The life insurance business is all about betting on how long you’re likely to live. Now, one company is turning to the hot, but still unproven, field of epigenetics to try to make that bet more scientific.
GWG Life, which buys life insurance policies from people who don’t want or can’t afford them anymore, last month started requiring those people to turn over a saliva sample. Its quarry: patterns of DNA methylation. In layman’s terms, it analyzes the samples to see whether certain genes are switched on or off at hundreds of specific spots.
In theory, that could help the company predict your life span. In theory.
“Is that more predictive than whether someone smokes or drinks or has a hobby of alligator wrestling? I don’t see that,” said Mark Rothstein, a bioethicist at the University of Louisville who studies the use of genetic and epigenetic information.
GWG is just the latest in a rush of entrepreneurs peering into DNA for clues about how fast people are aging.
Several companies have started marketing mail-order tests to measure the lengths of people’s telomeres. (Those are the caps of DNA at the end of chromosomes; frayed telomeres have been linked to disease risk.) A California biotech called Zymo Research last year launched a service that uses DNA methylation analysis to help researchers determine the biological (as opposed to the chronological) age of a sample.
Then there’s GWG, which hopes that its model will not only boost its own business but transform the entire life insurance industry.
“We just think that we may have stumbled on something that has some pretty broad and important applications for a much larger industry,” GWG chief executive Jon Sabes said.
Based in Minneapolis, GWG operates in most states and bought 315 life insurance policies last year. Its sales pitch: We’ll pay you up front for your policy. We’ll pay your premiums for the rest of your life. And when you die, we get the insurance money.
To make a profit, it’s crucial for companies like GWG that operate in this somewhat macabre niche to accurately predict how long each policy holder is likely to live. (They don’t want to pay out too much up front to people who are likely to stay alive for many years, postponing the company’s payday.)
The problem: They’re bad at such predictions.
“For the past 10, 15, 20 years, they’ve been very poor at judging how long people are living. People have lived a lot longer than the [companies that] bought those policies thought they would,” said Steve Weisbart, an economist at the Insurance Information Institute who studies life insurance.
Sabes was convinced there must be a better way. He tasked his team with finding an “InsUber” — a technology that would do for life insurance what the ride-sharing app has done to transportation.
That quest led Sabes to Steve Horvath, the UCLA biostatistician behind a predictive method now known as “Horvath’s clock.” Horvath reported in 2013 that he had developed a statistical model to estimate the biological age of tissue from noting whether chemical tags known as methyl groups are attached at 353 spots in a person’s DNA. The model was seen in the scientific community as intriguing — but still preliminary.
Last September, Horvath coauthored a meta-analysis evaluating a handful of epigenetic clocks (some developed in his lab) to see how well they predicted longevity. They identified one algorithm as the best of the bunch — and GWG quickly swooped in to option it.
A key developer of that algorithm and the lead author of the meta-analysis was Brian Chen, a trained epidemiologist who has worked closely with Horvath. Chen recently signed on with GWG. The company isn’t using the saliva samples it’s collecting to set prices yet; it’s still refining how to deploy its prediction models.
“What we’re trying to do is like precision medicine, but ‘precision insurance’ — and so you get more customized, personalized rates,” Chen said.
Several independent scientists questioned whether the technology is ready for prime time.
“I would doubt whether it gives a prediction of life expectancy … that is by itself accurate to any useful extent,” said John Greally, a geneticist at Albert Einstein College of Medicine who has studied DNA methylation.
As for Horvath, he told STAT by email that he’s “not a businessman and cannot comment on the commercial utility” of his work.
Sabes and Chen acknowledged that the algorithm isn’t perfect. But Sabes said it doesn’t have to be.
“The level of accuracy, so to speak, has to be at such a high level before it has economic value in most applications,” Sabes said. But in the life insurance industry, he said, “if you can just be a little bit better, it can have a pretty big implication, because we work in the law of averages.”
GWG currently tries to predict life expectancy by looking at policy holders’ medical records, reviewing their prescription drugs, and conducting phone interviews. (It plans to continue to take these factors into account even after it integrates epigenetics.) To get life insurance in the first place, consumers often need to fill out questionnaires about their family history, or submit to a urine or blood test.
Laws at the federal level and in most states prevent health insurers and employers from requiring genetic information or discriminating on that basis. (A bill moving through Congress, however, would let employers demand workers’ genetic test results, with financial penalties for those who refused.)
The bar is much lower for life insurers, which if challenged must simply convince state regulators that there’s a plausible scientific basis for using a certain factor in underwriting. So, for instance, if a customer’s medical records show she tested positive for a gene variant linked to breast and ovarian cancer, the insurer could take that into account when setting premiums.
In the past month, GWG has collected saliva samples using a sponge from about 40 of the people from whom it’s bought policies, Sabes said. Only a few refused.