Objective: To examine correlations of commercial health planperformance on Health Plan Employer Data and Information Set(HEDIS®) effectiveness-of-care measures with utilization rates, as aproxy for cost.
Study Design: Cross-sectional study of 254 commercial healthplans.
Methods: This report used data reported by commercial managedcare plans in the 2003 HEDIS dataset. Utilization measuresincluded access to care (the proportion of adults with at least 1 primarycare or preventive visit), outpatient use (the number of outpatientvisits per 1000 members per year), inpatient discharges (thenumber of inpatient discharges for medical conditions per 1000members per year), and inpatient days (inpatient hospital days formedical conditions per 1000 members per year). A compositequality score was calculated from HEDIS indicators. Estimates ofhealth plan membership demographics were identified fromConsumer Assessment of Health Plans (CAHPS) survey data. Of316 reporting plans, 254 reported sufficient data to be included inthis analysis. Bivariate correlations and multivariate regressions(controlling for health plan and membership characteristics) wereconducted.
Results: Quality was positively correlated with access to outpatientcare (= 0.46, < .001), negatively associated with inpatientdays (= -0.30, < .001), and not associated with total outpatientvisits (= 0.04, not significant). Regression results controlling forselected plan and member characteristics demonstrated similarfindings.
Conclusions: Although the mechanism of this cross-sectionalassociation is unclear, these data provide important startingpoints for further research on the interrelationships of quality andresource use.
(Am J Manag Care. 2005;11:521-527)
Recent trends in healthcare costs1 have raisedconcerns among purchasers and consumers ofcare about whether they are getting reasonablevalue for their increasing healthcare outlays. Studiesfocusing on the Medicare population demonstrate thatthere are strong regional differences in healthcarespending and that these differences in spending do notappear to be reflected in better access to or quality ofcare, nor with improved health outcomes and satisfaction.2-4 Indeed, recent research suggests that higherspending may correlate with poorer quality of care.5Isolated studies in other settings similarly show thatmore care and more costly care do not always result inhigher quality. For example, among Medicaid community-based providers, no relationship between cost andquality was found.6 In another study involving 18 medicalgroups, there was no consistent relationshipbetween performance on 21 ambulatory-care, process-orientedquality indicators and case-mix adjusted costsof care.7 We are aware of only 1 prior published studyaddressing the issue of costs and quality among commercialhealth plans: a report using early data fromHealth Plan Employer Data and Information Set(HEDIS®) found that health plans with higher qualityhad higher medical expense ratios, indicating that theyspent a higher proportion of their premium income ondirect services as opposed to administrative expenses.8
More detailed information about the relationshipbetween quality and utilization is critical, especially forthe employer-sponsored health insurance market whereemployees are bearing a larger share of healthcareexpenses.1 To examine these relationships for commerciallyenrolled populations at the health plan level, weused data from commercial HMO plans reporting HEDISmeasures to the National Committee for QualityAssurance (NCQA) to explore the relationship betweenquality and utilization (as a proxy for costs). First, weexamined the correlations of performance on effectiveness-of-care measures with measures of outpatient andinpatient utilization for adults aged 20 to 64 years. Wethen used regression analyses to determine whether therelationship between quality and utilization remainedafter controlling for a number of patient and plancovariates.
Data Sources and Study Group
This reports uses data from the healthcare utilizationand quality measures reported by commercial managedcare plans in the 2003 HEDIS dataset.9 Plans report ona standardized set of performance measures usingdetailed specifications and after undergoing an independentaudit.10 This study includes data from allreporting plans (including some plans not accredited byNCQA that submit data but do not allow public reportingof individual plan data). More than 66% of commercialhealth plans report to NCQA, representing 85% ofthe commercially enrolled managed care organization(MCO) population. Of the 316 commercial plans thatreported in 2003 (for care received during calendar year2002), a total of 254 representing 83% of commercialMCO enrollees were eligible for these analyses. Sixtyoneplans were excluded because of missing values forthe dependent variables (n = 28), member characteristicsderived from Consumer Assessment of Health Plans(CAHPS®) 3.0H (n = 10), or values for 4 or more qualitymeasures (n = 23); 1 other plan was excludedbecause of extreme outlier observations. Submission ofHEDIS data (including CAHPS surveys) is voluntary,and some plans choose to submit data for only a portionof HEDIS measures. The excluded plans were less likelyto be accredited by NCQA (30.6% of excluded planscompared with 70.9% of plans in this analysis) and hadfewer members. Plans from the South Central andPacific regions were more likely to be excluded andplans from Mid-Atlantic were more likely to be in theanalysis group. The memberships of plans also differed:excluded plans had a higher proportion of memberswho belonged to a minority race (25.3% in the excludedgroup vs 18.6% in the analysis group) or who had pooror fair health status (9.9% in the excluded group vs 8.8%in the analysis group).
The utilization measures were restricted to adultplan members aged 20 to 64 years because the selectedquality measures reported by commercial plans aremost robust for the adult population in that age range.Commercial plans generally have small and highly variablenumbers of enrollees aged 65 years and over. Thisgroup was excluded to avoid one source of potentialbias.
Dependent variables addressed utilizationof care. HEDIS utilization-of-care measures includeoutpatient use, defined as the number of outpatient visitsper 1000 members per year; emergency department(ED) visits, defined as the number of ED visits per 1000members per year; inpatient discharges, defined as thenumber of inpatient discharges for medical conditionsper 1000 members per year; and inpatient days, definedthe number of inpatient hospital days for medical conditionsper 1000 members per year. To approximate asclosely as possible utilization that represents discretionaryvariations in care, we focused on inpatient medicalcare (excluding surgery, maternity, and mentalhealth/substance abuse care) because this categoryshowed the greatest variation across entities and marketsin previous studies.3,4 Likewise, the outpatient-visitmeasure excluded visits for ambulatory surgery/proceduresand observation-room stays that resulted in discharge,as well as visits for mental health or substanceabuse care.
Health plan quality indicators were limitedto those available in the HEDIS 2003 effectiveness-of-caredomain. Detailed specifications for these measurescan be found in HEDIS volume 2: 9 Preliminary bivariate correlations (data notshown) indicated a high degree of inter-correlationsamong a subset of the measures (eg, breast and cervicalcancer screening) and within measures with multipleindicators for a single illness (eg, diabetes care, cholesterolmanagement, antidepressant medication management,and follow-up for mental illness). For themeasures that were strongly correlated (ie, breast cancerscreening and cervical cancer screening, = 0.62),only 1 measure was chosen to be included in the analyses.For the measures with multiple indicators that alsowere moderately to highly correlated with each another,such as 6 comprehensive diabetes care measures,only 1 indicator was selected for inclusion. Where available,we selected outcome measures (such as a glycosylatedhemoglobin level < 9.5% for persons with diabetes)because these capture the full range of measured experience.A total of 10 quality measures were included. Anexploratory factor analysis suggested that the itemsloaded on a single factor. Thus, we developed a singlequality composite by taking the mean of scores acrossthe 10 quality measures. To retain as many observationsas possible, missing values for the quality measureswere substituted with the regional mean. Valueswere substituted for 2% to 18% of the plans, dependingon the quality measure. Missing values occurred whenplans did not have enough eligible members to reportthe measure (n = 42 plans for beta-blocker treatment)or have elected not to collect the information (n = 31 forblood pressure control).
The quality composite had a mean of 67.4% (SD =5.0%), range 51.2% to 79.0%, and the internal-consistencyreliability of the scale was good (Cronbach's alpha= 0.80). Descriptive statistics for the 10 quality measuresand the quality composite are shown in Table 1.
A limited number of variables are availableto describe the plans and their member populations.The percentage of female members and the agedistribution of plan members were derived from healthplan enrollment data. We also used data from CAHPS3.0H as an estimate of member-level characteristics ofrace, education, and health status. The CAHPS survey isadministered via mail and/or telephone surveys to arandom sample of health plan members following astandardized protocol, and the average response rate is42%.11 Using CAHPS data, we identified the proportionof health plan respondents who reported their race asminority (including black, Asian or Pacific Islander, andother) compared with those reporting their race aswhite; the proportion with any college education(including those with some college or a 2-year degree,those with a 4-year college degree, and those with amore-than-4-year college degree) compared with thosewho reported a high school education or less; and theproportion with fair or poor health status (based on asingle item self-rating health as excellent, very good,good, fair, or poor).
Other health plan characteristics are collected aspart of the NCQA data submission process. Plans indicatewhether or not they will allow public reporting oftheir HEDIS performance data, whether their tax statusis for profit or not for profit, and whether they offer aHMO product only versus offering a point-of-service(POS) product only, or both HMO and POS products.The geographic location of the plan's primary businesswas categorized by census regions. Because a smallnumber of plans included in this report (7.1%) havemembership in multiple regions, we randomly assignedsuch plans to a single region.
All analyses were conducted using SAS 8.0 software(SAS Institute Inc, Cary, NC). We examined the bivariaterelationship between the utilization measures andthe quality measures using unadjusted Spearman correlations.Multivariate analyses were conducted to estimatethe relationship of quality to utilization whilecontrolling for plan and member characteristics.Covariates include plan region and profit status; we didnot include plan accreditation or public reporting statusbecause these measures are known to be related to quality,12 and we wanted to control for measures that wouldbias comparisons of utilization and quality but not maskthem. Patient covariates included age, sex, minority status,and health status. Due to the skewness of the utilizationmeasures that are typical of utilizationdistributions, we used a logarithmic transformationof these measures and conducted a linearregression on the transformed dependent variables.13 To evaluate the overall quality of ourmodels, we calculated McFadden's 2 statistic,also known as the likelihood-ratio index.14 Itcompares the likelihood for the intercept-onlymodel to the likelihood for the model with all thecovariates.
Table 2 presents descriptive information onthe plans and their membership. Most planswere for profit (70.1%) and had POS or combinedHMO/POS products (69.7%), nearly allallowed public reporting of data (91.3%), andthe largest numbers of plans came from theeastern regions.
The unadjusted Spearman correlationsbetween individual quality measures and utilizationmeasures are shown in Table 3. Severalquality measures were positively correlatedwith outpatient visits, including advising smokersto quit (= .22, = .0004), asthma medicationmanagement (= 0.19, = .0032), breastcancer screening (= 0.20, = .0019), and thequality composite (= 0.19, = .0032). Severalquality measures were negatively associatedwith ED visits, including asthma medicationmanagement (= -0.24, = .0001), cholesterolcontrol (= -0.20, = .0014), acute-phase antidepressanttreatment (= -0.22, = .0004), flushots (= -0.23, = .0003), and the quality composite(= -0.18, = .0034). All quality measures exceptblood pressure control were significantly and negativelycorrelated with both mean inpatient dischargesand mean inpatient days per 1000 enrollees. Forexample, mean inpatient days per year had a correlationof -0.35 (< .0001) with the quality compositeand ranged from -0.16 (= .014) for mental health follow-up after hospitalization to -0.42 (< .0001) forantidepressant medication management.
Multivariate regression results controlling for theavailable plan-level and member-level variables areshown in Table 4. Plans with higher quality compositeshad fewer hospital discharges (beta = -0.6900, = .04)and fewer hospital days (beta = -0.7781, = .0207). Theassociations of quality with outpatient visits (beta =0.5702, = .13) and ED use (beta = -0.4735, = .37)were similar in direction but not statistically significant.The parameter estimates indicate that an improvementof 5 percentage points in the quality composite (about 1standard deviation) was associated with an approximately5% reduction in the average number of hospitaldays (4 days of care per 1000 enrollees). Although membercharacteristics were significantly associated withseveral of the utilization measures, there was not a consistenteffect across all of the measures. Regional effectswere seen for all utilization measures, with region havingthe most pronounced effect on the hospitalizationrates. For example, the Mountain, Northeast, andPacific regions were associated with less hospital use.
Commercial health plans that achieved higher performanceon measures of quality tended to have lowerhospitalization rates. Although plan quality scores werepositively correlated with outpatient use and negativelycorrelated with ED visits in bivariate analyses, theseassociations were not significant after adjusting for planand member characteristics in multivariate analyses.The finding of no association between quality and outpatientutilization may reflect our inability to separateprimary care visits from subspecialty care. Subspecialtyvisits were negatively correlated with quality in theMedicare program.5 These results suggest that someplans appear to be able to achieve similar levels of qualitywith lower utilization.
The inverse relationship between health plan qualityand hospital utilization is striking and supports theregion-level findings of Fisher et al in the Medicare population.3,4 The mechanism for this relationship isunclear. Strong correlations between quality measuresand hospital days were found for several measures thatcould be expected to directly reduce hospital days,such as medication management for asthma (= -0.30)and flu shots (= -0.30); however, a similarly strongrelationship was found for breast cancer screening(= -0.30). The strong relationship between antidepressantmedication management and hospital days(= -0.42) is interesting because the inpatient measurefocused on medical discharges and did not includestays for mental health and substance abuse care.However, a growing literature documents the negativeimpact of depression on the outcomes of chronic medicalconditions such as diabetes and heart disease,although studies to date have not demonstratedwhether improving depression treatment leads toimproved outcomes.15-18
Importantly, although this cross-sectional studydemonstrated correlations between higher qualitycare and reduced hospital days, it did not identify acausal relationship between these measures. Thereare several competing explanations for these findingsbased on plan management and patient selection. Itmay be that health plans achieving high levels of performanceon HEDIS quality measures are betterorganized, both in their management of hospitalizationdays and in the processes that contribute to highscores on HEDIS measures of clinical effectiveness.Organized systems that allow plans to identifypatients in need of preventive services, to assistpatients withchronic conditionsin obtainingappropriatefollow-up, andto plan andcoordinate carefor patients atrisk of hospitalizationcouldenable plans todeliver betterquality care andavoid costly inpatientdays.In addition,these findingsmay reflect theselective enrollmentof healthierpatients athealth plansthat achievehigher qualityscores. BecauseHEDIS data areaggregated atthe health planlevel, we reliedon CAHPS dataas a proxyfor the socioeconomicandhealth status ofthe health planmembership.This approachoffered onlylimited ability to control for differences in case mix.Further research is needed to identify potential mechanismsthat might explain the relationship between qualityand utilization seen here.
Regional differences in utilization reported hereare consistent with prior research that has demonstratedgreat variation in use of hospital care.Residence in areas with greater per capita numbersof hospital beds is associated with higher hospitalizationrates without positive benefits for mortalityrates.19 Although our analysis controlled for healthplan region in isolating the effect of quality on utilization,further work to disentangle a potential "qualityeffect" from the "supply effect" on utilization wouldhelp inform the policy debate on strategies to reduceunnecessary utilization and costs without sacrificingquality.
These results have several limitations. The cross-sectionalanalyses cannot sort out the temporal relationshipbetween quality and utilization. Given thelimited data available on utilizations, we cannot concludethat lower utilization rates lead to lower costs ofcare because intensity of care may differ. Although19.6% of plans reporting HEDIS data were ineligiblefor this analysis due to insufficient data, these exclusionshad small impact on the overall proportion ofthe managed care population represented and likelylimited the range of quality scores, making it more difficultto find a significant relationship between qualityand utilization.
Crossing the Quality Chasm,
These findings, despite their limitations, demandadditional attention to determine whether someplans are able to achieve high levels of quality forsimilar populations with less resource utilization, andif so, what distinguishes these plans from others, sothat we can work to encourage efficient, high-qualitypractices. If supported through further research andreplication, there is the potential to obtain the benefitsanticipated by the Institute of Medicine inwhich envisionedrestructuring the healthcare system to address bothquality and costs simultaneously.20 Certainly, theHEDIS performance data used in this report demonstrateample room for improving quality, with healthplans achieving high-quality performance only abouttwo thirds of the time. Although achieving higherquality may not be free,21 these data give hope thatimprovements in effectiveness of care may reduceboth the human costs of poor care and their financialimplications as well.
From the National Committee for Quality Assurance, Washington, DC.
Address correspondence to: Sarah Hudson Scholle, MPH, DrPH, National Committeefor Quality Assurance, 2000 L Street, NW Suite 500, Washington DC 20036. E-mail:email@example.com.
on Health Care Costs and Solutions.
1. Mercure S. Solutions for controlling cost: corporate employers report. 2003;2(3):1-2. Issue Brief.
Health Aff (Millwood).
2. Wennberg JE, Fisher ES, Skinner JS. Geography and the debate over Medicarereform. July-December 2002;Suppl Web Exclusives:W96-W114.
Ann Intern Med.
3. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. Theimplications of regional variations in Medicare spending. Part 1: the content, quality,and accessibility of care. 2003;138:273-287.
Ann Intern Med.
4. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. Theimplications of regional variations in Medicare spending. Part 2: health outcomesand satisfaction with care. 2003;138:288-298.
Health Aff (Millwood).
5. Baicker K, Chandra A. Medicare spending, the physician workforce and beneficiaries'quality of care. 2004;Suppl Web Exclusive:W184-W197.
6. Starfield B, Powe NR, Weiner JR, et al. Costs vs quality in different types ofprimary care settings. 1994;272:1903-1908.
Am J Manag
7. Solberg LI, Lyles CA, Shore AD, Lemke KW, Weiner JP. Is quality free? Therelationship between cost and quality across 18 provider groups. 2002;8:413-422.
J Health Care Finance.
8. Born P, Geckler C. HMO quality and financial performance: is there a connection?1998;24:65-77.
Employer Data and Information Set;
9. National Committee for Quality Assurance. Washington, DC: National Committee for Quality Assurance; 2003. vol 2.
Compliance Audit: Standards,
Policies, and Procedures.
Health Plan Employer Data and Information Set;
10. National Committee for Quality Assurance. Washington, DC: National Committee for QualityAssurance; 2003. vol 5.
Specifications for Survey
Health Plan Employer Data and Information Set;
11. National Committee for Quality Assurance. Washington, DC: National Committee for Quality Assurance; 2003.vol 3.
Manag Care Interface.
12. Bost JE. Managed care organizations publicly reporting three years of HEDISmeasures. 2001;14:50-54.
Categorical Data Analysis.
13. Agresti A. New York, NY: John Wiley & Sons; 1990.
Frontiers in Econometrics.
14. McFadden D. Conditional logit analysis of qualitative choice behavior. In:Zarembka P, ed. New York, NY: Academic Press;1974:105-142.
15. Katon WJ. Clinical and health services relationships between major depression,depressive symptoms, and general medical illness. 2003;54:216-226.
16. Blumenthal JA, Lett HS, Babyak MA, et al. Depression as a risk factor formortality after coronary artery bypass surgery. 2003;263:604-609.
J Clin Psychiatry.
17. Roose SP. Depression: links with ischemic heart disease and erectile dysfunction.2003;64(suppl):26-30.
18. Musselman DL, Betan E, Larsen H, Phillips LS. Relationship to diabetes types1 and 2: epidemiology, biology, and treatment. 2003;54:317-329.
Health Serv Res.
19. Fisher ES, Wennberg JE, Stukel TA, et al. Associations among hospital capacity,utilization, and mortality of US Medicare beneficiaries, controlling for sociodemographicfactors. 2000;34:1351-1362.
Crossing the Quality Chasm: A New Health System for
the 21st Century.
20. Institute of Medicine. Washington, DC: National Academy Press; 2001.
Quality Is Free.
21. Crosby PB. New York, NY: McGraw-Hill Book Company; 1979.