Potential Savings in the Cost of Caring for Alzheimer’s Disease

Potential Savings in the Cost of Caringfor Alzheimer’s DiseaseTreatment with Rivastigmine

A. Brett Hauber,1 Ari Gnanasakthy,2 Edward H. Snyder,3 Mohan V. Bala,4 Anke Richter1

and Josephine A. Mauskopf1

1 Research Triangle Institute, Research Triangle Park, North Carolina, USA2 Novartis Pharma AG, Basel, Switzerland3 Novartis Pharmaceuticals Corporation, East Hanover, New Jersey, USA4 Centocor, Inc., Malvern, Pennsylvania, USA

Abstract Objective: To estimate savings in the cost of caring for patients with Alzheimer’sdisease (AD) during 6 months, 1 year and 2 years of treatment with rivastigmine.An intermediate objective was to estimate the relationship between disease pro-gression and institutionalisation.Design and setting: We assessed the relationship between Mini-Mental StateExamination (MMSE) score and institutionalisation using a piecewise Cox pro-portional hazard model. To estimate cost savings from treatments lasting 6months, 1 year and 2 years, estimates of the probability of institutionalisationwere integrated with data from two 6-month phase III clinical trials of rivastigm-ine and a hazard model of disease progression.Main outcome measures and results:Our data suggest that savings in the over-all cost of caring for patients with mild and moderate AD can be as high as$US4839 per patient after 2 years of treatment. Furthermore, the probability ofinstitutionalisation increases steadily as MMSE score falls. Among our studyindividuals, age, race, level of education and marital status were significant pre-dictors of institutionalisation, whereas gender had little effect.Conclusions: Using rivastigmine to treat AD results in a delay in disease pro-gression for patients who begin treatment during the mild or moderate stages ofthe disease. By delaying the probability that a patient will be institutionalised,the cost of caring for AD patients can be significantly reduced.

ORIGINAL RESEARCH ARTICLE Pharmacoeconomics 2000 Apr; 17 (4): 351-3601170-7690/00/0004-0351/$20.00/0

© Adis International Limited. All rights reserved.

Alzheimer’s disease (AD) is a degenerative, of-ten fatal, disease of the brain, and is the most com-mon cause of dementia in the US. Treatment of ADplaces a considerable economic burden on the UShealthcare system, particularly when the patient isinstitutionalised. Cholinesterase inhibitors such asrivastigmine, tacrine and donepezil have been shownto have statistically and clinically significant ef-

fects on cognition.[1-6] Drug therapies that delaycognitive decline may result in significant savingsin the cost of caring for AD patients. The primaryobjective of this paper is therefore to estimate po-tential savings in the cost of caring for a repre-sentative AD patient in the US (exclusive of drugtreatment cost) when the transition to more severestages of AD is delayed by rivastigmine.

The instrument used to measure cognitive de-cline is the patient’s score on theMini-Mental StateExamination (MMSE).[7] We model the efficacy ofrivastigmine in delaying decline in a patient’sMMSE score with a parametric hazard model ofAD progression developed by Fenn and Gray[8]that uses data from two 26-week phase III clinicaltrials of rivastigmine efficacy.[1,6] We use thismodel to estimate the delay in cognitive declineattributable to rivastigmine over 3 time horizons: 6months, 1 year and 2 years.We then estimate the total cost of caring for AD

patients as a function of the patients’MMSE scoresusing a weighted average of the cost of caring fora patient in the community and the cost of caringfor a patient in an institution. The cost data arederived from Ernst et al.[9] The weights used inthese estimates are derived from a model of therelationship between a patient’s MMSE score andthe probability that the patient will be institu-tionalised, estimated using data from the Consor-tium to Establish a Registry for Alzheimer’s Dis-ease (CERAD). We then combine the increase incost associated with the transition to a more severestage of AD with the results of the model of cogni-tive decline to estimate the potential cost savings(exclusive of drug treatment costs) attributable torivastigmine.

Methods

Hazard Model of Cognitive Decline

Duration models of disease progression can beused to estimate the length of time during which apatient remains at his or her current state of health.In this section we describe a hazard model of ADprogression (measured by cognitive decline) de-veloped by Fenn and Gray[8] that is based on two26-week phase III clinical trials of rivastigmine.We use this model to estimate the duration (mea-sured in days) over which a patient is likely toremain at his or her current disease stage (mild,moderate or severe) with rivastigmine and withplacebo. We describe this model below. Further de-tails are available in Fenn and Gray.[8]

We measure a patient’s health state by theirscore on the MMSE. MMSE scores range from 0to 30, with 30 being a perfect score. MMSE as ameasure of disease progression was chosen for thisstudy primarily because of its widespread use as ameasure for cognitive changes. The 2 studies uponwhich our study draws[8,9] use this measure tomodel patient decline, the cost of care and the prob-ability of institutionalisation. Although other cog-nitive measures such as the Alzheimer’s DiseaseAssessment Scale-Cognitive Subscale (ADAS-Cog), the Global Deterioration Scale (GDS) andthe Blessed Information-Memory-ConcentrationTest (BIMC) might also have been used, MMSE isthe only assessment measure collected in both theclinical trials for rivastigmine and the CERADdatabase.In order to estimate the model, one must deter-

mine the timing of each 1-point drop in MMSEscore. Since a patient’s MMSE score is unlikely tohave dropped on exactly the day the examinationwas administered, Fenn and Gray[8] take observedMMSE scores from 2 consecutive examinations foreach patient in the clinical trials, and linearly inter-polate the date at which each 1-point drop oc-curred.To model a patient’s progression from one

MMSE score to the next lower MMSE score, Fennand Gray[8] use an accelerated failure time modeland assume a Weibull distribution for the hazard.For reasons of interpretation, themodel is specifiedin terms of survival functions rather than hazardfunctions. That is, rather than estimating the like-lihood that an individual will experience a drop inMMSE score at any given point in time, the modelestimates the likelihood that a patient will remainat his or her currentMMSE score at any given pointin time.The survival function takes the form:

S = exp(−(λt)α) (Eq. 1)

where λ, the scale parameter, reflects the averageprobability of transition and is specified as a log-linear function of patient age, MMSE score at thestart of the trial and treatment group (high or low

352 Hauber et al.

© Adis International Limited. All rights reserved. Pharmacoeconomics 2000 Apr; 17 (4)

dosage), and α, the shape parameter, determinesthe extent to which the hazard rises or falls overtime.Patient MMSE scores at the start of the trials

(baseline) ranged from a high of 26 to a low of 10.The maximum drop in MMSE score experiencedby a patient during the 26-week trials was 5points.[1,6] Therefore, Fenn and Gray[8] estimate afamily of 5 survival curves for each baselineMMSE score for patients in the rivastigmine treat-ment group, and a corresponding family of sur-vival curves for patients in the placebo arm of thetrial. Each curve in the set reflects the likelihoodthat a patient with a given baseline MMSE will notexperience an x-point drop in MMSE (where x = 1to 5) at a given point in time. A sample set of thesesurvival curves is presented in figure 1.We can interpret the vertical distance between

each survival curve as the proportion of the trialpopulation to whom this family of curves appliesthat will experience a 0-, 1-, . . ., >5-point drop inMMSE score at any given time. For example, at182 days, approximately 52% of these patients willhave experienced no decline in MMSE score fromthe baseline. During the same time period, approx-

imately 15% (67 to 52%) will have experienced a1-point decline from baseline, and so on. The ver-tical distance between the highest survival curveand the top of the figure at 182 days represents theproportion of these patients who will have experi-enced a decline in MMSE score of ≥5 points bythe end of the trial time period.Having estimated survival curves for each base-

line MMSE score for each of the 2 treatmentgroups, one can then compare the likelihood ofavoiding an x-point drop in MMSE score betweentreatment groups. This comparison is illustrated infigure 2.The higher survival curve in figure 2 corre-

sponds to that of a patient in the treatment arm ofthe trial, and the lower curve corresponds to a sim-ilar patient in the placebo arm of the trial. The factthat the survival curve is higher for the patient inthe treatment group indicates that, as a result oftreatment, this patient has a greater likelihood ofavoiding an x-point drop in MMSE score than hasa similar patient in the placebo group. The verticaldistance between the 2 survival curves at any givenpoint in time can be interpreted as the increasedprobability that a treated patient will not experi-

0.20 100 200 300 400

0.4

0.6

0.8

1.0

Like

lihoo

d of

avo

idin

g dr

op in

MM

SE

Time (days)

MMSE −1MMSE −2MMSE −3MMSE −4MMSE −5

Fig. 1. Survival to Mini-Mental State Examination (MMSE)thresholds (from Fenn & Gray[8]). Vertical dotted line indicatestrial duration (182 days).

0 500 1000

0.5

1.0

Like

lihoo

d of

avo

idin

g dr

op in

MM

SE

Time (days)

PlaceboHigh dosage therapy

Fig. 2. Treatment effect (from Fenn & Gray[8]). Vertical dottedline indicates trial duration (182 days). MMSE = Mini-MentalState Examination.

Rivastigmine in Alzheimer’s Disease 353


ence an x-point decline in MMSE score by thattime, relative to the untreated patient. Therefore,the area between the curves represents the numberof additional days a patient avoids the x-point de-cline in MMSE score as a result of treatment.Stated differently, the area between the 2 curves isa measure of the delay in cognitive decline result-ing from patient treatment with rivastigmine.Recall that the objective of our study is to esti-

mate the savings associated with delaying the tran-sition to more severe stages of AD. We define 3disease stages (mild, moderate and severe) as fol-lows: (i) mild: 21 ≤MMSE ≤30; (ii) moderate: 11≤MMSE ≤20; (iii) severe: 0 ≤MMSE ≤10.Fenn and Gray[8] then estimate a full set of sur-

vival curves for each baseline MMSE score foreach treatment group. A full set of survival curvesconsists of the full set of curves necessary to modelthe likelihood of successive drops in MMSE scorefrom the baseline disease stage through the severestage of AD. By estimating the average number ofdays saved for each baseline cohort for each dis-ease stage threshold, Fenn and Gray estimate thetotal delay in cognitive decline that results fromtreatment.The model specification, while based on the re-

sults of the clinical trials, assumes that the efficacyof treatment lasts beyond the 26-week trial period.In particular, the estimated proportional differencein the likelihood of decline between the treatmentand placebo groups is assumed to persist over allperiods greater than 26 weeks. When this frame-work is used, the effects of treatment do not lastindefinitely. As time passes and the disease pro-gresses, all patients are expected to enter the mostsevere stage of AD and, at that point, the propor-tional treatment effect declines to zero.

Cost of Alzheimer’s Disease as a Function ofMMSE Score

We estimate the cost of care for a patient with agiven MMSE score as a weighted average of thecost of living in the community for a given MMSEscore and the cost of institutionalisation. Thereforethe cost of care is expressed as:

C(M) = P(M) • CI(M) + [1 − P(M)] • CC(M) (Eq. 2)

where C(M) is the total cost of care for a patient forwhomMMSE =M, CI(M) is the cost of institutionalcare for a patient for whom MMSE = M, CC(M) isthe cost of community care for a patient for whomMMSE = M, and P(M) is the probability of insti-tutionalisation for a patient for whomMMSE =M.

US Cost EstimatesErnst et al.[9] find that the cost of caring for a

patient is only significantly related to MMSE scoreif the patient is living in the community; that is, fora patient living at home, the cost of care rises asMMSE score falls. They estimated this relationship(in 1993 $US) by regressing the natural logarithmof cost on MMSE score for patients living in thecommunity. Their results are:

E[CC(M)|MMSE] = exp(9.0339 − 0.0799 • MMSE)(Eq. 3)

They estimate the annual cost of institutionalcare (in 1993 $US) [CI(M)] to be $US30 284.

Probability of InstitutionalisationOnce we have estimated the cost of caring for a

patient living in the community as a function ofMMSE score, we must estimate the likelihood thata patient with a given MMSE score will actually beliving in the community. To do this, we use a piece-wise Cox proportional hazard model and data fromCERAD.

Adjusting Observed MMSE ScoresIn our model of the probability of institution-

alisation, we must determine each institutionalisedpatient’s MMSE score at the time of institution-alisation. To do this, we must adjust the last knownMMSE score for each institutionalised patient.This correction is necessary because the patient’slast MMSE examination was often administeredwell before the actual date of institutionalisation.During this elapsed time, the MMSE score is ex-pected to have decreased.The medical literature on AD reports an overall

decrease in MMSE score over time, as shown intable I. Although the decreases reported in each ofthe studies are similar, they all have high standard

354 Hauber et al.


deviations associated with them because of indi-vidual variation in cognitive decline and the possi-ble nonlinear progression of the disease. We calcu-lated the average decline in MMSE score for theAD patients in CERAD and obtained a meanchange of 3.1 points per year with a standard devi-ation of 3.9. As table I shows, this calculation iswell within the reported ranges. On the basis of thisresult, we adjust the MMSE scores for institu-tionalised patients by the average decline of 3.1points per year. In addition, because not everyonein the CERAD database has been institutionalised,the data set is censored. Those people who diedwithout being institutionalised were also consid-ered to be censored data points. Therefore, we es-timate our model with censored data.

Piecewise Cox Proportional Hazard ModelWe model the probability of institutionalisation

with a Cox proportional hazard model. This modelspecifies the hazard rate at time t for a given cov-ariate vector X as:

h(t) = h0(t)eβX (Eq. 4)

where h(t) is the hazard rate at time t when thecovariate is X, h0(t) is the hazard rate at time t whenthe covariate is 0, X is the vector of covariates, and βis a vector of unknown parameters to be estimated.In this model, the predicted hazard, h(t), is the

likelihood that an individual will be institu-tionalised at time t for a given value of the cov-ariates. The ‘time’ parameter is the patient’sMMSE score and, as mentioned above, the ‘event’

we are interested in is the patient’s institutionalisa-tion. The explanatory variables for each of the 758observations in our data include the following:• married: a dummy variable equal to 1 if the pa-tient is married, 0 otherwise

• education: an integer variable indicating thenumber of years of formal education, rangingfrom 0 to 26

• race: a dummy variable equal to 1 if the patientis White, 0 otherwise

• gender: a dummy variable equal to 1 if the pa-tient is male, 0 if female

• age: a dummy variable equal to 1 if the patientwas 72 years (the mean) or older and equal to 0if the patient was younger than 72 years at entryinto the study.Summary statistics for these data are presented

in table II.The Cox proportional hazard model described

above assumes that the ratio of the 2 hazard rates[h(t)/h0(t)] is independent of MMSE score. Wetested this assumption and found that it is notvalid in our case. Therefore, we estimate a piece-wise Cox proportional hazard model, dividing the

Table I. Average yearly decline in Mini-Mental State Examination (MMSE) score

Study n Base MMSE scorea Mean annual rate of change (points/y)a

Uhlmann et al.[10] 120 17.2 ± 7.4 2.2 ± 5.0Becker et al.[11] 44 21.3 ± 5.2 1.8b

Yesavage et al.[12] 30 16.5 ± 7.6 4.2 ± 4.8Salmon et al.[13] 55 18.5b 2.8 ± 4.3Teri et al.[14] 106 18.2 ± 6.7 2.8 ± 4.6Burns et al.[15] 96 10.0 ± 5.9 3.5b

CERAD database 1004 12.6 ± 7.4 3.1 ± 3.9a Mean ± standard deviation.b No standard deviation available.CERAD = Consortium to Establish a Registry for Alzheimer’s Disease.

Table II. Summary statistics for Consortium to Establish a Registryfor Alzheimer’s Disease (CERAD) data used to estimate the prob-ability of institutionalisation

Variable Mean Standard deviationAge (>72 years) 0.5264 0.4996Race (White = 1) 0.8654 0.3415Gender (male = 1) 0.4367 0.4963Married (= 1) 0.6887 0.4633



MMSE scores into 3 distinct ranges: 0 to 9, 10 to19, 20 to 30. The piecewise Cox model estimates adifferent coefficient vector β for each range. Thisassumes that the constant proportionality betweenthe 2 hazard rates holds only within each of the 3ranges and not for the entire range of MMSEscores. We calculate the eβX for each MMSE range,and then calculate the appropriate hazard rate foreach MMSE.Table III presents the estimated parameters of

the piecewise Cox proportional hazard model. Inall MMSE ranges some combination of age, raceand marital status is significant at the 0.05 level.For patients whose MMSE score ranged between 0and 9 (severely impaired), marital status is negativeand significant. This implies that, of severely im-paired patients, those who are married are lesslikely to be institutionalised. For patients whoseMMSE score ranged from 10 to 19, both age andrace are significant. The coefficients associatedwith these variables were both positive, implyingthat, of moderately impaired patients, those whoare White or over the age of 72 years are morelikely to be institutionalised. For patients whose

MMSE scores range between 20 and 30, race andmarital status are significant.The resulting probabilities of institutionalisa-

tion at each MMSE score, evaluated at the means ofthe explanatory variables, are presented in table IV.Using cost of care estimates from Ernst et al.[9]

(adjusted to 1997 $US), along with the proba-bilities of institutionalisation from table IV, we es-timate the expected annual cost of care as a func-tion of MMSE score. These costs are presented incolumn 1 of table V. We take the average of thecosts of care over the relevant MMSE range as ourestimate of the cost associated with each diseasestage. These estimates are presented in column 2 ofthe table. Furthermore, knowing the expected costof care at each disease stage, we can calculate theincremental cost associated with the transition to amore severe stage of AD as the difference betweenthe costs of care associated with 2 consecutive dis-ease stages. These incremental cost estimates arepresented in column 3 of the table.By converting the annual incremental cost of

moving from one disease stage to the next intodaily incremental cost, and multiplying these sav-ings by the number of days by which the transitionis delayed, we can determine the total cost savings

Table III. Results of the piecewise Cox proportional hazard model

Variable Coefficient

0 ≤≤≤≤MMSE ≤≤≤≤9Age (>72 years) 0.04Race (White = 1) 0.42Gender (male = 1) 0.13Married (= 1) −0.40*

10 ≤≤≤≤MMSE ≤≤≤≤19Age (>72 years) 0.45*Race (White = 1) 4.27**Gender (male = 1) −0.06Married (= 1) −0.17

20 ≤≤≤≤MMSE ≤≤≤≤30Age (>72 years) 0.87Race (White = 1) 9.14**Gender (male = 1) 0.25Married (= 1) −1.26*MMSE = Mini-Mental State Examination. * indicates statisticallysignificant at the 5% level or better; ** indicates statistically signifi-cant at the 1% level or better.

Table IV. Probability of institutionalisation [P(M)] at each Mini-Mental State Examination (MMSE) score, calculated at the meanof the explanatory variables

MMSE score P(M) MMSE score P(M)30 0.0000 15 0.323729 0.0000 14 0.387628 0.0000 13 0.461227 0.0000 12 0.575226 0.0000 11 0.759825 0.0041 10 0.774924 0.0090 9 0.787223 0.0239 8 0.802322 0.0331 7 0.818921 0.1005 6 0.830720 0.1320 5 0.853019 0.1507 4 0.873518 0.1767 3 0.890617 0.2068 2 0.910916 0.2774 1 1.0000

0 1.0000

356 Hauber et al.


resulting from treatment. By repeating this calcu-lation for each transition (mild to moderate, mod-erate to severe) for a given family of curves, wecan then calculate the total cost savings from treat-ment during the 6-month, 1-year and 2-year timeperiods.

Results

In this section, we present the results of the analy-ses described in the Methods section. We begin bypresenting estimates of the delay in cognitive de-

cline over the 6-month, 1-year and 2-year time hor-izons. We conclude this section by estimating thetotal cost savings attributable to this delay in cog-nitive decline.

Delay in Cognitive Decline

Using Fenn andGray’smodel[8] described in theMethods section, we first estimate the number ofdays by which cognitive decline is delayed (fig. 2)for each time horizon: 6 months, corresponding to

Table V. Weighted average cost estimates by Mini-Mental State Examination (MMSE) scorea

Disease stage MMSE score Annual total cost ($US) Average annual total cost ($US) Incremental costb ($US)Mild 30 836

29 90528 98027 1 06226 1 150 173425 1 24624 1 66823 2 09622 2 53121 4 861

Moderate 20 5 92919 6 68718 7 75917 8 83216 11 131 13 147 11 41415 12 50014 14 76913 17 01612 20 72111 26 129

Severe 10 26 4969 27 1498 27 5117 28 1506 28 503 29 476 16 3285 29 1244 29 7313 30 3222 30 8951 33 1770 33 177

a The Consumer Price Index was used to adjust all costs to 1997 $US.b Incremental cost for a given disease stage (S) is equal to the average total cost of the next disease stage [C(S+1)] minus the total cost

for the given disease stage [C(S)].



the trial time period, 1 year and 2 years. These re-sults are presented in table VI.The first column in table VI orders the results

by the cognitive status of each patient at the startof the trials. As noted above, the trials includedonly patients with baseline severities of mild ormoderate. The third, fourth and fifth columns oftable VI indicate the number of days by which eachtransition to a more severe stage of AD is delayedby treatment over each of the 3 time horizons. TableVI makes explicit the fact that patients in the mildbaseline cohort first experience a delay in the tran-sition to the moderate stage and then experience asubsequent delay in the transition from the moder-ate stage to the severe stage.The results presented in table VI reveal only a

modest delay in cognitive decline during the first6 months of treatment. In particular, patients whobegin treatment while in the mild stages of AD ex-perience an average delay in the onset of the mod-erate stage of AD of only 4 days. The results aresimilar for patients with moderate disease over thissame time horizon. Over the course of 1 year, pa-tients in the moderate baseline cohort are predictedto experience an average delay in the transition tothe severe stage of 22 days, while patients in themild baseline cohort are expected to experience adelay in the transition to the moderate stage of 25days. Likewise, those patients in the mild baselinecohort who would otherwise be expected to declineto the severe stage of the disease first experience25 additional days in the mild category and then anadditional day in the moderate stage, for a totaldelay of 26 days over the course of 1 year.The greatest delay in cognitive decline is expe-

rienced by the mild cohort at 2 years. In particular,this group of patients can expect to first put off thetransition to the moderate stage of AD for an aver-age of 56 days (approximately 8 weeks) and then

delay the subsequent transition to the severe stageby 69 days (nearly 10 weeks), for a total delay incognitive decline of almost 18 weeks over this timeperiod.

Cost Savings

Next, we combine the estimates of delayed cog-nitive decline with the incremental cost of movingto the next AD stage to determine total cost savings(independent of drug treatment cost) for each base-line cohort over each time horizon. We present ourestimates of total cost savings in table VII.The results presented in table VII reveal that

cost savings during the first 6 months of treatmentmay be only $US137 for patients with moderateAD and as little as $US132 for patients in the mildbaseline cohort. Likewise, total cost savings duringthe first year are predicted to be greater for thosepatients who begin treatment in the moderate stageof AD. These observations result from the fact thatpatients with lower baselineMMSE scores (greaterdisease severity) face a higher probability of insti-tutionalisation and a higher expected cost of careat baseline. Therefore, any delay in cognitive de-cline is expected to yield greater benefits than thesame delay experienced by a patient with a higherbaseline MMSE score.Most potential cost savings are realised begin-

ning at 2 years. The average total cost savings for

Table VI. Estimates of the number of days by which cognitive decline is delayed for each baseline cohort

Baseline severity Days saved 6 months 1 year 2 yearsMild Additional days in mild 4 25 56

Additional days in moderate 0 1 69Moderate Additional days in moderate 3 22 51

Table VII. Total cost savings as a result of delayed cognitivedeclinea

Disease severity atbaseline

6 months($US)

1 year($US)

2 yearsb

($US)Mild 132 836 4839Moderate 137 980 2290Average 134 907 3578a Estimates of cost savings are reported in 1997 $US.b Second-year cost savings are discounted at 3%.

358 Hauber et al.


the entire trial population are predicted to be$US3578 over this time horizon. Patients in themild baseline cohort are likely to realise cost sav-ings that are at least twice as great as those of pa-tients in the moderate baseline cohort. The relativecost savings between the 2 cohorts at 2 years differfrom those found at 1 year. In particular, it is themoderate baseline cohort that realises the greatercost savings at 1 year. This change in relative sav-ings between year 1 and year 2 can be attributedprimarily to the fact that the mild baseline cohortis expected to experience a much greater delay inthe onset of more severe stages of AD at 2 years(125 days) than the moderate cohort (51 days).Therefore, even though the incremental cost oftransition from mild to moderate is much less thanthat of the transition from moderate to severe (seetable V), the large number of days by which thetransition from mild to moderate is delayed resultsin greater cost savings for the mild cohort.

Discussion

The purpose of this study is to estimate potentialsavings in the cost of caring for Alzheimer’s pa-tients in the US (exclusive of drug treatment costs)attributable to rivastigmine. Our hypothesis is thatby delaying the progression of AD, rivastigminewill reduce the burden to society by delaying thecosts of care that accompany more severe stages ofthe disease. The largest direct cost of care in thelatter stages ofAD results from institutionalisation.Therefore, the primary component of the cost sav-ings attributable to delayed cognitive decline dur-ing the first year of treatment for moderate AD pa-tients is the reduction in expected expenditures forinstitutional care. For patients who begin treatmentin the mild stages of the disease, the greatest costsavings over time arise from the significant delayin the transition to the moderate and severe stagesof AD.These results indicate that, although rivastig-

mine may result in cost savings for moderately de-mented patients, the most cost-effective indicationmay be for patients in the early stages of AD. How-ever, it is important to note that the above conclu-

sion is dependent upon the life expectancy of pa-tients in each cohort, a feature which is not ad-dressed in this model. In particular, if patients haverelatively short life expectancies, greater overallcost savings may be achieved by treating only pa-tients with moderate disease.Although Fenn and Gray[8] extrapolate their re-

sult out to 3 years, we limited our study to the 2-year time horizon. Data from nonblind follow-uptrials validate the model results at 1 year. Specific-ally, Fenn and Gray find that using follow-up datafrom a nonblind study (255 patients who continuedhigh dosage treatment up to 52 weeks) does notinvalidate their estimated parameters. We believethat extrapolating the results beyond 2 years maynot be defensible, given only 1 year of validatingdata. The long term (>2 years) results, however, areworth mentioning. In particular, Fenn and Grayfind, as we do, that the cost savings from treatmentdo not become substantial until the 2-year timehorizon. In addition, they find that the cost savingscontinue to grow out to 3 years for patients whobegin taking rivastigmine in the mild stage of AD,whereas the cost savings for moderate patients be-gin to plateau after 2 years. This result again sug-gests that the appropriate indication for prescribingrivastigmine may be for patients in the early stagesof AD.One shortcoming of the model as it has been

applied in this study is that it may well understatethe cost savings associated with treatment in the6-month time horizon. This is in part an artifact ofthe way disease stages are defined. In particular,because the cost of caring for ADpatients increasesonly when a patient crosses the MMSE thresholdfrom one disease stage into the next, small declinesin MMSE score within a disease stage may resultin no increase in cost. Because only those who be-gin treatment with anMMSE score near a thresholdexhibit a transition from one disease stage to thenext in such a short time span, any cost savingsassociated with delaying the cognitive declinewithin a disease stage are ignored. Because, as thetime horizon increases, more patients exhibit atransition to a more severe stage of AD, the model



captures a larger portion of the potential cost sav-ings over longer periods of time. In addition, themodel may underestimate possible gains fromtreatment because it does not allow for the possi-bility of improvements in MMSE resulting fromtreatment.Caution is warranted in using the estimates pre-

sented in this study in a strict cost-benefit calculusbecause the costs estimated by Ernst et al.[9] andused in this study are derived from a relativelysmall sample of AD patients in one area of Califor-nia. Additional research may be needed to derivecost estimates corresponding to disease stagesfrom a larger, more diverse, patient population.

Acknowledgements

This study was funded by Novartis Pharmaceuticals Cor-poration, East Hanover, NJ, USA. Brett Hauber would alsolike to thank Paul Fenn of theUniversity of Nottingham (UK)and two anonymous referees for helpful comments. Dr Balawas employed at the Research Triangle Institute during thepreparation of this article.

References1. Corey-Bloom J, Anand R, Veach J, for the ENA713B352 Study

Group. A randomized trial evaluating the efficacy and safetyof ENA 713 (rivastigmine tartrate), a new acetylcholinester-ase inhibitor, in patients with mild to moderately severe Alz-heimer’s disease. Int J Geriatr Psychopharmacol 1998; 1:55-65

2. Farlow M, Gracon SI, Hershey LA, et al. A controlled trial oftacrine in Alzheimer’s disease. JAMA 1992; 268: 2523-9

3. Knapp MJ, Knopman DS, Solomon PR, et al. A 30-week ran-domized controlled trial of high-dose tacrine in patients withAlzheimer’s disease. JAMA 1994; 271: 985-91

4. Rogers SL, Friedhoff LT. The efficacy and safety of donepezilin patients with Alzheimer’s disease: results of a US multi-

centre, randomized, double-blind, placebo-controlled trial.Dementia 1996; 7: 293-303

5. Rogers SL, Farlow MR, Doody RS, et al. A 24-week, double-blind, placebo-controlled trial of donepezil in patients withAlzheimer’s disease. Neurology 1998; 50: 136-45

6. Rosler M, Anand R, Cicin-Sain A, et al., ENA 713 B303 StudyGroup. Safety and efficacy of rivastigmine in patients withAlzheimer’s disease: results of an international, 26-week,multicentre, randomised, placebo-controlled trial. BMJ 1999;318: 633-8

7. Folstein MF, Folstein SE, McHugh PR. Mini-mental state: apractical method for grading the cognitive state of patients forthe clinician. J Psychiatr Res 1975; 12: 189-98

8. Fenn P, Gray A. Estimating the long term cost savings from thetreatment of Alzheimer’s disease: a modelling approach.Pharmacoeconomics 1999; 16 (2): 165-74

9. Ernst RL, Hay JW, Fenn C, et al. Cognitive function and thecosts of Alzheimer’s disease: an exploratory study. Arch Neu-rol 1997; 54: 687-93

10. Uhlmann RF, Larson EB, Koepsell TD. Hearing impairmentand cognitive decline in senile dementia of the Alzheimer’stype. J Am Geriatr Soc 1986; 34: 207-10

11. Becker JT, Huff FJ, Nebes RD, et al. Neuropsychological func-tion in Alzheimer’s disease: pattern of impairment and ratesof progression. Arch Neurol 1988; 45: 263-8

12. Yesavage JA, Poulsen AB, Sheikh J, et al. Rates of change ofcommon measures of impairment in senile dementia of theAlzheimer type. Psychopharmacol Bull 1988; 24: 531-4

13. Salmon DP, Thal LJ, Butters N, et al. Longitudinal evaluationof dementia of the Alzheimer type: a comparison of 3 stand-ardized mental status examinations. Neurology 1990; 40:1225-30

14. Teri L, Hughes JP, Larson EB. Cognitive deterioration in Alz-heimer’s disease: behavioral and health factors. J Gerontol BPsychol Sci Soc Sci 1990; 45 (2): 58-63

15. Burns A, Jacoby R, Levy R. Progression of cognitive impair-ment in Alzheimer’s disease. J Am Geriatr Soc 1991; 39:34-45

Correspondence and offprints: A. Brett Hauber, ResearchTriangle Institute, Center for Economics Research, 3040Cornwallis Road, P.O. Box 12194, Research Triangle Park,NC 27709-2194, USA.

360 Hauber et al.


Documents

Potential Savings in the Cost of Caring for Alzheimer’s Disease