Some of the many other aging clocks developed since have become notable in their own right. Examples include the PhenoAge clock, which incorporates health data such as blood cell counts and signs of inflammation along with methylation, and the Dunedin Pace of Aging clock, which tells you how quickly or slowly a person is aging rather than pointing to a specific age. Many of the clocks measure methylation, but some look at other variables, such as proteins in blood or certain carbohydrate molecules that attach to such proteins.
Today, there are hundreds or even thousands of clocks out there, says Chiara Herzog, who researches aging at King’s College London and is a member of the Biomarkers of Aging Consortium. Everyone has a favorite. Horvath himself favors his GrimAge clock, which was named after the Grim Reaper because it is designed to predict time to death.
That clock was trained on data collected from people who were monitored for decades, many of whom died in that period. Horvath won’t use it to tell people when they might die of old age, he stresses, saying that it wouldn’t be ethical. Instead, it can be used to deliver a biological age that hints at how long a person might expect to live. Someone who is 50 but has a GrimAge of 60 can assume that, compared with the average 50-year-old, they might be a bit closer to the end.
GrimAge is not perfect. While it can strongly predict time to death given the health trajectory someone is on, no aging clock can predict if someone will start smoking or get a divorce (which generally speeds aging) or suddenly take up running (which can generally slow it). “People are complicated,” Horvath tells MIT Technology Review. “There’s a huge error bar.”
On the whole, the clocks are pretty good at making predictions about health and lifespan. They’ve been able to predict that people over the age of 105 have lower biological ages, which tracks given how rare it is for people to make it past that age. A higher epigenetic age has been linked to declining cognitive function and signs of Alzheimer’s disease, while better physical and cognitive fitness has been linked to a lower epigenetic age.
Black-box clocks
But accuracy is a challenge for all aging clocks. Part of the problem lies in how they were designed. Most of the clocks were trained to link age with methylation. The best clocks will deliver an estimate that reflects how far a person’s biology deviates from the average. Aging clocks are still judged on how well they can predict a person’s chronological age, but you don’t want them to be too close, says Lucas Paulo de Lima Camillo, head of machine learning at Shift Bioscience, who was awarded $10,000 by the Biomarkers of Aging Consortium for developing a clock that could estimate age within a range of 2.55 years.
None of the clocks are precise enough to predict the biological age of a single person. Putting the same biological sample through five different clocks will give you five wildly different results. LEON EDLER
“There’s this paradox,” says Camillo. If a clock is really good at predicting chronological age, that’s all it will tell you—and it probably won’t reveal much about your biological age. No one needs an aging clock to tell them how many birthdays they’ve had. Camillo says he’s noticed that when the clocks get too close to “perfect” age prediction, they actually become less accurate at predicting mortality.
Therein lies the other central issue for scientists who develop and use aging clocks: What is the thing they are really measuring? It is a difficult question for a field whose members notoriously fail to agree on the basics. (Everything from the definition of aging to how it occurs and why is up for debate among the experts.)