Volume 4 • Number 6 • 2001VALUE IN HEALTH

© ISPOR 1098-3015/01/$15.00/401 401–420

401

ISPOR Fourth Annual European CongressContributed Presentation Abstracts

Contributed Podium Presentations

SESSION I

HEALTH POLICY I

HP1

VALIDITY OF THE MINIMUM-DATA-SET-BASED QUALITY INDICATOR FOR DETERMINING THE PREVALENCE OF ANTIPSYCHOTIC MEDICATION USAGE AMONG LOW RISK NURSING HOME RESIDENTS: A STATEWIDE ASSESSMENT

Monroe DM

1

, Cortes LL

2

, Fouladi R

3

1

The University of Texas at Austin, Austin, TX, USA;

2

Texas Department of Human Services, Austin, TX, USA;

3

The University of Texas at MD Anderson Cancer Center, Houston, TX, USA

OBJECTIVE:

The Minimum Data Set (MDS) is an inter-national facility-reported database used to standardizethe assessment process and improve the quality of care innursing homes. Specific MDS assessments are used asquality indicators to determine the prevalence of antipsy-chotic prescribing as well as potentially inappropriate anti-psychotic usage. MDS classifies low-risk residents as receiv-ing an antipsychotic medication but having no documentedpsychotic or related condition and no cognitive impairmentwith behavioral symptoms. Without an audit of residentrecords, it is impossible to determine if the MDS accuratelyreports the prevalence of antipsychotic medications andwhether all low-risk residents represent inappropriateprescribing. The purpose of this study was to deter-mine MDS validity for identifying low-risk residents inTexas.

METHODS:

Facility-based MDS assessments were com-pared to independent on-site assessments. A team ofpharmacists reviewed the chart records of low risk resi-dents and determined whether a resident received an an-tipsychotic medication and the presence or absence of avalid indication. These findings were compared to thestatewide MDS quality indicator prevalence rate.

RESULTS:

Based on MDS assessments, 10,163 residentsfrom 1021 nursing homes were identified as low risk.The chart records revealed that 12.4% of these residentswere not receiving antipsychotic medication. Of the 8904residents who were receiving an antipsychotic, 48.2%had a valid diagnosis or indication and were mistakenlyidentified as low-risk by the MDS database. Of the re-maining 4610 residents, 84.4% were missing a diagnosisand 15.6% had an inappropriate indication.

CONCLUSION:

For this population of residents, a lackof accurate essential diagnostic information in MDS as-sessments and a high prevalence of inappropriate antipsy-chotic prescribing was found to be a widespread prob-lem. These findings indicate that nursing facilities wereinaccurate in their responses to MDS assessment itemsand therefore, the MDS had poor validity as a quality in-dicator for antipsychotic use.

HP2

TRANSLATING EVIDENCE TO PRACTICE IN BREAST CANCER

Bloom BS

1

, Jayadevappa R

1

, de Pouvourville N

2

, Chartre S

1

1

University of Pennsylvania, Philadelphia, PA, USA;

2

University of Paris Medical School, Paris, France

OBJECTIVE:

Practice guidelines are being implementedfor many diseases to improve outcomes, increase use ofevidence-based medicine, and to reduce medical errors.Rarely are these efforts systematically evaluated in clini-cal care. We chose breast cancer in women as one diseasethat has excellent scientific evidence on treatment efficacybut almost no data on actual use in physician practice.

METHODS:

We studied use of breast conserving sur-gery, breast reconstruction, adjuvant chemotherapy, radi-ation, taxoids, tamoxifen, and aromatase inhibitors in el-igible women. The population was diagnosed between1995 and 2000 in six surgical oncology practices in theeight counties around Philadelphia with nearly seven mil-lion people. Four thousand three hundred ninety fivewomen were enrolled and followed retrospectively up tofive years after diagnosis. We abstracted each patient’scomplete clinical record.

RESULTS:

Lumpectomy was provided to 51.0% ofwomen, among whom 2.3% also had reconstruction.Forty nine percent of women (49.0%) had mastectomywith 18.0% having breast reconstruction, 56.8% re-ceived radiation therapy and 32.3% received adjuvantchemotherapy. Among eligible women, 0.6% had aro-matase inhibitors and 12.3% were prescribed taxoids.Among estrogen and/or progesterone-receptor-positivewomen, 72.4% got tamoxifen, while 35.1% with nega-tive estrogen and progesterone receptors got tamoxifen.One woman refused all treatment, one refused any breastsurgery, five refused adjuvant chemotherapy, two refusedradiation. Eight percent of eligible women refused tamox-ifen, mainly because of elevated risk of adverse side effects,and one had an autologous bone marrow transplant. Lo-gistic regression found that breast reconstruction was in-

402

Abstracts

versely related to patient age (p

�

.0029), tamoxifen var-ied directly with patient age (p

�

.0002) but inverselywith physician age (p

�

.02), and adjuvant chemotherapyand taxoids varied inversely with patient age (p

�

.0001).

CONCLUSIONS:

Two decades of multiple randomizedcontrol trials and meta-analyses have led to only modestsuccess in the use of highly beneficial treatments forbreast cancer in these specific oncology practices, al-though non-beneficial treatments are almost never used.

HP3

REGIONAL VARIATION IN PRESCRIPTION USE IN THE UNITED STATES

Motheral BR, Cox ER, Mager D, Henderson R

Express-Scripts, Maryland Heights, MO, USA

OBJECTIVE:

To evaluate the factors influencing the utiliza-tion of prescribed medications across the United States.

METHODS:

Prescription claims data for a random sam-ple of commercially insured Express Scripts members, 18to 64 years of age and continuously eligible for at leasttwo months in 1999 (N

�

1,467,699), were geo-coded atthe block group level. Key independent variables includedage, gender, family size, median income, education, ur-ban/rural location, client type, and pharmacy benefit de-sign. Outcome measures included any use of prescrip-tions. Hierarchical modeling (HLM) was used to controlfor clustering of patients within health plans.

RESULTS:

Results from HLM indicate that after control-ling for demographics, client type, and pharmacy benefitdesign, variation in the probability of using a prescribedmedication was found across regions of the country.Members in the South have a greater likelihood of useand members in the Northeast and West having a lowerprobability of use than members in the Midwest. Agreater probability of use was associated with older age,female gender, higher income, family size

�

1, and enroll-ment in a managed care plan. Factors associated withlower probability of prescription use included presence ofa deductible, living in an urban area and having a manda-tory generic policy.

CONCLUSIONS:

While patient age and gender remainkey factors driving prescription utilization, socioeconom-ics, pharmacy benefit design, and where a person lives arealso predictors of prescription use. Future research isneeded to better understand the causative roles of diseaseburden, physician practice style, and consumer preferencesin prescription drug use variation across the United States.

CANCER I

CN1

BRIEF ASSESSMENT OF PRIORITY SYMPTOMS IN HORMONE REFRACTORY PROSTATE CANCER (HRPCA)

Yount SE

1

, Banik D

1

, Singh A

2

, Ashraf T

2

, Padley RJ

2

, Dalal M

2

, Cella D

1

1

Evanston Northwestern Healthcare, Evanston, IL, USA;

2

Abbott Laboratories, Abbott Park, IL, USA

OBJECTIVE:

The objective of this study was to derive abrief, clinically relevant symptom index for men withHRPCa.

METHODS:

An international sample of 44 medical on-cologists, radiation oncologists, and urologists were pre-sented with 29 disease-related symptoms derived fromthe Functional Assessment of Cancer Therapy-Prostate(FACT-P) quality of life (QOL) instrument and asked toselect the five most important symptoms in assessing thetreatment response for men with HRPCa. Symptoms en-dorsed at a frequency greater than chance were retainedfor the index. Retrospective validation of the index wasaccomplished using data from a multi-national, random-ized, double-blind, placebo-controlled clinical trial ofatrasentan, a selective, oral ET-A receptor antagonist.The FACT-P and European Organization for Researchand Treatment of Cancer (EORTC) QLQ-C30 adminis-tered at baseline, six weeks and final visit were used toevaluate QOL responses to treatment of 288 men withHRPCa.

RESULTS:

Experts initially selected eight priority symp-toms. Item-Response Theory analysis identified one mis-fitting item (worry), lending support for a seven-item in-dex (FPSI-7). Internal consistency was high for the seven-item index (0.75). Concurrent validity was establishedthrough correlations (

P

�

0.05) between the FPSI-7 andthe FACT-General (FACT-G;

r

�

0.56); FACT-P (

r

�

0.73); FACT Prostate Cancer Subscale (PCS;

r

�

0.88);Trial Outcome Index (TOI;

r

�

0.83); and QLQ-C30global score (

r

�

.56), symptom domain scores (

r

��

0.33 to

�

0.76), and generic domain scores (

r

�

0.36 to0.60). The FPSI-7 successfully differentiated patients byfunctional status, measured by the Eastern CooperativeOncology Group (ECOG) performance status rating(PSR,

p

�

.001), and clinical outcome, measured by dis-ease progression status (

p

�

.001). PSR and disease pro-gression are important outcomes for patients and clinicaltrials. The FPSI-7’s responsiveness to PSR change com-pared favorably to the FACT-G, PCS, FACT-P and TOI.

CONCLUSION:

The FPSI-7 is a valid and responsivesymptom index that can provide efficient, rapid symptomassessment to evaluate treatment outcome in clinical trials.

CN2

RESOURCE CONSUMPTION RELATED TO MUCOSITIS IN LYMPHOMA PATIENTS RECEIVING HIGH-DOSE CHEMOTHERAPY WITH AUTOLOGOUS PBPC TRANSPLANTATION

Chan WW, Lawless G

Amgen Inc, Thousand Oaks, CA, USA

OBJECTIVES:

To evaluate resource consumption relatedto mucositis in lymphoma patients receiving high-dosechemotherapy (HDC) with autologous PBPC transplan-tation.

Abstracts

403

METHODS:

Data on mucositis and clinical parametersof hospitalization, IV antibiotic use, fever, analgesic use,and parenteral nutrition were collected as part of a multi-center clinical trial conducted in Europe (n

�

234). Pa-tients were quantified as having mucositis if ulcerativelesions were present after receiving BEAM (BCNU, Eto-poside, Ara-C, Melphalan) HDC. Days of usage for clini-cal parameters were quantified as resource consumptionand compared in patients with and without mucositis.

RESULTS:

Mucositis showed an association with resourceconsumption. Days of analgesia had the highest correla-tion followed by antibiotic use, fever, hospitalization, andparenteral nutrition (r

�

0.4128*, 0.3003*, 0.2634*,0.2558*, 0.0973, respectively, *p

�

.0001). Patients withmucositis had two additional days of hospitalization andthree additional days of IV antibiotics. Mean

�

s.e. for re-source consumption is listed for patients (with mucositis,without mucositis, p-value), as follows: days of hospital-ization (26.8

�

0.4, 24.4

�

0.5, p

�

.001); days of IV an-tibiotics (12.6

�

0.7, 9.5

�

0.7, p

�

.002); days of anal-gesic use (9.2

�

0.7, 4.3

�

0.5, p

�

.001); days ofparenteral nutrition (1.8

�

0.3, 1.6

�

0.2, p

�

.668);days of fever (2.9

�

0.3, 2.0

�

0.3, p

�

.035).

CONCLUSIONS:

Patients with mucositis had signifi-cantly higher resource consumption. Mucositis impactson length of hospitalization, days of fever, IV antibiotic,and analgesic use.

CN3

AN ECONOMIC MODEL FOR THE PHARMACOLOGICAL MANAGEMENT OF ADVANCED NON-SMALL CELL LUNG CANCER IN THE US

Sorensen SV

1

, Chambers M

2

, Heyes A

3

1

MEDTAP International, Bethesda, MD, USA;

2

MEDTAP International UK, London, UK;

3

AstraZeneca UK Limited, Macclesfield, UK

OBJECTIVES:

The majority of patients with non-smallcell lung cancer (NSCLC) present with advanced disease,for which pharmacological therapy is the primary treat-ment option. Economic models are useful tools to helpdecision-makers determine the cost and health impact ofexisting and new therapies in their practice settings.

METHODS:

A decision-analytic model was developed torepresent the progression of advanced (stage IIIB/IV)NSCLC from diagnosis and first-line therapy to subse-quent lines of therapy, best supportive care and death.The model is a semi-Markov process with three-week cy-cles and a two-year time horizon. By selecting differentsets of model parameters, it may be used to estimate cost-of-illness (by pooling cost and health outcomes across arange of therapies), cost-effectiveness (comparison be-tween new or existing therapeutic options) and impact onhealth-care budgets. Data sources include published clini-cal trials, treatment patterns based on published studies,and a panel of US clinicians and US unit costs. In thiscase, the NSCLC model was used to examine the costs

and outcomes of a distribution of mono and combinationagents most commonly administered as first and second-line therapy in the US.

RESULTS:

Over one and two years respectively the aver-age cost of therapy was $42,319 and $48,151 per pa-tient, with an estimated survival of 40.2% and 7.4%.The estimated cost for a health-care purchaser with aprevalence of 100 patients with advanced NSCLC was$4.90 million per year. If all patients used carboplatinand paclitaxel as first-line therapy, this increased to$5.14 million. If a new therapy increased the median timeto progression for patients on first-line therapy by twomonths, the cost increased to $5.0 million per year.

CONCLUSION:

This model provides a flexible analyti-cal platform for a range of economic evaluations of alter-native therapeutic strategies in advanced NSCLC. This isvaluable for strategic planning in Phase II/III drug devel-opment.

CARDIOVASCULAR DISEASE I

CV1

ESTIMATING THE IMPACT OF ANTICOAGULATION QUALITYON EVENT RATES

Matchar DB, Samsa GP

Duke University, Durham, NC, USA

OBJECTIVES:

Warfarin anticoagulation has been shownto effectively decrease the rate of thromboembolism (TE),with an associated increased rate of bleeding. Maximiz-ing the benefit and minimizing the risk depends on main-taining patients in a narrow therapeutic range for pro-thrombin time (measured as INR). An increased percentageof time spent in target range (TTR) is predictive of betterclinical outcomes. However, strategies to increase TTR canbe costly. To allow local clinical policy makers to deter-mine whether these strategies can be justified, we developeda general approach to estimating bleeding and TE rates as-sociated with a given population distribution of INR.

METHODS:

We used data from a large cohort (Cannegi-eter, 1995) to derive a logistic equation for baseline abso-lute risk of bleeding as a function of INR. Based on a sep-arate large cohort of patients (Fihn, 1993) with varioustimes since initiation of therapy, we derived a risk ratiothat modifies the above baseline relationship that ac-counts for time since initiation of therapy. A similar strat-egy was used to estimate TE, using INR and indication(mechanical valve versus atrial fibrillation) as predictors.

RESULTS:

The model for bleeding was: logit(p)

�

�

8.84

�

0.83 INR

�

0.64 EARLY, where EARLY

�

1 if the pa-tient is in the first 3 months of anticoagulation, and 0otherwise. The final model for TE was: logit(p)

�

�

0.73

�

1.17 INR

�

0.64(AF), where AF

�

1 if the indication isatrial fibrillation and 0 if the indication is mechanicalheart valve. The integral of these logistic equations,weighted by a given population distribution of INR, pro-vides population estimates of annual bleed and TE rates.

404

Abstracts

CONCLUSION: Existing epidemiological data can pro-vide tailored estimates of concrete benefits resulting fromimproving the quality of anticoagulation.

CV2

A SIX YEAR FOLLOW-UP STUDY OF THE RELATIONSHIP BETWEEN MORTALITY, HOSPITALISATION AND ADHERENCE TO STATIN TREATMENT AFTER FIRST MYOCARDIAL INFARCTIONWei L, Wang J, Davey P, MacDonald TMEMO, University of Dundee, Dundee, UK

OBJECTIVE: To measure adherence to statins by pa-tients treated for secondary prevention after myocardialinfarction and to estimate the effect of adherence on out-come.METHODS: We used a cohort design in the populationof Tayside, Scotland. Patients who experienced their firstMI between January 1990 and November 1995 wereidentified from hospital discharge data. We used two out-comes: mortality from any cause; hospitalization for re-current MI. Adherence to statins was calculated as thenumber of days for which statins were supplied dividedby the total number of days in the study for each patient.Results were adjusted for age, sex, deprivation (as mea-sured by the Carstairs code), serum cholesterol level, dia-betes mellitus, cardiovascular drugs and other hospital-ization using a Cox regression model.RESULTS: Of 5590 patients enrolled in the cohort 1299(23.2%) died during the follow-up period and 717(12.8%) experienced at least one further MI. Only 7.7%of patients used statins, and in comparison with non-us-ers, these patients had more cardiovascular risk factors.Compared to those not using statins, the adjusted relativerisk of mortality (95% CI) by quintiles of adherence was0.65 (0.24–1.80) for the worst adherence quintile, 0.46(0.06–3.43) for the second, 1.03 (0.37–2.88) for thethird, 0.19 (0.03–1.37) for the fourth, and 0.20 (0.09–0.47) for the best adherence quintile. The adjusted rela-tive risks of readmission by quintiles of adherence were0.65 (0.24–1.79) for the worst adherence quintile, 0.47(0.06–3.51) for the second, 1.05 (0.37–2.94) for thethird, 0.20 (0.03–1.41) for the fourth, and 0.21 (0.09–0.48) for the best adherence quintile.CONCLUSIONS: Statins were used infrequently and usewas a marker of cardiovascular risk. Despite such confound-ing by indication, good adherence to treatment was associ-ated with lower risks of further MI and lower mortality.

CV3

VASOPEPTIDASE INHIBITOR REDUCESIN-HOSPITAL COSTS FOR CONGESTIVEHEART FAILURE PATIENTS: RESULTSFROM THE IMPRESS TRIALEisenstein EL1, Nelson CL1, Simon TA2, Smitten AL3, Lapuerta P2, Mark DB1

1Duke Clinical Research Institute, Durham, NC, USA; 2Bristol-Myers Squibb, Princeton, NJ, USA; 3Yale University School of Medicine, New Haven, CT, USA

OBJECTIVE: The IMPRESS clinical trial randomized pa-tients with congestive heart failure to a daily regimen ofeither omapatrilat or lisinopril. At 24 weeks, patientsrandomized to omapatrilat had a significant reduction inthe primary endpoint of death, hospitalization, or discon-tinuation of study drug for worsening heart failure (p �.035) and fewer cardiac events (p � .04). This studysought to determine the economic consequences of theomapatrilat patient’s lower event rates.METHODS: Economic outcomes were assessed in termsof hospitalization events and their medical costs. Hospi-tal event information was obtained via serious adverseevent forms, and hospital costs were evaluated by assign-ing each hospitalization a DRG-based average cost forphysician and hospital services. Emergency room visitsfor worsening heart failure were assigned costs equivalentto those at Duke University Medical Center. All costswere expressed in 1998 US dollars. Drug costs were notassessed.RESULTS: Patients in the omapatrilat (n � 289) and lisi-nopril (n � 284) arms were evenly matched with regardto baseline characteristics: age (both 64 years); NYHAclass III or IV heart failure (36% versus 38%); ejectionfraction (both 28%). There was no difference between studyarms in all-cause mortality. However, there was a trend to-ward a greater number of all-cause hospitalizations inthe lisinopril versus omapatrilat patients (0.275 versus0.215, p � .07). Differences in cardiac hospitalizations be-tween lisinopril and omapatrilat were significant (0.208versus 0.145, p � .03). There was a trend toward re-duced medical costs at 24 weeks follow-up in omapatri-lat-treated patients (US$1930 versus US$2002, p � .09).Considering only cardiac medical costs, this trend towardreduced medical costs became significant (US$1240 ver-sus US$1442, p � .03).CONCLUSIONS: In the first study to compare economicoutcomes in congestive heart failure patients treated withomapatrilat and lisinopril, we found fewer hospitaliza-tions and lower medical costs for omapatrilat patients at24 weeks.

COST ESTIMATION

CE1

IMPACT OF CENSORED COST DATA ON THE OUTCOMES OF ECONOMIC EVALUATIONSOostenbrink JB, Al MJ, Rutten-van Mölken MMInstitute for Medical Technology Assessment, Rotterdam, Netherlands

OBJECTIVE: Patients in a clinical trial who withdrawbefore the scheduled end date are a serious problem ineconomic evaluations. The method to deal with datafrom these patients can have important impact on out-

Abstracts 405

comes in terms of resource use and costs, especially whenwithdrawal rates differ between treatment groups. Theaim of this study was to compare the impact of variousmethods for dealing with censored data on the total costsand on the difference in costs between treatment groups.METHODS: Five methods for dealing with censoreddata were applied to data from 519 patients with chronicdisease participating in a one-year randomized clinicaltrial. These five methods are complete case analysis, lin-ear extrapolation, hot-decking, predicted regression, andmultiple imputation.RESULTS: Fifteen percent of the patients in treatmentgroup A and 21% of the patients in treatment group B with-drew from the study before the scheduled end date. Meancosts per patient varied from €889 (SE: 94) in the complete-case analysis to €1400 (SE: 189) after predicted regression.Cost differences between treatment groups varied from €14in the complete-case analysis, to €243 after multiple impu-tation, to €372 after predicted regression. Hot-decking,multiple imputation and predicted regression were sensitiveto the selection of covariates.CONCLUSION: The various methods had a consider-able impact on total costs and on the difference in costsbetween treatment groups. In economic evaluations moreattention should be paid to methods for dealing with cen-sored patients and the impact of these different methodson the CE-ratio.

CE2

HANDLING MISSING DATA IN STOCHASTIC COST-EFFECTIVENESS ANALYSIS: THE IMPACT OF IMPUTATION METHODS ON ESTIMATES OF THE PHYSICAL QUANTITIES OF MEDICAL CARE RESOURCE USEBell T1, Liu J1, Backhouse M2

1Research Triangle Institute, Raleigh-Durham, NC, USA; 2Research Triangle Institute, Manchester, UK

OBJECTIVE: An issue that has recently received atten-tion from health economists is how to handle the prob-lem of missing data in stochastic cost-effectiveness analy-sis. The purpose of this paper is to highlight the impactthat different approaches to the imputation of missingdata can have on estimates of the physical quantities ofmedical care resource use.METHODS: Medical care resource use data were col-lected prospectively in a six-month RCT comparing twotreatments for a chronic condition that is characterizedby acute episodes. Two approaches of the multiple impu-tation were used to address the problem of missing data.Method A relied on imputing missing data for total costsand then estimating the physical quantities of medicalcare resource use. Method B relied on imputing missingdata for the physical quantities of medical care resourceuse and then estimating total costs. Results for physicianand nurse visits and days in the hospital were reported.RESULTS: The two multiple-imputation approaches pro-duced different estimates of medical care resource use.

The average number of physician and nurse visits anddays in the hospital between the two groups were 5.7 vs.5.3 physician visits, 1.0 vs. 0.9 nurse visits, and 4.0 vs.4.7 days in the hospital determined with method A. Theaverage number of physician and nurse visits and days inthe hospital between the two groups were 6.0 vs. 6.3physician visits, 1.2 vs. 1.3 nurse visits, and 4.0 vs. 5.0days in the hospital using method B.CONCLUSIONS: Medical care resource use estimatesare sensitive to the imputation approach. Method Bbuilds prediction models specifically for the utilizationcomponents under the imputation, and results from theimputed data sets may be less biased. It also providesmore flexibility for analyzing the cost components.

CE3

THE COST OF UPPER GASTRODUODENAL ENDOSCOPY: AN ACTIVITY-BASED APPROACHCrott R1, Makris N2, Barkun A3, Fallone C4

1University of Montreal and EORTC, Brussels, Belgium; 2University of Montreal, Montreal, QC, Canada; 3McGill University Health Center, Montreal, QC, Canada; 4McGill University Health Center, Montreal, QC, Canada

OBJECTIVE: The cost of medical procedures is often un-known, but is nevertheless crucial for setting reimburse-ment and health-care policies. This study investigated thecost of an upper gastrointestinal endoscopy in ambula-tory adults in a large academic hospital in the province ofQuébec, Canada from the perspective of the hospital.METHODS: An activity-based costing methodology wasused to break down the endoscopy procedure into a numberof primary tasks to which were allocated resources used atthe department level (labor, equipment, and materials). Unitcosts per activity were calculated from detailed tracking ofitems and factors used for performing each task.RESULTS: The direct cost of performing an endoscopyranged from 62$Can (1Can$ � 0.75 EUR) for an unse-dated, unbiopsied patient to 89$Can for a sedated, biop-sied patient. Not included in this amount are separate re-imbursement fees of 15$Can for biopsy analysis and50$Can professional fees for the performing physician,which are charged directly to the Ministry of Health. Acost-volume function was constructed under two differ-ent hypotheses of divisibility (sharing between clinicalunits) or undivisibility of fixed equipment. This showedan optimal unit cost per procedure starting at around3000 procedures a year for the installed equipment. In-corporating institutional overhead raises the cost of theprocedure substantially by an amount of 41$ as does theuse of non-reusable biopsy forceps, which adds about63$Can to the total cost of the procedure.CONCLUSION: Given the high proportion of overallhospital-wide overhead in the total cost of the procedure,allocation methods for these overheads in current hospi-tal accounting systems should be improved in order toobtain more precise estimates of the full cost of medicalprocedures like upper gastrointestinal endoscopy. The

406 Abstracts

use of single-use equipment rather than reusable equip-ment also increases the variable unit costs significantly.

SESSION II

HEALTH POLICY II

HP4

PRICING OF PHARMACEUTICALS IN CANADA AND EUROPE: A COMPARATIVE ANALYSISPalmer WNPalmer D’Angelo Consulting Inc, Ottawa, ON, Canada

OBJECTIVES: This paper compares pricing and reim-bursement of pharmaceuticals between Canada and se-lected European countries.METHODS: We examined the pricing and reimburse-ment schemes and cost-containment initiatives adoptedin Europe and Canada and their impact on prices of ex-isting and new medications. Published data from the Pat-ented Medicine Prices Review Board (PMPRB) providesinternational price comparisons for patented medicinesavailable in Canada. Prices for 13 new active substancesfirst marketed in Canada in 2000 were assembled forCanada and six European countries. For comparativepurposes, prices in the United States were also included inthe analysis.RESULTS: Drug prices in Canada are typically lowerthan prices in Germany, Sweden, Switzerland, and theUK but higher than prices in France and Italy. PatentedDrugs were 8% lower than the international median asmeasured by the PMPRB. However, for new drugs intro-duced in 2000, prices were, on average, 8.3% higher thanthose in Europe, although most were within the range ofEuropean prices. Prices in Canada and Europe were al-most always lower than those in the United States.CONCLUSIONS: The health-care systems in Canadaand Europe are fairly similar, as are the drug pricing andreimbursement schemes. However, local policies (e.g.,reference pricing, generic substitution, pharmacoeconom-ics) can result in significant international price variationsand the potential for parallel importation. Cost-effective-ness and international price comparisons play an impor-tant role in most major markets. The prices of pharma-ceuticals in Canada are comparable to those in Europealthough there is a trend for Canadian prices to behigher. Although pharmaceutical prices in Canada andEurope appear to be lower than in the United States,there are important differences in the health-care systemsas well as in the reimbursement mechanisms that makeprice comparisons with the US problematic.

HP5

USE OF THE SOJA METHOD FOR CHOOSING A HMG CO-A REDUCTASE INHIBITOR (STATIN)IN IRELANDHeerey A, McGowan B, Barry MNational Centre for Pharmacoeconomics, Dublin, Ireland

OBJECTIVE: The System of Objectified JudgementAnalysis (SOJA) method is a recognized, standardizedsystem, which aims to distinguish a medication of choicefrom a class of drugs. It combines expert opinion on pri-orities during prescription, with clinical evidence per in-dividual drug. We used the SOJA method in the Irish set-ting for statins and compared the results to their actualprescribing patterns in Ireland.METHODS: A questionnaire was sent to 1,705 Irish gen-eral practitioners (GPs) asking them to rank between 0and 10 the relative importance of 11 criteria when pre-scribing statins. These average results were assigned tothe SOJA method. The higher the overall SOJA score permedication, the better the statin performed and the moreevidence to support the prescription of this drug overothers in the same class. The SOJA results were com-pared to the number of statin prescriptions for public pa-tients at the time the questionnaire was completed (Aug/Sept 2000).RESULTS: There was a 42% reply rate. The average or-der of importance, as rated by the GPs were clinical effi-cacy (9.26), side effects (8.24), drug interactions (7.93),dosage frequency (7.53), clinical endpoint studies (7.50),number of approved indications (7.07), documentation(6.96), cost-effectiveness (6.95), bioavailablility (6.69),absorption (6.46) and number of formulations (5.55).The resulting SOJA score (with number of prescriptions)were simvastatin 64.6 (7,150), atorvastatin 59.5 (19,770),pravastatin 59.4 (34,736), cerivastatin 56.3 (3,500) andfluvastatin 51.4 (4,134).CONCLUSION: This study indicated that simvastatinshould be the statin of choice in Ireland followed by ator-vastatin or pravastatin. However, pravastatin is by farthe most widely prescribed statin. This suggests that non-rationale factors such as marketing may be influencingprescribing in Ireland. Additionally, despite the fact thatstatins represent one of the most expensive long-termtherapies, Irish GPs rank cost-effectiveness low on the listof priorities during their prescription.

HP6

THE IMPACT OF PIPELINE DRUGS ON UNITED STATES DRUG EXPENDITUREGROWTH TRENDSMullins CD, Wang J, Palumbo FB, Stuart BUniversity of Maryland, Baltimore, MD, USA

OBJECTIVE: Our study had two objectives. The firstwas to disaggregate historical drug expenditure increasesinto three components: price increase, utilization in-crease, and product shift. The second objective was todemonstrate the potential impact of pipeline drugs on an-nual drug-expenditure increases, factoring in what hasbeen observed historically for newer drugs.METHODS: The method for calculating pure price infla-tion involved a fixed set of drugs and measured the utili-zation-weighted mean of annual percentage increases.

Abstracts 407

The calculation of utilization increases reflects the changein the aggregate number of prescriptions filled. The cal-culation of annual increase in expenditures reflects thechange in expenditure from an old set of drugs to a newset. To estimate the impact of pipeline drugs, case scenar-ios were performed around HCFA’s baseline projectionsfor future drug trends. The compounded, annual, growth-rate decrease of new drug approvals was used as a lowerbound, while a constant increase in the pipeline projectionprovided the upper bound.RESULTS: Historically, slightly less than one-third of theincrease in drug expenditure was due to price increaseson existing drugs, approximately one-third was due to in-creases in the utilization of existing drugs, and slightlymore than one-third was due to product shift. Based onour data, the historical increase in expenditures was 15%for all drugs, 25% for newer drugs and 7% for olderdrugs. Our lower and upper bound scenarios projectednational drug expenditures of $191.56 billion and$215.62 billion, respectively, for 2004.CONCLUSION: The impact of pipeline drugs on futuredrug trends is significant. There is considerable variationsurrounding the impact of pipeline drugs on future drugexpenditures based on several plausible scenarios.

CANCER II

CN4

THE IMPACT OF THE DIAGNOSIS OF CANCER ON OUT-OF-POCKET HEALTH-CARE EXPENDITURES MADE BY THE US ELDERLYLanga K1, Hayman J1, Chernew M1, Kabeto M1, Slavin M2, Fendrick AM1

1University of Michigan, Ann Arbor, MI, USA; 2Ortho Biotech, Raritan, NJ, USA

If the “total” costs of a disease, such as cancer, are to beaccurately quantified, out-of-pocket expenditures (OOPE)made by patients and caregivers must be added to costsincurred by third party payers (direct medical) and em-ployers (lost productivity). Prior studies of OOPE by theelderly used small geographically restricted samples andnarrow patient subgroups.OBJECTIVES: To rigorously quantify OOPE for individ-uals older than 70 years in the United States. To empha-size cancer-related OOPE, three patient cohorts were ex-amined: 1) no cancer (No CA); 2) history of cancer butnot undergoing treatment (CA/No Tx), and 3) undergo-ing active cancer treatment (CA/Tx).METHODS: Data from the Asset and Health DynamicsStudy, a nationally representative, longitudinal survey ofcommunity-dwelling elderly were used. Respondents de-noted cancer status and reported OOPE over two yearsfor: 1) nursing home/hospitals; 2) doctor visits; 3) pre-scription drugs, and 4) “special” services. Using a multi-variable two-part regression model to control for differ-

ences in co-morbidity, health status, living situation, andsociodemographics, the additional cancer-related OOPEwas estimated.RESULTS: Of the 6576 respondents, 5553 (84%) re-ported No CA, 843 (13%) reported CA/No Tx, and 180(3%) reported CA/Tx. Cancer diagnosis and current can-cer treatment were significant predictors of increasedOOPE compared to no cancer. The mean annual OOPEfor No CA, CA/No Tx, and CA/Tx groups was US$1900,US$2400, and US$3300, respectively (p � .001). Hospi-tals (US$1400/yr) and prescription drugs (US$1100/yr)were the largest OOPE components for the CA/Tx group.The incremental OOPE for CA/No Tx and CA/Tx pa-tients approximates US$1.5 billion annually.CONCLUSIONS: OOPE for elderly individuals with ahistory of cancer or ongoing therapy are substantial andsignificantly greater than for those without cancer. IfOOPE remain unaccounted for, total costs of cancer willbe consistently underestimated. Economic evaluations ofinterventions aimed at cancer prevention and treatmentmust account for OOPE.

CN5

A SPECIFIC QUALITY OF LIFE SCALE INUPPER LIMB LYMPHOEDEMA: THEULL-27 QUESTIONNAIRELaunois R1, Mègnigbêto A1, Le Lay K1, Alliot F2

1Rees France, Paris, France; 2Clinique Hartmann, Neuilly Sur Seine, France

OBJECTIVE: The aim of this study was to validate a self-completed questionnaire in Upper Limb Lymphoedema.METHODS: A qualitative survey was conducted to iden-tify patients’ complaints. This questionnaire was admin-istered to 154 patients. Principal component analysis wasused to identify dimensions. A validation study was con-ducted in 304 patients. Six instruments have been used inthe case report form: volume differences between thehealthy and the affected arms; composite symptom scalescompleted by clinicians from patient interviews; ULL-27and SF-36 scales completed by patients; overall opinionof doctors and patients. Internal validity was checkedthrough factorial analysis. Trait validity was investigatedby correlating the domains rated with ULL-27 with theSF-36 scale. Nomologic validity was tested by comparingthe means of the ULL-27 subscales across severity stages.Sensitivity was tested only in patients with progressivedisease between D0 and D28 by comparing mean sub-scoresfor the ULL-27 scale and by calculating the effect size.RESULTS: Three hundred four patients were included inthe study. Factorial analysis isolated three dimensions:physical (15 items), psychological (7 items), and socialwithdrawal (5 items). The Cronbach coefficients aregreater than 0.80 for all dimensions. The Spearman cor-relations clearly distinguish the different life domainsfrom each other. At D0 the physical and social dimen-sions of ULL-27 scale were significantly correlated withseverity of illness but it was not the case for the psycho-

408 Abstracts

logical dimension. Correlation coefficients in patientsclinically stable between D0 and D28; were all greaterthan 0.84 for all dimensions of ULL-27. The sensitivityanalysis between D0 and D28 in patients with active dis-ease demonstrated significant differences between meanscores for all ULL-27 dimensions.CONCLUSION: Volume of oedema poorly reflects theimpact of the illness upon the patient. The ULL- 27 scaleseems to be a consistent instrument.

CN6

MEN’S PREFERENCES FOR THE CONSERVATIVE MANAGEMENT OF NON-METASTATIC PROSTATE CANCER: THE USE OF CONJOINT ANALYSISSculpher M1, Bryan S2, Emberton M3, Fry P3, de Winter P3, Payne H3

1University of York, York, UK; 2Birmingham University, Birmingham, UK; 3Middlsex Hospital, London, UK

OBJECTIVES: Selecting conservative therapies for menwith non-metastatic prostate cancer involves trade-offsbetween treatment attributes. An interview-based surveyusing conjoint analysis was undertaken to establishwhich treatment attributes are important to men in se-lecting treatments, and how attributes are traded off.METHODS: On the basis a pilot study, eight treatment-related attributes were selected for the survey: diarrhea;hot flushes; ability to maintain an erection; breast swell-ing/tenderness; physical energy; sex drive; life expect-ancy, and out-of-pocket personal costs. A discrete choicepreference elicitation mechanism was used. One hundredeighty men with non-metastatic prostate cancer from twoLondon hospitals were invited to participate. Of these,129 men, mean age of 70 years, 58% T-stage 1 or 2 at di-agnosis, were interviewed. Data were analyzed using ran-dom effects probit models.RESULTS: On average, men’s responses to the conjointquestions were sensitive to variation in the levels of all at-tributes (p � .01) and coefficient signs on all attributeswere as expected. A statistically significant interactionwas shown which indicated that the attribute ability ‘tomaintain an erection’ was less important to older men (p �.001). Most men were willing to make trade-offs betweenavoiding side effects and both losses in life expectancyand out-of-pocket costs. In terms of the former, theywere, on average, most willing to forgo life expectancy toavoid limitations in physical energy (mean of 3.01months), and least willing to trade life expectancy toavoid hot flushes (mean of 0.58 months to move from‘moderate’ to ‘mild’ or ‘mild’ to ‘none’).CONCLUSIONS: Men with prostate cancer are willingand able to participate in a relatively complex exercisethat weighs-up the benefits and harms of various conser-vative treatments for their condition, and to make trade-offs between attributes. The results provide an indicationof the relative importance of different aspects of treat-ment to patients with prostate cancer.

CARDIOVASCULAR DISEASE II

CV4

COMPARISON BETWEEN INVESTIGATOR AND PATIENT’S GLOBAL HEALTH ASSESSMENTS USING CALCULATED HUI-III AND SF-36 UTILITY VALUESHatoum HT1, Brazier JE2, Akhras KS3, Boyer JG3, Tooley JF3

1Hind T Hatoum & Company, Chicago, IL, USA; 2University of Sheffield, Sheffield, UK; 3Pharmacia Corporation, Skokie, IL, USA

OBJECTIVE: To compare patients’ own global health as-sessment item of the SF-36 (SF-1) and investigators’ glo-bal health assessment (GLBHLT) with values calculatedfor Health Utilities Index (HUI-III) and SF-36 preference-based (SF-6D), in patients with coronary artery disease(CAD). SF-1 is considered a coarse measure for patients’own health assessment while GLBHLT is a widely usedclinical-trials endpoint.METHODS: Baseline data of the SF-36, HUI-III andGLBHLT were collected for 331 patients enrolled in adouble-blind, multinational, phase III clinical trial. Boththe SF-1 and GLBHLT rate patients’ health on a scale ofone to five, where one is excellent and five is poor.RESULTS: Correlation coefficients (r) between the SF-1and HUI-III, and the SF-1 and SF-6D were 0.501, 0.508,respectively (p � .001). An r � 0.27 between the SF-1 andGLBHLT was found significant, albeit the magnitudewas almost half of those calculated for the SF-1 and SF-6Dor the SF-1 and HUI-III. Calculated SF-6D and HUI-IIIvalues for GLBHLT � 1 (excellent) were 0.74 and 0.74compared with the corresponding SF-1 values of 0.81and 0.83, respectively. Also the SF-6D and HUI-III valuesfor GLBHLT � 5 (poor) were 0.59 and 0.43 comparedwith the corresponding SF-1 values of 0.54 and 0.21, re-spectively.CONCLUSION: The SF-1 as a rough estimate of the pa-tient’s own health, yielded a stronger correlation withutilities calculated for HUI-III and SF-6D while GL-BHLT, considered a routine measure in clinical trials,yielded much weaker correlation. Confirmation of thesefindings is needed to assess if GLBHLT is a fair represen-tation of the health of patients with CAD.

CV5

ORLISTAT IN OBESE TYPE 2 DIABETIC PATIENTS: ASSESSMENT OF LONG TERM OUTCOMES AND COST-EFFECTIVENESSLamotte M1, Annemans LJ2, Lefever A3, Nechelput M3, Masure J31HEDM, Meise, Belgium; 2Ghent University, HEDM, Meise, Belgium; 3Roche Belgium, Brussels, Belgium

OBJECTIVES: Obesity is a very common condition intype 2 diabetic patients. Treating obesity may enhancehypoglycemic treatment and, thus, may contribute to areduction in long-term microvascular and macrovascular

Abstracts 409

complications. Orlistat, a drug for achieving and main-taining weight loss, reduces cardiovascular risk factors inrandomized clinical trials. The objectives of this studywere to estimate the resulting long-term clinical conse-quences and cost-effectiveness of treating obese type 2 di-abetic patients with orlistat.METHODS: A Markov model was developed to predict,over a 10-year period, the complication rates and mortal-ity with and without a two-year orlistat treatment, as-suming a five-year weight catch-up period after treatmentaccording to epidemiological evidence. A stepwise ap-proach was used to apply published clinical data into pre-diction of long-term outcome. As a first step, the impactof orlistat on HbA1c, blood pressure and cholesterol wasobtained. In a second step, the impact of decreasing theserisk factors on mortality and micro- and macrovascularcomplications was obtained. Four subgroups, defined bythe presence of arterial hypertension and/or hypercholes-terolemia, were studied.RESULTS: Cost-effectiveness varies between 3,462 Euro/life year gained (LYG) for obese diabetic patients with hy-pertension and hypercholesterolemia, and 19,986 Euro/LYG for obese diabetic patients without other risk fac-tors. Sensitivity analyses on all relevant variables indicatethat these results are very robust.CONCLUSIONS: Our results suggest that orlistat is cost-effective in the management of obese type 2 diabetic pa-tients, especially in those with concomitant hypercholes-terolemia and/or hypertension.

CV6

EXCESS MORBIDITY AND COST OF FAILING TO ACHIEVE TARGETS FOR BLOOD PRESSURE CONTROL IN THE ELDERLYLloyd AC1, Hansson L2, Anderson PM1, Kopp ZS3, Buch J31Fourth Hurdle Consulting Ltd, London, UK; 2University of Uppsala, Uppsala, Sweden; 3Pfizer Inc, New York, NY, USA

OBJECTIVE: Despite the existence of effective therapy,millions of Europeans have blood pressure (BP) above in-ternationally agreed targets for control of cardiovascularrisk. There is particular resistance to aggressive manage-ment of the elderly. We estimated the acute health-carecosts resulting from failure to achieve BP targets in olderadults in France, Germany, Italy, Sweden and the UK.METHODS: We constructed a burden of disease modelto estimate the costs of uncontrolled hypertension in thisgroup. Prevalence of uncontrolled hypertension wastaken from the MONICA study and published surveys.The relationships between three cardiovascular (CV)events (symptomatic acute myocardial infarction (AMI),congestive heart failure (CHF) and stroke) and BP wereestimated from a large prospective study (the HOT trial).Costs came from government sources and published stud-ies. We estimated the acute medical costs of these eventsat current prevalence of uncontrolled hypertension andexpected number of events and cost if BP were treated to

target levels. Stochastic simulation was used to constructconfidence intervals.RESULTS: Among adults older than 65 years in the fivecountries, an estimated 15.2m have BP above 160/95mmHg and a further 12.9m have BP in the range 140/90–160/95 mmHg. The model estimated that 505,000 CVevents (AMI–91,000; CHF–190,000, stroke–224,000)occur each year in those older than 65 in these countries,resulting in annual acute hospital costs of Euro2.3Bn(95% CI Euro2.04–2.53Bn). One hundred twenty eightthousand of these events (AMI–9,000; CHF–55,000,stroke–64,000) could be avoided if BP targets were met.The annual cost of these avoidable events was estimatedto be Euro560m (95% CI Euro300–783m), representing24.2% of the acute medical cost of these CV events inthis population.CONCLUSION: Failing to implement existing guidelinesfor BP management in the elderly contributes substan-tially to the total human and economic burden of CV dis-ease.

PRACTICAL DESIGN ISSUES

DS1

PATIENT-REPORTED OUTCOMES: A COMPARISON OF TWO DATA-CAPTURE METHODSHufford MR1, Noe L2

1Invivodata, Inc, Pittsburgh, PA, USA; 2Ovation Research Group, Highland Park, IL, USA

OBJECTIVE: The reliable and valid capture of patient-reported outcomes (PROs) are becoming increasingly im-portant endpoints for the pharmaceutical industry. PROsare data reported by the patient, and can include health-related quality of life, functional status, symptoms, pain,bother, satisfaction, work loss, reduced productivity, andthe use of medical resources. These data, when includedin studies, have traditionally been collected using paper-based methods. Technological advances in electronicdata capture, such as handheld computers, internet-basedsolutions, and interactive voice-response systems, facili-tate collecting PROs and are important new tools forclinical and outcomes researchers. Our objective was toexamine compliance rates between a handheld computerdiary system and a paper-based diary.METHODS: Eighty pain patients were randomly as-signed to complete a three-week, diary-monitoring proto-col using either a compliance-enhanced electronic diarysystem or a paper diary. The paper diary was covertly in-strumented to allow for objective determination of whenthe diary was opened or closed.RESULTS: Participants submitted diary cards corre-sponding to 89% of assigned assessment times (�15min). However, the electronic record indicated that ac-tual compliance with the paper diary was only 11%, indi-cating a high level of faked compliance. On 32% of allstudy days the paper diary was not opened, yet reported

410 Abstracts

compliance for these days exceeded 90%. Evidence forback and forward filling of paper diary cards was ob-served. For the compliance-enhanced electronic diary, theactual compliance rate was 93%.CONCLUSIONS: Data from paper-based diaries are ofquestionable validity, given that many of their entries arenot completed as required by the protocol. Science-basedelectronic diaries can produce high rates of patient com-pliance in the field. Improved methods for data collectionshould encourage researchers in the pharmaceutical in-dustry to aggressively evaluate electronic PRO (ePRO)data to help differentiate their products.

DS2

POWER CALCULATIONS FOR WIDELY USED PATIENT-REPORTED OUTCOMES (PRO) MEASURES IN WOMEN’S HEALTH TRIALSAbetz L1, Brandman J2, Plante M3

1Mapi Values, Cheshire, UK; 2Pfizer Pharmaceutical Group, New York, NY, USA; 3Pfizer Inc, Ann Arbor, MI, USA

OBJECTIVE: Increasingly, federal authorities are request-ing power calculations for secondary endpoints in clinicaltrials, including patient reported outcomes (PRO). How-ever, most PRO measures do not provide power calcula-tions in their manuals; if provided, they are often based onmixed samples of males and females. It is well docu-mented that female and male PRO scores often differ.Thus when designing women’s health trials, it may beworthwhile to conduct power calculations using women’sPRO scores and standard deviations. This study presentsthe power and sample-size calculations for a variety ofquestionnaires used in women’s health studies.METHODS: The Menopause Quality of Life question-naire (MENQOL), Women’s Health Questionnaire (WHQ),Psychological General Well-Being Index (PGWB), andShort Form 36 and 12 (SF-36/SF-12) were assessed. Pub-lished information on scores and standard deviations infemale populations were used to determine sample sizesneeded to detect differences between two experimentalgroups, post-intervention.RESULTS: Results varied by questionnaire, due in part tovarying score ranges across questionnaires. For example, toachieve 90% power with a ten-point difference the follow-ing sample sizes per treatment arm were required: 158women when using the MENQOL vasomotor score (range:0–100); 47 women when using the WHQ total score (range:0–102); 70 women when using the PGWB total score(range: 22–132); 24 and 21 women when using the SF-36and SF-12 Physical Component Summary (no floor/ceiling).CONCLUSION: When calculating sample sizes, it is nec-essary to keep in mind the questionnaire’s possible scorerange in order to ensure that the power calculation isbased on a clinically meaningful difference between treat-ment groups. These results may be used to help calculatesample sizes needed to achieve sufficient power to detectstatistically significant differences in women’s health tri-als for these widely used measures.

DS3

DESIGN AND ANALYSIS OF UNIT COST ESTIMATION STUDIES: HOW MANY TYPES OF HOSPITALIZATION? HOW MANY COUNTRIES?Glick HA1, Orzol SM1, Tooley JF2, Polsky D1, Mauskopf J31University of Pennsylvania, Philadelphia, PA, USA; 2Pharmacia, Skokie, IL, USA; 3Research Triangle Institute, Research Triangle Park, NC, USA

OBJECTIVES: The availability of reliable methods for im-puting hospital costs (e.g., relative values from the U.S. diag-nosis-related group (DRG) payment system, cost weightsand average length of stay (LOS) estimates from the Austra-lian refined DRG system, or average LOS from the FrenchPatient Homogeneous Groups), allows one to address twodesign questions related to the collection of hospital unitcost data for use in multinational clinical trials with eco-nomic endpoints: 1) In individual countries, for how manytypes of hospitalization should estimates be obtained? 2)For how many countries should they be obtained?METHODS: We addressed these questions by assumingthat unit cost estimates for 47 types of hospitalizationcollected in four western European countries representedthe universe of hospitalizations. This assumption pro-vided a population estimate against which we could mea-sure the error associated with imputations in samplesdrawn from the population. To answer the first question,we: 1) randomly sampled subsets of hospitalization typesfrom this population and used them to develop imputa-tion regressions; 2) used the results of the regressions toimpute costs from the remaining hospitalization types,and 3) estimated measures of imputation error withineach sample. To answer the second question, we per-formed a similar analysis, but instead sampled countries.RESULTS: We found that the imputation error decreasedas the number of types of hospitalization and countriessampled increased, but that the rate of reduction in errorshrank. We also found that error was minimized by ob-taining estimates for fewer types of hospitalization frommore countries than the reverse.CONCLUSION: The availability of reliable methods forimputing hospitalization costs allows one to economizeon data collection. Our experiment suggested that col-lecting a small number of estimates (in our data, approxi-mately 25) in as many countries as is feasible minimizesimputation error.

SESSION III

DIABETES

DB1

ESTIMATING HEALTH-CARE COSTS ASSOCIATED WITH DIABETES-RELATED COMPLICATIONS USING DATA FROM THE UNITED KINGDOM PROSPECTIVE DIABETES STUDY (UKPDS)Gray A, Clarke PUniversity of Oxford, Oxford, UK

Abstracts 411

OBJECTIVES: Patients with type 2 diabetes are knownto make increased use of health-care resources, but theimpact of specific macro and microvascular complica-tions on costs is unclear. Here we use regression basedmethods to estimate the immediate and long-term impactof six diabetes-related complications on hospital costs,using data from the UKPDS, a large (n � 5102) and long-term (median duration 10.3 years) clinical trial.METHODS: Data on the occurrence and precise timingof pre-defined diabetes-related complications, and on allhospitalisations with associated specialties, lengths ofstay, and procedures, were collected routinely for all pa-tients during the trial. Panel data regression analysis wasused to estimate the immediate impact (i.e. in the yearevent occurred) and long-term impact (i.e. in each subse-quent year) of the following six diabetes-related compli-cations on hospital costs: fatal and non fatal myocardialinfarction (n � 828); fatal and non-fatal stroke (n �271); heart failure (n � 166); angina (n � 319); blindnessin one eye (n � 166); amputation (n � 67). Hospitalcosts were calculated using national average specialty-specific costs per inpatient day, expressed in 1999 £s UK.RESULTS: All six diabetes-related complications had astatistically significant impact on hospital costs. In theyear in which the complication occurred, diabetes-relatedcomplications were associated with increased hospitalcosts ranging from £995 for loss of sight in one eye to£5478 for an amputation. In subsequent years the annualmagnitude of this effect ranged from 13% to 33% of thecosts in the initial year.CONCLUSIONS: Regression analysis on a large andwell-validated, patient-specific data set yields plausibleempirically based estimates of the hospital cost conse-quences of diabetes-related complications. These will beof use to other economists and health service researchers,particularly those interested in assessing the costs of dia-betes and the cost-effectiveness of interventions within amodeling framework.

DB2

BOTTOM UP VERSUS TOP DOWN COST ESTIMATES FOR TYPE 2 DIABETESKoopmanschap MA, Redekop K, Niessen LErasmus University, Rotterdam, Netherlands

OBJECTIVES: To compare top down and bottom upcost estimates of direct health-care costs for diabetes type2 in the Netherlands.METHODS: For the top down cost estimates we usedcomprehensive Dutch national age and sex-specific costof illness estimates for diabetes and diseases related to di-abetes complications. Costs for diabetes type 2 were sep-arated by age (age 35 and older) and by prevalence ratesfor types 1 and 2. The costs of complications were esti-mated using costs and prevalence rates for cardiovasculardiseases, neuropathy, nephropathy and retinopathy, com-bined with relative risks for these complications in pa-tients with diabetes type 2. The bottom up costs were es-

timated using a sample of 1371 type 2 diabetes patientsfor whom their GP reported the total medical consump-tion related to diabetes and its complications during theprevious six months.RESULTS: Total medical costs for diabetes type 2 in1998 were Euro 567 mln according to the bottom up es-timate versus Euro 519 mln for the top down estimate, orless than 10 % difference in cost. The costs for in hospitalcare, ambulatory care and equipment were very compa-rable. The cost of medication was higher according to thebottom up study. The bottom up study identified a largeramount of cardiovascular and lipid lowering drugs. Bothcosting methods show that complications, especially car-diovascular, are responsible for a substantial portion oftotal health-care costs.CONCLUSIONS: For diabetes type 2 it was demon-strated that using comprehensive top down disease costscombined with sound epidemiological data on complica-tions, can yield valid cost estimates that are quite compa-rable with bottom up cost estimates.

DB3

A DYNAMIC, THREE-PART MODEL FOR PREDICTING HOSPITAL COSTS IN TYPE 2 DIABETIC PATIENTSWang J, Morris A, Davey PMEMO, University of Dundee, Dundee, UK

OBJECTIVE: To develop a model for predicting hospitalcosts in patients with diabetes mellitus.METHODS: We have previously shown that, in compar-ison with standard cost models, better prediction can beachieved with a two-part model that independently pre-dicts risk of hospitalization and cost of hospitalization. Inthis analysis we have developed a three-part model byadding mortality to the original two-part model, becausepatients who die do not necessarily incur hospital costs inthe year of death. Furthermore we have extended thethree-part model to an autoregression model. Age, gen-der, any hospitalization in the last year, log-hospital costsin the last year and log-mean costs over all previous yearswere included in the model. A Bayesian forecastingmethod was used to obtain a predictive distribution ofcosts in the next year.RESULTS: We identified 5672 type 2 diabetic patients inthe Tayside area and analyzed annual hospital costs be-tween 1988 and 1995. The fitted three-part model showedthat increasing age was associated with increasing mor-tality and with increasing costs per hospital episode.However, age was not associated with the risk of hospi-talization. Cost of hospitalization in previous years waspositively associated with risk of hospitalization (OR �1.51/ln(mean costs), CI � 1.48, 1.53) and mortality (OR �1.22/ln(mean costs), CI � 1.19,1.25). An example ofBayesian cost forecasting showed that for female patientsaged 50 to 59, increasing hospital costs in the previousyear increased the probability of having a high-cost hos-

412 Abstracts

pital admission in the next year but paradoxically it alsoincreased the probability of having zero hospital costs dueto increased risk of death without hospitalization.CONCLUSIONS: Costs incurred in previous years can beused to predict costs and outcomes in the future. This dy-namic three-part model clarifies the relationship betweenrisk of hospitalization, cost of hospitalization and mortality.

INFECTIOUS DISEASE

IN1

ESTIMATING THE POTENTIAL HEALTH GAIN AND COST CONSEQUENCES OF INTRODUCING A PRE-SCHOOL DTPA PERTUSSIS BOOSTER INTO THE UK CHILD VACCINATION SCHEDULEStevenson M1, Beard S1, Finn A2, Brennan A1

1University of Sheffield, Sheffield, UK; 2University of Bristol, Sheffield, UK

OBJECTIVE: To establish the likely health and cost im-pacts of a pre-school booster vaccination for Bordetellapertussis, when added to existing UK primary vaccina-tion schedules assuming that a diphtheria, tetanus, andacellular pertussis, (DTPa) booster would replace the cur-rent diphtheria and tetanus (DT) booster.METHODS: A transition state model of pertussis infec-tion in a closed population representative of England andWales, comprised of eight age bands with susceptible, in-fected and immune population sub-groups. Herd-immu-nity was explicitly modeled. Epidemiological service useand cost data were sourced from routine statistics, pub-lished literature and, where necessary, clinical estimates.The number of pertussis cases, hospitalizations anddeaths were forecast for a five-year period. Quality of lifegains were not explicitly calculated.RESULTS: Introducing a pre-school pertussis boosterand achieving 84% coverage is predicted to cost an addi-tional £14.32m over a five-year period, assuming £5marginal cost between DTPa and DTP or £8.60m assum-ing a £3 marginal cost. Offset against this are the cost ofreduced hospitalizations and GP consultations, which areexpected to total between £4.21m and £4.60m. The re-turn on this investment would be a reduction in up to1700 hospitalizations, between 5000 and 27,000 pertus-sis cases depending on the level of under-reporting andassuming a vaccine waning period, and one infant death.CONCLUSIONS: The introduction of a pre-schoolbooster is predicted to significantly reduce the number ofhospitalizations and pertussis cases contracted. The esti-mated marginal cost of this strategy is £10m over a five-year period, assuming a £5 difference between DTP andDTPa or £4m, were the difference only £3.

IN2

THE ECONOMIC BURDEN OF VIRAL RESPIRATORY INFECTION IN THEUNITED STATES

Fendrick AM1, Sarnes MW2, Nightengale B2, Monto AS1

1University of Michigan, Ann Arbor, MI, USA; 2Applied Health Outcomes, Tampa, FL, USA

Viral respiratory infection (VRI) is the most commonly oc-curring illness in man, imposing considerable burden onpatients and on society. However, to date, no formal studyof the economic impact of VRI has been performed.OBJECTIVES: To rigorously quantify the economic im-pact of viral respiratory disease (VRI) in the US in termsof health-care resource utilization (direct costs) and pro-ductivity losses (indirect costs).METHODS: Publicly available databases including theUS Census, National Health Interview Survey (NHIS),the Medical Expenditure Panel Survey (MEPS) and theNational Ambulatory Medical Care Survey (NAMCS)were used. From these databases, projections regardingpopulation characteristics, physician and emergency-roomencounters, prescription and over-the-counter drug utiliza-tion, and productivity losses related to VRI were made.Data obtained from primary epidemiological research andprospective randomized clinical trials were used to esti-mate the incidence of VRI in the general population andthe rate of secondary clinical complications associatedwith VRI.RESULTS: Nearly 500 million episodes of VRI occur an-nually in the US alone. Direct costs associated with VRIare estimated to be $16.8 billion annually and are brokendown as follows: physician visits, $6 billion; complica-tions, $3.8 billion; prescription and over-the-counter medi-cations, $4.8 billion. Indirect costs for employed individu-als approximate $7.6 billion per year. Physician encountersvia a telephone and the internet, productivity losses incurredby caregivers (i.e. parent) of VRI-infected individuals, andcosts associated with diminished productivity while at workor home were not included, suggesting that this projectionof total VRI costs—$25 billion annually in the US alone—isvery likely an under-estimate.CONCLUSIONS: Viral respiratory infections impose asignificant clinical and economic burden to society, ap-proaching or surpassing the aggregate costs of manycommon chronic diseases. The resultant clinical and costramifications attributable to this common acute condi-tion warrants increased attention from health-care pro-viders and policy makers.

IN3

THE OBSERVED COSTS AND HEALTH-CARE USE OF CHILDREN IN A RANDOMIZED CONTROLLED TRIAL OF PNEUMOCOCCAL CONJUGATE VACCINERay GT1, Butler JC2, Black SB1, Sheinfield HR1, Lieu TA3

1Kaiser Permanente, Oakland, CA, USA; 2Centers for Disease Control and Prevention, Anchorage, AK, USA; 3Harvard Pilgrim Health Care and Harvard Medical School, Boston, MA, USA

OBJECTIVE: Pneumococcal conjugate vaccine for in-fants has recently been found effective against meningitis,

Abstracts 413

bacteremia, pneumonia, and otitis media, but costs morethan previously introduced vaccines. We determined thesavings in medical costs over 36 months of life attribut-able to the use of the vaccine in healthy infants in a largerandomized trial.METHODS: We analyzed the actual costs and utilizationfor 36,471 children involved in a randomized trial of theheptavalent pneumococcal conjugate vaccine conductedin the Northern California Kaiser Permanente MedicalCare Program (KP). Costs were analyzed for all childrenrandomized in the trial (intent-to-treat) who were mem-bers of the health plan at any time during the follow-up.All clinic- and pharmacy-related costs were included, aswere those hospital costs associated with conditionsdeemed to be potentially pneumococcal related. The costof the vaccine and vaccine administration were excluded.Confidence intervals around cost savings were calculatedusing bootstrap replications.RESULTS: Compared with the control group, the vacci-nated group incurred $78 less in medical costs (CI: $5 to$158) per child during the first 36 months of life, exclu-sive of the cost of the vaccine. This represented savings ofabout 3% of total medical costs for these children duringthat time period.CONCLUSION: The pneumococcal conjugate vaccinereduced medical costs in children in the first 36 monthsof life, before factoring in the cost of the vaccine and vac-cine administration. These cost savings, however, are un-likely to offset the cost of the vaccine at its current price.

MENTAL HEALTH I

MH1

SOAP-51: A QUALITY OF LIFE SURVEY FOR COMMUNITY-RESIDING INDIVIDUALSWITH SCHIZOPHRENIABarr JT1, Schumacher GE1, Mason E1, Ohman S1, Hanson A2

1Northeastern University, Boston, MA, USA; 2 Division of Medical Assistance, Boston, MA, USA

OBJECTIVES: To establish the concurrent and discrimi-natory validity, and relative item importance of the previ-ously developed client-centered, 51-item SchizophreniaOutcomes Assessment Project (SOAP-51) survey, a self-administered, health-related quality of life instrument forindividuals with schizophrenia living in the community.METHOD: We asked 1500 community-based clientswith schizophrenia, and 150 of their caregivers (using theshortened objective version), in five ethnically and geo-graphically diverse areas within the US to complete theSOAP-51 four times over a three-month period; client re-tention was 84.2%. Average age was 42.8 years, 60%males. For concurrent validation, clients rated the impactof their condition on each of SOAP-51’s eight factors;this was compared to their factor score. For discriminatevalidity, the caregivers assigned clients to one of fourquartiles describing their perception of the client’s abilityto objectively function in each factor area. Client factor

scores were compared to caregiver-reported functionallevels. An additional 300 clients rated the relative impor-tance of each item.RESULTS: Cronbach’s alpha for the eight factors was0.71–0.88, test-retest reliability, 0.78–0.99. Client factorscores lowered as clients felt the condition had more neg-atively impacted on their lives (p � .001). Client factorscores improved with improving caregiver-perceived as-sessment of client functionality (p � .001). Client scoresfrom the lowest to highest assigned quartile had averagescores of 42.7%, 49.3%, 54.7%, and 60.0%, respec-tively. Ninety two percent of both clients and caregiversconsidered the survey valuable for monitoring clientprogress. No significant difference (p � .05) was ob-served in importance weights for the eight factors, or 51items, age or gender of client, caregiver or client re-sponses.CONCLUSIONS: Concurrent and discriminatory valid-ity show that the SOAP-51 meets psychometric charac-teristics for use as a client-centered outcomes measure inpatient monitoring and management, and health policydecision-making.

MH2

EFFICACY OF NURSE TELEHEALTH CARE AND PEER SUPPORT IN AUGMENTING TREATMENT OF DEPRESSION IN PRIMARY CAREHunkeler EM1, Meresman JF2, Hargreaves WA3

1Kaiser Permanente, Oakland, CA, USA; 2Kaiser Permanente, Northern California, Santa Clara, CA, USA; 3University of California, San Francisco, San Francisco, CA, USA

OBJECTIVES: Because clinical outcomes of depressiontreatment in primary care settings tend to be poor, we de-veloped and evaluated the efficacy of two augmentationsto antidepressant treatment to be delivered by primarycare nurses.METHODS: We conducted a randomized trial compar-ing usual care, telehealth care, and telehealth care pluspeer support for depressed patients seen in primary carein an HMO setting. Assessments were conducted at base-line, six weeks and six months after study enrollment attwo managed-care, adult primary-care clinics. Partici-pants included 303 patients recently started on antide-pressants. The intervention consisted of: telehealth care;emotional support and focused behavioral interventionsin 10 seven-minute calls over four months by speciallytrained primary-care nurses and peer support; telephoneand in-person supportive contacts by trained Health Planmembers recovered from depression. Primary outcomemeasures were the Hamilton Rating Scale for Depression,Beck Depression Inventory, Mental and Physical Func-tioning (Short Form 12), and treatment satisfaction andmedication adherence questionnaires.RESULTS: Nurse-based telehealth patients with or with-out peer support more often experienced 50% improve-ment on the Hamilton at six weeks (50% vs. 37%, P �.01) and six months (57% vs. 38%, P � .003), and on

414 Abstracts

the Beck at six months (48% vs. 37%, P � .05). Greaterquantitative reduction in symptom scores on the Hamil-ton at six months (10.4 vs. 8.1, P � .006) were observed.Telehealth care improved mental functioning at six weeks(47.1 vs. 42.6, P � .004) and treatment satisfaction at sixweeks (4.41 vs. 4.17, P � .004) and six months (4.20 vs.3.94, P � .001). Medication adherence was the same inall groups and adding peer support to telehealth care didnot improve the main outcomes.CONCLUSION: Nurse Telehealth Care improves clinicaloutcomes of antidepressant treatment, improves patientsatisfaction, and fits well in primary care. The nurse tele-health care program has been implemented in Maine,Ohio and Southern California.

MH3

RECENT WEIGHT GAIN AND THE COST OF ACUTE SERVICE USE AMONG INDIVIDUALS WITH SCHIZOPHRENIAWeiden PJ1, Mackell JA2, McDonnell DD3

1SUNY Health Sciences Center at Brooklyn, Brooklyn, NY, USA; 2Pfizer Pharmaceuticals Group, Pfizer, Inc, New York, NY, USA; 3Consumer Health Sciences, Princeton, NJ, USA

OBJECTIVE: Newer antipsychotics have been associatedwith increased weight gain. There is also mounting evi-dence that this leads to noncompliance and a lower qual-ity of life. Gaining weight is also undesirable for healthreasons and may lead to increased use of health-care re-sources. This study considers the association betweenweight gain and acute service use for patients withschizophrenia.METHODS: Questionnaires were mailed to people withschizophrenia identified through National Alliance forthe Mentally Ill and the National Mental Health Associa-tion in spring 2000 (n � 390). Data presented here arefrom the 345 respondents who reported weight loss (n �94, 27%), no weight change (n � 106, 31%), someweight gain (1–14lbs; n � 70, 20%), and significantweight gain (�15lbs; n � 75, 22%) within the last sixmonths. Acute service use was defined as emergencyroom (ER) visit or hospitalization. Cost values werethose reported in Ernst and Hay (1994). For each individ-ual, total costs were computed by summing across cate-gories.RESULTS: The group reporting significant weight gainwas significantly more likely to use acute services thanthe other three groups (p � .001 for hospitalization, p �.005 for ER visit). The association remained significantwhen controlling for other variables in multivariate anal-yses, including age, gender, ethnicity, and overall distress.Overall costs were highest for those who gained 15 ormore pounds ($9,486). Those who lost weight incurredcosts of $7,400, those who did not change weight in-curred costs of $4,095, and those who gained 1–14pounds incurred costs of $3,647.DISCUSSION: Our preliminary results suggest that re-cent weight gain is associated with greater use of acute

services and higher costs. There are several plausible ex-planations. For example, physicians might change medi-cations for people doing poorly (e.g., start a new medica-tion after an acute psychiatric episode). Another possibilityis that acute medical services are more likely to be neededafter an episode of rapid weight gain.

QUALITY OF LIFE

QL1

CAN HEALTH STATE VALUES BE PREDICTED FROM HEALTH-RELATED QUALITY OF LIFE MEASURES?Svensson K1, Szende A2, Lundbäck B3

1AstraZeneca R&D, Lund, Sweden; 2MEDTAP International, Amsterdam, The Netherlands; 3University Hospital, Umeå, Sweden

OBJECTIVE: To predict Health State Values (HSV) fromhealth-related quality of life (HRQL) assessments could,if possible, be a fruitful way to use HRQL values forhealth-economic evaluations. With this in mind, we in-vestigated to what degree values from the EuroQol in-strument, the EQ-5D index, and the EQ-VAS visual ana-log scale, could be predicted using HRQL measures forpatients with respiratory diseases.METHODS: Data from two surveys of patients with res-piratory disease were used for this evaluation. The firstdata set was from 206 patients in Hungary suffering fromasthma, and the other was from 120 patients in thenorthern part of Sweden with COPD. Both surveys in-cluded patients with different severities of the diseases.The HRQL instruments used in both surveys were the SF-36, a generic instrument, and St George’s RespiratoryQuestionnaire (SGRQ), a disease-specific instrument. Thetwo data sets were analyzed separately using a multiplelogistic regression model in a stepwise manner to predictEQ-5D and EQ-VAS from the eight domains of SF-36 andthe three domains of SGRQ, after transformation of EQ-5D and EQ-VAS to a 0–1 range.RESULTS: The amount of variation in both the EQ-5Dand EQ-VAS that could be explained from the combinedHRQL measures was at most 56%. EQ-5D had largervalues than EQ-VAS. Using SF-36 domains only as pre-dictors gave marginally lower values. The two domainsfrom SF-36 with best predictability explained about 90%of the reduction achieved with all 11 domains togetherfrom the two HRQL-measures, indicating correlation be-tween the different domains.CONCLUSION: Values for EQ-5D and EQ-VAS pre-dicted from SF-36 and SGRQ have moderate precisionand should be used cautiously.

QL2

ESTIMATING PATIENTS’ PREFERENCES IN TREATMENT CHOICES INVOLVING RISK: A NEW MODIFIED STANDARD GAMBLE METHODWalshe RUniversity of Cologne, Cologne, Germany

Abstracts 415

OBJECTIVE: Cost-effectiveness, -utility and -benefitanalyses have made a substantial impact on heath care.Current approaches are methodologically coherent andprovide a basis for improved resource allocation. As hasbeen noted before, they are based on expected (average)outcome and do not take into account patients’ risk pref-erences, which are important aspects of clinical decision-making. From a patient’s perspective, a treatment associ-ated with a lower level of expected clinical benefit may bepreferred if it has better worst-case scenario than a treat-ment with higher expected (average) benefit. We analyzedecision-scientific methodology in order to incorporaterisk preferences into the framework of outcome and util-ity studies.METHODS: Concepts of risk preferences in decision-making from economics and management science are de-scribed, and their applicability to a clinical context istested using a decision tree model. Methods analyzed in-clude the Bayes-Principle of expected value (m-principle),the m-s-principle, risk restriction with given outcome andthe Bernoulli-Principle. The absence of risk preferences inthe QALY (quality adjusted life years) and HYE (healthyyears equivalent) concepts is shown mathematically.RESULTS: Current methodological concepts used in out-comes research do not adequately incorporate patients’risk preferences. On the basis of a modified standardgamble approach a method of obtaining risk preferencesfor given treatment outcome is developed. It leads to thedetermination of relative marginal utilities and may beused in choosing health-care interventions by consideringrelative marginal costs.CONCLUSION: Patients’ risk preferences should be takeninto account in outcome and utility analyses involving sub-stantial risks. The approach proposed here may improvethe empirical estimation of patients’ preferences and thequality of resource allocation in health care.

QL3

PSYCHOMETRIC PERFORMANCE OF THE MEDICAL OUTCOMES STUDY SLEEP SCALE IN THE US GENERAL POPULATIONHays RD1, Martin SA2, Sesti AM2, Spritzer KL1

1UCLA, Los Angeles, CA, USA; 2Pfizer, Ann Arbor, MI, USA

OBJECTIVE: Support for the reliability and validity ofthe Medical Outcomes Study (MOS) Sleep Scale was pro-vided in the MOS sample of 3445 individuals withchronic illness. We sought to extend this work by evalu-ating the psychometric properties of the MOS-Sleep Scalein the US general population.METHODS: The MOS-Sleep Scale is a 12-item, self-re-ported survey that yields six subscales: sleep disturbance,snoring, awaken short of breath or with headache, quan-tity of sleep, sleep adequacy, and somnolence as well as anine-item sleep problem index. The subscales and prob-lem index are scored on a 0 to 100 range, with higherscores indicating more of the domain being measured.We administered the MOS measure by telephone to a na-

tionally representative sample of 1011 US adults aged 18and older in January 2001.RESULTS: The average age of the sample was 46; 51%were female and 74% were white. Internal consistencyreliability estimates for the MOS-Sleep scales tended tobe adequate: sleep disturbance (4 items, alpha � 0.80);sleep adequacy (2 items, alpha � 0.82); sleep somnolence(3 items, alpha � 0.63); and nine-item sleep problems in-dex II (alpha � 0.83). Adjusting for age and gender,MOS patients reported significantly more quantity ofsleep (t � 3.27, P � .002), but significantly worse sleepdisturbance (t � 5.08, P � .001), snoring (t � 2.16, P �.05), shortness of breath (t � 4.59, P � .001), sleep ade-quacy (t � �2.39, P � .05), somnolence (t � 5.10, P �.001), and sleep problems (t � 3.27, P � .002) than thegeneral US population.CONCLUSIONS: The MOS-Sleep Scale was found tohave good internal consistency reliability and to discrimi-nate between patients with chronic illness and the USgeneral population. Further work is needed to comparethe MOS-Sleep Scale results with objective measures ofsleep such as polysomnography.

SESSION IV

MENTAL HEALTH II

MH4

SLEEP DISORDERS AND HEALTH RELATED QUALITY OF LIFE—AN EPIDEMIOLOGICAL SURVEYIsacson D, Bingefors KUppsala University, Uppsala, Sweden

OBJECTIVE: To analyze health-related quality of lifeamong people with sleeping problems from an epidemio-logical perspective.METHOD: A cross-sectional survey on a sample agedbetween 20 and 84 years in the county of Uppland, Swe-den yielded a response in 5404 patients (68%). A recallperiod of two weeks was used for sleeping problems anduse of sleeping medication. The SF-36, used to measureHRQoL, covers eight domains of health: physical func-tion (PF); role limitation because of physical health (RP);bodily pain (BP); general health perception (GH); vitality(VT); social functioning (SF); role limitation because ofemotional health problem (RE), and mental health (MH).Linear regression analysis was employed for the multi-variate analyses.RESULTS: In all, 20.3% of the population reportedsleeping problems. Sixteen percent had experienced sleep-ing problems but had not used sleeping medication while4.3% had used medication. The prevalence of sleepingproblems was of the same magnitude between the ages of20 to 74 years (around 20%) but higher among thoseaged 75 to 84 years (29.5%). Sleeping problems weremore prevalent among women (23.8%) than men (16.1%).The use of sleeping medication increased by age. Among

416 Abstracts

persons aged 20 to 34 years, 1.0% had used sleepingmedication and among those aged 75 to 84 years 16.1%did. Use of sleeping medication was more common amongwomen (5.3%) compared to men (3.0%). Individuals withsleeping problems scored significantly lower (p � .05) thanthose without sleeping problems on all of the eight do-mains in SF-36 � RE � 20.3; VT � 19.3; MH � 16.8;GH � 16.2; SF � 15.6; RP � 15.1; BP � 13.4 and PF �7.9. Among individuals with sleeping problems, those us-ing sleeping medication scored lower on seven domainsof the SF-36 � RE � 10.5; MH � 10.4; SF � 8.8; GH �8.1; BP � 7.8; VT � 6.1 and PF � 5.8.CONCLUSION: Sleeping problems are common in thepopulation and lead to a significantly decreased qualityof life among affected individuals.

MH5

COST OF THE FIRST, SECOND AND SUBSEQUENT EPISODE OF DEPRESSION IN POLANDKiejna A1, Czech M2, Faluta T2, Pachocki R2, Corcaud S3

1Medical Academy of Wroclaw, Wroclaw, Poland; 2Servier Polska, Warsaw, Poland; 3Les Laboratoires Servier, Neuilly sur Seine Cedex, France

OBJECTIVE: Depression is a common recurrent disease.Usually, each successive episode of depression is moredifficult to cure. The objective of this study is to assesscosts, methods of treatment, and outcomes of the first ep-isode of depression compared to the second and subse-quent episodes.METHODS: Study was designed as a one-year prospec-tive, natural observation. Psychiatrists who took part inthis trial recruited one patient with a first episode of de-pression, one with a second episode after a period of re-mission and one with a third or next episode after a pe-riod of remission. During the study, carried out in 1999with three months tolerance, data concerning treatmentmethods, their outcomes, changes in health state, andeconomic and social consequences of depression werecollected. A social perspective was applied. The capitalcost method was used for indirect costs.RESULTS: Five hundred thirty two patients were in-cluded, and 94% of them were followed to the end of thestudy (one-year observation). The average direct medicalcost per patient with a first episode of depression was2097 PLN (1 Euro � 3.50 PLN), with a second was 2241PLN and with a third and the following 3357 PLN. Di-rect medical costs included: the cost of hospitalization(49%); doctors consultations (29%); non-pharmacologi-cal treatment (14%); pharmacological treatment (8%);laboratory and diagnostic tests (1%). Indirect costs perpatient with a first episode reached 10,834 PLN, with asecond, 13,802 PLN, and with a third and following,15,012 PLN. The distribution of indirect cost was as fol-lows: sick leaves 40.5%, sickness pensions 59%, and sui-cides 0.5%. Total costs in the three study groups were12,931 PLN, 16,043 PLN and 18,369 PLN respectively.

CONCLUSIONS: Treatment of a third episode of de-pression costs 60% more than the first episode. Directmedical costs account for only 16% of total costs.

MH6

MODELLING THE COSTS OF ILLNESS AND THE COSTS OF RELAPSE IN THE MANAGEMENT OF SCHIZOPHRENIA IN THE UKLaunois R1, Présenté G1, Hansen K2, Toumi M3

1Rees France, Paris, France; 2Lundbeck, Paris, France; 3Lundbeck SA, Paris, France

OBJECTIVE: Schizophrenia is probably one of the mostcostly mental-health illnesses in terms of its impact on thesociety, on the health-care system and on patients andtheir families. We aim to develop a framework for ana-lyzing schizophrenia through the estimation of the costsof illness as well as the costs of relapse in the manage-ment of schizophrenia in a UK care setting.METHODS: A model that utilizes a Markov structure waselaborated in order to calculate the lifetime cost of illnessand the principal cost driver in schizophrenia, the cost ofrelapse. The model included the main confounding factorssuch as compliance, dropout, and treatment location,which have been shown to be important in the literaturewhen constructing a model. The costs were estimated for a10-year period as well as for the lifetime of a patient, andconsider all types of treatment structures in the UK.RESULTS: The average 10-year cost of illness for a patientin a UK treating environment was estimated at £ 66 600,and the lifetime cost per patient at £ 98 510. As for the costof relapse of a schizophrenic patient, the 10-year cost was£ 51 630 and the lifetime cost per patient was £ 75 890.CONCLUSIONS: This model enables the identificationand the calculation of the main cost drivers in the treat-ment of schizophrenia. This model can therefore be usedto evaluate the cost implication of treatment options andpolicy choices, by generating budgetary provisions. Hence,this model can help decision-makers in policy-making formental health-care organizations in the management ofschizophrenia.

NEUROLOGICAL DISORDERS

ND1

INDIRECT COSTS DUE TO BACK PAIN IN THE UNITED STATESMychaskiw MA, Thomas III JPurdue University, West Lafayette, IN, USA

OBJECTIVE: Back pain is a leading cause of absence anddisability in the workplace. Costs associated with lostproductivity due to back pain are significant and may beas high as the costs of medical care for this condition.However, there has been limited study of the indirectcosts of back pain in the United States. The objective ofthis study was to determine the indirect costs due to backpain in the US population.

Abstracts 417

METHODS: A retrospective analysis of the 1996 portionof the Medical Expenditure Panel Survey (MEPS) wasconducted. The MEPS provided data from a nationallyrepresentative sample of 22,601 respondents and datafrom respondent’s medical care and health-insuranceproviders and employers. Data included medical condi-tions and employment information comprised of hourlyearnings, hours worked, and disability days. Back painpatients who incurred disability days were identified us-ing International Classification of Diseases (ICD-9-CM)codes determined by an expert panel of physicians andcoders as indicative of back pain and variables denotingdisability days. Indirect costs were calculated for backpain patients who missed workdays using the humancapital approach. Sample estimates were weighted andprojected to the population and 95 percent confidence in-tervals for estimates were calculated using the Taylor ex-pansion method.RESULTS: Total indirect costs for back pain patientswho missed workdays were $18,533,583,620. Mean in-direct costs were $4586 per back pain patient whomissed workdays (95% CI � $3852 to $5321). Relativeto the entire population, the mean indirect costs were$68.92 per person.CONCLUSIONS: With losses in productivity greater than$18.5 billion, indirect costs significantly contribute to thetotal cost of back pain. The indirect costs of back pain tolabor underscore the need to prevent back injury therebypromoting employee health and maintaining productivity.

ND2

ASSESSING THE QUALITY OF LIFE IN PARKINSON’S DISEASESpottke AE1, Reuter M1, Smala A2, Berger K2, Athen O3, Köhne-Volland R3, Meyer D3, Oertel WH1, Dodel RC1

1Philipps-University Marburg, Marburg, Germany; 2MERG - Medical Economics Research Group, Munich, Germany; 3Metronomia, Munich, Germany

OBJECTIVES: To assess the influence of depression, de-mentia and disease severity on the health-related qualityof life in Parkinson’s disease (PD).METHODS: Health-related quality of life was assessedin 91 patients with PD (male: 62; female: 29; mean age:67 � 9) using the Parkinson’s disease questionnaire(PDQ-39) and the Parkinson’s disease Quality of lifequestionnaire (PDQL). Dementia was evaluated by usingthe MMSE and depression was rated with the Beck’s De-pression Inventory (BDI). Disease severity and motor dys-function were recorded with the Hoehn and Yahr scaleand the UPDRS II-IV. Pearson’s correlation coefficientwas used to calculate correlation scores.RESULTS: Both specific disease QoL rating scales corre-lated with depressive symptoms (r � 0.73, p � .0001).UPDRS III (r � 0.69, p � .0001) and HY stage (r � 0.59,p � .001) also correlated with decreased QoL. However,motor fluctuations and dyskinesias had a minor impacton QoL (r � 0.40, p � .001). Dementia or cognitive im-

pairment had a minor correlation with QoL (r � 0.30, p �.001).CONCLUSIONS: We found a high correlation betweenQoL and depression by using PDQ-39 and PDQL. Incontrast, a minor correlation was observed betweenMMSE scores and health-related QoL. Treating depres-sive symptoms in PD patients may considerably improvetheir quality of life.

ND3

COST-EFFECTIVENESS OF CABERGOLINE COMPARED TO LEVODOPA IN EARLY PARKINSON’S DISEASE IN SWEDENLindgren P1, Jönsson B2

1Stockholm Health Economics Consulting, Uppsala, Sweden; 2Stockholm School of Economics, Stockholm, Sweden

OBJECTIVE: 25% to 30% of patients with Parkinson’sdisease treated with levodopa present motor complica-tions after three years. The health-care consumption bypatients with these symptoms is roughly twice as high asthe costs for patients without them. A clinical trial com-paring cabergoline (with the possibility if adding levodopaif needed) to levodopa has shown that the use of caber-goline delays the onset of motor complications. The pur-pose of this study is to compare the cost-effectiveness ofthe two treatment strategies in a Swedish setting.METHODS: A Markov model was constructed withstates based on the Hoehn and Yahr (HY) stages with anadditional stage for patients with complications. Transi-tion probabilities between the HY stages were calculatedfrom the clinical trial data using ordered probit regres-sion, while the probability of developing complicationswas based on the frequency of this event at each timepoint in the trial. Direct costs were derived from a Swed-ish observational study. Years-of-complications-avoidedwas used as the measure of effectiveness in this analysis.RESULTS: The incremental cost-effectiveness ratio forcabergoline was 4000 USD per year of motor complica-tions avoided. Inclusion of indirect costs would influencethe results in younger patients. The literature suggeststhat the costs of indirect care would decrease the cost-ef-fectiveness ratio, so that the number reported here mayrepresent an overestimate of the total cost of preventingmotor complications.CONCLUSION: Based on our data, Cabergoline wasfound to represent a cost-effective means for reducing theimpact of motor dysfunction if the willingness to pay ex-ceeds 4000 USD per year of motor complications avoided.

POPULATION COST & HEALTH ISSUES

CH1

DEVELOPMENT OF A MINIMUM DATASET (MDS)-BASED COMORBIDITY INDEX FOR A NURSING HOME POPULATIONGilsenan ARTI Health Solutions, Research Triangle Park, NC, USA

418 Abstracts

OBJECTIVE: To develop two comorbidity indexes con-structed from the Minimum Dataset (MDS) disease infor-mation predictive of 18-month mortality and �2 hospi-talizations in a prevalence cohort of 1424 white, femalenursing-home residents that could be used as part of arisk adjustment method for quality-of-care outcomes.METHODS: A split-sample approach was taken for de-velopment and cross validation of the indexes. Multivari-ate logistic regression techniques were employed to iden-tify the MDS diseases most likely to predict each outcomeafter controlling for age in the development sample.Weights equal to the parameter estimates were assignedto each disease retained in the multivariate model to cre-ate a single comorbidity index variable for each outcome.Using the validation samples, the predictive validity ofthe MDS-based comorbidity indexes was determined andcompared to other measures of comorbidity (CharlsonIndex [ChI], Chronic Disease Score [CDS], count of dis-eases) as well as to Morris’ activities of daily living (ADL)index for each outcome.RESULTS: The MDS-based mortality index includedweighted variables for atherosclerotic heart disease (ASHD),dysrhythmias, congestive heart failure (CHF), respiratorydisease and depression and had a c-statistic of 0.62 in thedevelopment sample, and 0.60 in both validation samples.Similar c-statistics were found for the comparison of co-morbidity and functional status measures. The c-statisticsfor the MDS-based hospitalization index (with weightedvariables for hemiplegia, glaucoma, peripheral vasculardisease (PVD), CHF, diabetes and hypertension) were 0.66and 0.59 for the development sample and validation sam-ple respectively and were similar to those found for theother measures.CONCLUSIONS: Presence or absence of diseases mea-sured at baseline are weakly predictive of subsequenthospitalization or mortality in this cohort. Additional pa-tient risk factors such as disease severity, or change infunctional status should be investigated in future researchaimed at defining risk adjustment methods in this popu-lation.

CH2

TRANSITION IN HEALTH-RELATED QUALITY OF LIFE IN U.S. ELDERLYZhang JXThe Chinese University of Hong Kong, Shatin, Hong Kong

OBJECTIVE: Health-related quality of life (HRQOLs)are important health-status indicators. The longitudinalaspect of transitions in HRQOL is not well known. Thisstudy examines how HRQOL changes over time using aMarkov probability model.METHOD: We analyzed 10,492 elderly persons aged 65years or older and followed from 1991 to 1994 in MCBS,a nationally representative sample of the US elderly pop-ulation. We measured HRQOL by global health percep-tion, Katz Index of Activities of Daily Living (ADL), and

Instrumental Activities of Daily Living (IADL). We esti-mated the probability of transition from one state ofHRQOL to another using a Markov probability model.We investigated the transition matrix of HRQOL by in-corporating the competing risk of death and socio-demo-graphic variables including age, race, gender and the in-teraction terms among them.RESULTS: Including death as a worst outcome, 1411persons (13%) had a steady deterioration in global healthperception in the four-year time period. One thousandone hundred six (1106) (11%) had steady improvementof health perception, 1047 (10%) had a constant state,and 6928 (66%) experienced fluctuations in health per-ception. Similar patterns were found in ADL and IADLmeasures. The transition probabilities for a female with abase age of 70 indicated that one had 34% to 44%chance of remaining in the same state of general health,while the probability steadily decreases as the health per-ception decreases. A male aged 70 years has a probabilityof 32% to 53% to remain at the same state. As HRQOLdecreases by one unit, the probability of death increasessteadily from less than 1% to 12% for a female and from5% to 21% for a male across all HRQOL measures.CONCLUSION: The study shows the patterns in transi-tion of HRQOL in the elderly population. The probabili-ties of transition in three measures of HRQOL are gener-ally consistent and interpretable.

CH3

TRENDS IN POPULATION-BASED HEALTH-RELATED QUALITY OF LIFE: IMPLICATIONS FOR INTERPRETATION OF CHANGEErickson PPennsylvania State University, State College, PA, USA

OBJECTIVES: This analysis presents trends in thehealth-related quality of life of the US population from1984 to 1994 and examines the hypothesis that fluctua-tions in population health levels are related to changes ineconomic status.METHODS: Data on health status and quality of life aswell as personal and household characteristics are fromthe US National Health Interview Survey and the US Vi-tal and Health Statistics System. To examine the relation-ship between health and economic well being, these dataare linked to economic data, including medical care in-puts.RESULTS: Health-related quality of life and quality-ad-justed life years rose from 1984 to 1990 and declined af-ter this period. Specifically, for the total population,quality-adjusted life years rose from 63.4 in 1984 to 64.5years in 1990 and declined to 64.0 in 1994. Similar pat-terns were observed across genders and ethnic subgroups.The changes in quality of life and mortality coincidedwith periods of economic growth and recession over thesame interval. Regression analyses indicate that per cap-ita income levels and unemployment rates are important

Abstracts 419

in explaining variations in population health within agiven year as well as across years within this time period.Population health levels are further analyzed to examinewhether changes are related to personal and householdcharacteristics, economic environment, and availabilityof medical care inputs.CONCLUSIONS: These analyses indicate that changes inhealth-related quality of life can occur due to factorsother than health-care interventions, a finding which hasimportant implications for the selection and use of nor-mative data for interpretation of findings.

RESPIRATORY DISEASE/DISORDERS

RP1

DOES ILLNESS-RELATED LOST PRODUCTIVE WORK TIME VARY BY JOB DEMAND OR JOB CONTROL: A CASE STUDY OF ALLERGIC RHINITIS (AR)Stewart WF, Ricci J, Leotta CR, Chee EAdvancePCS, Hunt Valley, MD, USA

OBJECTIVE: To determine if lost work time differs byjob demand and control among workers with AR. Workloss estimates were derived for missed workdays, missedhours, and reduced productivity on days at work whilenot feeling well. METHODS: Three different phone in-terviews were developed to quantify illness-related workloss. Two different recall periods, at one week and fourweeks, were used for each interview. A convenience sam-ple (n � 20,088) of adult residents from the Baltimore,MD and Chicago, IL areas was contacted by phone, ofwhom 7,691 met occupation eligibility criteria. The inter-views included questions on job demand and control, andhealth questions to screen for AR symptoms and deter-mine AR severity.RESULTS: Mean lost work time per week was calculatedamong respondents with jobs categorized as high demand-high control (HH), high demand-low control (HL), lowdemand-high control (LH), and low demand-low control(LL). Among AR cases (n � 1596), job demand-controlwas significantly (p � .05) related to the number of missedworkdays per week for health reasons, but not to the meantotal hours of lost productive time per week at work whilenot feeling well. AR cases in HL jobs missed significantlymore workdays for health reasons than non-AR controls(1.3 days versus 0.38 days; p � .0001). This was not ob-served for AR cases in the other demand-control job cate-gories, or for mild AR cases in HL jobs. Among moderate-to-severe AR cases in HL jobs, the mean number of missedworkdays increased to 1.74 days per week.CONCLUSION: AR symptoms are associated with lostwork time. However, the way in which work time is lostappears to vary with the level of job demand and control.Given little latitude to adjust work performance to ac-commodate AR symptoms, individuals in high demand-low control jobs may be more likely to miss work.

RP2

THE NET COST OF ASTHMA TO A US PUBLIC MEDICAID PROGRAMPannicker S1, Nielsen K2, Martin BC1

1University of Georigia, Athens, GA, USA; 2GlaxoSmithKline, RTP, NC, USA

OBJECTIVE: To determine the net direct medical costand utilization patterns of asthma in a public Medicaidprogram.METHODS: A retrospective matched case-control designwas used. Administrative claims data were obtained forpersons eligible for Medicaid benefits in North Carolinafrom May 1996 through April 1998. Asthma cases weredefined as persons who had at least one inpatient claimfor asthma (ICD-9-CM � 493.) or had two or more out-patient asthma claims in the first year. A group of controlsubjects matched 1 to 1 on age, gender, and race was ob-tained. Persons who had a diagnosis for emphysema,COPD, or cancer or who were not between 1 and 65years of age were excluded. Net costs were determinedusing the second year (May 1997 to April 1998) follow-ing the recruitment period. Adjusted differences were ob-tained from a generalized least square model using HuberWhite heteroscedasticity consistent variance–covariancematrix with terms to control for comorbidity burden,Medicare eligibility, and duration of eligibility.RESULTS: Thirteen thousand two hundred fifty fiveasthma cases were identified with a mean age of 16 years,50% of who were black, and 56% were female. Asthmacases had a higher comorbidity burden than controls,particularly for diabetes, psychoses, hypertension, andanemia. The average total annual cost was $3226 forasthma cases compared to $2015 (p � .05) for controls.This translated to a net un-adjusted annual cost of $1210or $111 per member per month (PMPM). After multi-variate adjustment, the net annual cost of asthma was es-timated at $821 and $77 PMPM. Twice as many caseshad emergency room visits and inpatient hospitalizationsthan did controls and prescription expenditures wereover 100% higher than controls.CONCLUSION: Caring for indigent persons withasthma costs approximately 50% more than for similarpersons without asthma. Inpatient and prescription ex-penditures account for approximately 50% of the addi-tional expense.

RP3

PREPARING THE ASTHMA SYMPTOM UTILITY INDEX FOR INTERNATIONAL USE: TRANSLATION AND LINGUISTIC VALIDATIONFlood EM1, Eremenco SL2, Schmier JK1, Mörk AC3, Stahl E3, Arnold B2, Hudgens S2, Leidy NK1

1MEDTAP International, Inc, Bethesda, MD, USA; 2Evanston Northwestern Healthcare, Evanston, IL, USA; 3AstraZeneca R&D, Lund, Sweden

420 Abstracts

OBJECTIVE: The Asthma Symptom Utility Index (ASUI) isan 11-item asthma-specific health utility instrument devel-oped in the US. The purpose of this study was to translateand linguistically validate the ASUI for international use.METHODS: Seventeen languages were identified: Bul-garian; Chinese (traditional and simplified); Dutch; Finn-ish; French; Hebrew; Hungarian; Indonesian; Italian;Malay; Norwegian; Portuguese; Romanian; Spanish;Swedish; Turkish. The following method was used foreach translation: two forward translations by nativespeakers; one reconciliation of the two forward transla-tions; one back-translation by an English-speaker fluentin the target language; three reviews by independent bi-lingual experts, and final reconciliation by a native-speaking language coordinator. Pilot-testing and cogni-tive debriefing interviews were performed with 10 to 15native-speaking asthma patients in 19 countries. Item andtotal score distributions and internal consistency levels

were examined. US-derived item weights were used to es-timate utility scores.RESULTS: A total of 290 patients participated in theevaluation. Mean age of the sample was 48.1 (� 16.8)years, and 60% were female. Internal consistency esti-mates (alpha) ranged from 0.74 to 0.90 (mean � 0.86).Mean utility scores ranged from 0.42 (� 0.20) for pa-tients in Norway (Norwegian; n � 10) to 0.88 (� 0.12)for patients in Israel (Hebrew; n � 15). Consistent withthese scores, the Norwegian sample also experiencedmore severe symptoms, with more than half reportingmoderate to severe wheezing (5/10) or dyspnea (8/10)during the previous two weeks compared with 2/15 and2/15 in the Israeli group, respectively.CONCLUSIONS: Results suggest the 17 ASUI transla-tions are culturally acceptable, linguistically valid, andconceptually equivalent to the original, paving the way forinternational asthma-specific, utility-estimation studies.