Study: Sui­cides hide in over­doses

Johns Hop­kins re­searcher con­cludes up to third of opi­oid deaths not ac­ci­dent

Baltimore Sun Sunday - - FRONT PAGE - By Mered­ith Cohn

Tens of thou­sands of peo­ple fa­tally over­dose each year on opi­oids and other drugs. Some­times med­i­cal ex­am­in­ers la­bel them ac­ci­dents, and some­times they don’t know what to call them.

But where hu­mans wa­ver, a com­puter pro­gram us­ing a kind of ar­ti­fi­cial in­tel­li­gence finds that many are likely sui­cides — pos­si­bly a third of them, ac­cord­ing to a study by a Johns Hop­kins School of Medicine re­searcher who part­nered with a Utah high school stu­dent.

The in­for­ma­tion could bring sharper fo­cus to the scope of epi­demics of both opi­oid abuse and sui­cides, as well as the need for re­sources.


“If we’re try­ing to pre­vent deaths in the com­mu­nity, we have to fig­ure out why peo­ple are dy­ing,” said Dr. Paul Nes­tadt, the Johns Hop­kins as­sis­tant pro­fes­sor of psy­chi­a­try and be­hav­ioral sciences who was one of the study’s au­thors.

“If peo­ple are dy­ing by ac­ci­den­tal over­dose, the best in­ter­ven­tions — nalox­one avail­abil­ity and treat­ment in the com­mu­nity for ad­dic­tion — are dif­fer­ent from those for sui­cide, like the avail­abil­ity of hot­lines and an­tide­pres­sants,” he said.

There have been other at­tempts to iden­tify the sui­cides hid­den in the

over­dose num­bers. This study, re­cently pub­lished in the jour­nal Sui­cide and Life-Threat­en­ing Be­hav­ior, backs up some of those find­ings. The Hop­kins research, how­ever, may be the first at­tempt to use so-called ma­chine learn­ing to ver­ify the es­ti­mates.

Un­der the model, the re­searchers en­tered in­for­ma­tion about over­dose deaths in Utah from 2012 to 2015 into an al­go­rithm. The data in­cluded age, sex, race, his­tory of men­tal ill­ness, and stres­sors like job loss, Nes­tadt said. The al­go­rithm, taught to rec­og­nize the im­por­tance of risk fac­tors, com­bined the data to de­ter­mine the prob­a­bil­ity that a case was a sui­cide.

Few over­dose cases na­tion­ally are la­beled sui­cides be­cause, ab­sent a sui­cide note, ex­am­in­ers are of­ten un­sure of the drug user’s in­tent. That’s led to over­dose cases largely be­ing called ac­ci­dents or “un­de­ter­mined.”

Over­doses quadru­pled over two decades to about 70,000 in 2017, mostly due to opi­oids that in­clude pre­scrip­tion painkiller­s and il­licit heroin and fen­tanyl. Sui­cides have risen by about a quar­ter to 47,000 in about the same time frame.

“There are two epi­demics, sui­cide and opi­oids,” said Dr. Ian Rockett, who has been re­search­ing the un­der­count­ing of sui­cides for a decade. “They tend to be treated sep­a­rately, when there is con­sid­er­able over­lap.”

Rockett, who was not in­volved in the com­puter study, said it may have found a way to as­cer­tain more ac­cu­rate sui­cide num­bers among the over­doses, mak­ing it im­por­tant.

The study, which could end up in­flu­enc­ing pol­icy na­tion­wide, grew out of a science fair project by a fresh­man at West High School in Salt Lake City. Daphne Liu was work­ing on a project in­volv­ing ma­chine learn­ing, and some­one in Utah’s health depart­ment sug­gested she tap statewide over­dose data.

The idea res­onated with her; her brother had lost a friend to a drug over­dose.

Liu un­cov­ered the likely sui­cides. In 2018, her project won first place in the Na­tional In­sti­tute on Drug Abuse’s Ad­dic­tion Science Award. Nes­tadt read about the award and con­tacted Liu’s science teacher to sug­gest that they work to­gether to re­fine and pub­lish the find­ings.

Liu, now a high school ju­nior, said the re­searchers hoped to con­tinue mod­i­fy­ing the al­go­rithm to eval­u­ate fa­tal over­doses from other states, which could make the data more use­ful. In Utah, about 20% of over­dose cases had an un­de­ter­mined man­ner of death. Next up may be Mary­land, where 80% of cases were la­beled un­de­ter­mined, by far the high­est num­ber of any state.

“The big­gest thing I hope comes of this is aware­ness,” Liu said. “We’re al­ready told the sui­cide rates are su­per high. But that’s not the whole pic­ture. … It’s up to pol­i­cy­mak­ers to de­cide what to do with it.”

Rockett said more de­tailed au­top­sies may be needed to find the sui­cides among over­dose cases. The gold stan­dard would be “psy­cho­log­i­cal au­top­sies” in which fam­ily, friends, co-work­ers and oth­ers are in­ter­viewed, a prac­tice that orig­i­nated in the 1950s in Cal­i­for­nia but was largely aban­doned be­cause of the ex­pense, said Rockett, pro­fes­sor emer­i­tus in West Vir­ginia Univer­sity’s Depart­ment of Epi­demi­ol­ogy.

“Sui­cide gets ne­glected by in­ves­ti­ga­tors for prac­ti­cal rea­sons,” Rockett said. “The trou­ble is, though, if you’re not mea­sur­ing sui­cides ac­cu­rately then you’re not pre­vent­ing them.”

He said the com­puter study had lim­i­ta­tions, many noted by the re­searchers them­selves. The model is only as good as the data, and many states don’t col­lect good in­for­ma­tion on the lives of each over­dose case, Rockett said.

He also said the com­puter didn’t re­view cases that were la­beled ac­ci­dents, as­sum­ing that they were la­beled prop­erly, and ex­am­in­ers likely missed sui­cides in this cat­e­gory.

Fur­ther, some cases may be more nu­anced, Rockett said, with drug users not in­tend­ing to die that day but still in­tend­ing to harm them­selves.

Still, he said, the com­puter study pro­vided use­ful in­for­ma­tion and could help draw at­ten­tion to ris­ing over­dose sui­cide cases that may be un­der pub­lic radar.

An­other re­searcher, Michael Schoen­baum, said the com­puter may be over­reach­ing, and that Utah specif­i­cally may not be a good proxy for other states.

But Schoen­baum, a se­nior ad­viser for men­tal health ser­vices, epi­demi­ol­ogy and eco­nom­ics at the Na­tional In­sti­tutes of Men­tal Health, said strict death cer­tifi­cate la­bels may not mat­ter so much when it comes to treat­ment. Sui­cides and ac­ci­den­tal over­doses are both “deaths of de­spair,” he said.

“If some­one presents to­day with an ac­ci­den­tal in­jury, to­mor­row they may be a sui­cide risk, and we need to be more proac­tive in find­ing peo­ple with any com­bi­na­tion of risk,” he said. “The out­come you want for pa­tients is not be­ing dead.”

That is an area of agree­ment on a con­tro­ver­sial sub­ject, said Dr. Maria Oquendo, pres­i­dent of the Amer­i­can Foun­da­tion for Sui­cide Preven­tion and a past pres­i­dent of the Amer­i­can Psy­chi­atric As­so­ci­a­tion.

Her own re­view of research in 2018 con­cluded that at least 30% of all over­dose deaths were likely sui­cides, though the num­ber could be closer to 40%. Her find­ings took into ac­count emer­gency depart­ment data that in­cluded in­ter­views with over­dose sur­vivors about their in­tent.

Oquendo, chair of the Univer­sity of Penn­syl­va­nia’s psy­chi­a­try depart­ment, said ef­forts are un­der­way to im­prove the data, in­clud­ing de­vel­op­ment of na­tional re­port­ing stan­dards for med­i­cal ex­am­in­ers, coro­ners and oth­ers who cat­e­go­rize over­dose deaths. Com­puter mod­el­ing could help.

There is value in get­ting the data right be­cause there are proven life-sav­ing in­ter­ven­tions, she said.

“It’s def­i­nitely worth know­ing,” she said. “It’s def­i­nitely worth try­ing to save a life.”

Find­ing peo­ple be­fore they over­dose is key, re­searcher and clin­i­cians say. They sug­gested, among other ef­forts, univer­sal screen­ing in emer­gency de­part­ments and else­where for those who have con­sid­ered sui­cide; then seek­ing treat­ment and tak­ing steps such as elim­i­nat­ing pills, al­co­hol and firearms from the homes of those deemed at risk.

Dr. Ken­neth Stoller, di­rec­tor of the Johns Hop­kins Broad­way Cen­ter for Ad­dic­tion, agrees that treat­ment works for peo­ple with ad­dic­tions and men­tal health dis­or­ders, and there is over­lap.

He said those with sub­stance use dis­or­ders of­ten live on a con­tin­uum be­tween want­ing to live to want­ing to die.

Their feel­ings can wa­ver de­pend­ing on whether they are in­tox­i­cated or in with­drawal and in the throes of de­pres­sion, for ex­am­ple.

Men­tal health dis­or­ders and ad­dic­tion are both diseases of the brain that are “very tied to­gether in a very dan­ger­ous way,” he said.

He said treat­ment gen­er­ally in­volves med­i­ca­tion and coun­sel­ing, as well as com­pas­sion and em­pa­thy.

As for la­bel­ing over­dose deaths as sui­cides or ac­ci­dents, Stoller said he wasn’t sure those buck­ets were enough.

“When I get a chance to ask pa­tients who sur­vive their drug use, they tell me it’s about es­cap­ing,” he said. “Whether some­one wanted to es­cape per­ma­nently or ab­so­lutely wanted it to be tem­po­rary, or some­where in be­tween, it may be tough to tell.”



Daphne Liu, now a ju­nior at West High School in Salt Lake City, worked with Dr. Paul Nes­tadt, a Johns Hop­kins School of Medicine re­searcher, to re­fine and pub­lish her ma­chine learn­ing study that found up to a third of over­doses were likely sui­cides.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.