Here’s a formula for actuaries:
Take the amount of money a person spends on TV and accoutrements like cable fees and recording devices. Divide that number by disposable income. What do you get?
A marker for diabetes, according to Chris Stehno, a director for Deloitte.
Actuaries have become adroit at finding surprising correlations in big data. The connection between credit scores and driving experience may be the most famous, and the use of telematics devices in cars to divine driving patterns may be one of the fastest growing.
But property-casualty actuaries’ search for correlations could increasingly extend into the world of health, as a pair of speakers explored at the CAS Ratemaking and Product Management Seminar in Orlando. Their presentations showed how actuaries and other quantitative researchers are finding innovative ways to use emerging data.
The presenters, Stehno and Lin Xing, FCAS, who works at Verisk Analytics, took different approaches to the subject. Stehno showed how commonplace data — buying TV equipment — can signal a high potential for disease. Xing showed how health data might potentially be useful in providing discounts for homeowners’ or auto insurance.
Stehno tapped a database that gave him facts about 230 million adult Americans —their ages, their shopping patterns, where they live — information culled from public and private sources.
“I know their buying patterns,” he said. “That data isn’t just good for marketing; it actually has strong ties to health.”
The health records were accumulated with the consent of consumers, often given as part of an exam for a life insurance policy.
Electronic health records became widely available after the American Recovery and Reinvestment Act helped pay for doctors to make records electronic. Consumers control their own records, Stehno said, and they often make them available to third parties, which is how his database was formed.
The records are most revealing for people older than 50, he said. Younger people do not see the doctor as often.
When he pairs the medical information with shopping and other information, he learns interesting things:
- People with shorter commutes are generally healthier.
- People who have been through bankruptcy are generally less healthy.
- People who watch a lot of television are generally less healthy.
For an extended example, Stehno used himself. He applied an algorithm he developed using medical claims data to try to find people who are at greatest risk for depression.
Based on his own circumstances, the model said he has about a nine percent chance of suffering from depression, just below the claims dataset average of 9.5 percent. Were he to divorce, that likelihood would rise to 13 percent. If in addition he became a renter, the likelihood would rise to 15 percent. If his income fell substantially, 17 percent.
In his example, he continued to pile woe unto himself. He became a heavy user of mail order; he gave away his pet; his credit card spending surged; he showed no interest in retirement products, and so on. Eventually the model showed he had a 36 percent chance of having a medical claim for depression.
Stehno has created 150 disease and medical condition algorithms like the depression model.
Originally the models were used in financial areas, helping health and life insurers hone their rates. More recently, life insurers used the information to simplify the underwriting process, and health insurers used the information to encourage healthy behaviors.
In the past, health incentives were broad — so broad they were ineffective. If you offer a discount to a health club, Stehno said, you mainly subsidize people who would have bought a membership anyway.
Now an algorithm pinpoints a person likely to get skin cancer. An insurer could send the person a visor and coupons for sunscreen.
Or an algorithm can prompt a doctor’s treatment. Families with pets run greater risk of having medical complications associated with asthma. Knowing about the pet, the family doctor can order a spirometry test to assess whether their young children’s lungs are healthy.
P&C insurers are kicking the tires on the new data. At least two major insurers have applied for patents to use biometric information to help create a driving score, Xing stated.
In her research, Xing concluded that casualty actuaries could potentially use such data, presumably voluntarily provided by policyholders via smartphones, Fitbits and the like, to help predict auto results.
Xing described several consumer products that can monitor wellness. An earbud tracks the movements of its wearer, their body temperature, heart rate and blood pressure. Smart contact lenses determine glucose levels. A sweat sensor can tell if the wearer needs a drink of water. Some life and dental insurers are already providing these types of devices to policyholders and offering discounts for greater data sharing.
These products are relatively new, Xing said. To see whether the information they gather could work in P&C insurance, Xing studied data from the Centers for Disease Control and Prevention’s (CDC) Behavioral Risk Factor Surveillance System. The CDC asks more than 400,000 people per year about their health and related behaviors, and they publish data at the county level.
Xing linked 17 health characteristics from the CDC database into auto insurance data from Verisk’s ISO subsidiary. She looked at the effects of factors including mental health, weight, smoking and drinking habits of a county’s residents. She also looked at the presence of chronic health conditions, like asthma or diabetes. She paired the health data for a county with the loss experience of its residents to evaluate how behavioral incentives by insurers could potentially help improve results.
One surprising result: Smokers appeared to have lower auto liability losses.
Though her results are tentative, Xing has found research from the 1990s that showed smoking makes drivers more alert, hence better drivers.
An ethical question immediately rises: “Should we reduce rates for smokers?” Xing asked. “That is an eye-opening question.”
Xing said her wellness model does a good job of predicting losses for the four major auto coverages: bodily injury, property damage, collision and comprehensive.
Issues do remain. For example, as with other new data sources such as vehicle telematics data, regulators would expect insurers to demonstrate proper protections are in place for individual privacy. There may also be cybersecurity concerns about the data.
The next step may be to develop a model that helps provide individuals with the right incentives to improve their wellness. Later, more models could be built around other lines of business such as workers’ compensation. Actuaries, Xing said, can help insurers “be on the forefront of innovation, taking advantage of the powerful tools of the wellness industry to create a happier, healthier and more profitable property-casualty book.”
James P. Lynch, FCAS, is chief actuary and director of research and information services for the Insurance Information Institute in New York.