Us­ing big data to pre­vent drug er­rors

Modern Healthcare - - INNOVATIONS - By Sabriya Rice

Dr. Gidi Stein re­calls how dis­turbed he was when he heard about a mal­prac­tice case in­volv­ing the death of a pe­di­atric pa­tient that was the re­sult of a med­i­ca­tion er­ror. The child’s doc­tor had ac­ci­den­tally se­lected the wrong med­i­ca­tion from hun­dreds of drugs listed on the drop-down menu of a com­put­er­ized or­der-en­try form.

Stein, a pro­fes­sor of medicine and molec­u­lar imag­ing at Tel Aviv Univer­sity in Is­rael, was dis­mayed that the tragic in­ci­dent was de­scribed as an “un­avoid­able er­ror.” With his back­ground in soft­ware en­gi­neer­ing and com­pu­ta­tional bi­ol­ogy, he re­jected that think­ing as wrong-headed and dan­ger­ous. “Even if it’s a one-in-a-mil­lion chance, hun­dreds could die like that,” he said. “It’s com­pletely un­ac­cept­able.”

In 2012, Stein co-founded an Is­rael-based com­pany called MedAware. The firm of­fers a big-data soft­ware plat­form that in­te­grates with a hos­pi­tal’s elec­tronic health-record sys­tem to de­tect pre­scrip­tion er­rors be­fore they hap­pen. It draws from pat­terns in mil­lions of pa­tient records to flag med­i­ca­tion-or­der out­liers.

If a physi­cian chooses a drug that doesn’t match any con­di­tion in the pa­tient’s record or di­verges from how other pa­tients with sim­i­lar his­to­ries have been treated, the dis­crep­ancy is flagged. The sys­tem blocks the drug or­der un­til the doc­tor con­firms its ac­cu­racy or can­cels and re-en­ters the or­der.

MedAware is catch­ing the at­ten­tion of safety lead­ers work­ing to re­duce med­i­ca­tion er­rors, who say they aren’t aware of other prod­ucts tak­ing this ap­proach. “It’s a nice added fea­ture to what’s out there,” said Matthew Grissinger, di­rec­tor of er­ror re­port­ing pro­grams for the In­sti­tute for Safe Med­i­ca­tion Prac­tices, a Philadelphia-based not-for-profit fo­cused on drug er­rors.

Med­i­ca­tion er­rors are gain­ing more at­ten­tion with the broader dis­sem­i­na­tion of EHRs and com­put­er­ized pre­scrip­tion or­ders. A re­cent study in the jour­nal BMJ Qual­ity and Safety found that of more than 1 mil­lion med­i­ca­tion er­rors re­ported to the U.S. Phar­ma­copeia MEDMARX re­port­ing sys­tem be­tween 2003 and 2010, more than 63,000 were re­lated to com­put­er­ized provider-or­der en­try.

Many er­rors arose from user is­sues, in­clud­ing typ­ing and pull-down menu er­rors, as well as ig­nor­ing or over­rid­ing alerts.

Typ­i­cally, EHR alert sys­tems are pro­grammed to spot dan­ger­ous drug in­ter­ac­tions, higher-than-nor­mal doses and du­pli­cate pre­scrip­tions. But these sys­tems have not used large vol­umes of ag­gre­gated data to de­ter­mine in real time the like­li­hood that the wrong drug was se­lected for a par­tic­u­lar pa­tient.

“This has a huge ben­e­fit,” Grissinger said. He noted that phar­ma­cists as well as doc­tors can make mis­takes. As the first few letters of a drug name are typed into an or­der­ing sys­tem, a list of drugs that be­gin with the same letters ap­pears. Se­lect­ing the wrong one hap­pens fre­quently.

MedAware uses the same out­lier ap­proach that credit card com­pa­nies use to de­tect fraud. Con­sumers typ­i­cally ex­hibit pre­dictable pat­terns such as shop­ping at par­tic­u­lar stores and spend­ing within a cer­tain range. If a larger-than-usual charge sud­denly ap­pears in a dif­fer­ent state or coun­try, “it would be a com­plete out­lier to your spe­cific be­hav­ior,” Stein ex­plained. “Credit card com­pa­nies will call you up.”

Draw­ing from mil­lions of pa­tient records, MedAware uses a math­e­mat­i­cal model to pre­dict the like­li­hood of par­tic­u­lar types of pa­tients be­ing pre­scribed spe­cific drugs. If the soft­ware de­tects an out­lier when a clin­i­cian en­ters a pre­scrip­tion or­der—such as pre­scrib­ing a drug for an in­fant that gen­er­ally is used for se­nior cit­i­zens, for ex­am­ple—an alert mes­sage im­me­di­ately ap­pears on the or­der screen.

Although the soft­ware is not cur­rently used in hos­pi­tals, pre­lim­i­nary find­ings on MedAware’s ef­fec­tive­ness were pre­sented at the Healthcare In­for­ma­tion and Man­age­ment Sys­tems So­ci­ety’s an­nual meet­ing in April. Ret­ro­spec­tive analy­ses of more than 44 mil­lion filled pre­scrip­tions from two large hos­pi­tals and one HMO in Is­rael gen­er­ated alerts for more than 7,000 pa­tients, with a low rate of false alarms.

Brigham and Women’s Hos­pi­tal in Bos­ton is test­ing MedAware’s ef­fec­tive­ness in flag­ging er­rors through a ret­ro­spec­tive look at filled pre­scrip­tions in more than 748,000 pa­tient records, rep­re­sent­ing more than 9 mil­lion pre­scrip­tions. Find­ings are ex­pected by the end of the sum­mer.

Dr. David Bates, the hos­pi­tal’s chief in­no­va­tion of­fi­cer, said he’s ex­cited about MedAware’s po­ten­tial. “It’s the kind of ap­proach that can get you to a higher level of safety,” he said.

Grissinger said MedAware holds great prom­ise when used in con­junc­tion with es­tab­lished drug-in­ter­ac­tion, dos­ing and du­pli­ca­tion alerts. But providers still need to have other safety pre­cau­tions in place, such as ev­i­dence­based clin­i­cal pro­to­cols and staff train­ing in fully and ac­cu­rately en­ter­ing pa­tient in­for­ma­tion into EHRs, he added.

“You can have all the data min­ing in the world,” Bates said. “But if the needed fields in the EHR are empty, you’re go­ing to get noth­ing out of it.”

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.