Sarah:** My first instinct is that it falls, because the environment just took four centimetres off everyone. But let me check the spread. If everyone loses the same four centimetres, the tallest person is still the tallest, and the gaps between people haven't changed. The total variance is still fifty, and the genetic part is still forty. So the heritability stays at 0.8.

**Kiffer:** Exactly. The average height in the town dropped by four centimetres, and the heritability didn't move at all. That's the clearest way I know to show that a high heritability says nothing about whether the environment matters for the average. It's the Dutch height example from the lesson, run in reverse.

**Sarah:** And what if the bad harvest hit some families hard and left others alone?

**Kiffer:** Then the environment adds spread. Say the uneven harvest adds thirty units of environmental variance. The genetic part is still forty, but the total is now eighty, so the heritability is forty divided by eighty, which is 0.5. Same people, same genes, same trait, and the number fell from 0.8 to 0.5 because the environment became more unequal.

**Sarah:** That's a strange property for a number people quote as if it were a fixed fact about a trait. It also means a high heritability can partly reflect a fairly equal environment.

**Kiffer:** Yes, and it leads straight to the misuse the lesson warns about most strongly, which is comparing groups. Heritability is calculated within one population. Two groups can each have a heritability of 0.8 for height and still differ in average height entirely because of nutrition, the way our town did before and after the harvest. The within-group figure carries no information about the cause of a gap between groups. That error sat at the centre of a lot of eugenic reasoning, and it still turns up in popular writing about genetics.

**Sarah:** Question two. What can a polygenic risk score do for one person, and what can it do for a population? On the individual side, the lesson gives the example of a high score that moves someone's lifetime risk of type 2 diabetes from about ten percent to about fifteen percent.

**Kiffer:** And it helps to hear that change two ways. As a relative change, it's a fifty percent increase, which sounds alarming. As an absolute change, it's five more people in every hundred, and eighty-five of every hundred people with that high score will still never develop the condition. A person deciding what to do with a result needs the absolute numbers.

**Sarah:** The population side is where I think these scores get oversold, so here's a made-up example. A health authority serves twenty thousand adults. Their average lifetime risk of type 2 diabetes is ten percent, so about two thousand of them will eventually develop it. The authority proposes to give everyone a polygenic score and to offer an intensive prevention program only to the top fifth, the four thousand people with the highest scores. In that top fifth, the lifetime risk is fifteen percent.

**Kiffer:** Here's the question for everyone. Of the two thousand people who will eventually develop diabetes, how many are outside the high-score group the program would serve? Take a few seconds.

*(Pause)*

**Sarah:** The top fifth is four thousand people at fifteen percent, which is six hundred cases. The total is two thousand, so the other sixteen thousand people account for fourteen hundred cases. Seventy percent of future cases are in the group the program would leave out.

**Kiffer:** And what is the lifetime risk for the sixteen thousand people outside the high-score group?

**Sarah:** Fourteen hundred divided by sixteen thousand is 0.0875, or a little under nine percent. So the high-score group's risk is about 1.7 times higher. That's a real difference for each person in it, but most of the cases still come from the much larger group at ordinary risk.

**Kiffer:** That's the pattern Geoffrey Rose described in his 1985 paper, Sick Individuals and Sick Populations. He contrasted a high-risk approach, which protects susceptible individuals, with a population approach, which goes after the causes of disease in everyone. A measure that brings a large benefit to the community can offer little to each person, and a measure that helps a few high-risk people a great deal can do little for the population. This is a large part of why the lesson argues that, outside cancer care, population-wide measures such as healthier food environments and blood pressure control often prevent more disease than genetically targeted ones.

**Sarah:** Now let's add the second idea from the lesson, the ancestry gap. Martin and colleagues reported in 2019 that, in people of European ancestry, the scores they studied were about twice as accurate as in people of East Asian ancestry and about four and a half times as accurate as in people of African ancestry.

**Kiffer:** So suppose the same health authority also serves a made-up community of twenty thousand adults in which the score works less well. The average lifetime risk is still ten percent, but the top fifth by score has a risk of only twelve percent. How many of that community's two thousand future cases fall in the top fifth?

**Sarah:** Four thousand times twelve percent is four hundred and eighty. That leaves fifteen hundred and twenty cases, or seventy-six percent, outside the top fifth, and the risk for everyone else is fifteen hundred and twenty divided by sixteen thousand, which is 9.5 percent. So the high-score group's risk is only about 1.3 times higher. The program would be putting its resources into a group that barely differs from everyone else.

**Kiffer:** That's the equity problem from the lesson, worked out in numbers. The same program would do less in the communities where the score is less accurate, and those can be communities that already carry more diabetes and have less access to care. If I were advising that health authority, I'd ask how well the score performs in its own population before using it to target anything, and I'd want the population-wide measures in place either way.

**Sarah:** One caution about our example. We kept it simple by giving everyone in each group the same risk, and real scores sort people along a continuum. The general point still holds. Ask what share of future cases the targeted group contains, and ask how well the score works in the people you're applying it to.

**Kiffer:** Question three. What makes a condition right for newborn screening, and what does a positive screen mean for a family? Newborn screening is the lesson's main example of genetics in public health that works, and the four criteria explain why it works for some conditions and fails for others. Let's classify three made-up conditions. Condition A is a hypothetical enzyme disorder. Babies look healthy at birth, brain damage begins in the second or third week, a special diet prevents it, and the heel-prick test in the first three days detects it reliably.

**Sarah:** And the other two?

**Kiffer:** Condition B is a hypothetical neurological disease that begins in middle age. It can be detected from the same blood spot, and no treatment changes its course. Condition C is a hypothetical metabolic crisis that makes babies seriously ill on the first day of life, before a heel-prick result could come back from the lab. It's treatable once doctors recognize it.

**Sarah:** Let me try. A meets all four criteria. It's serious, there's a window between birth and harm, early treatment works, and the test is accurate. B is serious and detectable, but there's no effective treatment, so finding it at birth hands the family information decades before anything can be done, and the child never gets a say in whether they wanted to know. B fails.

**Kiffer:** And what about Condition C, which is serious and treatable and seems to pass?

**Sarah:** I was going to say it passes. But the screening result would arrive after the baby is already sick, so there's no preclinical phase for screening to catch. Those babies will be diagnosed in the hospital from their symptoms, and screening adds nothing for them.

**Kiffer:** Right, and it's the criterion people forget most often. Screening helps only when it finds a condition earlier than doctors otherwise would, and early enough that treatment changes the outcome. That's also why most adult-onset single-gene conditions stay off newborn panels, however serious they are.

**Sarah:** Now the family's side. Across the whole panel, roughly one baby in every thousand screened turns out to have one of the conditions.

**Kiffer:** So here's a made-up regional screening lab that tests fifty thousand newborns a year. Suppose the panel picks up nearly every one of the babies who have a condition, at about one in a thousand, and it also flags 0.3 percent of unaffected babies by mistake. Every family with a positive screen gets a call and a follow-up test. How many families get that call, and what share of them have an affected baby? Take a few seconds.

*(Pause)*

**Sarah:** Fifty thousand divided by a thousand is fifty affected babies. That leaves 49,950 unaffected babies, and 0.3 percent of that is just under a hundred and fifty. So about two hundred families get a call, and fifty of them have an affected baby. That's one in four.

**Kiffer:** So about three of every four calls go to families whose baby turns out to be fine. That's an accepted cost of a program built to miss as few affected babies as possible, and it's a reason the way those calls are made matters. Every condition added to a panel also brings its own false alarms, so expanding a panel has costs for families as well as benefits.

**Sarah:** Question four, and this is where Kiffer and I disagree. Where does reproductive choice end and eugenics begin? The case I want to use is polygenic embryo screening. During in vitro fertilization, some companies now offer to score embryos with polygenic risk scores so that parents can choose the one with the lowest predicted risk. The first child known to have been selected this way was born in May 2020, after testing by a company called Genomic Prediction.

**Kiffer:** It's worth putting the Canadian history beside it first, because that history gives the word eugenics its weight. Under Alberta's Sexual Sterilization Act, in force from 1928 to 1972, the Eugenics Board approved 4,725 people for sterilization, and 2,834 operations were carried out. Leilani Muir was sterilized in 1959, at fourteen, after being told she was having her appendix removed. She sued the province and won in 1996.

**Sarah:** My position is that polygenic embryo screening follows eugenic logic even with no government involved. It ranks potential children by predicted traits, using scores we just showed are weak for individuals and weaker still outside European ancestry. And when a market offers to optimize children, the pressure to use it tends to grow. I'd restrict it.

**Kiffer:** I agree the science doesn't support the product. In 2022, the European Society of Human Genetics said that using polygenic scores to choose embryos is unproven, and that it would be premature at best. Where I'd push back is on using the word eugenics for every reproductive choice. In Alberta, the state decided, often without consent, who could have children at all. A couple choosing among their own embryos is a different kind of act, and protecting reproductive choice is one of the things we learned from that history.

**Sarah:** The lesson's reflection on prenatal screening makes the counterpoint, though. Individual choices can add up to a population effect, and the message about whose lives are worth living is similar whether a government or a market sends it.

**Kiffer:** That's a fair point, and it's the disability rights argument at its strongest. I'd answer it by regulating the claims and the context. That means no selling a test on evidence that doesn't exist, honest genetic counselling, and real support for families raising children with disabilities. Canada already draws some lines. The Assisted Human Reproduction Act prohibits procedures to select an embryo's sex, except to prevent, diagnose or treat a sex-linked disorder, and it bans changes to the genome that could be passed on to descendants.

**Sarah:** So where do you land, given that we agree about the science and disagree about the label?

**Kiffer:** I'd keep the line the lesson draws between state coercion and individual choice, and I'd watch closely for the ways commercial pressure can blur it. I'd leave embryo testing for serious single-gene conditions available, and I'd keep polygenic scoring of embryos out of clinics until there's evidence that it predicts what it claims to.

**Sarah:** I'm more worried than Kiffer about where choice ends, but we agree on a practical rule. When you assess any reproductive genetic technology, ask who decides, what pressure surrounds the decision, and whether the evidence supports what's being sold.

**Kiffer:** Question five. When someone takes a direct-to-consumer genetic test, whose privacy is involved, and what does Canadian law protect? The answer to the first part is everyone they're related to. First cousins share about twelve and a half percent of their DNA on average, so one person's sample says something about dozens of relatives who never agreed to anything.

**Sarah:** The 23andMe breach shows how far that reaches. In 2023, attackers used reused passwords to get into about fourteen thousand customer accounts. Through a feature that links customers with their genetic relatives, they reached information about roughly 6.9 million people.

**Kiffer:** That works out to about five hundred people exposed for every account that was actually broken into. Then the company went bankrupt, and in 2025 a court approved the sale of the company and its customers' data to a nonprofit led by its co-founder, for 305 million US dollars.

**Sarah:** So let's test the Genetic Non-Discrimination Act on some made-up situations.

**Kiffer:** Here are three. First, a life insurer tells an applicant she must share the results of an old ancestry test before it will sell her a policy. Second, a testing company's terms let it share customer data with a drug company that is a research partner. Third, police upload DNA from a crime scene to a genealogy database and identify a suspect through his cousin's profile. Which of these does the Act address? Take a few seconds.

*(Pause)*

**Sarah:** The first one. The Act makes it an offence to require someone to take a genetic test, or to disclose test results, as a condition of providing goods or services or entering into a contract, and an insurance policy is a contract. The Supreme Court of Canada upheld the Act in 2020, in a five to four decision. The second is only partly covered. The Act also bars disclosing someone's test results without their written consent, but a customer who accepted the company's terms may already have given that consent, so in practice it comes down to the terms and to privacy law. Police use of genealogy databases in the third situation falls outside the Act.

**Kiffer:** Right. The Act protects people from being forced to hand over results. Once you've shared your data with a company yourself, your protection depends on privacy law and the company's terms, and your relatives never had a say. My advice to anyone thinking about one of these tests is to read the data terms first, think about the family members your sample describes, and take any health result that worries you to a genetic counsellor before acting on it.

**Sarah:** That last point matters for the whole family, since one person's decision to send in a sample is also a decision about everyone they're related to.

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

**Sarah:** First, heritability describes the spread of a trait within one population and one environment. It can stay the same while the average changes, and it tells you nothing about one person's trait or about the cause of differences between groups.

**Kiffer:** Second, for any genetic score or test, ask for the absolute numbers and ask about the population. That means asking what share of cases the targeted group holds, how well the tool works in the people it's used on, and what share of positive results are true.

**Sarah:** And third, bring the history with you. For any use of genetics, ask who decides, under what pressure, on what evidence, and whose information, including relatives', is being used.

**Kiffer:** If you'd like more practice, rework the diabetes prevention problem from question two with a risk of twenty percent in the top fifth, and work out what share of future cases the program would still miss.

**Sarah:** Next time, it's Lesson eight, Health Behaviors, Mental Health, and Social Influence, where we look at why people do what they do for their health and how the people around them shape it.

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

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