by geraint lewis, heather kirkham, ian duncan, and rhema ... · a 2010 study found that the average...

By Geraint Lewis, Heather Kirkham, Ian Duncan, and Rhema Vaithianathan

How Health Systems Could Avert‘Triple Fail’ Events That AreHarmful, Are Costly, And ResultIn Poor Patient Satisfaction

ABSTRACT Health care systems in many countries are using the “TripleAim”—to improve patients’ experience of care, to advance populationhealth, and to lower per capita costs—as a focus for improving quality.Population strategies for addressing the Triple Aim are becomingincreasingly prevalent in developed countries, but ultimately success willalso require targeting specific subgroups and individuals. Certain events,which we call “Triple Fail” events, constitute a simultaneous failure tomeet all three Triple Aim goals. The risk of experiencing different TripleFail events varies widely across people. We argue that by stratifyingpopulations according to each person’s risk and anticipated response toan intervention, health systems could more effectively target differentpreventive interventions at particular risk strata. In this article wedescribe how such an approach could be planned and operationalized.Policy makers should consider using this stratified approach to reducethe incidence of Triple Fail events, thereby improving outcomes,enhancing patient experience, and lowering costs.

The “Triple Aim” of health care is toimprove individual patients’ expe-riences of care, advance populationhealth, and reduce per capitahealth care costs.1 A central tenet

of the Triple Aim is to restructure care in waysthat lead to improvements across all three ofthese goals.The Institute for Healthcare Improvement has

worked with organizations in many countries toimplement populationwide interventions tofoster the Triple Aim.2 Examples are programsthat encourage self-management of chronic con-ditions,3 promote e-mail communication be-tween patients and physicians,4 and encouragegreater use of primary care.5

Other organizations have adopted a moretargeted approach to achieving the Triple Aim.For example, a Commonwealth Fund case studyfound examples of organizations that werefocusing on improving access and care for

individual patients who had low incomes, wereuninsured, or had complex chronic conditions.3

Indeed, several authors have argued that successwill require both population health and in-dividually focused strategies, such as thoseemployed by Genesys Health System in Flint,Michigan.1,3,6,7 For example, Genesys increasedits primary care capacity (that is, a populationapproach) and offered health navigators to itshigh-risk patients (that is, a targeted approach).The objectives of this article are to propose a

third, stratified approach to tackling the TripleAim and to explore some of the ethical chal-lenges that this new approach presents. Thestratified approach to the Triple Aim involvesidentifying and prioritizing subpopulations ac-cording to their risk of experiencing healthencounter failures—what we call “Triple Fail”events—and according to their likelihood ofbenefiting from preventive care.6,8

Wedefine aTriple Fail event as ahealthoutcome

doi: 10.1377/hlthaff.2012.1350HEALTH AFFAIRS 32,NO. 4 (2013): –©2013 Project HOPE—The People-to-People HealthFoundation, Inc.

Geraint Lewis ([email protected]) is chief data officerof the National HealthService, in London, England.

Heather Kirkham is a managerin the Clinical Outcomes andAnalytics Department atWalgreens, in Deerfield,Illinois.

Ian Duncan is the vicepresident in the ClinicalOutcomes and AnalyticsDepartment at Walgreens.

Rhema Vaithianathan is asenior research fellow at SimKi Boon Institute, SingaporeManagement University, anddirector of the Centre forApplied Research inEconomics, University ofAuckland, in New Zealand.

April 2013 32:4 Health Affairs 1

Quality & Governance

that is recorded in administrative data and arisesfrom the health care process. Such events simul-taneously have three failures: They are costly,represent a suboptimal health outcome, andare a poor patient experience. To generate thelist of Triple Fail events in Exhibit 1, we appliedour definition to the published literature to con-firm each example’s failure on all three TripleAim goals.

Triple Fail EventsTriple Fail events include untimely nursinghome admissions,9 unplanned hospital re-admissions,10–12 inappropriate initiations of

hemodialysis,13–15 and surgeries for lowback painin patients not offered decision support.16 Forexample, nursing home admission was cited asa common fear among older people,9 was oftenpreceded by poor physical and emotionalhealth,17 and was estimated to cost $203–$243 billion annually in the United States in2009.18

Other Triple Fail events are more controver-sial, such as overmedicalized death, which maybe defined as receiving “life-sustaining treat-ments, such asmechanical ventilation, thatmostbeneficiaries indicate they prefer to avoid whenfaced with less than a year to live.”19

Exhibit 1

Examples Of Triple Fail Events

Event Quality of care Patient experience Cost

Unplanned hospitalreadmission within30 days

A readmission may indicate complications,a premature discharge, a failure tocoordinate and reconcile medications,inadequate communication, or poordischarge planninga

Higher 30-day risk-standardized hospitalreadmission rates areassociated with lowerpatient satisfactionb

A 2009 study found that 30-dayrehospitalizations cost Medicare$17.4 billion annuallyc

Nursing home admission Predictors of nursing home admissioninclude low self-rated health status andfunctional and cognitive impairmentd

Loss of independence andnursing home admissionare two of the major fearsof older peoplee

The cost of long-term care in theUnited States in 2009 wasestimated to be $203–$243 billionf

Inappropriate initiation ofhemodialysis

Peritoneal dialysis patients experienceda lower adjusted relative risk of deathcompared with those beginninghemodialysis;g late and early dialysisinitiation appear to be associated withequal outcomesh

Peritoneal dialysis patientsreported better quality oflife and better satisfactionwith dialysis carei

Median annual health care costs in2004 were $43,510 higher forhemodialysis patients thanperitoneal dialysis patientsj

Wrong-site surgery Wrong-site surgery is one of the NationalQuality Forum’s “never events”k

Wrong-site surgery can bea devastating experiencefor the patientl

A 2010 study found that the averagecompensation paid to victims ofwrong-site surgery was $47,216m

Intentional injury ormaltreatment of a child

Child maltreatment involving physicalabuse is the leading cause of infantdeath from injuryn

Child abuse has beenassociated with a widerange of psychologicalsymptoms in the victimo

A 2012 study found the annual acutemedical costs and annual societalcosts of childhood abuse in theUnited States were $2.9 billion and$80 billion, respectivelyp

Overly invasive treatmentfor a preference-sensitivecondition

Patients offered a decision aid for apreference-sensitive condition reportedsignificantly improved outcomesq

Patients offered a decisionaid reported significantlygreater satisfaction withtheir selected treatmentq

Patients offered a decision aid were21–44% less likely to choosecostly, aggressive surgeryq

SOURCE Authors’ analysis. Sources are listed by individual entries in this exhibit. NOTE A preference-sensitive condition has multiple treatment options with roughlyequivalent risks and benefits, which means that the patient’s preference should determine which option is selected. An example is low back pain, which can betreated by surgery, physical therapy, or analgesics. aSee Note 10 in text. bSee Note 11 in text. cSee Note 12 in text. dSee Note 17 in text. eSee Note 9 in text. fSeeNote 18 in text. gMehrotra R, Chiu YW, Kalantar-Zadeh K, Bargman J, Vonesh E. Similar outcomes with hemodialysis and peritoneal dialysis in patients with end-stage renal disease. Arch Intern Med. 2011;171(2):110–8. hSee Note 13 in text. iSee Note 14 in text. jSee Note 15 in text. kMichaels RK, Makary MA, Dahab Y,Frassica FJ, Heitmiller E, Rowen LC, et al. Achieving the National Quality Forum’s “Never Events”: prevention of wrong site, wrong procedure, and wrong patientoperations. Ann Surg. 2007;245(4):526–32. lMulloy DF, Hughes RG. Wrong-site surgery: a preventable medical error. Chapter 36 in: Hughes RG, editor. Patientsafety and quality: an evidence-based handbook for nurses. Rockville (MD): Agency for Healthcare Research and Quality; 2008 Mar. p. 381–94. mStahel PF, SabelAL, Victoroff MS, Varnell J, Lembitz A, Boyle DJ, et al. Wrong-site and wrong-patient procedures in the universal protocol era: analysis of a prospective database ofphysician self-reported occurrences. Arch Surg. 2010;145(10):978–84. nOverpeck MD, Brenner RA, Trumble AC, Trifiletti LB, Berendes HW. Risk factors for infanthomicide in the United States. N Engl J Med. 1998;339(17):1211–6. oGross AB, Keller HR. Long-term consequences of childhood physical and psychologicalmaltreatment. Aggress Behav. 2006;18(3):171–85. pGelles RJ, Perlman S. Estimated annual cost of child abuse and neglect [Internet]. Chicago (IL): Prevent ChildAbuse America; 2012 Apr [cited 2013 Feb 27]. Available from: http://www.preventchildabuse.org/downloads/PCAA_Cost_Report_2012_Gelles_Perlman_final.pdf. qSeeNote 16 in text.


2 Health Affairs April 2013 32:4

Approaches To Preventive CarePopulation And Targeted Approaches Thetwo leading approaches to preventive care—thepopulation strategy and the targeted (high-risk)strategy—were described in a seminal article bythe British epidemiologist Geoffrey Rose.20 Thepopulation strategy seeks to shift the distribu-tion of risk within an entire population toward alower range—for example, by decreasing theamount of salt in the typical diet to reduce thepopulation’s average blood pressure. The tar-geted strategy aims to truncate the risk distri-bution by identifying high-risk individuals andoffering them interventions to reduce their indi-vidual susceptibility—for example, by screeningblood pressure among the population and offer-ing medication to people with hypertension.An important advantage of the population

strategy is its potential to make widespread im-provements inpublic health.20However, theben-efit to each individual is relatively small usingthis approach. Most people experience no par-ticular improvement in their health—a resultknown as the “prevention paradox.”20,21 In con-trast, the targeted strategy provides customizedcare that maximizes outcomes for individual pa-tients. But such customized care will rarely becost-effective for all patients, andhence it shouldnot be offered universally.The choice of preventive strategies should de-

pend on the cost and effectiveness of the pro-posed intervention. Both of Rose’s strategiesmay be cost-effective when the intervention costis low.Butpopulationwideapproaches tend tobemore appropriate when the intervention’s effectis large,while targeted approaches aremore suit-able when accurate predictive models are avail-able and an optimal risk threshold for interven-tion is used. For example, treatment may bedetermined by a risk score cut-off point, whichis based on a simulation of the costs and benefitsand is designed to maximize the cost-effective-ness of the intervention.22

Rose’s hypothesis was formulated prior to thedevelopment of accurate multivariable risk-prediction tools.23 More recent analysis of pop-ulation data suggests that targeted, high-riskapproaches may be particularly advantageousunder certain circumstances, such as when theinterventionhas a degree of disutility—for exam-ple, if the intervention is considered a poor pa-tient experience, which might include cost, losttime, or an adverse lifestyle change.23

The Stratified Approach Our novel, strati-fied approach to the Triple Aim could generateadditional value by combining some of the ad-vantages of the other two approaches.8 This thirdapproach is best adopted by organizations withresponsibility for a population’s health, such as

accountable care organizations. It involves ana-lyzing medical claims, pharmacy claims, elec-tronic health record information, and other ad-ministrative data to predict individuals’ risks ofdifferent Triple Fail events. The organizationwould next estimate each person’s likely re-sponse to a range of preventive programs andthen assign people to different interventions ac-cording to their likely benefit.Compared with the population at large, the

subpopulation of each risk stratum would berelatively homogeneous, allowing the interven-tion to be customized to meet the needs of thepatients in that stratum. For example, a programaimed at preventing hospital readmission mightoffer case management, telephonic care, and re-mote monitoring, with more intensive interven-tions offered to patients in higher risk strata.Werecommend that an ethics committee approveany algorithms used for targeting interventionsprior to implementation and that excluded pa-tients be considered aspart of a feedback loop forprogram evaluation and improvement.Within a high risk stratum, the average per-

son’s risk of experiencing the Triple Fail eventwill be higher than the average population risk.As a result, a higher proportion of individuals inthese groups could benefit from preventive care.In other words, this stratumwould have a higherpositive predictive value, which in turn wouldincrease the cost-effectiveness of the preventiveintervention, all other things being equal. How-ever, the stratified approach is beset by a numberof challenges, principally those relating to theethical aspects of risk stratification, which aredescribed below.23

The stratified approach to the Triple Aim de-scribed in this article includes three phases. Aplanning phase would involve conducting an op-portunity analysis, developingpredictivemodelsand impactibility (alsoknownas intervenability)models. The latter are models that seek to iden-tify subgroups of high-risk people who are mostlikely to engage with and respond to variouspreventive interventions, such as case manage-ment. The planning phase would also include anethical review to ensure its compliance with ouradaptation of James Wilson and GunnerJungner’s prerequisites, described below.An operational phase would use the predictive

models and impactibility models to identifyhigh-opportunity patients—those who are bothat risk and amenable to an intervention—andoffer thempreventive interventions. An ongoingfeedback phase would refine the predictive mod-els and impactibility models—for example, byprioritizing patients with characteristics similarto those of patients who responded well to theintervention.


Implementing The StratifiedApproachOpportunity Analysis The starting point for ahealth system interested in pursuing the strati-fied approach to the Triple Aim is to undertake adetailed analysis of where the greatest opportu-nities exist for improving care. Known in thestrategic management literature as opportunityanalysis, this process would, in the case of theTriple Aim, involve analyzing historical popula-tion data to identify Triple Fail events and gaug-ing how responsive each such event might havebeen to a cost-effective preventive interventionidentified from the literature.24

Once a high-opportunity subpopulation ofpatients has been identified in historical data,people in the population with these character-istics need to be identified prospectively to de-termine who should be offered an intervention.Therefore, a key requirement for the stratifiedapproach is the ability to identify patients whoare at risk of future Triple Fail events.6

Predictive Modeling Predictive risk modelsare statistical algorithms based on relationshipsin historical population data. They may be ap-plied in an automated fashion to routinely col-lected data to estimate the probability that aperson will experience a Triple Fail event in aspecified future timeperiod.25,26 For example, thecombined predictive model is used by NationalHealth Service organizations in England to cal-culate a person’s risk of unplanned hospitali-zation in the next twelve months, according tofactors recorded in the previous two years’worthof primary care electronic health record data andhospital claims data.27

A predictive model for identifying vulnerablepeople could potentially be built for any type ofTriple Fail eventwhose occurrence is recorded inroutine data. For example, researchers in NewZealand have developed a model that predictsthe risk within five years of intentional injuryormaltreatment to an individual child.28 Amodeldeveloped in the United Kingdom predicts therisk of admission to a nursing home withintwelve months.29 And Canada uses a model topredict hospital readmissions within thirtydays.30

Impactibility Modeling A recognized limita-tion of using predictive risk models to organizecare is that some of the patients identified asbeing at high risk may not be amenable to theproposedpreventive intervention.25 In response,many organizations have developed impactibil-ity models that seek to identify the subgroups ofhigh-risk people who are most likely to engagewith and respond to various preventive inter-ventions.31,32 This additional filter is intendedto improve the cost-effectiveness of preventive

programs.Several approaches to impactibility modeling

have been described elsewhere.33 First, healthcare organizations may prioritize people withconditions known to be responsive to preventivecare, such as patients with an ambulatorycare–sensitive condition, who may be particu-larly likely to respond to hospital avoidanceinterventions.33

Second, organizations may prioritize patientswhose care appears suboptimal, such as patientswithmultiple “gaps in care.” An example of sucha gapwouldbenot givingbeta-blocker therapy toa patient with heart failure.34

Third, some organizations report that theyplace lowerpriority onpatientswhoareexpectedto respond poorly to preventive care, such aspeople with cognitive or other mental healthdisabilities and those who have language bar-riers. Or an organization may exclude all ofthe very highest-risk patients, because such pa-tients are sometimes regarded as being less ame-nable than others to preventive care.34

The first two approaches to impactibility mod-eling may help reduce health care disparities,because both suboptimal care35 and ambulatorycare–sensitive conditions36 tend to be morecommon in people with low incomes. In con-trast, the third approach—placing lower priorityon those less likely to benefit from preventivecare—raises serious ethical concerns. This ap-proach would probably exacerbate health caredisparities, and it may be illegal in some coun-tries.37 Finally, because patients in very high riskstrata have such a high propensity for Triple Failevents, expending resources to identify the fewwho can be affected is usually worth the effort.

Ethics Of Screening Using PredictiveRisk ModelsAs Rose noted, the individual approach to healthimprovement is hampered by the “difficultiesand costs of screening.”20(p36) A screening testseeks to identify people who are at sufficientlyhigh risk of an adverse outcome towarrant offer-ing them a diagnostic test or recommending aprophylactic treatment. The stratified approachto the Triple Aim likewise requires screening apopulation to identify subpopulations that are atsufficiently high risk of a Triple Fail event andsufficiently amenable to a preventive interven-tion to justify further action.Any screening test has the potential to cause

more harm than good, such as by exposing pa-tients to false positive and false negative results.Therefore, strict ethical guidelines are requiredto safeguard against the inappropriate use ofscreening.38 The World Health Organization



published ten prerequisites, proposed byWilsonand Jungner, that should be met by any ethicalscreening program.38,39 Among these pre-requisites are that the condition being screenedfor should be an important health problem; thatthere should be a detectable early stage whentreatmentwould be ofmorebenefit than itwouldbe later; and that the risks, both physical andpsychological, should be less than the benefits.Because the stratified approach to the Triple

Aim involves population screeningusing routinedata, we suggest that equivalent caveats shouldapply as part of the planning phase. AdaptingWilson and Jungner’s prerequisites,39 we pro-pose the following ethical criteria for stratifyingpopulations according to risk for Triple Failevents.

Prerequisites For Stratification Theevent being predicted should be an importanthealth problem. There should be an interventionthat can mitigate the risk of the event; resourcesand systems for timely risk stratification andpreventive interventions; sufficient time for in-tervention between stratification and the occur-rence of the event; a sufficiently accurate predic-tive risk model for the event, which—togetherwith the impactibility model—is acceptable tothe population at large; and an accepted policyabout who should be offered the preventiveintervention.In addition, the natural history—that is, the

practices and processes that typically lead to thistype of Triple Fail event—should be adequatelyunderstood by the organization offering the pre-ventive intervention. The cost of stratificationshould be “economically balanced,” meaningthat it should not be excessive relative to the costof the program as a whole. And stratificationshould be a continuous process, not just a “onceand for all” occurrence.

Access For Certain High-Risk PopulationsAnother important ethical concern relates to theuse of impactibility modeling. Although certainsubpopulations are at high risk, they may bedenied preventive care because they are not ex-pected to respond to it. For instance, peoplewithpersonality disorders or alcohol dependencymight not be amenable to programs aimed atpreventing hospital readmission. The questionis whether such people should be denied preven-tive care on this basis, which corresponds toWilson and Jungner’s requirement that therebe an accepted policy about who should be of-fered the intervention.NancyKass argued that for reasons of distribu-

tive justice, programs should not exclude indi-viduals on thebasis of nonclinical characteristicssuch as race and sex.40 However, Andrew Smart,Paul Martin, and Michael Parker argued that

such discriminationmay be justified on the prin-ciple that it is permissible to treat people differ-ently if there is some ethical justification. Forexample, programs that target low-income oruninsured people because of social justice con-siderations rightly treat different peopledifferently.41

DiscussionEstablishing A Feedback Loop A feedbackloop is necessary for assessing the impact ofthe preventive programon outcomes.42 This stepmay be valuable both for evaluating the programand for refining the impactibility model.For example, a regression analysismight show

that patients with certain characteristics weremore likely than others to respond to an inter-vention. This knowledge could then be used toadjust the impactibility model to ensure that pa-tients with these characteristics were prioritizedin the future, unless that adjustment violatedethical considerations.Current Policy Context The Triple Aim is

becoming increasingly important to policy mak-ers in developed countries as their populationsage, chronic diseases increase in prevalence,and funding constraints become pressing.43

Predictive modeling is now widely accepted inthe United States,44 but the use of impactibilitymodels is less extensive.New financial and quality rules are giving

hospitals and accountable care organizations in-centives to prevent Triple Fail events such asavoidable readmissions. However, many suchorganizations are relying on population strate-gies to achieve these goals, including better carecoordination and improvements in informationtechnology.45 Ultimately, success will probablyrequire targeting specific subpopulations aswell.45

Recommendations The following recommen-dations could promote the appropriate use of thestratified approach to the Triple Aim.▸USE PILOTS: First, Triple Aim pilots should

be established in which demonstration sites arerequired to compare population, individual, andstratified approaches; conduct opportunityanalyses; and apply predictive risk and impacti-bility models.▸REDUCE DATA LAGS: Second, the use of pre-

dictive modeling should be promoted by reduc-ing lags in the availability of information fromthe Medicare limited data set as well as filesavailable for specific uses, such as those foraccountable care organizations. The Medicarelimited data set lags by more than a year, andthe specific-use files lag by severalmonths. Sincesome factors recorded in routine data are


strongly predictive of an imminent Triple Failevent, reducing the time lag between the occur-rence of such factors and their availability inthe data improves the accuracy of the predictivemodel.▸CONDUCT AN ETHICAL REVIEW: Third, there

should be appropriate ethical reviews of both thedesign and implementation of predictive andimpactibility models.▸RECORD MORE EVENTS: Finally, the use of

database indicators should be expanded to covermore potentially adverse events. For example,the Hospital Episode Statistics database inEngland records whether each admission waselective or not, thereby indicating whether theevent was a likely Triple Fail event.Focus Of Future Work There is a growing

bodyof evidence for theability ofpredictivemod-els to identify and classify adverse outcomes.46

However, relatively few studies have assessedthe amenability of people to respond to different

interventions.32 Mathematical simulation mod-els may help clinicians and administratorschoose between different approaches foraddressing the Triple Aim.22 Given the currentfocus on comparative effectiveness research, itmight be helpful if future trials of interventionsdesigned to prevent Triple Fail events examinedthe effect of impactibility models on populationoutcomes and on disparities.Conclusion The Triple Aim is being used in

many countries to improve the quality, experi-ence, and cost-effectiveness of health care. Astratified approach to the Triple Aim offers anumber of potential advantages but requirescareful planning, monitoring, and adaptation.By increasing the use of predictive modeling toidentify Triple Fail events before they occur, thisapproach can contribute to improved individualand population health in a cost-conscious healthcare environment.

▪

Geraint Lewis, Heather Kirkham, and IanDuncan were employees of Walgreens atthe time this article was written.Walgreens uses some of the techniquesdescribed in this article, such assupporting the accountable careorganizations of which the company is amember. Lewis received researchfunding from the Commonwealth Fund,the UK Department for Communitiesand Local Government, the UKDepartment of Health, the NuffieldTrust, the UK National Institute forHealth Research, the University ofAuckland Faculty Research Fund, andthe Basque Institute for HealthcareInnovation. Duncan holds an ownership

share in SCIOinspire Corp., a privatelyheld company that performs predictivemodeling for clients on a consultingbasis, of 1.0–1.5 percent. RhemaVaithianathan received research fundingfrom the University of Auckland FacultyResearch Fund, the UK NationalInstitute for Health Research, andvarious Regional District Health Boardsin New Zealand. Lewis receivedreimbursement of travel expenses tospeak at conferences relating topredictive modeling. The funding camefrom the Commonwealth Fund, theBasque Institute for HealthcareInnovation, the US Department ofVeterans Affairs, the University of

Auckland, the German Sickness FundAOK, the Scottish government, theRoyal College of Physicians of London,the Society of Physicians in Wales, thePortuguese National School of PublicHealth, the Borås Health Region inSweden, and two of his previousemployers: the Nuffield Trust andWalgreens. The authors are grateful toBradford Gray and Robin Osborn, whoprovided helpful comments on aprevious draft. The views expressed inthis article are those of the authorsalone and do not necessarily representthe views of their employingorganizations.

NOTES

1 Berwick DM, Nolan TW,WhittingtonJ. The Triple Aim: care, health, andcost. Health Aff (Millwood). 2008;27(3):759–69.

2 Institute for HealthcareImprovement. The Triple Aim:project prospectus for the UnitedKingdom and Europe [Internet].Cambridge (MA): The Institute;2010 Apr [cited 2013 Feb 27].Available from: http://www.ltu.se/cms_fs/1.64419!/ta%20international%20prospectus.pdf

3 McCarthy D, Klein S. The Triple Aimjourney: improving populationhealth and patients’ experience ofcare, while reducing costs [Internet].New York (NY): CommonwealthFund; 2010 Jul 22 [cited 2013Feb 27]. Available from: http://www.commonwealthfund.org/Publications/Case-Studies/2010/Jul/Triple-Aim-Improving-Population-Health.aspx#citation

4 Zhou YY, Kanter MH, Wang JJ,

Garrido T. Improved quality atKaiser Permanente through e-mailbetween physicians and patients.Health Aff (Millwood). 2010;29(7):1370–5.

5 Sweeney S, Bazemore A, Phillips RLJr., Etz RS, Stange KC. A re-emergingpolitical space for linking personand community through primaryhealth care. Am J Prev Med. 2012;42(6 S2):S184–90.

6 Beasley C. The Triple Aim. HealthcExec. 2009;24(1):64–6.

7 Martinez-Vidal E, Gauthier AK,DiVincenzo A. Helping states en-hance health care quality throughtechnical assistance. Health Aff(Millwood). 2010;29(3):558–62.

8 Manuel DG, Lim J, Tanuseputro P,Anderson GM, Alter DA, Laupacis A,et al. Revisiting Rose: strategies forreducing coronary heart disease.BMJ. 2006;332(7542):659–62.

9 Quine S, Morrell S. Fear of loss ofindependence and nursing home

admission in older Australians.Health Soc Care Community.2007;15(3):212–20.

10 Berenson RA, Paulus RA, KalmanNS. Medicare’s readmissions-reduction program—a positive al-ternative. N Engl J Med. 2012;366(15):1364–6.

11 Boulding W, Glickman SW, ManaryMP, Schulman KA, Staelin R.Relationship between patient satis-faction with inpatient care and hos-pital readmission within 30 days. AmJ Manag Care. 2011;17(1):41–8.

12 Jencks SF, Williams MV, ColemanEA. Rehospitalizations among pa-tients in the Medicare fee-for-serviceprogram. N Engl J Med. 2009;360(14):1418–28.

13 Cooper BA, Branley P, Bulfone L,Collins JF, Craig JC, Fraenkel MB,et al. A randomized, controlled trialof early versus late initiation ofdialysis. N Engl J Med. 2010;363(7):609–19.



14 Kutner NG, Zhang R, Barnhart H,Collins AJ. Health status and qualityof life reported by incident patientsafter 1 year on haemodialysis orperitoneal dialysis. Nephrol DialTransplant. 2005;20(10):2159–67.

15 Berger A, Edelsberg J, Inglese GW,Bhattacharyya SK, Oster G. Costcomparison of peritoneal dialysisversus hemodialysis in end-stagerenal disease. Am J Manag Care.2009;15(8):509–18.

16 O’Connor AM, Llewellyn-ThomasHA, Flood AB. Modifying unwar-ranted variations in health care:shared decision making using pa-tient decision aids. Health Aff(Millwood). 2004;23:var-63–72.DOI: 10.1377/hlthaff.var.63.

17 Luppa M, Luck T, Weyerer S, KönigH-H, Brähler E, Riedel-Heller SG.Prediction of institutionalization inthe elderly: a systematic review. AgeAgeing. 2010;39(1):31–8.

18 Frank RG. Long-term care financingin the United States: sources andinstitutions. Appl Econ PerspectPolicy. 2012;34(2):333–45.

19 Barnato AE, Farrell MH, Chang CC,Lave JR, Roberts MS, Angus DC.Development and validation of hos-pital “end-of-life” treatment inten-sity measures. Med Care. 2009;47(10):1104.

20 Rose G. Sick individuals and sickpopulations. Int J Epidemiol.1985;14(1):32–8.

21 Rose G. High-risk and populationstrategies of prevention: ethicalconsiderations. Ann Med. 1989;21(6):409–13.

22 Ahern J, Jones MR, Bakshis E, GaleaS. Revisiting Rose: comparing thebenefits and costs of population-wide and targeted interventions.Milbank Q. 2008;86(4):581–600.

23 Zulman DM, Vijan S, Omenn GS,Hayward RA. The relative merits ofpopulation-based and targeted pre-vention strategies. Milbank Q. 2008;86(4):557–80.

24 Duncan I. Opportunity analysis.Unpublished paper.

25 Cousins MS, Shickle LM, Bander JA.An introduction to predictive mod-eling for disease management riskstratification. Dis Manag. 2002;5(3):157–67.

26 Duncan I. Healthcare risk adjust-ment and predictive modeling.Winsted (CT): Actex Publications;2011.

27 Wennberg D, Siegel M, Darin B,Filipova N, Russell R, Kenney L, et al.

Combined predictive model: finalreport [Internet]. London: King’sFund; 2006 Dec [cited 2013 Feb 28].Available from: http://www.kingsfund.org.uk/sites/files/kf/field/field_document/PARR-combined-predictive-model-final-report-dec06.pdf

28 Vaithianathan R, Maloney T, JiangN, De Haan I, Dale C, Putnam-Hornstein E, et al. Vulnerable chil-dren: can administrative data beused to identify children at risk ofadverse outcomes? [Internet].Auckland: University of AucklandCentre for Applied Research inEconomics; 2012 Sep [cited 2013Mar 12]. Available from: http://www.msd.govt.nz/documents/about-msd-and-our-work/publications-resources/research/vulnerable-children/auckland-university-can-administrative-data-be-used-to-identify-children-at-risk-of-adverse-outcome.pdf

29 Bardsley M, Billings J, Dixon J,Georghiou T, Lewis GH, SteventonA. Predicting who will use intensivesocial care: case finding tools basedon linked health and social care data.Age Ageing. 2011;40(2):265–70.

30 Van Walraven C, Dhalla IA, Bell C,Etchells E, Stiell IG, Zarnke K, et al.Derivation and validation of an in-dex to predict early death or un-planned readmission after dischargefrom hospital to the community.CMAJ. 2010;182(6):551–7.

31 Meek JA. Affordable Care Act: pre-dictive modeling challenges and op-portunities for case management.Prof Case Manag. 2012;17(1):15–21.

32 Lewis GH. “Impactibility models”:identifying the subgroup of high-riskpatients most amenable to hospital-avoidance programs. Milbank Q.2010;88(2):240–55.

33 Billings J, Dixon J, Mijanovich T,Wennberg D. Case finding for pa-tients at risk of readmission to hos-pital: development of algorithm toidentify high risk patients. BMJ.2006;333(7563):327.

34 Weber C, Neeser K. Using individu-alized predictive disease modeling toidentify patients with the potentialto benefit from a disease manage-ment program for diabetes mellitus.Dis Manag. 2006;9(4):242–56.

35 Hart JT. The inverse care law.Lancet. 1971;1(7696):405–12.

36 Billings J, Mijanovich T, Dixon J,Curry N,Wennberg D, Darin B, et al.Case findings algorithms for patients

at risk of re-hospitalisation: PARR 1and PARR 2 [Internet]. London:King’s Fund; [updated 2006 Feb 22;cited 2013 Feb 28]. Available from:http://www.kingsfund.org.uk/sites/files/kf/field/field_document/PARR-case-finding-algorithms-feb06.pdf

37 National Health Service. Equalityand diversity in the NHS: what theEquality Act 2010 means for you[Internet]. London: NHS; 2012; [lastreviewed 2012 Sep 25; cited 2013Feb 28]. Available from: http://www.nhs.uk/NHSEngland/thenhs/equality-and-diversity/Pages/equality-and-diversity-in-the-NHS.aspx

38 Andermann A, Blancquaert I,Beauchamp S, Déry V. RevisitingWilson and Jungner in the genomicage: a review of screening criteriaover the past 40 years. Bull WorldHealth Organ. 2008;86(4):317–9.

39 Wilson J, Jungner G. Principles andpractice of screening. Geneva: WorldHealth Organization; 1968.

40 Kass NE. An ethics framework forpublic health. Am J Public Health.2001;91(11):1776–82.

41 Smart A, Martin P, Parker M.Tailored medicine: whom will it fit?The ethics of patient and diseasestratification. Bioethics. 2004;18(4):322–42.

42 Epstein L, Gofin J, Gofin R,Neumark Y. The Jerusalem experi-ence: three decades of service, re-search, and training in community-oriented primary care. Am J PublicHealth. 2002;92(11):1717–21.

43 Martini EM, Garrett N, Lindquist T,Isham GJ. The boomers are coming:a total cost of care model of the im-pact of population aging on healthcare costs in the United States byMajor Practice Category. Health ServRes. 2007;42(1 Pt 1):201–18.

44 Feder JL. Predictive modeling andteam care for high-need patients atHealthCare Partners. Health Aff(Millwood). 2011;30(3):416–8.

45 Burns LR, Pauly MV. Accountablecare organizations may have diffi-culty avoiding the failures of inte-grated delivery networks of the1990s. Health Aff (Millwood). 2012;31(11):2407–16.

46 Farley TA, Dalal MA, Mostashari F,Frieden TR. Deaths preventable inthe U.S. by improvements in use ofclinical preventive services. Am JPrev Med. 2010;38(6):600–9.


ABOUT THE AUTHORS: GERAINT LEWIS, HEATHER KIRKHAM,IAN DUNCAN & RHEMA VAITHIANATHAN

Geraint Lewis ischief data officerof the NationalHealth Service inEngland.

In this month’s Health Affairs,Geraint Lewis and coauthorspropose a new rubric for improvinghealth care: predicting andpreventing so-called Triple Failevents, which simultaneouslyreflect poor patient experience andhigh costs while constitutinginferior patient care—the oppositeof the Triple Aim. They note thatthe risk of experiencing differentTriple Fail events varies widelyacross individuals, and they arguethat by stratifying populationsaccording to each person’s risk andanticipated response to a healthintervention, health systems couldmore effectively target differentpreventive services, such as casemanagement. The authors alsodiscuss how the approach ofstratifying the population andtailoring interventions could beplanned and operationalized.

Lewis, now chief data officer ofthe National Health Service inEngland, was senior director ofclinical outcomes and analytics atWalgreens. He is a fellow of theRoyal College of Physicians and a

fellow of the UK Faculty of PublicHealth. Lewis holds a master’sdegree in public health from theLondon School of Hygiene andTropical Medicine and a medicaldegree from the University ofCambridge.

Heather Kirkham isa manager in theClinical Outcomesand AnalyticsDepartment atWalgreens.

Heather Kirkham is a manager inthe Clinical Outcomes andAnalytics Department at Walgreens.She is responsible for researchinvolving Walgreens’ health systempartners. Kirkham earned amaster’s degree in public healthfrom the George WashingtonUniversity and a doctorate inpublic health epidemiology fromWalden University.

Ian Duncan is thevice president inthe ClinicalOutcomes andAnalyticsDepartment atWalgreens.

Ian Duncan is the vice presidentin the Clinical Outcomes andAnalytics Department at Walgreens.He is responsible for Walgreens’outcomes, research andpublications, and custom analytics.He is also an adjunct professor ofactuarial statistics at the Universityof California, Santa Barbara, andan adjunct research professor ofhealth care administration atGeorgetown University. Duncanfounded Solucia Consulting, aprovider of analytical andconsulting services to the healthcare financing industry. He holds adegree in economics from BalliolCollege, Oxford.

RhemaVaithianathan is asenior researchfellow at Sim KiBoon Institute,SingaporeManagementUniversity.

Rhema Vaithianathan is a seniorresearch fellow at the Sim Ki BoonInstitute, Singapore ManagementUniversity, and director of theCentre for Applied Research inEconomics and an associateprofessor of economics at theUniversity of Auckland, NewZealand. She has a doctorate ineconomics from the University ofAuckland.



by geraint lewis, heather kirkham, ian duncan, and rhema ... · a 2010 study found that the average...

Documents