Spencer Platt / Getty

It can be hard to comprehend the true scope of something as disastrous as the opioid epidemic. Perhaps that’s why it’s been compared with falling 747s and crashing cars. But in fact, knowing exactly how many people have perished is crucial to stopping the deaths.

That’s why Elaine Hill and Andrew Boslett, economists at the University of Rochester, were so concerned when they found that many potential opioid deaths aren’t counted as such. In the fall of 2018, Hill and Boslett were studying how deaths from overdoses of opioids, such as heroin or OxyContin, were influenced by the decline of coal mining and the rise of shale gas fracking. But when they began looking at death records of Americans who had died of drug overdoses, they noticed that in more than 20 percent of the cases, the record said the type of drug could not be specified, perhaps because an autopsy had not been performed. In other words, the person had died of a drug overdose, but the death record didn’t say which drug.

Hill and Boslett realized that such a high rate of unknowns wouldn’t work for the phenomenon they were trying to study. “Our lab wants to make as strong of a claim as possible, given evidence that maybe an economic shock … had an effect on drug-overdose rates,” Boslett says. “We want to know that the estimates we’re using on local drug-overdose rates are correct, or as correct as possible.”

So the researchers set out to try to determine the real causes behind those unspecified drug overdoses. In the process, they uncovered something unsettling about how deaths are tracked in the U.S.: The way that a given county investigates deaths matters, and it could be dramatically shifting our nationwide estimates of the number of people who die of everything including opioids, childbirth, and the new coronavirus.

Hill worked with Boslett and a doctoral candidate, Alina Denham, to come up with a model to estimate how many of those unspecified drug overdoses were caused by opioids. To do it, they set aside some of the death records in which the type of drug was known and created a model that would predict the drug, given other things that were known about that person: the county they lived in, their sex, where they died, other health conditions that contributed to the person’s death, and so on. For opioid deaths, that meant factoring in whether the person had other characteristics typically associated with opioid overdose, such as being addicted to opioids or having chronic pain. By applying the model to the “unspecified” overdose deaths, they were able to predict that 72 percent of those were actually from opioids.

In fact, they estimate in a new study in the journal Addiction, there were more than 99,000 additional opioid deaths from 1999 to 2016 than had been previously documented, raising the national death toll by about 28 percent, to 453,300. What’s more, the discrepancies varied widely by state. In Alabama, Mississippi, Pennsylvania, Louisiana, and Indiana, Hill and her team estimated that the number of deaths from opioid overdoses was actually double the previous estimates.

Addiction

“This paper is a very strong one,” says Atheendar Venkataramani, a health-policy professor at the University of Pennsylvania, who was not involved in the study. It suggests that “if you just follow the vital statistics alone, we’re probably underestimating the true number of opioid deaths.”

Hill and her team suspect that’s because of differences in how counties across the U.S. investigate deaths. In essence, whether a given county uses a coroner or a medical examiner to investigate deaths matters. Medical examiners are doctors specially trained in pathology and forensics, but coroners can be general practitioners or even laypeople with no medical training. For coroners, “in many places, like the state of Pennsylvania, the only requirements are to be a legal adult with no felony convictions who has lived in the county for one year and to complete a basic training course,” Jordan Kisner wrote this week in The New York Times Magazine. Meanwhile, as Kisner pointed out, the United States has a dire shortage of medical examiners.

Because of this low standard of training, Denham explains, “you would think [coroners] would not be able to identify opioid involvement in a death as well as a medical doctor trained in it would.” That inference seems to be held up by data: The states that had a lot of unclassified drug-overdose deaths, Hill and her colleagues found, tended to use coroners in their death investigations.

The undercounting of opioid deaths is important because “you need to know the scale of a problem to know how to intervene in the problem,” Venkataramani says. Dealing with a crisis like opioid addiction—or coronavirus, for that matter—requires lawmakers and public-health workers to make choices about where to direct precious funding and resources. If the severity of the opioid epidemic is underestimated, local public-health departments could be shortchanged, and even more lives could be lost. This is particularly important in the case of infectious diseases like coronavirus, where knowing the total number of deaths can help public-health officials estimate its lethality.

Especially in the case of addiction, so much of illness happens outside the public eye that it’s sometimes only when someone dies that her neighbors or the government see exactly what she was going through. The tragedy of epidemics like opioid abuse is that nothing can be done to help the dead. But the dead can help others—if the things that killed them are accurately reported. Having a better grasp of just how many people are dying from various ailments is crucial for policy makers to help those who are still living.

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.