Last year, a life insurance agent came to Nathan DeWall’s Lexington, Kentucky, home to weigh him, take his blood pressure and ask a litany of health- and life-related questions to predict when the 34-year-old would die. A few weeks later, DeWall received an envelope in the mail, containing the result: He would live to be 88.
“What does that number really mean?” asks DeWall, a psychologist at the University of Kentucky who studies how people cope with the prospect of their own mortality. Would a few extra slices of pizza push him down to 87.7? Would a bit more time on the treadmill move the needle to 88.3?
Life insurance companies have long relied on an equation that mixes actuarial tables with specific facts about one’s health, age, lifestyle, hazardous activities or hobbies—and even criminal, financial or driving records—to calculate an approximate date of death. But mortality calculators are about to get a lot better, and a lot more complicated.
Behold big death data.
From lethal disease to murders, to deadly workplace accidents, suicides, fatal domestic violence incidents and natural disasters, researchers are now harnessing vast amounts of data to more specifically forecast mortality. These death algorithms, based on health and prescription drug records, social media, cellphone trails, crime statistics and beyond, drive data-based intuition for police departments, communities, hospitals and corporations to deploy at will—and for all of us at home to digest at our own emotional risk.
“Prediction of death is about as sensitive as it gets,” says Eric Siegel, author of Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die. Imagine the public’s dismay if the death and funeral industries started using big data to determine who would perish on a particular day. Picture eerie ads for personally engraved headstones popping up on your Facebook pages. Or Google sidebars asking if you ever thought of having your cremains pressed into a vinyl record.
The fifth century B.C. Greek tragedian Euripides wrote, “No one can confidently say that he will still be living tomorrow,” and for most of human existence, we’ve lived life with that edict in mind. But what if we were wrong? What if big data could help us determine our day of death down to the minute, as Tikker, a Kickstarter-backed “happiness watch” that counts down to your death, claims? How would our behavior change? “When we think of our own mortality, the mind is very resilient,” says DeWall, who published a study a few years back in Psychological Science that showed people who were asked to think about their deaths, and then given a series of incomplete thoughts or phrases, were more likely to fill those in with positive words—joy, baby, kiss, beach, puppy—than negative ones. “We immediately seek out things that will give us a sense of meaning and belonging in our lives.”
But thinking we know our check-out time can also be detrimental. “There is the impression that if we can look at data and discern hidden patterns, it’s going to give us some privileged access” into mortality, says Sheldon Solomon, a social psychologist at Skidmore College. That impression, he says, is flawed. The more we think we know about how and when we will die, the more we will grasp for psychological strands of control. For some, this means a last-ditch effort to live more healthfully; for others, it could mean seeking out distractions in an attempt to banish death from their thoughts. All of it is just another attempt to deny the inevitability of death instead of accepting it, he says, not so different from immortality researchers or cryogenics. Solomon is one of the psychologists behind Terror Management Theory, which has shown that when people are subconsciously exposed to ideas of death they show less forgiveness, less tolerance, and a greater tendency to drive aggressively, go on shopping sprees or cling to religion.
With big data, “rather than appealing to a higher power, we’re appealing to tables of probabilities,” Solomon says. Either way, “it conveys a false sense of security.”
There are, after all, multitudes of ways to die. The writer David Southwell has documented 1,001 bizarre deaths throughout history, like a man who electrocuted himself on the toilet while fixing a television. Euripides is said by some to have been killed by a pack of hunting dogs when they mistook him for a boar while he was walking in the woods of Macedonia in search of inspiration. It is hard to imagine an algorithmic sense could ever grasp all pathways leading to the grave.
Instead, says Siegel, the idea is to figure out how to forestall the preventable causes of death—or for some companies, to make big bucks off of our demise.
Will You Die at Work?
The Chinese construction worker introduced himself as Jim. He told Chuck Pettinger, a U.S. safety management expert at Predictive Solutions who was visiting the factory outside of Beijing in 2006, that his real name would be too difficult for him to pronounce. During Pettinger’s two weeks in China, where he was rolling out a new program—predictive analytics for job safety—Jim took him out to eat, drove him around, translated and taught him the inner workings of the construction company. Pettinger went home with Jim and met his family. He learned that Jim’s salary was supporting not only his immediate family but also his parents, his three siblings and his siblings’ kids.
A few days after Pettinger returned to his office in San Francisco, the CEO came over to him and put his arm around him. “I hate to tell you this,” he said, but Jim had been killed when scaffolding malfunctioned and tipped over.
The news shook Pettinger. If only he had implemented the safety analytics program a few months earlier, he thought, perhaps the accident could have been avoided. “His family might still have him there,” Pettinger says.
Predictive Solutions has examined over 130 million safety inspections prepared for the Occupational Safety and Health Administration (OSHA), alongside data compiled by the companies it works with, in order to predict the likelihood that someone might die in a work-related accident, such as a machine entanglement, slip or trip, electrocution or construction site fall. The goal is to decrease the approximately 4,500 work-related deaths that, according to OSHA, occur in the U.S. each year. Data analyzed by Carnegie Mellon University has found that Predictive Solutions can predict injury and death with 70 to 90 percent accuracy, Pettinger says.
Its database digests inspections of workplace histories, and job site employees also input safety information into their iPads, noting, for example, whether an employee’s eyes are on the task, or if he is wearing the right boots or glasses, or lifting too much weight, or if machinery is up to snuff. The metrics are analyzed, including whether every safety standard is too perfectly met—that could imply somebody is just carelessly checking off boxes.
“It’s not necessarily about going to the guy turning a wrench and saying, ‘How does that make you feel?’ It’s more about going to the guy turning a wrench and saying, ‘He’s looking really rushed, is he under pressure?’” says Pettinger, who has a degree in industrial psychology. Rushing is a risk factor, and specific life or death predictions can be made from those very assessments when cross-referenced with other analytics.
“That’s where big data comes in,” Pettinger says. “Our vision for our company is to eliminate death on the job by the end of the century.”
Will Someone Kill You?
In 2012, Philadelphia law enforcement handled 107,093 reports and 8,535 calls, and made 5,568 arrests, all for domestic abuse, according to Women Against Abuse. Over a six-year period, from 2007 to 2013, an average of 28 people a year died in Philadelphia from relationship violence.
The numbers are bleak, but according to Richard Berk, a professor of criminology and statistics at the Wharton School of the University of Pennsylvania, they are also ripe for predictive analysis. Berk believes he can pinpoint an abuser or murderer before he or she strikes. He has been working with Philadelphia police to analyze 100,000 domestic violence records for patterns that might reveal which abusers will reoffend or even kill, and stop them.
Berk has also collaborated with Maryland’s department of family services, analyzing hundreds of thousands of profiles of kids and their households to forecast which children are at risk for abuse or violent death, and comparing the profiles with coroners’ reports of kids who have died. The records include “internal documents of family agencies that report suspected or actual child abuse,” Berk says. “There’s a trained algorithm that looks at household risk to see whether there needs to be special attention.”
Berk’s involvement in predicting domestic violence and child abuse stems from his work on forecasting murder among probationers and parolees, published in the Journal of the Royal Statistical Society in 2009. As he explains, “It boils down to developing profiles. These are essentially fancy search engines looking through hundreds of thousands of cases and finding clusters of people alike in backgrounds.”
Profiles are cross-analyzed with historical data about crime trends and statistics. In Philadelphia, for example, Berk analyzed 60,000 crimes and then, using an algorithm, forecast some offenders who, it turned out, were charged with homicide or attempted murder within two years of being released from prison.
If these tools are put into place across the country, Berk predicts, “we will be much more effective at preventing violence.”
Not everyone thinks this is a solution. Kade Crockford, director of the Technology for Liberty Project of the ACLU of Massachusetts, says to understand the worst-case scenario for predictive policing, people just need to watch Minority Report, the 2002 Steven Spielberg film (based on a sci-fi short story by Philip K. Dick) that depicted an intrusive world in which police could foresee crimes. The film raises the concern that predicting crimes might, in some cases, drive the act of crime itself, creating a self-fulfilling policy. In the real world, Crockford says, officers should refocus on gumshoe detective work, “instead of acting like God and preempting human behavior using fancy computer technology that definitely impacts people’s privacy in a negative way.”
Nevertheless, predictive tools are being rolled out in cities nationwide. The Los Angeles Police Department is also using big data technology to patrol some of the most violent and gang-ridden neighborhoods, predicting where burglaries and robberies will happen. Officers cruise around using a program monitoring data from a company, PredPol, created by a UCLA professor, Jeff Brantingham, and a Santa Clara University scientist, George Mohler. Red boxes appear on screen maps, indicating hot spots—calculations based on previous crimes down to the half-block—to predict where the next ones will occur, and officers are deployed to survey those areas. In parts of Los Angeles, big data has led to a 25 percent drop in some crime.
Atlanta and Chicago have been using the same technology to predict homicides and gun violence, Brantingham says. Though the data is not efficient in predicting lone wolf shootings, like the recent rampage near the University of California, Santa Barbara that claimed seven lives, it can predict crime and gun-related homicides by studying similar crimes. At least 45 police departments are now using PredPol in some capacity across the U.S., U.K. and Uruguay.
Will You Kill Yourself?
A decade ago, one of Chris Poulin’s best friends posted a suicide note online, then killed himself. Poulin, who spent much of his career building e-commerce search engines wondered, Could big data technology analytics have helped save him?
In 1897, a French sociologist, Emile Durkheim, did his version of data mining by studying the diaries and notes of suicide victims to determine patterns of behavior and social factors. Poulin knew that in the digital age there were cyber-trails that could quantitatively measure a person’s isolation, depression and suicidal tendencies in real time.
He teamed up with Veterans Affairs and the Department of Defense to obtain anonymous medical records of patients, doctors’ notes, drug prescriptions, religious and social behaviors, psychiatric care and treatment to discern suicidal patterns among them. The Durkheim Project launched in 2012. Poulin and his co-researchers then turned to social media, teaming up with Facebook and Twitter to apply that predictive model to the online activity of military veterans who agreed to be part of the experiment. The project is now gathering thousands of data points in order to better understand when negative social media expression is an indicator of serious suicide risk.
The next step is to secure funding to put a team of psychologists and emergency service personnel in place to step in if someone is deemed suicidal. Twenty-two veterans kill themselves a day, and if the system helps save their lives, Poulin says, it could also hold promise for bullied kids, depressed teens, the LGBT community and other populations at risk for taking their own lives.
What Disease Will Nail You?
From a profit-making perspective, relying on sophisticated death analytics makes sense, particularly for life insurers, who are in the business of putting prices on our heads. Gen Re, a global life, property and casualty reinsurance company owned by Warren Buffett’s Berkshire Hathaway Inc., draws from a complex data pool that goes beyond basic biometrics, to include medical studies, lab test results and disability claims in Germany; banking behavior data in the U.K.; salary rates, wellness surveys and socioeconomic neighborhood statistics in South Africa; and laboratory mortality models in the U.S.—all to better understand patterns of death.
Beyond money-saving measures and marketing interests, some public-spirited big data efforts are trying to identify people who are going to die, and save them. IBM’s Smarter Planet initiative is using big data to tackle chronic diseases by building the world’s largest cancer registry, and the company’s Watson supercomputer—which beat two Jeopardy! champions—is being adapted for use in hospitals to synthesize patient histories, electronic medical records and scientific research to come up with more targeted medical diagnoses and treatments.
Meanwhile, Eric Horvitz, a researcher at Microsoft, believes he’s come up with a way to predict the next cholera outbreak by studying 70 years of headlines from The New York Times and other data sources. And Google thinks it can help us cheat death with its launch of Calico, a firm devoted to dramatically extending our life spans by using data processing to study the genetic causes of aging and disease, and figuring out how all of it can be defeated with drugs or other treatments.
With the profusion of self-monitoring devices for people to keep track of all of the physiological inner workings of their bodies—embedded in watches, vests, walking canes and even beds—the day is coming when the analysis of such data will be better suited to help diagnose and treat patients in real time, says Charlie Schick, director of business development and health care at the technology company Atigeo.
But Schick also says that the health care industry still has a long way to go to catch up to the business world when it comes to using predictive analytics. Hospitals need to better understand how to collect, store and process data, he says, and especially to understand it in a meaningful way to make strides “in reducing pain, reducing suffering, reducing death.” Those shifts will come from analyzing health records, but also from documenting financial and social data, like identifying patients who do not have families at home to care for them, which predictive analysis has shown plays a role in hospital readmittance rates.
There has been wide discussion and progress in using big data to study and prevent “the three big killers: obesity, congestive heart failure and diabetes,” Schick says. “The next steps will come in preventing unique ailments, or in working with the mental health system (like predicting which patients will be violent), as well as in understanding how comorbidities are linked.
Death by Dog Walk
Receiving his own expiration date in the mail did little to change Nathan DeWall’s behavior. It was the unexpected death of his mother that became his catalyst for a lifelong priority shift. At 60, his mother had been healthy; he always figured her death “number” would have been higher. But one day while walking the dog, she tripped in the driveway, hit her head and died a few days later.
“For me, this random fluctuation of variability in my mom’s life completely changed me,” DeWall says. “It motivated me to take care better of myself. That’s when I started running.”
DeWall now runs ultra-marathons. He doubts an algorithm could have predicted that series of events. Proof, he says, that insights and forecasts, no matter how confident, still must surrender to chance.