Federal data severely underestimates Pa. and N.J. opioid deaths. So what?

Deaths from heroin and other opioids are far higher than federal statistics indicate in New Jersey and even more in Pennsylvania, according to a new analysis that examined information that was missing from records and estimated what it likely represented. Nearly 50 percent of Pennsylvania's drug overdose deaths in federal data carried no indication of the types of drugs involved, as did 30 percent of those in New Jersey.

The new study examined 2008 and 2014 drug mortality data from the Centers for Disease Control and Prevention, which receives data from counties via the states. When information on drug type was missing, it applied a formula based on those that were complete, weighted by various demographic information. Reporting by most states improved over that time period but still varied widely, with information missing from 0.4 percent of cases in Rhode Island vs. 52.4 percent in Alabama in 2014.

Nationally, nearly 20 percent of reports were incomplete. That translated to 24 percent more deaths than reported from all opioids and 22 percent more from heroin. Opioids include heroin, most prescription painkillers, and fentanyl, which can come in both pharmaceutical and illicit forms.

"The most striking example was Pennsylvania," study author Christopher Ruhm, a health economist at the University of Virginia's Frank Batten School of Leadership and Public Policy in Charlottesville, wrote in the online version published Monday in the American Journal of Preventive Medicine. The state's adjusted ranking for opioid deaths went from 17th to sixth; heroin went from 20th to fifth. The new rankings were more in line with the overall drug overdose fatality rate, which in 2014 had put the state in seventh place.

New Jersey was better than average in overall drug deaths that year. But the study's corrected data showed its opioid-death-rate ranking moved from 14th to 10th and for heroin, from 15th to 12th.

But the practical meaning of the finding, while dramatic, was unclear. County coroners and medical examiners for several years have been reporting disturbingly high numbers of deaths due to opioids such as heroin, fentanyl, and oxycodone — information that apparently has not made it into federal databases.

The Drug Enforcement Administration's Philadelphia Division, for example, which collected statistics directly from coroners throughout Pennsylvania, came up with very different numbers than the CDC did in 2014. The local division found 1,293 heroin-related deaths statewide while the federal CDC database showed 503 — an even greater gap than what the University of Virginia study calculated.

The new study did not include county data but a query of the same CDC database it used illustrates the issue. It shows 11 heroin deaths in Philadelphia in 2014. The city Medical Examiner's Office reported 349, according to the DEA's local division.

Still, it was not at all clear why the local and CDC data would be so different. Both originate at the county level.

Among the possibilities: "We typically don't put toxicology reports on death certificates. There's not enough lines to list all those drugs," said Lycoming County Coroner Charles Kiessling Jr., president of the Pennsylvania State Coroners Association.

Death certificates are the best source of data for cause of death but are notoriously imprecise. How they are filled out and by whom varies widely from state to state, as does the design. Pennsylvania's contains just four lines for contributing causes of death when often there would be far more than that, Kiessling said.

A drug overdose is fairly easy to determine but the toxicology tests needed to find heroin, fentanyl, or hydrocodone can take months.

Carol Gifford, a spokeswoman for the Pennsylvania Department of Drug and Alcohol Programs, said she could only guess at the reasons for the missing information, but pointed out another possibility: Testing is expensive.

Ruhm, the health economist, said he was concerned that federal policy — and possibly funding — might be based on incomplete information.