ACOs De­mand Data

Ex­ten­sive in­for­ma­tion on pa­tients is es­sen­tial to en­sure providers aren’t ‘fly­ing blind’ as they seek big­ger re­wards.

Health Data Management - - CONTENTS - BY MAG­GIE VAN DYKE

To de­ter­mine their risk pro­files, ACOs re­quire deeper, bet­ter data than any EHR can pro­vide.

Baystate Health, an in­te­grated health sys­tem in Springfield, Mass., has ap­prox­i­mately $60 mil­lion per year in net pa­tient rev­enue at stake as a Medi­care Next Gen­er­a­tion ACO. “That’s a big swing, plus or mi­nus $60 mil­lion. If you don’t do well, the loss can be a real prob­lem,” says CIO Joel Vengco, se­nior vice pres­i­dent and CIO.

The “Next Gen” pro­gram is for pro­gres­sive providers will­ing to as­sume siz­able two-sided risk to care for a Medi­care pop­u­la­tion, in­clud­ing out­pa­tient, acute and post-acute ser­vices. On the up­side, ACOs get to keep 80 to 100 per­cent of any sav­ings achieved over a set an­nual spend­ing tar­get, plus a 5 per­cent in­cen­tive. On the down­side, ACOs must cover 80 to 100 per­cent of any ex­cess spend­ing.

Baystate is among a mi­nor­ity of or­ga­ni­za­tions tak­ing on this level of risk. Ac­cord­ing to a 2017 sur­vey, 70 per­cent of provider or­ga­ni­za­tions in­volved in value-based pay­ment had less than 20 per­cent of rev­enues at risk. In ad­di­tion, 38 per­cent were in up­side-only risk ar­range­ments.

One rea­son for the slow up­take is the fi­nan­cial risk in­volved. Baystate Health’s ACO, Pi­o­neer Val­ley Ac­count­able Care, has done well in Next Gen, earn­ing more than $4.6 mil­lion in shared sav­ings in 2016, while main­tain­ing qual­ity. But seven of the orig­i­nal 18 Next Gen ACOs suf­fered losses, rang­ing from $63,570 to $6.15 mil­lion. “It’s a tough jour­ney,” Vengco says. “To be able to man­age a pa­tient pop­u­la­tion, you have to re­ally know your pa­tients.”

This in­volves lev­er­ag­ing so­phis­ti­cated data an­a­lytic tools tra­di­tion­ally used by

pay­ers, says Ja­son Joseph, CIO at Spec­trum Health, a 15-hos­pi­tal sys­tem based in Grand Rapids, Mich. “You need a strong an­a­lyt­i­cal ca­pa­bil­ity that helps you un­der­stand prospec­tively, ‘What is the risk of this pop­u­la­tion?’ Then, the pop­u­la­tion needs to be ef­fec­tively man­aged to en­sure ap­pro­pri­ate uti­liza­tion and high-qual­ity care. With­out these fun­da­men­tal ca­pa­bil­i­ties, you are fly­ing blind.”

Prov­i­dence St. Joseph Health, based in Ren­ton, Wash., has been busy build­ing a com­pre­hen­sive an­a­lyt­ics plat­form. The health sys­tem has seen a 150 per­cent growth in lives cov­ered un­der risk-based ar­range­ments over the past three years. When asked to pin­point the most im­por­tant data an­a­lytic ca­pa­bil­i­ties for manag­ing risk, Deepak Sadagopan, group vice pres­i­dent of pop­u­la­tion health in­for­mat­ics and govern­ment pro­grams, out­lines four types—busi­ness de­ci­sion sup­port, care man­age­ment, per­for­mance im­prove­ment and op­er­a­tional in­tel­li­gence.

Busi­ness de­ci­sion sup­port

“Health­care ex­ec­u­tives are hav­ing to make de­ci­sions around the type of pay­ment mod­els their or­ga­ni­za­tions will par­tic­i­pate in,” Sadagopan says. “But there’s a lack of ad­e­quate in­fra­struc­ture to sup­port them in mak­ing those de­ci­sions. All the anal­y­sis we’ve done over the past few years has been on spread­sheets.”

To ad­dress this short­com­ing, Prov­i­dence St. Joseph Health is build­ing a busi­ness de­ci­sion sup­port struc­ture that en­ables ex­ec­u­tives to proac­tively weigh the po­ten­tial ROI from par­tic­i­pat­ing in risk-based pay­ment mod­els. For in­stance, many of Prov­i­dence St. Joseph’s 50 hos­pi­tals and af­fil­i­ated med­i­cal groups wanted to par­tic­i­pate in Medi­care’s new vol­un­tary Bun­dled Pay­ment for Care Im­prove­ment (BPCI) Ad­vanced model. But a key de­ci­sion needed to be made: Which of the 32 clin­i­cal episodes in BPCI Ad­vanced—which range from car­diac valve pro­ce­dures to kid­ney fail­ure—were the or­ga­ni­za­tions best pre­pared to take on?

To pro­vide in­sight, Sadagopan and his team built an an­a­lyt­i­cal frame­work us­ing three years of his­tor­i­cal claims data to find the av­er­age cost of care for BPCI Ad­vanced episodes across Prov­i­dence St. Joseph’s provider groups. The anal­y­sis also com­pared providers’ his­tor­i­cal costs against the bench­mark tar­get set by CMS. “We were able to cal­cu­late the prob­a­bil­ity of each provider group be­ing able to meet spe­cific tar­get prices,” Sadagopan says.

The model is cur­rently be­ing ex­panded to han­dle what-if sce­nar­ios. For in­stance, how would a provider group per­form in BPCI Ad­vanced over time if it could lower post-acute costs for a par­tic­u­lar episode by 10 per­cent? Build­ing this model in­volved in­te­grat­ing enor­mous amounts of claims data and tag­ging crit­i­cal pieces of in­for­ma­tion, such as en­counter costs and read­mis­sions, be­fore stor­ing it in a data mart for anal­y­sis.

Care man­age­ment

Strat­i­fy­ing a pa­tient pop­u­la­tion by health risks and other traits is a key part of suc­cess­ful care man­age­ment. This of­ten be­gins by iden­ti­fy­ing pa­tients with high health risks, such as those with mul­ti­ple chronic dis­eases, as well as those on the verge of hav­ing a se­ri­ous health event.

“High-risk pa­tients, who make up 4 to 6 per­cent of any given pop­u­la­tion, can be as much as 75 per­cent of health­care de­liv­ery costs,” Vengco says. “The op­por­tu­nity is in iden­ti­fy­ing those who aren’t yet in that high-risk pool, some­times re­ferred to as ris­ing risk, so you can help them be­fore they get chron­i­cally sick.”

While a rich source of in­for­ma­tion, the EHR may not in­clude data on ser­vices that pa­tients seek out­side a health sys­tem or net­work. “An im­por­tant part of manag­ing pa­tients is get­ting real-time ac­cess to en­counter no­ti­fi­ca­tions,” Vengco says. “If you can get alerts when pa­tients are ad­mit­ted to an in­pa­tient fa­cil­ity or go to the com­mu­nity health cen­ter down the street, you have a bet­ter chance of en­gag­ing them im­me­di­ately and per­haps triag­ing them to the right level of care.”

Baystate Health is one of the founders of the Pi­o­neer Val­ley In­for­ma­tion Ex­change (PVIX), which in­cludes most health­care or­ga­ni­za­tions in western Mas­sachusetts. “We use the plat­form to ac­cess the holis­tic clin­i­cal record of the pa­tient, en­abling us to view all the care the pa­tient has re­ceived,” Vengco says. “We can see all their meds, lab re­sults, al­ler­gies and prior clin­i­cal notes, re­gard­less of whether it’s a Baystate Health clinic or not.”

But clin­i­cal data is only one type of data needed to un­der­stand pa­tient risk. Re­search gen­er­ally shows that so­cial de­ter­mi­nants of health, such as food in­se­cu­rity and lack of trans­porta­tion, ac­count for about 50 per­cent of health out­comes. Rec­og­niz­ing this, Baystate has started ask­ing Medi­care ACO pa­tients about any so­cial, eco­nomic or en­vi­ron­men­tal chal­lenges they face.

“If one of our Med­ic­aid pa­tients is un­able to see a doc­tor be­cause of trans­porta­tion is­sues, we need to know that so we can try to find a way to get them to the clinic, or per­haps they are a can­di­date for our tele­health ser­vices,” Vengco says.

Com­mu­nity health work­ers are in­ter­view­ing pa­tients face-to-face, us­ing a care needs as­sess­ment app. “That

data is then stored in PVIX and can be re­viewed by clin­i­cians in the ACO and the re­gion when those pa­tients end up in their care,” Vengco says.

To fur­ther en­gage pa­tients, Baystate is part­ner­ing with Pa­tien­tBond to in­cor­po­rate psy­cho­graphic seg­men­ta­tion into its pa­tient registry. Pa­tien­tBond uses a va­ri­ety of con­sumer data, as well as a 12-ques­tion sur­vey, to pin­point pa­tients’ com­mu­ni­ca­tion pref­er­ences (for ex­am­ple, whether they pre­fer con­tact by text, email or phone) as well as what tends to mo­ti­vate them in manag­ing and im­prov­ing their health.

“Pop­u­la­tion health is re­ally about manag­ing be­hav­ior change,” Vengco says. “It’s not un­like what re­tail­ers like Ama­zon and Tar­get have been do­ing. They un­der­stand what mo­ti­vates peo­ple, and they use that knowl­edge to en­gage them and mod­ify their be­hav­iors. We need to take lessons from other in­dus­tries on how to lever­age var­i­ous data types be­yond health­care to truly know our pa­tients and ul­ti­mately bet­ter man­age pa­tient be­hav­iors and, mostly im­por­tantly, their health.”

Per­for­mance im­prove­ment

Spec­trum Health has its own health plan, Pri­or­ity Health, which has proven ben­e­fi­cial in mov­ing to­ward risk-based pay­ment. “As a health plan, we’re in the risk busi­ness al­ready,” Joseph says.

The health sys­tem has built a so­phis­ti­cated an­a­lyt­ics frame­work that in­cor­po­rates sta­tis­ti­cal pack­ages, in­clud­ing SAS and Python, and the vi­su­al­iza­tion tool from Tableau. It has also de­vel­oped risk-scor­ing and care-man­age­ment pro­cesses that lever­age pre­dic­tive an­a­lyt­ics.

In ad­di­tion to aid­ing pop­u­la­tion man­age­ment, these an­a­lyt­i­cal ca­pa­bil­i­ties help data sci­en­tists un­cover driv­ers of vari­a­tion in pa­tient care. “We use the data to go af­ter im­prove­ments that will lower the cost or im­prove the qual­ity of care,” Joseph says.

One les­son learned is the im­por­tance of fo­cus­ing an­a­lyt­i­cal ef­forts on spe­cific high-pri­or­ity is­sues. “Data an­a­lyt­ics is not ‘a build it and they will come’ strat­egy,” Joseph says. “In other words, don’t build a data ware­house with all the data any­one could ever want hop­ing that some­body will come and use it one day.”

Spec­trum Health is us­ing an it­er­a­tive ap­proach. “We start with, ‘Here’s the prob­lem we’re try­ing to solve.’ Then we build an­a­lyt­ics around that prob­lem and pull to­gether data to solve the prob­lem,” he says.

For ex­am­ple, af­ter Spec­trum Health iden­ti­fied hos­pi­tal through­put as an im­prove­ment pri­or­ity, staff gath­ered and an­a­lyzed data to iden­tify the bot­tle­necks and op­por­tu­ni­ties to im­prove per­for­mance across the sys­tem. That led to a re­design of how pa­tients were triaged, moved and put in exam rooms. A lot of use­ful data for the anal­y­sis was al­ready be­ing col­lected via the process log in the health sys­tem’s EHR, in­clud­ing the time when pa­tients are checked in, see a physi­cian, are moved through the or­ga­ni­za­tion and are dis­charged.

“We fol­low the prin­ci­ple, ‘Let’s start with an­a­lyt­ics that we can de­velop based on data we al­ready have be­fore we talk about an­a­lyt­ics that will re­quire ad­di­tional data,’ ” Joseph says. “Some­times data fields are there al­ready, but we haven’t built the an­a­lyt­i­cal tools yet to lever­age that data. As we move up the an­a­lyt­ics ma­tu­rity scale into pre­dic­tive an­a­lyt­ics and lev­er­ag­ing ma­chine learn­ing, we will need to en­sure our data is well or­ga­nized and well man­aged.”

Op­er­a­tional in­tel­li­gence

Sadagopan be­lieves the great­est an­a­lyt­ics op­por­tu­nity lies in in­te­grat­ing in­tel­li­gence into the clin­i­cian and care man­ager work­flow. This in­cludes pro­vid­ing physi­cians with vis­i­bil­ity into any care gaps pa­tients have, in­clud­ing those re­lated to qual­ity met­rics, such as pa­tients need­ing im­mu­niza­tions, as well as lists of high-risk pa­tients who would ben­e­fit from care man­age­ment.

“If we make com­mit­ments in spe­cific payer con­tracts, but do not con­duct those ac­tiv­i­ties, we will not com­ply with the terms of the con­tract and may leave money on the table,” Sadagopan says.

One chal­lenge in im­ple­ment­ing payer con­tracts is the dis­par­ity in how ad­min­is­tra­tive data, in­clud­ing mem­ber ros­ters and claims data, is for­mat­ted and shared. “We see for­mats and struc­tures all over the place,” Sadagopan says. “Some pay­ers send us data as flat de­lim­ited files. Oth­ers send us ad­di­tional bits of in­for­ma­tion in an Ex­cel file. In some cases, we get rich clin­i­cal data with risk codes. In other cases, we don’t get any rich data like that.”

This Wild West sit­u­a­tion can make even sim­ple tasks dif­fi­cult. “We even strug­gle to iden­tify pa­tients at­trib­uted to Prov­i­dence St. Joseph Health,” Sadagopan says.

Re­solv­ing this sit­u­a­tion will re­quire col­lab­o­ra­tion across providers and pay­ers to stan­dard­ize ad­min­is­tra­tive data ex­change, sim­i­lar to how clin­i­cal in­for­ma­tion ex­change has be­come more uni­form. Prov­i­dence St. Joseph has started to part­ner with pay­ers to achieve this. ☐

“Data an­a­lyt­ics is not a ‘build it and they will come’ strat­egy; don’t build a data ware­house and hope that some­body will use it.”

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.