THE STATE OF PRE­DIC­TIVE AN­A­LYT­ICS

IN U. S. HEALTH­CARE

Modern Healthcare - - NEWS -

With the ex­pan­sion of risk in health­care, the abil­ity to pre­dict needs and out­comes is more im­por­tant than ever. Min­ing data, fore­cast­ing prob­a­bil­i­ties and trends, and ul­ti­mately man­ag­ing risk, is a bur­geon­ing area for health­care through pre­dic­tive an­a­lyt­ics.

Six ex­ec­u­tives from across the coun­try gath­ered in Philadel­phia on Aug. 30, 2016, to share how their or­ga­ni­za­tions are ap­proach­ing pre­dic­tive an­a­lyt­ics through pro­cesses, em­ployee train­ing, depart­ment struc­tures and more. Here are the valu­able in­sights— and suc­cesses— they shared.

HOW ARE YOUR OR­GA­NI­ZA­TIONS US­ING PRE­DIC­TIVE AN­A­LYT­ICS?

Haines: In our emer­gency depart­ment, we started iden­ti­fy­ing for clin­i­cians the pa­tients who are a po­ten­tial read­mis­sion. When a pa­tient vis­its the ED, it shows up in a real-time dash­board whether the pa­tient was ad­mit­ted within the last 30 days. This helps the clin­i­cian think about how they could pos­si­bly treat this pa­tient dif­fer­ently to avoid a read­mis­sion.

Wro­bel: For us, it’s iden­ti­fy­ing abusers of the sys­tem. We sift through our data to find the few per­cent who in­crease the cost of care. We can re­move those abusers from the net­work, and start pre­dict­ing those who will be­come abusers based on early be­hav­ior. In many cases we ac­tu­ally re­port them to law en­force­ment for fraud.

Peele: Once a pa­tient con­sumes a lot of care and we re­al­ize they need care man­age­ment, it is too late to im­pact the sit­u­a­tion. We have a very ex­cit­ing pre­dic­tive model called Project Flash­light, which pre­dicts which pa­tients are go­ing to need high-touch care man­age­ment as they en­ter the UPMC Health Plan.

How do we do this? We pur­chase pub­licly avail­able data, and ret­ro­spec­tively an­a­lyze who con­sumed a lot of care. Then, we com­pare it to our data to run a prospec­tive model, pre­dict­ing who would con­sume a lot of care based on the en­tire pop­u­la­tion. It works beau­ti­fully. The math is still on my white­board; be­cause it is such a thing of beauty, I can­not bring my­self to erase it.

Dunn: I spend a lot of time think­ing about where hos­pi­tal tal­ent will come from in the fu­ture. Where are they now? Where will they be in five years? So we do a lot of pre­dic­tive an­a­lyt­ics around re­place­ment of key clin­i­cal staff, turnover met­rics and more. For in­stance, we can­not pre­dict ex­actly when peo­ple are go­ing to re­tire, but by flag­ging em­ploy­ees who have run their re­tire­ment and pen­sion plan num­bers three times in one year, that’s a pre­dic­tor for me to start look­ing for that po­si­tion’s re­place­ment.

HOW ARE YOUR OR­GA­NI­ZA­TIONS CUR­RENTLY CHAL­LENGED WITH DATA?

Haines: Gath­er­ing the data and be­ing able to use it has been a chal­lenge. Health­care is in a time of merg­ers and ac­qui­si­tions, and ac­quir­ing providers on dif­fer­ent tech­nol­ogy sys­tems ex­poses the prob­lem that we have too many dis­parate sources of data, and no com­mon def­i­ni­tions.

Peele: As an in­dus­try, we need more stan­dard­iza­tion. The root of the prob­lem is that all the data we’re us­ing now was cre­ated to do some­thing else. For in­stance, claims data are bills to be paid. But then we grab it, and pre­tend to do pop­u­la­tion health man­age­ment with it.

Driver: We too are chal­lenged by the state of to­day’s data. Our ap­proach is to ac­tu­ally ac­cept the world as it is with dis­parate data sources, and look at it as a bowl of in­di­vid­u­ally wrapped can­dies. We cre­ated a sys­tem that acts as a “shrink wrap” around all of those can­dies—that data—to trans­late the dis­parate sources into one lan­guage. We built this with fund­ing from our in­sur­ance com­pany and in con­junc­tion with a ven­dor.

Sav­age: That is a vi­able ap­proach, for now. But we still don’t have a na­tional stan­dard for health­care data. When banks started to process elec­tronic trans­ac­tions, a sin­gle stan­dard was cre­ated. Health­care data is far too valu­able to not treat it the same way.

WHAT CAN PRE­DIC­TIVE AN­A­LYT­ICS AC­COM­PLISH?

Driver: What we’re re­ally jazzed about is tak­ing de­ci­sion an­a­lyt­ics to the pa­tient level. We have the tech­nol­ogy and are steer­ing it to­ward end-of-life care, as this is a very costly area in health­care but has al­most no plan­ning or struc­ture around it. (Re­search also shows there’s a com­plete dis­cord be­tween the pa­tient’s think­ing about prog­no­sis and the clin­i­cian’s.)

So, we are de­vel­op­ing our pro­pri­etary sys­tem to scan and “read” the in­for­ma­tion con­tained in a med­i­cal record over a life­time, as well as per­sonal doc­u­ments, to dis­cover what that pa­tient’s val­ues are. Then, we will com­pare that against the quan­ti­ta­tive in­for­ma­tion and re­search avail­able about their prog­no­sis—gen­er­at­ing a frame­work of the pa­tient’s val­ues against what is ac­tu­ally pos­si­ble in medicine. This will help the pa­tient and physi­cian make keener de­ci­sions.

Sav­age: The end-of-life ap­pli­ca­tion is a valu­able one, be­cause the cul­ture of health­care right now is to pro­vide what­ever care is nec­es­sary for as long as pos­si­ble to sus­tain life to 100 years old. That mind­set needs to change. The ac­tu­ar­ies of Medi­care re­cently re­leased that by 2025, $5.6 tril­lion will be needed for the cost of care. That’s un­ten­able. Wro­bel: I agree, this is a huge is­sue. It is where a lot of our ex­pen­di­ture goes.

Peele: We’ve learned that so­cial de­ter­mi­nants are in­cred­i­bly im­por­tant in pre­dict­ing out­comes. Pa­tients who live alone, or have ex­pe­ri­enced a sig­nif­i­cant life event within the last year—sit­u­a­tions such as these af­fect clin­i­cal out­comes. A very im­por­tant piece of data we use in pre­dic­tive model­ing is the de­pri­va­tion in­dex, which cap­tures the so­cial re­sources avail­able to pa­tients within their ZIP code. This is free in­for­ma­tion. We find that it cor­re­lates with the topol­ogy of chronic con­di­tions, and helps us iden­tify where ad­di­tional re­sources need to be de­ployed based on where a pa­tient lives to im­pact out­comes.

This is im­por­tant, be­cause start­ing next year in Penn­syl­va­nia the in­surer will be re­spon­si­ble for more than just for­mal­ized care. They will also be tasked with mak­ing the pa­tient’s home ac­ces­si­ble, ex­ter­mi­nat­ing pests, putting in chair rails and more. A scope of ser­vices far be­yond medicine.

HOW IS YOUR OR­GA­NI­ZA­TION’S CUL­TURE CHANG­ING TO ADOPT DATA ANAL­Y­SIS AND PRE­DIC­TIONS?

Wro­bel: We’re try­ing to fig­ure out how to take a sys­tem that’s bro­ken, dis­jointed and in­fla­tion­ary, and how to make it more ef­fi­cient. And we are go­ing to, but it will take time. I be­lieve we will see a group of in­cre­men­tal wins that will fuel pre­dic­tive model­ing, lead­ing to more change.

Peele: As an in­dus­try, we’re not even think­ing about the fact that fi­nan­cial risk has been pushed onto providers who are not trained to han­dle or man­age it. In an ef­fort to man­age it, clin­i­cians are turn­ing to data. We’re not push­ing it down their throats—they’re com­ing with their plates hun­gry for it, be­cause they need that in­for­ma­tion to un­der­stand where they are against their fi­nan­cial risk.

Dunn: When de­part­ments re­ceive data and are ready to fix a prob­lem, some­times the in­for­mal in­ter­nal sys­tems will not al­low these changes to hap­pen. You will need some­one, or a group, to change the cul­ture and pro­cesses that al­low for change.

Haines: Our or­ga­ni­za­tion has ac­tu­ally re­ceived data so well that the de­mand in­creased im­me­di­ately. And we didn’t have a process in place to screen for what data is be­ing asked. Now, we ask in­ter­nal stake­hold­ers whether pulling and an­a­lyz­ing a cer­tain set of data is go­ing to add value to the or­ga­ni­za­tion. If you’re ask­ing for the data, is it ac­tu­ally chang­ing some­thing? Or is it just that you want to “know”? We have lim­ited re­sources and need to think very se­ri­ously about where we spend them.

Driver: What we’ve done is gain ac­knowl­edge­ment for the use­ful­ness of data and then marry it with de­sign think­ing. This in­volves bring­ing groups of peo­ple to­gether to de­cide what the data is telling them and what are they go­ing to do about it— but then in­stead of reach­ing for cookie- cut­ter so­lu­tions, en­gi­neer so­lu­tions that ac­tu­ally work in the cul­ture and the place where the prob­lem ex­ists.

CAN PRE­DIC­TIVE AN­A­LYT­ICS BE DONE WRONG?

Wro­bel: Yes. Health­care is in­creas­ingly learn­ing to value data; the prob­lem is when peo­ple in­ter­pret the in­for­ma­tion with­out enough train­ing. Cer­tain dis­ci­plines and pro­fes­sion­als asked to in­ter­pret data with­out the back­ground to fully un­der­stand it is very prob­lem­atic.

Sav­age: In­deed, un­aided an­a­lyt­ics is very, very dan­ger­ous. It has lim­ited the power of pre­dic­tive an­a­lyt­ics be­cause when you have some­one us­ing it in­ap­pro­pri­ately, it’s in­ef­fec­tive.

Dunn: There is a qual­i­ta­tive com­po­nent to this. In health­care we do well with hard data, but no dif­fer­ent than merg­ers and ac­qui­si­tions, it can fail be­cause of soft com­po­nents like cul­ture changes and com­mu­ni­ca­tion is­sues.

WHAT SKILLS, TRAIN­ING AND PRO­FES­SION­ALS WILL PRO­PEL HEALTH­CARE IN PRE­DIC­TIVE MODEL­ING?

Peele: The value of an­a­lyt­ics to an or­ga­ni­za­tion is the amount of in­flu­ence those an­a­lyt­ics have on de­ci­sion-mak­ing. So, I have two jour­nal­ists in my shop whose job is to make the data con­sum­able to the end user. I have 30 em­ploy­ees now, and we are grow­ing to 50 in the next year. I credit the jour­nal­ists en­tirely for the ap­proval in ex­pan­sion be­cause they’ve prop­erly com­mu­ni­cated the value.

Dunn: We need to look at the abil­ity to use data in de­ci­sion-mak­ing as a core com­pe­tency for on­board­ing and hir­ing lead­ers. Too many or­ga­ni­za­tions are not screen­ing lead­ers’ back­grounds and how they un­der­stand and read data—we just leave it up to ran­dom groups of peo­ple who un­der­stand how to do it. Mov­ing for­ward, these must be con­sid­ered core com­pe­ten­cies.

Wro­bel: As an ac­tu­ary, I value what ac­tu­ar­ies bring to the ta­ble: pro­fes­sion­al­ism and reg­u­la­tory knowl­edge. We don’t want “data pullers” in our or­ga­ni­za­tion. Geisinger errs on the side of train­ing rather than re­cruit­ing in this area, so we’ve built a cul­ture around re­ward­ing those who pass ac­tu­ar­ial ex­ams. We em­pha­size the im­por­tance of the ac­tu­ar­ial cre­den­tial for an­a­lyt­ics.

WHAT DOES THE FU­TURE OF PRE­DIC­TIVE AN­A­LYT­ICS LOOK LIKE?

Haines: To me, it is pair­ing the right clin­i­cian with the right pa­tient. We need to start look­ing at out­comes based on per­son­al­i­ties, ed­u­ca­tion and physi­cian pref­er­ences. Some physi­cians are bet­ter at tak­ing care of cer­tain types of pa­tients. So, the out­come that’s de­sired by both pa­tient and physi­cian—not nec­es­sar­ily de­sired by so­ci­ety or the health plan—that’s an ex­cit­ing fron­tier.

Driver: The law of ac­cel­er­at­ing re­turns tells us that even though we are at the be­gin­ning of pre­dic­tive an­a­lyt­ics in health­care, we will see faster re­turns as time goes on. Think about how in the 1960s we were us­ing cal­cu­la­tors, and to­day we have phones that com­mu­ni­cate with our watches. The suc­cesses in pre­dic­tive an­a­lyt­ics will be copied and am­pli­fied across the in­dus­try, cre­at­ing faster adop­tion. There is no doubt that pre­dic­tive model­ing will be wo­ven into the fu­ture of health­care.

City Hall, Philadel­phia

KURT WRO­BEL Chief Ac­tu­ary, Geisinger Health Plan

JIM DUNN Park­land Health and Hos­pi­tal Sys­tem

CEO, Health­care Intelligence Chair­man, Health An­a­lyt­ics In­sti­tute at Iona Col­lege PAUL SAV­AGE

JEFFREY DRIVER CEO, The Risk Author­ity Stan­ford Health Care

CAROL HAINES VP, Clin­i­cal Trans­for­ma­tion Thomas Jef­fer­son Univer­sity Hos­pi­tals

Moder­a­tor, Mod­ern Health­care Cus­tom Me­dia CHRISTINA GALOOZIS

PAMELA PEELE UPMC Health En­ter­prises

CAROL HAINES, VP OF CLIN­I­CAL TRANS­FOR­MA­TION AT THOMAS JEF­FER­SON UNIVER­SITY HOS­PI­TAL, RIGHT, EX­PECTS PRE­DIC­TIVE AN­A­LYT­ICS TO BET­TER MATCH PHYSI­CIANS AND PA­TIENTS IN THE FU­TURE BASED ON THEIR PER­SON­AL­I­TIES, VAL­UES AND OTHER TRAITS—IN HOPES OF BET­TER ACHIEV­ING MU­TU­ALLY DE­SIRED OUT­COMES.

JEFFREY DRIVER, CHIEF RISK OF­FI­CER FOR STAN­FORD HEALTH CARE, RIGHT, IS AP­PLY­ING PRE­DIC­TIVE AN­A­LYT­ICS TO END-OF-LIFE CARE. “THIS IS A VERY COSTLY AREA IN HEALTH­CARE BUT HAS AL­MOST NO PLAN­NING OR STRUC­TURE AROUND IT,” HE SAYS.

PAMELA PEELE, CHIEF AN­A­LYT­ICS OF­FI­CER FOR UPMC EN­TER­PRISES, EM­PHA­SIZES THE IM­POR­TANCE OF HER AN­A­LYT­ICS “SHOP” OP­ER­AT­ING IN­DE­PEN­DENTLY AND RE­PORT­ING DI­RECTLY TO THE CEO.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.