An abstract interstitial image featuring a clock and syringe.

Time to Death

Eve Zelickson

Algorithms can now predict mortality with startling accuracy, threatening a new form of digital redlining.

The geneticist Steve Horvath likes to recall a pact he made with his identical twin at the end of high school in the late 1980s—they would dedicate their careers to extending the human lifespan. Horvath stuck to his word. After getting degrees in mathematics and biology, he started building statistical models hoping to answer an intimate, terrifying question: how long do we each have to live before we die?

Horvath began his work on that question in the mid-2000s, as a professor at the University of California, Los Angeles. He was working in a relatively new field called epigenetics, the study of how biological processes alter gene expression without changing our underlying DNA. If you imagine DNA as a ribbon of patterned cloth, epigenetic processes can be thought of as a sort of embroidery that accentuates or conceals its design. 

To illustrate the power of epigenetics, some scientists point to the differences among females in the colonies of certain ant species, including Camponotus floridanus, the Florida carpenter ant. All the females in a floridanus colony share the exact same set of genes, but their behaviors and lifespans are dramatically different. Some serve as foragers, others as soldiers, living about a year and never reproducing; only one ant rules as queen, laying eggs and living for up to thirty years. 

Horvath began looking for epigenetic processes that might correlate with aging in humans. One was a process called DNA methylation, in which small molecules of carbon and hydrogen attach to particular parts of a person’s genetic code, with the ability to activate or suppress the underlying genes. In 2011, Horvath measured methylation in DNA from saliva samples and found that the more cells a person had with a particular methylation pattern at three DNA sites, the older a person tended to be. The sample was small, but the correlation was strong. DNA methylation predicted age to within five years in more than 85 percent of the subjects.

To find out how robust this relationship was, Horvath began gathering publicly available datasets that included information about subjects’ age and DNA methylation patterns. By 2012, he had gathered eighty-two datasets with eight thousand samples in them. The data came from a wide range of people of different ages, and from different parts of the body: cord blood from newborns in various parts of the world; brain, stomach, lung, liver, breast, and uterine tissue; sperm; immortalized B cells from people with a rare genetic disease. He even gathered data on chimpanzees. 

Horvath used about half of these datasets to train an algorithm to look for associations between DNA methylation and age. Then he tested the algorithm on the other half of the datasets. What he found was so remarkable that several academic journals rejected his results out of hand. The pattern of methylation at just 353 of the three billion or so pairs of DNA nucleotides in the human genome corresponded to a person’s age with 96 percent accuracy, an unprecedented degree of correlation between a biomarker and the process of aging. After Horvath finally convinced a journal to publish his results, other researchers began to replicate them. 

Horvath referred to his successful algorithm as an “epigenetic clock”: insert some body tissue, and it would spit out your epigenetic age. He hypothesized that people with a higher epigenetic age than chronological age (the one we mark with birthdays and greeting cards) might be at a higher risk of early death. With a group of fellow researchers, he set out to develop a new algorithm that would analyze methylation patterns at DNA sites associated with mortality and compare them to the age of death of subjects in large longitudinal studies. It turned out that the 5 percent of people with the highest epigenetic age relative to their chronological age were twice as likely to die prematurely as an average person of their chronological age. Horvath and his colleagues called this new epigenetic clock GrimAge, after the grim reaper, because it seemed to have the disturbing potential to tell if you were headed for a premature death.

This was a scientific breakthrough with an uncommonly wide range of applications, some of them quite unsettling. Almost as soon as Horvath began publishing the research on his clocks, people began to realize their potential to shape domains from law enforcement to healthcare. Scientists started to investigate whether the algorithms could be used to help solve crimes by using genetic material left at crime scenes—blood, hair, skin cells, bodily fluids—to determine the age of unknown victims or perpetrators. Online, a half-dozen or more companies sprang up selling versions of Horvath’s clocks for around $200, promising prophetic insights to largely affluent customers concerned with prolonging their youth. These companies are like 23andMe, but instead of purporting to reveal your ancestral past, they claim to divine, and allow you to intervene in, your biological future.

This rush to commercialize epigenetic clocks involves a number of potential pitfalls. Some industries are likely to use the clocks to wittingly or unwittingly entrench various forms of social inequality. Other commercial applications of the clock are based on a fundamental and possibly willful misunderstanding of the science that shifts responsibility for social and structural problems on to individuals. At the same time, this commercialization obscures something more hopeful: the democratic potential of the clocks.

Distributing Death

The industry that has so far taken the largest interest in epigenetic clocks—and where the dangers of those clocks are most clear—is the four-trillion-dollar-a-year life insurance industry. Life insurance offers an important form of financial security to the heirs of people who can afford it. For example, if I die before paying off my mortgage, life insurance might mean my kids can still keep a roof over their heads.

The industry’s business model involves using troves of personal and population-level data to bet on how long individuals will live, and charge them accordingly. Underwriters collect data from applicants on things like personal and family medical history, occupation, lifestyle and hobbies, and then gather data on the backend from driving records, criminal records, prescription history, credit reports, and more. Underwriters feed a portion of this data into an algorithm, which helps them determine a risk score that measures how likely an applicant is to die during the term of the policy. 

You pay for the risk you represent, no matter its source. This principle is known in insurance as “actuarial fairness.” Those who are young and healthy can get more financial protection for less money, while the older or more at risk of death you are deemed to be, the more expensive and less accessible that protection becomes. Maybe you’re a wealthy suburbanite who likes racing ATVs on the weekend, or a formerly incarcerated person who now works on a fracking rig, or a single mother with asthma living near a superfund site—all of this helps determine your risk score and how much insurance you can access. Life insurance, in short, is a system built on discrimination. 

A reliable test for how one’s cells are aging could be a transformative new tool for the industry. It’s illegal to explicitly discriminate against applicants on the basis of race, gender, or other categories protected by the law—life insurance companies used to deny coverage to Black people and women, but allowed masters to insure their slaves—but epigenetic underwriting could make such discrimination possible through a new form of algorithmic redlining. That’s because epigenetic tests can measure chronic environmental and psychological stresses that often map on to race, class, and gender. For example, one study used one of Horvath’s epigenetic clocks to examine blood samples from 392 Black adults and found that high lifetime stress correlated with accelerated epigenetic aging. Similarly, populations living in highly polluted areas, oftentimes the poor, will have distinctive methylation patterns that could allow the life insurance industry to further discriminate based on class.

There’s a second way that epigenetic tests are likely to be used in the life insurance industry, one that carries a related risk. Epigenetics has been heralded within both academia and the media as the answer to the longstanding nature-versus-nurture debate: proof that our surroundings and behaviors influence us on a molecular level. Using this logic, the life insurance industry—along with companies offering consumer epigenetic tests—is likely to use epigenetic clocks to offer personalized health insights and behavior-modification programs to its customers. High epigenetic age? Trying eating more kale! This could also lead to discount programs that reward policyholders who successfully lower their epigenetic age. Similar discount schemes are commonplace in auto insurance, where companies like TrueMotion allow policyholders to download an app that tracks their actions behind the wheel and gives discounts or makes price hikes based on safe or risky driving.

The problem with this, of course, is that nurture is nothing like driving. Nurture is the air quality and poverty levels in the neighborhood you grew up in; where you went to school and what you ate for lunch; the way you were treated by your parents and society; the life stresses you have experienced. Like aging processes, all of this is gendered, classed, and racialized. Despite what the life insurance and other industries might claim about personalized behavior-modification programs based on epigenetic tests, you can’t meditate your way out of structural inequality. But this misunderstanding about the nature of nurture is useful to such industries—it generates profits by shifting risk and responsibility for wellbeing from societies, corporations, and governments onto individuals. 

That’s not to say nothing is under our control. Some things surely are, like diet and smoking. Be that as it may, we have very little idea how precisely these choices affect epigenetic processes of aging. A 2016 study by Brian Chen, a scientist who worked with Steve Horvath and is now at a life sciences company that is trying to pioneer epigenetic underwriting, found that the correlation between early mortality and methylation remained even after controlling for lifestyle factors such as exercise levels and smoking status. This suggests that epigenetic clocks capture some aspect of biological aging that is divorced from—and can’t be altered by—lifestyle. As Chen recently told me, “We don’t know, as the scientific community, because the science is so new, what will alter one’s epigenetic clock.”

Proof of Harm

Though our genes differ, shared experiences and circumstances shape us in similar ways. The life insurance industry may want to use epigenetic tests in a manner that entrenches inequality and pernicious myths of personal responsibility, but the science can also be used to strengthen the case for action on social and systemic problems.

“The health impact of environmental and social inequalities, phenomena that occur outside the body, is now identifiable, measurable, and potentially treatable within the body,” Charles Dupras, a bioethicist at the University of Montreal, has written. There are many situations in which molecular-level proof of harm could bolster advocacy work and the case for regulations. For example, multiple existing environmental policies—such as the Clean Air Act, the Safe Drinking Water Act, and the Toxic Substances Control Act—require agencies to assess whether an atmosphere harms human health, and epigenetic tests could be an important source of proof. 

Epigenetic tests could also lead to safer working environments. Under the Occupational Safety and Health Act, the cornerstone federal law regulating workplace safety, employees are entitled to a workplace free from “recognized hazards that are causing or are likely to cause death or serious physical harm.” Mark Rothstein, the Director of the Institute for Bioethics, Health Policy and Law at the University of Louisville School of Medicine, argues that epigenetic tests could capture the adverse effects of everything from toxins at work to a hostile work environment.

When I asked Dupras why he believes that we need epigenetics to promote findings we already know to be true—pollution is bad for your health, exercise can help you stay healthy—he said that epigenetics meets the needs of both the social scientist pushing for equitable policy and the biotech company looking to make advances in precision medicine. Sometimes these things seem at odds—we hunt for cellular therapies when we should be working to change the environmental conditions that cause illness in the first place—but there’s no reason epigenetic science can’t help us do both. 

As the field matures and we learn more about how our environment impacts our biology, we should be skeptical of market solutions promising individualized interventions. As consumers, we should recognize epigenetic testing for what it is: exciting molecular confirmation of the fact that healthy environments promote longer lives. As a society, we should acknowledge it for what it could be: a discovery that will help us to draw roadmaps toward a better collective future.

Eve Zelickson is a writer and researcher at Data & Society.

This piece appears in Logic's issue 13, "Distribution". To order the issue, head on over to our store. To receive future issues, subscribe.