Products & Services

A Return on Investment Evaluation of the Citibank, N.A., Health Management Program

Description
A Return on Investment Evaluation of the Citibank, N.A., Health Management Program
Published
of 14
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  T H E ~ C; i E N C E OF H EA LT HPROMOTION Methods, Issues, and Results in Evaluation and Research A Return on Investment Evaluation of theCitibank N.A. Health Management Program Ronald J. Ozminkowski, Rodney L. Dunn, Ron Z. Goetzel, Richard I. Cantor, Jan Murnane, Mary Harrison  bstract PURPOSE Objectives. Citibank, N.A., initiated a comprehensive health, demand, and diseasemanagement rogram in 1994, using program services offered by Healthtrac, Inc., ofMenlo Park, California. Program components ncluded an initial screening of employees,computerized riage of subjects into higher and lower risk intervention programs, extensivefollow-up with the higher risk subjects, and general health education and awareness build-ing. The objective of this study was to estimate the financial impact of this program onmedical expenditures.Methods. A quasiexperimental design was applied comparing medical expenditures be-fore vs. after the intervention for program participants and nonparticipants. The 22, 838subjects (11,194 program participants and 11, 644 nonparticipants) were follozoed for average of 38 months before and after administration of a Healthtrac health risk apprais-al (HRA) nstrument that triggered the start of the program. To adjust for selection bias the extent possible with these data, multiple regression models were used to estimate thesavings in medical expenditures associated with program participation. The resulting dol-lar savings were compared o program costs to estimate the economic return on the compa-ny’s investment in the program.Results. The return on investment (ROI) was estimated to be between 4.56 and 4.73saved per dollar spent on the program, depending on the discount rate applied. These re-sults are similar to published evaluations of Healthtrac programs mplemented with otherpopulations.Conclusions. Despite limitations inherent in any retrospective observational study, thestrong, positive ROI shown here suggests that a well-designed health management rogram(HMP), which focuses interventions on high risk populations, can result in monetary sav-ings to an organization. (A m J Health Promot 1999;1411]:31-43.)Key Words: Accent Program, Financial Analysis, Health Management Pro-grams, Health Risk Appraisal, Medical Expenditures, Return on Investment Ronald J. Ozminkowski, PhD, is a Senior Economist and Research Manager at The MED-STAT Group, Inc., Ann Arbor, Michigan. Rodney L. Dunn, MS, is a Statistician at theUniversity of Michigan, Ann Arbor, Michigan. Ron Z. Goetzel, PhD, s Vice President, Con-sulting and National Practice, at The MEDSTAT Group, Inc., Washington, DC. Richard LCantor, MD, s Vice President and Medical Director, and Jan Murnane, MEd, MBA, wasVice President and Director, Health Promotion and EAP Services, at Citibank, N.A., NewYork at the time this study was done. Mary Harrison, JD, is a Client Manager with TheMEDSTAT roup, Inc., Stamford, Connecticut. Send reprint requests to Ronald J. Ozminkowski, hD, Senior Economist nd Research Manager, The MEDSTAT roup, nc., 777 East Eisenhower arkway, Suite 500, ,Man Arbor, MI 48108. This manuscript as ubmitted ctober 0, 1998; evisions were equested anuary 8, 1999; he ~nanuscript as c- cepted or publication uly 7, 1999. Am Health Promot 999;14(1):31-43.Copyright 1999 y American ournal f Health romotion, nc,0890-1171/99/ 5. 0 + 0 Corporate worksite health man-agement, health promotion, andwellness programs have often beensold to senior management with thepromise that they will save money.The rationale for savings is derivedfrom the intuitive belief that if em-ployees improve their health habitsand lead healthier lifestyles, they willbecome sick less often, use healthcare benefits infrequently, and spendmore time at work being productive.Increasingly, program supportersrecognize the need for better re-search to support this economic ar-gument for corporate health man-agement? There are, however, sever-al obstacles that stand in the way.Health management advocates andtheir customers, senior management,often assume that health manage-ment programs (HMPs) are standardand likely to achieve uniform, posi-tive financial outcomes. In fact, theseprograms vary tremendously in theirdesign, comprehensiveness, intensity,and impact.2 Rigorous econotnic re-search is difficult, requiring expertisethat is frequently not resident at ei-ther vendor or sponsor organizations.Good research also requires trackingseveral thousand subjects over an ex-tended period to establish programeffects that are statistically valid.Nonetheless, business leaders fre-quently require hard economicdata to support their investment inthese programs and provide justifica-tion for program expansion.In response to this demand fordata, the volume of peer-reviewedstudies investigating the economicimpact of worksite health manage-ment has grown dramatically over September/October 999, Vol. 14, No. 131  the past two decades. In Pelletier’sthree reviews of 77 studies focusedon work.,;ite programs, he reachedthe conclusion that the majority ofthese programs are both health- andcost-effective. 1.3,4While all of the studies reviewedin those articles were published inpeer-reviewed journals, most wereperformed using only descriptivemethods. Few employed control orcomparison groups, and fewer stillapplied multivariate methods to sta-tistically control for alternative expla-nations o.f results. As a point of con-trast, evaluation studies with relativelystrong designs and a financial out-comes focus that were cited by Pelle-tier include those conducted forJohnson ;and Johnson,5,6 DuPont,the Bank of America,8,9 Tenneco,°Duke University,~1 and the CaliforniaPublic Retirees System.~2 Other nota-ble studies include those conductedfor Procter and Gamble~3 and Chev-ron Corporation.4Generally, when financial analysesof worksi~e HMPs are performed,cost savings are rarely weighedagainst program expenditures to de-termine whether the programs arecost-beneficial. Moreover, when re-turn on investment (ROI) figures arecited, they are often derived fromself-report survey research15 or simu-lation studies built upon a set of of-ten questStonable assumptions (J. Ter-borg, unpublished data, 1988). In hismost recent review, which was limitedto worksite-based cardiovascular riskmanagement interventions, Pelle-tier1~ noted a trend toward favorablecost outcomes but a lack of rigorousmethods ~:o assess cost-effectiveness.The investigation of Citibank’sHMP attempted to improve on priorresearch shortcomings. This evalua-tion tracked the medical expendi-tures of 2:2,838 Citibank employeesover a mu.ltiyear period, comparingthe cumulative experience of pro-gram participants to that of nonpar-ticipants, iProgram savings were basedon differences over time in medicalexpenditures by program participantsand nonparticipants. To limit selec-tion bias, regression methods wereemployed to adjust for the effects ofpotential confounders in participantand nonparticipant groups, such asdemographics and employment sta-tus. (It is possible, however, thatsome inherent self-selection bias re-mains that cannot be fully eradicatedby the multivariate approach.) Final-ly, cost savings were compared toHMP expenditures to determinewhether the program was cost-benefi-cial. METHODSDesign A pre-post participant vs. compar-ison group design was used to quan-tify the financial impact of the Citi-bank HMP. The evaluation comparedHMP participants to nonparticipantsin terms of net medical expendituresper person-month. Participants in-cluded everyone who completed ahealth risk appraisal (HRA) survey,regardless of the level of health riskassessed by the survey. Nonpartici-pants did not complete any HRA ur-veys and received no interwentions.Net medical expenditures includ-ed payments by Citibank for its self-insured indemnity and point-of-ser-vice health care plans. Expendituresfor copayments and deductibles thatwere paid directly by the researchsubjects were excluded, becausethese expenditures were no~ directlyrelevant to the corporate ROI focusof the analysis. Expenditure datawere not available for members of Ci-tibank’s health management organi-zation (HMO) plans because pay-ments for HMO ervices were madeon a capitated basis rather than on atransaction basis. Estimates of the im-pact of program participation onmedical expenditures were applied toestablish Citibank’s economic ROI inthe HMP. Sample Those considered for the evalua-tion were all 47,838 active Citibankemployees eligible for medical bene-fits and for the HMP, or the periodJanuary 1, 1994, through December31, 1996. Of the 47,838 eligible em-ployees, 25,931 participated in theHMP, a 54.3% participation rate.Staff from all business units exceptInternational Staff Services (ISS)were included in the evaluation ofthe HMP. The ISS staff (n = 1480)were excluded becau~e not all medi-cal claims were available from thesestaff members, who were stationedoverseas. Others who were excludedwere those for whom enrollmentdata were missing (n = 180) andthose who were erroneously assignedto more than one risk category attheir first HRA n = 25).Subjects were dropped from anal-yses of medical expertditures if theywere retired, over age 64, had optedout of medical cover~.ge, had HMOcoverage, or had less than 6 monthsof enrollment data before or afterthe first HRA was taken. We also ex-cluded pregnant worcten and thosewho died sometime during the studyperiod because their ’atilization wasnot expected to be influenced by theCitibank HMP. The resulting samplesize for the medical expenditure anal-yses was 22,838, including 11,194HMP participants andl 11,644 non-participants. Intervention: The Citibank I-IMP The Citibank HMP was adminis-tered by Healthtrac,s’vt Inc., of MenloPark, California. The program beganin 1994, and its evaluation continuedthrough 1997. The program was de-signed to do the following:¯ help Citibank employees improvetheir health practices and behav-iors, thereby reducing the preva-lence of preventable disease;¯ help them better manage theirchronic medical conditions; and¯ reduce the demand for unneces-sary and inappropriate health ser-vices.In addition to these goals, the pro-gram aimed to demonstrate a posi-tive economic ROI. The HRA and Scoring t~’ocess. The Ci-tibank HMP made substantial use ofa standard HRA survey developed byHealthtrac. This HRA orm request-ed self-reported inforraation fromemployees on demographic factors,smoking status, alcohol use, vitality,height, body weight, exercise levels,seat belt use, nutrition habits, stresslevels, use of medications, chronichealth problems, and perceptions of 32American ournal of Health Promotion  one’s health status and most impor-tant health problem.Most of the questions chosen forthe Healthtrac HRA were taken fromthe Health Assessment Questionnairedeveloped by the Centers for DiseaseControl in 1980. The reliability andvalidity of these HRA uestions havebeen studied in several applications;a review of these studies was recentlypublished by Eddington et al.~7 Theynoted that there is general agree-ment that the HRA has a high de-gree of face validity and that [r]eliability issues are not a majorproblem with HRA calculations, sincethe results are minimally affected byminor changes in answers to mostsurvey questions (1999:136).Eddington et al.17 also noted thatthe reliability and validity of HRA p-plications should be assessed as HRAinstruments change. They recom-mended this especially for studies ofrelationships between risk and medi-cal expenditures. The HealthtracHRA has been promoted as a meansof identifying those at high risk forpoor health or high medical expen-ditures, and it was used for both pur-poses in its Citibank application. Theassumption here was that high riskfor poor health would be associatedwith lower health status, which inturn would be associated with highermedical expenditures. If this is true,and if an intervention such as the Ci-tibank HMP an be used to identifyhigh risk people and motivate themto better manage their health, medi-cal expenditures may decrease forthose who participate in the inter-vention. This evaluation of the Citi-bank HMP herefore focused on itsimpact on medical expenditures. Lat-er studies will focus on the relation-ship between the HMP and changesin risk over time.Relationships between risk factorsfor poor health and subsequent poorhealth status have been illustrated inseveral research studies in the past,but relationships between risk factorsand subsequent medical expendi-tures are less well documented?7 Toverify the validity of its claim that theHealthtrac HRA could be used toidentify those at risk for high medi-cal expenditures, Healthtrac staffconducted an unpublished study ofrelationships between medical expen-ditures and several predictive factors,many of which were taken from theirHRA nstrument. The validity studywas conducted using a data set with24,626 observations. This externaldata set was developed before con-tracting with Citibank and thereforedid not include Citibank data.To test the validity of the Health-trac HRA or use in predicting medi-cal expenditures, the 24,626-observa-tion external data set was split ran-domly into two data sets by Health-trac staff. One of these random splitswas labeled the learning data set (n = 12,303), and the other was de-noted the test data set (n 12,323). The learning data set wasused for stepwise multiple regressionanalyses to identify a good cost pre-diction model. Independent variablesfor the cost prediction model includ-ed sample members’ self-reportedage and sex, an aggregate risk scorecalculated from responses abouthealth risks using the HRA nstru-ment, the self-reported existence ofspecific medical conditions noted onthe HRA, the reported number ofphysician visits used in the prior year,the reported number of medicationsbeing taken, total health care expen-ditures in the previous year, and out-patient expenditures in the previousyear. The resulting stepwise regres-sion model was then applied to thetest data set to generate predictionsof medical expenditures.To assess the validity of theHealthtrac HRA or predicting medi-cal expenditures, the predicted ex-penditures obtained from the testdata set were then compared to actu-al health care expenditures by decile.Specifically, Healthtrac staff sortedthe distribution of predicted medicalexpenditures from lowest to highestand then broke that distribution into10 equally sized (roughly) groups.The mean values of predicted expen-ditures within each decile were thencompared to the mean values of ac-tual expenditures for the same decileusing a paired t-test. The differencesbetween these two sets of mean val-ues were always less than $40 in abso-lute value, and the paired t-testsshowed no significant differences inthe mean values for 9 of the 10 dec-iles. The one exception was the low-est cost decile, and even in that dec-ile, the predicted and actual meanexpenditure values differed by only$28.17.In light of these results, and sincemany of the significant predictors ofmedical expenditures were obtainedfrom information reported on theHRAs, he predictive validity of theHealthtrac HRA has been estab-lished, at least for the populationstudied in that unpublished research.It should be noted, however, that allvalidity tests are always situation-spe-cific; validity is not a universal fea-ture of any intervention.~S Thus, thepredictive validity of the HealthtracHRA hould be studied in other con-texts as well, even though the earlyevidence from the Healthtrac study iscompelling.Partly because of its demonstratedpredictive validity in that unpub-lished study, the Healthtrac HRA wasused as a basis for the interventionsapplied in the Citibank HMP. HRAdata were used to classify Citibankemployees as being at lower or high-er risk for high medical expendi-tures, using the expenditure regres-sion model described above. Higherrisk persons were labeled as thosewhose predicted expenditures wereabove the 80th percentile of the pre-dicted expenditure distribution. Citibank HMP mplementation. The Ci-tibank HMP was implemented as fol-lows. The standard Healthtrac HRAquestionnaire was offered to all47,838 Citibank employees in 1994and again in 1996. As an incentive tocomplete the HRA and participate inthe program, subjects were given a$10 credit, which was applied to theirportion of the cost of the company’semployee benefits program. Basedon the initial HRA esults, thosefound to be at high risk for highmedical expenditures were invited toparticipate in a high risk interven-tion program called Accent. Those who self-reported the exis-tence of specific conditions includingasthma, arthritis, back pain, highblood pressure, chronic lung disease,diabetes, heart disease, tobacco use,and high body weight were also invit-ed to participate in Accent programs September/October 999, Vol. 14, No. 133  tailored to those conditions. Accentprogralns were also offered to thosewho had multiple risk factors (e.g.,stress, overweight, poor nutrition,and sedentary lifestyle), even whenany one of those risk factors, by it-self, was not considered seriousenough to lead to high risk status.For these individuals, the combina-tion of risk led to their assignment tohigh risk status.Each Accent high risk programincluded three additional HRA ques-tionnaires, offered at about 3-monthintervals, after the initial HRA wastaken. These additional HRAs weretailored to the risks that led to Ac-cent program participation. Howev-er, some risks are common to prob-lems such as diabetes and obesity orto tobacco use, lung disease, andheart disease, so the various Accentmodule HRAs included some of thesame questions. Following each Ac-cent module questionnaire, partici-pants received a personalized letterand report, identifying their mostcurrent health risks and recom-mending simple actions to reducethose risks and improve health. Inaddition: after each questionnaire,participants received pamphlets andother health education materialsspecific to their chronic conditionsor lifestyle risks. Accent programparticipants identified in 1994 alsoreceived a standard HRA n 1995and two standard HRAs in 1996,along with follow-up reports. A ran-dom sample of Accent participantsfrom 1996 also received one tele-phone counseling call to check ontheir progress in reducing risk andto support risk-reduction behaviors.In 1996, Citibank also introduced aninbound telephone counseling ser-vice through a Healthtrac, Inc., sub-contractor. This service was offeredto high risk participants who calledto obtain advice from a nurse re-garding their Accent modules. Theinbound counseling service also of-fered telephone access to an audiohealth library of tapes.All HI~A participants (includingthose at lower risk) received a confi-dential letter and report identifyingtheir individual health and lifestylerisks, based on the information sub-mitted on their HRA forms. Low riskHRA articipants received their ini-tial HRA, a follow-up report, andgeneral health education materials.All participants, regardless of risk,were provided access to Healthtrac’scustomer service toll-free telephoneline, to ask any questions aboutHealthtrac program services.Finally, all HRA articipants weremailed their choice of one of threepopular self-care books that providedadvice and instruction on personalhealth care, child health care, in-formed use of health care services,and healthy lifestyles. Analysis Data Sources. Three databases wereintegrated for the Citibank evalua-tion. Healthtrac provided data onparticipation in the Citibank HMPand the high risk Accent modules. Inaddition, data on medical expendi-tures and health plan enrollmentwere provided by Citibank healthplan administrators. These data wereindependently processed and mergedfor analysis by the authors. Statistical Methods. The evaluation ofthe Citibank HMP was operational-ized under the assumption thatmonthly medical expenditures de-pended on HMP participation andother factors such as demographics,job type, type of health care coverage(single vs. family), type of healthplan (indemnity vs. managed care),etc., as noted below. Two-part multi-ple regression models were used toestimate relationships between HMPparticipation and medical expendi-tures per person-month, controllingfor these other factors. The two-partmodel approach was developed andtested by Rand Corporation econo-mists and statisticians in the 1980s)Goetzel et al)4 illustrated the use ofthis model in the context of wellnessprogram evaluations.The two-part regression modelswere estimated twice, once for theperiod before the first HRA was tak-en (the pre-HRA period) and oncefor the period after the first HRAwas taken (the post-HRA period).Each pair of analyses included a lo-gistic regression designed to estimatethe impact of program participationon the odds that any medical expen-ditures were incurred during the pre-or post-HRA period of interest. Eachpair also included an. ordinary least-squares regression designed to esti-mate the impact of F, rogram partici-pation on the magnitude of any ex-penditures incurred.Within each pre- and post-HRAtime period, the total impact of par-ticipating in the Citibank HMP wascalculated mathematScally as theproduct of two numbers that wereobtained from the regression analys-es: (1) the estimated impact of pro-gram participation on the probabilitythat any medical expenditures wereincurred (obtained h’om the logisticregression), and (2) the estimatedimpact of participation on the aver-age number of dollars incurred bythose who had any expenditures (ob-tained from the ordinary least-squares regression). ~ffter performingthis calculation for each time period,the difference in the values obtainedfor the pre- and post-.HRA periodswas calculated to produce an overallestimate of the impact of participat-ing in the Citibank HIME Details of the Regressi~m EstimationProcedure. Within the two-part re-gression model framework, the logis-tic regression analysis was used topredict the probability that medicalexpenditures would he incurred byHMP participants and nonpartici-pants, controlling for the influenceof potentially confounding variables.The ordinary least-squares regres-sion analysis was used to predict theamount of medical expenditures perperson-month that would occur forparticipants and non’participantswho had any expendittures, control-ling for the influence’, of the con-founding variables.The natural logarithm of medicalexpenditures was used as the depen-dent variable in the ordinary least-squares regression analysis, to smooththe impact of outlier values and helpassure that the statistical assumptionsbehind regression analyses would bemet. The use of the logged depen-dent variable required that a retrans-formation factor be applied to gener-ate predictions of medical expendi-tures that could be expressed in actu- 34American ournal of Health Promotion  al dollar terms rather than inlog-dollar terms. The application ofthis retransformation factor, knownas the smearing estimate,’~° correctedfor the understatement that wouldhave occurred if a simple exponen-tial transformation of log dollars hadbeen applied.The general format for the regres-sion analyses can be summarized asfollows: Pre-HRA Regressions: Log Odds of Incurring Any = a0 + a~Low Risk + a2High Risk/No Accent + a:~High Risk/Some Accent + a4High Risk/Full AccentMedical Expenditures + ai Z xi + elLog Medical Expendituresper Month, if Any=b0 + b~Low Risk + b2High Risk/No Accent + b:~High Risk/Some Accent + b4High Risk/Full Accent + bi Z Xi +Post-HRA Regressions: Log Odds of Incurring Any = go + g~Low Risk + g2High Risk/No Accent + g:~High Risk/Some Accent + g4High Risk/Full AccentMedical Expenditures + gi ~ Xi + e5 Log Medical Expendituresper Month, if Any=h0 + h~Low Risk + h2High Risk/No Accent + h3High Risk/Some Accent + h4High Risk/Full Accent -~ hi Z Xi ~- e6 With this format, ao, bo, go, and ~ arethe intercept terms for the four regres-sions. The rest of the ag, b~, g~, and hgare coefficients that were estimated forthe HRA participation and risk terms andthe other variables included in the regres-sion models (denoted by X~). £ is thesummation operator denoting linear rela-tionships between the other variables andthe outcome variables. The e’s refer to er-ror terms for each regression that wereneeded because there is always at leastsome random error involved in modelinghuman behavior.The major independent variables in-cluded in the regressions denoted wheth-er participation in the HMP occurred andthe resulting level of risk assigned to eachparticipant. The variable labeled LowRisk was a binary indicator denotingwhether the HRA respondent was foundto be at low risk for high medical expen-ditures. The variable labeled High Risk/No Accent was a binary indicator denot-ing whether respondents were found tobe at high risk but decided not to partici-pate in any Accent modules. The variablelabeled High Risk/Some Accent denot-ed whether those who were found to beat high risk decided to participate insome Accent activities without completingthe Accent module. Finally, the variablelabeled High Risk/Full Accent denotedwhether those who were found to be athigh risk completed the Accent modulesthey started. The remaining independentvariables, summarized by the X~ terms, aredescribed below.The logistic regression analyses wereused to address the probabilities thatmedical expenditures were incurred ineach time period. Logistic regression wasbased on the premise that a linear rela-tionship existed between the log of theodds that expenditures were incurred andthe predictors of those events. For eachobservation in the data set, the logistic re-gression coefficients were multiplied bythe values of their independent variables,and the results were summed, aloug withthe intercept of the equation, to generatepredictions that log-odd values of expen-ditures were incurred for program partici-pants and nonparticipants. These logodds were then transformed mathemati-cally into probabilities that were used insubsequent calculations.Similarly, the coefficients of the inde-pendent variables used in the ordinaryleast-squares regressions of medical expen-ditures were multiplied by the values ofthose independent variables. The resultswere summed, along with the interceptterms, to generate predicted values of logdollars for each observation. These log-dollar values were then exponentiatedand multiplied by their respective smear-ing estimates, to generate predictions inactual dollar terms rather than in logterms. Duanz° presented the details of thesmearing estimate derivation and its ap-propriate use to convert log dollars backinto actual dollar values.Once the predicted probabilities andthe predicted dollar values were obtained,the probabilities were multiplied by theirassociated dollar values to estimate the to-tal medical expenditure values for eachobservation. The means of these total val-ues were then calculated for each timeperiod for Healthtrac participants andnonparticipants.The difference in the mean valuesover time was calculated for Healthtracparticipants at various levels of risk, aswell as for nonparticipants, to estimatethe overall impact of Healthtrac participa-tion. For the medical expenditures analy-sis, the overall impact of participation wasdefined as a weighted average of thechange in expenditures for program par-ticipants at all levels of risk minus the av-erage change in expenditures for nonpar-ticipants. Thus, even though the study in-corporated information for people at dif-ferent levels of risk, the financial impactof the program as a whole was estimatedby comparing all participants and nonpar-ticipants who met the inclusion criteriafor the study. The math for the impactcalculation is summarized as follows:Impact = {(Proportion of observa-tions at Low Risk)(Mean $ for LowRisk post-HRA - Mean $ for LowRisk pre-HRA) + (Proportion observations at High Risk who com-pleted an Accent program)(Mean for High Risk/Full Accent post-HRA - Mean $ for High Risk/FullAccent pre-HRA) + (Proportion observations at High Risk withsome Accent participation) (Mean for High Risk/Some Accent post-HRA - Mean $ for High Risk/Some Accent pre-HRA) + (Propor-tion of observations at High Risk September/October 999, Vol, 14, No. 135
Search
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks