• cost-effectiveness analysis;
  • decision analysis;
  • Parkinson's disease;
  • systematic review


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Objective:  To give an insight into the structural and methodological approaches used in published decision-analytic models evaluating interventions in Parkinson's disease (PD) and to derive recommendations for future comprehensive PD decision models.

Methods:  A systematic literature review was performed to identify studies that evaluated PD interventions using mathematical decision models. Using a standardized assessment form, information on the study design, methodological framework, and data sources was extracted from each publication and systematically reported. Strengths and limitations were assessed.

Results:  We identified eight studies that used mathematical models to evaluate different pharmaceutical (n = 7) and surgical (n = 1) treatment options in PD. All models included economic evaluations. Modeling approaches comprised mathematical equations, decision trees, and Markov models with a time horizon ranging from 5 years to lifetime. All based progression on the evolution of clinical surrogate endpoints. Treatment effects were either modeled via reduction of symptomatic progression and/or initial symptomatic improvement or via reduction of adverse effect rates. No model is currently available that encompasses both the underlying biologic disease progression and the spectrum of all relevant complications and also links them to patient preferences and economic outcomes.

Conclusions:  Models have been successfully applied to evaluate PD treatments. However, currently available models have substantial limitations. We recommend that a comprehensive, generic, and flexible decision model for PD that can be applied to different treatment strategies should consider a large spectrum of clinically relevant outcomes and complications of the disease during a sufficiently long time horizon, include PD-specific mortality, systematically evaluate uncertainty including heterogeneity effects, and should be validated by independent data or other models. Approaches to model treatment effects included reduction of symptomatic progression, initial symptomatic improvement, or reduction of adverse effects. We believe that structural bias could be avoided if underlying disease progression and treatment effects on symptoms are modeled separately.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Parkinson's disease (PD) is one of the most common neurodegenerative disorders with a prevalence of 66 to 258 per 100,000 [1–5]. The clinical picture is characterized by (resting) tremor, slowness of movement (bradykinesia), rigidity, and impairment of balance reflexes (postural instability). One crucial milestone during disease progression is the advent of motor complications [6], which occur in approximately 50% of the patients after disease duration of 3 to 5 years [7]. Other complications include psychiatric disturbances such as depression, dementia, and hallucinations and gastrointestinal symptoms, which may have a high impact on patient's activities of daily living and health-related quality of life [8].

Medical treatment aims to restore the failure of dopamine production by using levodopa, dopamine agonists, or inhibitors of dopamine degradation (MAO-B inhibitors, COMT-inhibitors) [9–11]. In addition, anticholinergic drugs as well as N-methyl-D-aspartate (NMDA) antagonists (amantadine) are used. Recently, a resurgence of interest in surgical procedures such as pallidotomy and deep brain stimulation has arisen [12,13]. Although pharmaceutical and surgical therapies can alleviate signs and symptoms of PD, the disease relentlessly progresses. Therefore, the long-term effect of treatment on clinical symptoms and health-related quality of life (QoL) is paramount in evaluating PD interventions. However, most clinical trials have a short time horizon and use surrogate endpoints or focus only on specific complications instead of assessing the entire spectrum of the disease and the impact on the patient's QoL. Therefore, mathematical models must be used to link the short-term clinical outcomes from clinical trials with evidence for the long-term progression of the disease, patient preference data, and costs to guide clinicians and health-care policy decision-makers [14–18]. However, modeling intervention effects on PD is highly complex. In particular, the wide spectrum of PD symptoms and treatment complications and the lifelong progression of disease require the combination of different health outcomes and the extrapolation of them beyond the time horizon of clinical trials.

To date, several decision-analytic or other types of mathematical models for interventions in patients with PD have been published, but to our knowledge, there are no published reviews of PD models. Therefore, we sought to provide decision analysts and clinicians interested in formal medical decision making with an insight into the structural and methodological approaches used in PD decision modeling including their strengths and limitations. This inspection is intended to help them to improve further modeling or to select the optimal structural approach to answer their specific research question. We did not intend to systematically compare the effectiveness or cost-effectiveness of interventions. Finally, we derive recommendations for the development of comprehensive decision models evaluating interventions in PD.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Identification of Relevant Studies

We performed a systematic literature search in electronic databases using a combined search strategy, including the following terms: “decision analysis,”“decision-analytic,”“decision model,”“health care model,”“health care evaluation model,”“decision tree,”“Markov model,”“cost-effectiveness,”“cost-utility,”“cost-benefit,”“cost-minimiz(s)ation,”“QALY,” and “Parkinson.” The search was first performed on the databases Medline and PreMedline (1966 to July 2002), Current Contents (all editions, 1993 to July 2002), Embase (1991 to July 2002), EconLit (1969 to July 2002), and the Cochrane's database of systematic reviews (to 2nd quarter 2002). A more broadly specified search was performed on the databases DARE (Database of Abstracts of Reviews of Effectiveness), NHS EED (Economic Evaluation Database), and HTA (Health Technology Assessment) of the United Kingdom's NHS Centre for Reviews and Dissemination ( In addition, we examined previous reviews and reference lists of identified studies. There was no restriction on language or publication time.

One of the inclusion criteria for this review was that the study was based on a decision-analytic model or any other type of mathematical health-care model evaluating therapeutic interventions for PD. Regarding the term “model,” we used a modified definition recently given by the “ISPOR Task Force on Good Research Practices—Modeling Studies”[19], and defined health care evaluation model as an analytic methodology that accounts for health outcomes over a defined time and across a defined population, whose purpose is to estimate the effects of an intervention on valued health consequences and/or costs. This definition includes decision trees, Markov models, and models based on mathematical equations. Purely descriptive studies or studies using models only as an illustration or in a tutorial were excluded. Only published studies were included in the review. Studies published only as abstracts without providing full information about the model were not included [20–22].

Data Extraction and Model Assessment

We used a standardized assessment form, which was based on guidelines and recommendations for decision modeling and cost-effectiveness analysis [17,23,24] to extract data from included studies. The form comprised the following domains defined prior to the study:

  • • 
    Study reference;
  • • 
    Target population, country, setting;
  • • 
    Decision-analytic framework;
  • • 
    Data sources;
  • • 
    Model validation;
  • • 
  • • 
  • • 
  • • 

The study reference included authors, year of publication, reference, and the setting (country and year). The decision-analytic framework comprised information about target population, study question/objectives in the author's words, study type, compared strategies, time horizon, outcome measures, brief model design, statistical analysis, perspective, and discounting (annual discount rate). The domain data sources listed data sources for epidemiology/natural history, efficacy, quality of life/utilities, costs, and other data. Furthermore, key structural and numeric assumptions as well as efforts of model validation were stated. Base-case results for effectiveness, costs, and the cost-effectiveness relation as well as results from sensitivity analyses were extracted. We extracted major limitations of each publication, as recognized by the investigators or as commented upon by the assessors. Finally, the author's main conclusions were summarized.

All data were independently assessed by three of the authors (Siebert, U, Bornschein, B, Walbert, T) using the standardized assessment form. The extracted data were summarized in evidence tables. Furthermore, a model and study assessment was performed in a qualitative fashion and the assessor's comments on model features and study quality were summarized in a brief commentary for each publication. If extracted information or judgment of model assessment differed between two of the reviewers, the study was completely reassessed by an expert in methodology and neurology (Dodel, RC) and the issue was discussed among the group to achieve agreement.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Systematic Literature Search

The systematic literature search was performed in July 2002 and yielded 18 hits in Medline and 20 hits in Current Contents. Searches in the other databases did not lead to additional studies. Only 5 of the 18 studies identified in the Medline search were modeling studies; they are included in the review. The other 13 were excluded because they addressed neuropsychological [25–29], physiological [30], and nursing aspects [31]. Several studies were not included because they were randomized clinical trials (RCT) or cost studies which were not based on a model [32–36] or they used a model only for illustration purposes [37].

In the Current Contents search, 3 of the 20 identified studies were modeling studies. All 3 had been previously identified in the Medline search. The remaining 17 studies were excluded because they addressed neuropsychological [26–28], biological [30], or nursing aspects [31,38]. Two articles presented RCT results [33,34], 7 did not contain decision models [36,39–44], and 2 described models only for illustration purposes [37,45].

The manual search of reference lists did not lead to additional documents, but expert advice and an updated search in February 2003 revealed 3 additional relevant studies. One study was published in a recently founded journal [46] and the other 2 were published in October 2002 [47] and February 2003 [48]. Overall, 8 studies were included in the systematic review and formal data extraction [46–53].

Systematic Description and Assessment of Studies

Table 1 summarizes the methods and findings of the 8 studies included in the systematic assessment. Table 2 lists the analytic framework and model features. In the following, a brief narrative description is given for each study, highlighting methodological strengths and limitations and quality of input data. We also highlighted features we thought to be of specific methodological interest or from which we can learn for future modeling.

Table 1.  Summary of methods and findings of included studies
Study No.Authors, year, countryTarget populationResearch question/objectiveStudy typeData sourcesMain resultsConclusions
  1. Abbreviations: Au$, Australian dollar; CEA, cost-effectiveness analysis; COI, cost of illness-study; CUA, cost-utility analysis; CMA, cost-minimization analysis; DBS, deep brain stimulation; HY, Hoehn & Yahr stage; ICER, incremental cost-effectiveness ratio; ICUR, incremental cost-utility ratio; MC, Monte Carlo (simulation); PD, Parkinson's disease; QALY, quality-adjusted life years; QoL, quality of life; RCT, randomized clinical trial; US$, US dollar; VAS, visual analog scale.

1Hoerger et al. 1998, USA [49]Levodopa-naive and levodopa-treated PD patientsIncremental C/E of pramipexole vs. no treatment and pramipexole + levodopa vs levodopaCUAEffectiveness based on clinical trials and manufacturers’ data Utilities from VAS study National cost data, therapy/demand patterns from survey, clinical trials, expert interviewsA) Pramipexole vs. no treatment using direct costs (total costs): ICUR US$34,423/QALY (8837) B) Pramipexole + levodopa vs. levodopa alone: ICUR US$31,528/QALY (12,294)Pramipexole is cost-effective, even with very conservative assumptions
2Davey et al. 2001, Australia [50]PD patients HY I–IIIIncremental C/E of pergolide vs. bromocriptine in PDCEAEffectiveness from single RCT Progression from cohort study Cost data from expert panel, cohort study, and official national chargesPergolide results in longer time in HY I–III and is cost-saving. Savings: AU$1028 per patient. Dominance of pergolide was not altered when effect duration beyond RCTtime horizon (26 weeks) was varied.Pergolide is more efficacious than bromocriptine and cost-saving
3Nuijten et al. 2001, The Netherlands [51]PD patients with severe end-of-dose motor fluctuations (> 25% off time), HY I–IV, levodopa-responsiveIncremental C/E of complementary therapy with entacapone vs. usual care (i.e., levodopa for all, add-on of pergolide, bromocriptine, selegiline or amantadine for some) in patients with end-of-dose motor fluctuationsCUA and CEAEffectiveness from RCTs Progression from survey on duration of disease Utilities from prospective US study Cost data from official sources, resource utilization from RCT and German cross-sectional studyComplementary entacapone yields higher effectiveness regarding both time without severe fluctuations and QALYs.Costs are similar to those for usual care. Entacapone yielded 0.63 incremental years without severe fluctuations and increased QALYs by 6%.Complementary entacapone is cost-effective and dominates over usual care
4Tomaszewski & Holloway, 2001, USA [52]Late stage PD, HY III–V with intractable motor fluctuations, age > 50 yearsIncremental C/E of DBS compared with the best medical management in late stage PD (i.e., typical treatment pattern in PD patients before DBS as reported in literature)CUAEffectiveness primarily based on case series Utilities from small VAS study National and local cost dataDBS vs. best medical management: ICUR: US$49,000/QALY ICUR < US$100,000/QALY if QoL improves by 18% or moreDBS is cost-effective, if QoL improvement is ≥18% compared with best medical management
5Shimbo et al. 2001, Japan [53]Male PD patients HY II–V, already receiving levodopaIncremental C/E of bromocriptine or pergolide compared with l evodopa aloneCUAEffectiveness/clinical parameters from RCT Utilities from survey performed alongside the study Cost data from RCT, cost-of-illness study, national health statistics and expert opinion (long-term care)Neither dopamine agonist is cost-effective in HY II. Both dopamine agonists are more efficacious and less expensive in HY III–V (dominance). Cost and effectiveness have large influence on ICUR, e.g., generic bromocriptine (less than 50% of the original price) is dominant even in HY IIBromocriptine or pergolide therapy is cost-effective (dominant) only in later stages of disease. Efficacy and cost substantially influence ICER with effects even on earlier disease stages.
6Linna et al. 2002, Finland [46]PD patients with motor fluctuationsReliability of incremental C/E of adjunctive entacapone to levodopa, any other additional PD medication allowedCUAEffectiveness/clinical data from 2 RCTs Cost and utility data from naturalistic COI studyEntacapone as an adjunctive treatment to levodopa is both cost-saving and increases QoL. Uncertainties of parameter estimates, especially progression rates, significantly affect C/E results.Adjunctive entacapone is cost-effective, even under assumptions strongly varying from base case.
7Palmer et al. 2002, USA [47]PD patients who experience off-time, mostly between HY 1.5–3Incremental C/E of complementary entacapone to levodopa in patients with end-of-dose motor fluctuations (other PD medication allowed but not specified)CUAEffectiveness from RCTs Utilities from prospective US study Cost data from official tariff lists and national statistics; resource utilization from RCT, epidemiologic study, and US expert panelAdjunctive entacapone seems to be cost-effective. Model is sensitive to dosage of entacapone, rate of disease progression, and magnitude of initial improvement by entacapone.Additional entacapone to standard therapy is cost-effective from a social and third-payer perspective.
8Iskedjian & Einarson, 2003, Canada [48]PD patients HY 1–3 without dyskinesias, only short-term pretreatmentAre additional cost of the more expensive drug ropinirole (alone or together with additional levodopa) off-set by an associated lower rate of dyskinesiasCMAEffectiveness and clinical (safety) data from RCT and expert panel and unpublished study data from manufacturer Cost and resource utilization data from official tariff lists and expert panelRopinirole (alone or together with adjunctive levodopa) is cost-saving when cost for productivity loss is included Result is considered robust against several assumptionsIn an early stage of PD use of ropinirole instead of levodopa is cost-saving because of reduced dyskinesia-associated downstream costs
Table 2.  Summary of analytic framework and model features of included studies
Study No.Authors, yearTime horizonHealth outcomesAnalytic approachStatistical analysisPerspectiveUncertaintyModel validation
  • *

    Model was published earlier by Nuijten et al. 2001.

  • Abbreviations: SA, sensitivity analyzes; n.r., not reported.

1Hoerger et al. 1998 [49]LifetimeUPDRS scores over time, QALYSet of mathematical equations: UPDRS scores as function of treatment and time; lifetime costs and QALYs as function of UPDRS scores and covariatesMultivariate regression analysis, mathematical simulationSocietal and third party payerSeveral deterministic 1-way SAn.r.
2Davey et al. 2001 [50]10 yearsTime in HY I–IIIMarkov model. Markov states: HY stages I–V and deathCohort simulationHealth care systemSeveral deterministic 1-way SA; scenarios analyzes varying duration of treatment effectn.r.
3Nuijten et al. 2001 [51]5 yearsQALY, time with ≤ 25% off-time/dayMarkov model. Markov states: Off-time ≤ 25% and >25%, deathCohort simulation and Monte Carlo simulationSocietalSeveral deterministic 1-way SA, 1st order Monte Carlo simulation based on target population (n = 7000)n.r.
4Tomaszewski & Holloway, 2001 [52]LifetimeQALYMarkov model embedded in decision tree for surgical complications. Markov states: in nursing home, outside nursing home, deadCohort simulationSocietalSeveral deterministic 1-way SAn.r.
5Shimbo et al. 2001 [53]10 yearsQALYMarkov model. Markov states: HY I–V and deathCohort simulationSocietalSeveral deterministic 1-way SAn.r.
6Linna et al. 2002 [46]5 yearsQALYMarkov model. Markov states: modified HY stages 1.0–5.0 and deathMonte Carlo resampling; cohort simulationSocietalProbabilistic multiway SA using Monte Carlo simulationn.r.
7Palmer et al. 2002*[47]5 yearsQALY, time with ≤25% off-time/dayMarkov model. Markov states: Off-time ≤ 25% and >25%, deathCohort simulationSocietal and third-party payerSeveral deterministic 1-way SAn.r.
8Iskedjian & Einarson, 2003 [48]5 yearsAdverse eventsDecision-tree (case management pathway tree) based on event probabilitiesCohort simulationSocietal and third-party payer (Ministry of Health)Several deterministic 1-way SAn.r.

Hoerger et al. 1998.  Hoerger et al. [49] performed a cost-utility study evaluating the use of the dopamine agonist pramipexole in two settings. The authors compared 1) pramipexole with no levodopa therapy in the early stages of PD; and 2) pramipexole plus levodopa with levodopa monotherapy in advanced PD. Two separate models for early and advanced PD were based on a set of mathematical equations. United Parkinson's Disease Rating Scale (UPDRS), which includes information on patient's history, physical examination, and complications, was used as surrogate for progression of disease and modeled as a function of treatment and time from onset of PD. In these equations, introduction of pramipexole and/or levodopa was assumed to instantaneously improve UPDRS scores and partially attenuate UPDRS-based progression rates. In further equations, quality-adjusted life years (QALY) and lifetime direct and indirect costs were expressed as a function of UPDRS scores. Applying multivariate regression analysis to clinical trial data and survey data on UPDRS scores, costs, and QoL fit the mathematical equations. The authors used probit models to predict hospital visits and working status, logarithmic regression for costs, and linear regression for utilities.

For both comparisons, the incremental cost-utility-ratio (ICUR) was below US$35,000 per QALY gained. The authors concluded that the use of pramipexole is cost-effective in either early-stage PD or in later stages in combination with levodopa. The authors mentioned that the model might be improved by including complications and other modern therapies. They also discussed the limited predictive value of the regression equations used (adjusted R-square 5–26%).

In this transparently reported study, costs and health outcomes are represented by a detailed set of mathematical equations. One strength of the authors’ use of mathematical equations is the clear and transparent presentation and the explicit regression-based parameterization under several conservative assumptions, which have been tested in extensive sensitivity analyses. The fact that the model is based on equations reflecting UPDRS scores limits its flexibility and applicability for treatments that have an effect on specific disease components or events not sufficiently included in or adequately weighted by the UPDRS.

Davey et al. 2001.  Davey et al. [50] performed a cost-effectiveness study comparing the dopamine agonists pergolide and bromocriptine in patients with PD. The authors developed a Markov model with a 10-year time horizon based on Hoehn and Year (HY) stages. HY staging is a well-established PD classification scheme, evaluated in the presence (HY on) or absence (HY off) of treatment. This Markov model included HY-on stages I–V and death and was based on HY-on-specific progression rates derived from an Australian 12-year cohort study. Efficacy was taken from one randomized clinical trial, which was selected among 5 identified RCTs. Resource utilization was based on an expert panel survey. The clinical outcome (time in stages HY-on I–III) of the comparator bromocriptine was based on the initial distribution of HY-on stages and progressions from the cohort study. The HY-on distribution of the pergolide branch was modeled using a 19% absolute risk difference for HY improvement, which was reported in the RCT. The authors assumed no treatment effect on progression rates conditional on HY-on stage. In the base case, HY-on distributions of both treatments were set to be equal after 2 years, varied from 0.5 to 10 years in scenario analyses.

Results showed improved effectiveness and lower health-care costs for pergolide (dominance). The authors concluded that pergolide is cost-saving and more efficacious than bromocriptine. Savings due to prevented or deferred complications offset the small additional drug costs of pergolide.

The model and its parameters were described in detail. The strength of the model and its applicability in other evaluations is given by the implementation of HY-on stages, which is a clinical classification applied in many studies. HY-on, however, is not a marker for the disease-underlying natural progression and empiric treatment efficacy data were restricted to HY-on stages I–III. Therefore, the authors were forced to extrapolate progression for HY-on IV–V. In addition, the initial (i.e., before treatment) HY-on distribution remained unclear. However, sensitivity analyses adequately addressed the uncertainty on duration of efficacy. No sensitivity analyses were performed on the magnitude of efficacy, which was assumed to be additive (absolute risk difference) regardless of the baseline risk. Neither background mortality rates of the general population nor QoL data were considered in the model. To adapt probabilities and risk differences to the 6-month cycle length, probabilities were simply divided by the number of intervals instead of using the declining exponential approximation of life expectancy (DEALE) approach within intervals.

Nuijten et al. 2001. Nuijten et al. [51] performed a cost-utility and cost-effectiveness study on the effect of the complementary use of the COMT-inhibitor entacapone in PD patients suffering from severe fluctuations in the Netherlands. The authors used a Markov model with a 5-year time horizon and Markov states based on severity of fluctuations expressed by off time (≤25% vs. >25% per day). Markov state-specific data on duration of the disease, which were obtained from a cross-sectional study, were transformed into progression rates using the DEALE approach. Based on RCT data for entacapone, improvement was restricted to the first 6 months in the model. Utilities were estimated from patient preference data assessed in a US study, which was designed for this project. Resource utilization patterns were based on a German retrospective cost study and drug costs were assessed along RCTs.

Entacapone was more effective and less costly than usual care (dominance). These results were robust in extensive sensitivity analyses performed on the relevant model parameters.

This is a well-performed study with explicit, plausible, and conservative assumptions. A major strength of this model is its simplicity and parsimony. Also, the Markov state classification based on daily off time reflects a very relevant aspect of QoL. However, utilities were assessed in two movement disorder clinics and one private neurology practice in the US and the transferability to patients in the Netherlands may be questionable. The fact that progression rates were estimated from a cross-sectional study was adequately addressed in sensitivity analyses. The model is limited by the binary off-time health state, which does not reflect the entire spectrum of disease severity and also does not reflect natural history of disease stages. Therefore, QoL effects and costs of specific adverse effects, complications, and concomitant illness cannot be analyzed with this version of the model. Finally, the 5-year time horizon may be too short to reflect the societal perspective.

Tomaszewski and Holloway 2001. Tomaszewski and Holloway [52] performed a cost-utility study for a new surgical treatment option, deep brain stimulation (DBS), in the advanced stage of PD. The study was performed at an early stage in the life-cycle of this technology. The authors developed a lifetime Markov model based on states reflecting care within the nursing home versus outside of it. The Markov model was embedded in a decision tree reflecting surgical complications. In the model, DBS improves QoL (based on UPDRS score improvement) in the initial period and then the effect gradually declines approaching the level of patients with best medical treatment.

The base case ICUR for DBS is US$49,000 per QALY. The authors consider DBS a cost-effective therapy alternative, given a surplus of QoL of at least 18%. However, data for this new technique are extremely sparse and cover only about 2 years of follow-up. QoL data rely on a small sample of patients, but have a very strong impact on ICUR. Therefore, these preliminary results indicate that further data with special respect to utilities are needed prior to recommendation of this technology.

In this carefully performed study, modeling is used in the situation of sparse clinical, economic, and outcome data. ICUR relied strongly on QoL values of DBS patients. QoL over time and several other crucial parameters (e.g., DBS-related complications and generator replacement intervals) were based on expert opinion. Therefore, results should be interpreted with caution and the authors indicated the need for further utility data. The strength of this model lies in its transparent modeling assumptions in the situation of sparse treatment effect data and sensitivity analyses underlining the uncertainty. The model is limited by the use of UPDRS scores as the relevant outcome. Side effects of the DBS were modeled as surgical death and permanent and temporary complications and accordingly adjusted for QoL. The effect of the anatomic stimulation target (subthalamic nucleus vs. globus pallidum internum) on costs and clinical effect was not adequately considered in the model.

Shimbo et al. 2001. Shimbo et al. [53] performed a cost-utility study comparing the adjuvant use of the dopamine agonists bromocriptine or pergolide with levodopa monotherapy. The authors applied a 10-year Markov model consisting of 6 Markov states (HY-on stages I–V and death). The model assumed an improvement of one HY-on stage in a certain proportion of a cohort of patients with HY-on stages II–V associated with the introduction of the dopamine agonist. Data on resource utilization and utilities were determined in a survey accompanying the study.

The results of this study indicated that both dopamine agonists were cost-effective only in the later stages of the disease, i.e., HY-on stages III or higher. Generic preparations of bromocriptine with 50% lower costs compared to the original formulation were more effective and cost saving even in HY-on stage II (dominant) when compared to levodopa alone. In sensitivity analyses, drug costs and effectiveness showed substantial influence on ICURs.

This is a carefully performed study with mostly conservative assumptions. However, some aspects of clinical routine were not considered, such as the increase in drug dose was not modeled. This study is especially relevant for the Japanese health-care context, because specific dosage patterns and prices were carefully assessed in Japan. Without substantial modifications, though, the model has limited use for evaluations outside of Japan, because parameters are not generalizable. For example, the model does not include dopamine agonist monotherapy, because this was not approved in Japan. In addition, there were substantial differences when comparing dosage and therapy costs for Japan versus other countries.

Linna et al. 2002.   Linna et al. [46] performed a cost-utility study investigating not only the cost-effectiveness of adjunct entacapone to levodopa, and any further PD medication, but also the extent of uncertainty associated with the result by Monte Carlo resampling methods. They developed a Markov model consisting of 8 Markov states (modified HY-on stages and death), which used treatment- and HY-specific transition probabilities from two RCTs with a follow-up of 6 months. Assuming a constant treatment effect over time, these transition probabilities were extrapolated to a time horizon of 5 years.

This study showed that adjunctive entacapone is cost-saving and associated with higher QoL values in 87.5% of the simulations. Variation in transition probabilities contributed to most of the overall uncertainty associated with ICUR.

The model parameters used in this study were not sufficiently described in the publication. Although data sources and data assessment methods were thoroughly explained and up-to-date evidence was considered, no model input parameters were reported for initial cohort distribution, transition probabilities, utilities, and costs. As strength, the investigators extensively explored uncertainty by Monte Carlo simulation and displayed cost-effectiveness planes with 95% confidence ellipses as well as cost-utility acceptability curves. One-way sensitivity analyses addressing systematic rather than random variation were not sufficiently reported. Along with the Finnish 15D, a new generic instrument for determination of health-related quality of life was used. This instrument differs conceptually and in QoL dimensions from other well-established instruments [54]. Therefore, the comparability of the ICUR with that of other studies may be problematic.

Palmer et al. 2002.  Palmer et al. [47] performed a cost-utility and cost-effectiveness study for adding entacapone to levodopa and standard treatment in US patients who experience off-time (re-emergence of PD symptoms). The authors modified the Markov model developed by Nuijten et al. [51] to adapt it to the US setting and to add the third payer perspective in addition to the societal perspective. For further details, see the description of the model in the section Nuijten et al.

In the base case, adding entacapone had an ICUR of US$9327/QALY for the societal perspective (i.e., considering total costs) and US$21,213/QALY for the third payer perspective (direct medical costs only).

This was a well-performed study with plausible and conservative assumptions. The model and its parameters were clearly described. However, the robustness of these model results is limited because the progression rate for PD patients was estimated from a single source, although the sensitivity analysis indicated that results were sensitive for this parameter. The authors mentioned as a further limitation the representation of off time by only two levels (i.e., ≤ 25% vs. >25% off-time). These levels neither discriminate the clinical course accurately nor fully reflect the underlying biologic progression of disease.

Iskedjian and Einarson 2003. The study conducted by Iskedjian and Einarson [48] is a cost-minimization study in PD which examined the economic impact of reducing dyskinesias using the dopamine agonist ropinirole instead of levodopa plus benserazide in treatment-naïve (de novo) PD patients. The authors modeled along one large RCT using 5-year clinical and economic data. Assuming equivalent effectiveness, only costs triggered by dyskinesias and hallucinations were considered. Costs over a time horizon of 5 years were compared adopting the societal perspective and the third-payer perspective of the Ministry of Health in Ontario/Canada. Whereas the perspective of the Ministry of Health considered only drug costs and other health-care costs, the societal perspective additionally included productivity costs and caregiver costs. The specific study question was whether savings due to avoided cases of dyskinesias offset the added cost of ropinirole. The calculation was based on a decision-tree approach, that is, the authors used a case-management pathway-tree derived from empiric data and expert estimates on the age- and regimen-specific frequencies of dyskinesias and hallucinations, which triggered health care, productivity, and caregiver time.

From the societal perspective, ropinirole was cost-saving. The authors concluded that savings due to avoided cases of dyskinesia offset the added cost of ropinirole. Loss of productivity was the major factor offsetting the drug costs followed by caregiver-associated costs. From the perspective of the Ministry of Health, ropinirole was more expensive than levodopa.

This study is unique in modeling the impact of adverse effects of the drugs ropinirole and levodopa on costs. However, the authors mention that this was not a comprehensive analysis because only costs triggered by adverse effects were considered and differential impacts of the compared regimens on QoL were not considered. Although clinical data were derived from a large RCT, substantial cost- and resource-utilization data were estimated by four expert panelists, which could limit generalizability. However, most assumptions were conservative, i.e., modeled against ropinirole treatment. Although this model is restricted to the evaluation of dyskinesias and hallucinations, the careful evaluation of drug effect profiles might be a promising approach for future cost-utility models that consider both the effectiveness and adverse effects as factors influencing the outcome QALYs. An additional strength of this study was its inclusion of cost due to lost productivity, which was based on the human capital approach. Further sensitivity analyses considering actual employment rates, around 60%, would have added valuable information to the study. The fact that information such as age distribution was not explicitly reported may prevent interested investigators in replicating the modeling results. The practice of using discounting only in sensitivity analyses and not in the base-case analysis is not in concordance with current recommendations for cost-effectiveness studies [17]. A more extensive use of sensitivity analyses to quantify uncertainty would have further increased the strength of this study.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

This systematic review and assessment of model-based studies of treatment strategies in PD patients includes 8 studies. All were published within the last 5 years and reflected the health-care systems of different countries—Australia, Canada, Finland, Japan, the Netherlands, and the USA (see Table 1). The target populations ranged from early-stage to advanced PD patients and—depending on the study—included treatment-naive patients as well as patients on levodopa treatment. Seven studies evaluated strategies for drug therapies using dopamine agonists (pramipexole, pergolide, bromocriptine, ropinirole), or the COMT-inhibitor entacapone, either alone or as an adjunct to levodopa. One study comparing deep brain stimulation to best medical management while this technology was in an early state of its life-cycle based effectiveness on data from case series. Seven studies [46,47,49–53] evaluated clinical benefits as well as economic outcomes and compared different strategies, and therefore, belong to the category of complete economic evaluations. In 6 studies [46,47,49,51–53], the authors reported cost-utility ratios in costs per QALY gained. In the two remaining studies, which were a cost-effectiveness [50] and a cost-minimization study [48], the clinical outcome was time free of progression to HY-on stages IV or V and costs per patient per year, respectively. In all studies, costs were based on resource-utilization data derived from short- to intermediate-term clinical or cost-of-illness studies. Long-term costs were based on national administrative databases or expert opinion. In cost-utility studies, health-state-specific utilities were transformed into QALYs and used in the denominator of the cost-utility ratio. The evaluated time horizon ranged from 5 years to lifetime and the perspective of either society or the health care payer was adopted. All studies showed that the investigated strategy was cost-effective or even cost-saving (i.e., more effective and less costly) compared to the current standard of care, at least for some alternative strategies or for some of the examined scenarios. However, the intention of this paper was not to compare the effectiveness or cost-effectiveness of interventions. Rather, our intention was to give an insight into the “architecture” of published decision models and to derive recommendations for future modeling in PD.

Type of Models and Methodological Approaches

Different model approaches and designs were used in the assessed studies. The first published model consisted of a mathematical equation system in which the equations described the progression of disease expressed through deterioration of UPDRS scores [49]. In further regression analyses, QoL and costs were linked to UPDRS scores. A later model used decision-tree techniques to calculate costs of illness when taking adverse drug effects into account [48]. All remaining models were Markov models with Markov states defined by HY stages rated during treatment (HY-on stages), on/off-time, or nursing home status (in/out) [46,47,50–53].

The authors used different methodological approaches to model the effect of treatment. Some assumed an initial improvement of the new intervention compared to standard care, some used reduced progression rates with respect to the symptomatic scales (e.g., HY, UPDRS), some combined both approaches, and one study used reduced adverse effect probabilities. Some of the models were limited in reflecting the continuous character of the progression of disease, because they included only few Markov states. Two studies classified patients by using the Markov states “time with ≤ 25% off time” and “time >25% off time”[47,51]. Although this is an elegant approach to focus on a relevant outcome, it may not fully reflect the spectrum of the disease and the related QoL. Another study used the surrogate outcome UPDRS score as a core model variable [49].The study on deep brain stimulation used the binary classification care inside or outside nursing home [52], which seems to be a very crude measure of defining the outcome of treatment. The cost-minimization study focused exclusively on the economic impact of adverse effects (e.g., dyskinesias) [48]. The remaining three studies used HY stages under adequate treatment (HY-on) to define the current health state of the person [46,50,53]. There are several advantages of this approach. First, the entire spectrum of disease (i.e., early to advanced disease with its specific complications) can be considered more precisely and progression over time can be modeled in greater detail. Second, HY-on stages are a standard clinical rating system that is measured in many clinical trials [55]. Thus, sufficient data are available to apply such a model to different interventions and settings. The third advantage of using health states based on HY-on is that the HY-on stages correlate with other clinical scores such as the UPDRS, complications, QoL, and costs [5,56–59].

However, there is a fundamental problem with using HY-on as a marker for the progression of disease. By definition, HY-on is a classification for patients under adequate treatment, and therefore is affected simultaneously by both the natural biologic progression of disease and the symptomatic response to treatment. Even valid and precise measurement of HY-on for each treatment over time does not allow for the isolation of the natural progression from the treatment effect for both the new treatment as well as the comparator (e.g., levodopa treatment). This causes two methodological problems. First, it is difficult—if not impossible—to state explicit and independent assumptions about progression effects and treatment effects and to evaluate the impact of these assumptions in sensitivity analyses, which normally represent one of the most important tools in decision-analytic modeling. Second, the mean transition time from one HY stage to the next takes about 2 to 3 years [60,61], but in most clinical studies the time horizons are too short to yield sufficient data for progression to advanced HY stages. Therefore, extrapolations for the treatment effect beyond the clinical evaluation period must be made. Although this is a common procedure in decision analysis, assumptions about treatment alone are not sufficient to project the future course of disease effects (i.e., the difference in the progression between one treatment and the other). The treatment effect must still be applied to the progression in the compared strategy (i.e., the baseline progression). The use of HY-on-based Markov states in a model forces the investigators to simultaneously extrapolate both baseline progression and treatment effect, which poses the risk of making the extrapolation speculative or introduce structural bias. Weinstein et al. stated that, in general, structural bias is avoided by modeling underlying disease states and then by calibrating outputs to data on observed clinical status [19].

If we apply this recommendation to PD, an alternative modeling approach that could solve this problem is defining the Markov states based on both HY-off and HY-on data. Transition rates for HY-off stages, which are available from progression studies for HY-off I–V [60,61], reflect the natural progression of the disease and therefore allow evidence-based modeling of the progression excluding a treatment effect. In addition, either model outputs have to be calibrated to observed HY on data or data on the association of HY-on and HY-off under different treatment options have to be assessed. Only when the underlying HY-on stages are validly linked to HY-on outputs, such a model could be used to predict long-term consequences for new interventions. Once sufficient knowledge about the interaction of pathophysiologic progression and clinical outcome under PD treatment is available, models may also be based on histologically or pathophysiologically defined health states.

This issue is especially important for the evaluation of possible benefits gained from early diagnosis (e.g., with imaging techniques such as single photon emission computed tomography [SPECT] or positron emission tomography [PET]) in combination with emerging neuroprotective treatments [62]. Early interventions can only be evaluated with a model that links clinical health states (e.g., HY-on stages) to an underlying chain of natural progression in the absence of treatment.

Time Horizon

Whereas four studies used a time horizon of only 5 years [46–48,51], 4 others used a time horizon of 10 years to lifetime [49,50,52,53]. The time horizon chosen can have a substantial impact on health effects and costs, and should be at least considered in sensitivity analyses. Only a time horizon that is long enough to cover even advanced stages of PD (i.e., 10 years and larger) can 1) combine potential tradeoffs between short-term gains and long-term losses due to drug adaptation effects, and 2) yield a reliable incremental cost-effectiveness ratio that allows a fair comparison with other medical technologies [24].

Adverse Events and Complications of the Disease

Adverse events play a substantial role in the treatment of PD. Although Nuijten et al. [51] and Palmer et al. [47] mentioned that adverse drug events may influence costs substantially, none of the research groups except Iskedjian and Einarson [48] included branches or health states for adverse drug events in their model. The effects of adverse events on QoL should also be considered in future models. If data are lacking, sensitivity analyses should evaluate the need for specific pharmacoepidemiologic studies. Tomaszewski and Holloway present an excellent example of integrating adverse effects after the implantation of electrodes for subthalamic stimulation (“new” technology) [52]. As the authors modeled adverse effects only in the “new” technology and not in the comparator (i.e., the adverse events from pharmacological therapy), their results are conservative in the sense that they may overestimate the cost-effectiveness ratio of deep brain stimulation. Conservative modeling of the new technology is usually the preferred analytic approach when data are lacking.

As outlined earlier, several complications may occur during the course of PD and these may have a strong impact on health-care utilization and QoL. These include behavioral disturbances and gastrointestinal symptoms. None of the assessed studies included these complications in their modeling approach. As clinical data concerning frequency, severity, and costs of these outcomes are sparse, their implementation in the model may cause considerable problems and needs further research. However, any modeling should attempt to make plausible and explicit assumptions regarding these outcomes in order for the results to be a valid representation of the entire disease process and not only of parts of it.

Neuroprotective and Mortality Effects

Finally, all models relied on the assumption that the evaluated treatments do not affect mortality or have a neuroprotective effect. Further basic and clinical research is necessary to address these issues in order to investigate not only the question of how to treat PD patients but also when to start treatment for optimizing PD care.


All Markov models covered in this review included an explicit Markov state for death, making it possible to model mortality. The modeling of background mortality causes other than PD is essential to yield valid cost-effectiveness estimates. However, none of the studies explicitly addressed PD-specific mortality. Although it may be reasonable to assume no treatment effect on mortality, the consideration of disease-specific mortality may still alter cost-effectiveness estimates in models with a long time horizon through reduced duration of life. It has been argued that life expectancy may be reduced in patients with PD compared to the general population [60,63,64]. Therefore, future investigators may want to consider this option in sensitivity analyses.

Health-Related Quality of Life

All but two [48,50] models included QoL and used QALYs as the clinical outcome. This raises the question whether the use of QALYs is an indispensable model feature. Davey et al. showed in their analysis that pergolide results in a longer stay in HY stages I–III than bromocriptine and yields cost-savings of more than Au$1000 per patient [50]. In this case, it may be justified that further QALY calculations were suspended because one intervention dominated the other. However, this dominance would only hold for the outcome QALY if it could be assumed that QoL is mainly determined by the HY stage and that drugs have little differential influence on QoL within the HY stages. In any case, assumptions such as those should be stated explicitly and be based on evidence or tested in a sensitivity analysis on drug-related QoL effects. The other study that did not apply QALYs was the cost-minimization analysis of Iskedjian and Einarson [48]. This type of analysis does not contrast incremental costs to incremental clinical outcomes and therefore implicitly assumes identical clinical outcomes for both compared strategies. Given the fact that adverse drug effects differed between the strategies and also may affect the QoL of patients under treatment, the inclusion of QALYs in the model would have enhanced the clinical relevance of this study.

Cost Analysis

Most of the 8 studies focused on direct costs. Only two evaluated costs due to lost productivity (indirect costs) and implemented them in their models [48,49]. Two others [52,53] explicitly followed recommendations for cost-utility analyses [23], which suggest for the reference case analysis not to consider costs due to productivity losses in the numerator of the cost-utility ratio but rather to use a QoL measure that already implicitly or explicitly incorporates the effects of morbidity on productivity time in non-monetary terms. This would avoid double counting of costs due to lost productivity. Nuijten et al. stated that they did not consider indirect costs because the average age of PD patients was close to retirement age [51]. Following this example, whatever approach is chosen regarding productivity losses should be explicitly reported. In any case, the description of expected costs due to productivity losses is of additional value for judgments on societal costs on a macroeconomic level.

Heterogeneity Bias

All models implicitly assumed homogeneity of progression and homogeneity of treatment effects in the patient population. It has been argued that patients who progress more slowly in the early disease stages also progress more slowly in the more advanced disease stages and vice versa [65,66]. If this were true, all models would be biased through the “heterogeneity of progression” bias. It has been shown that this phenomenon may lead to substantial bias in favor of the more effective therapy [67]. If this were also the more expensive therapy, as in most of the studies included in this review, the cost-effectiveness ratio calculated with the biased model would be underestimated. Similarly, if “heterogeneity of treatment effects” is not random but determined by specific inherent and fixed patient characteristics (e.g., genetic factor or subtype of disease), ignoring this heterogeneity may lead to bias. However, in this case, the bias would tend to work against the new treatment [68]. If the treatment effect is associated with genetic factors of the patient, this phenomenon is also called “pharmacogenomics bias.” Sensitivity analyses in subgroups with higher and lower progression rates and treatment effects across a plausible range should be performed to evaluate the related uncertainty.


Most decisions in health care are based on imperfect information and sensitivity analysis must be performed to assess the robustness of the results or indicate areas where further research may be valuable [15]. According to the recommendations of the Panel on Cost-Effectiveness in Health and Medicine [23] and the recently published “Principles of Good Practice for Decision Analytic Modeling in Health Care Evaluation”[19], every pharmacoeconomic modeling study should include extensive sensitivity analyses of key parameters.

In only two models was the variability of input data analyzed by Monte Carlo simulation (see Table 2), thus making possible extensive multiple sensitivity analyses and determination of probabilistic uncertainty [46,51]. Although there is no doubt about the value of such an analysis covering the overall uncertainty of all relevant models [69], this type of sensitivity analysis should not replace extensive and thorough deterministic one-way and two-way sensitivity analyses over plausible ranges of uncertain parameters [23]. If only the results of the Monte Carlo simulation are reported in the publication, the physician or decision-maker may not be able to interpret the results in the context of his or her own opinion or for a specific patient, if one of these differs from the base-case setting of the presented analysis.


Two studies [46,52] did not make any positive or negative statement about funding or conflicts of interests. The remaining six studies were at least partially funded by the industry manufacturing the pharmaceuticals for the examined intervention. Transparency regarding conflicts of interests could be improved by explicitly stating the grant sponsor and its input into study design, data analysis, manuscript preparation, and publication decisions.

Model Validation

Models are only as good as the quality of the input data. Therefore, after thorough technical testing (“debugging”), models should be subjected to internal validation, between-model validation, and external validation [19,70]. Internal validation assesses the concordance of the model results and outcome data from other sources. External validation includes the comparison with independent outcome data, which are ideally collected prospectively. Especially, transition probabilities between disease states must often be extrapolated beyond the time horizon of a clinical trial, and therefore, the long-term model results for disease progression should be externally validated using epidemiological data from long-term observational studies or registries. Even if such data are not yet available for the new treatment examined in a clinical trial, the investigators should examine the model validity for the standard treatment branch.

None of the studies included in this review reported results from internal or external model validation. We hope that the recommendations on validation recently published by the ISPOR Task Force on Good Research Practices [19] will motivate decision-modelers to perform and report validation. Once different modeling groups evaluate the same pharmaceutical or surgical treatment strategies, the comparison of these results will also help to judge the overall credibility. To date, most of the treatment scenarios have not been evaluated using different models. This limits the feasibility of between-model validation. However, as new insights from future clinical trials and observational long-term studies change some of the assumptions or data on which a model is built, it should not come as a surprise when models do not exactly predict the results from clinical studies. The ability of models to adapt to new evidence should not be considered as a weakness but as strength of decision-analytic models.

Complexity versus Simplicity

In this review we have discussed many strengths and limitations of PD models in general and in specific. Some models have useful features that others are lacking. However, it has to be emphasized, that it is not the ultimate goal of modeling to reflect the most granular aspect of nature nor are complex models always better models. It has been repeatedly recommended that a decision model should avoid unnecessary complexity to be transparent and to aid understanding by decision-makers [71–73]. The ISPOR Task Force on Good Research Practices–Modeling Studies suggested that “the structure of a model should be as simple as possible, while capturing underlying essentials of the disease process and interventions”[19]. However, important states or events (e.g., adverse events in clinical trials) should not be omitted because of lack of data or statistical power, unless the links to these events or states are contradicted by available evidence [19].

We think that decision-analytic modeling is a useful tool in clinical decision-making and the economic evaluation of interventions in PD. It cannot replace clinical studies but rather complements them to better inform physicians and policymakers about the potential long-term effectiveness and costs of new and promising interventions.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Based on our systematic review of decision-analytic models for the evaluation of treatment strategies for Parkinson's disease, we conclude that models have been successfully applied to evaluate pharmaceutical as well as surgical treatments in PD. In most cases, Markov models have been used to reflect this chronic and progressive disease. However, most of the models were lacking at least one important feature of an adequate decision-analytic framework in PD.

We recommend that models in PD consider a lifelong time horizon, a sufficient spectrum of clinical outcomes, relevant events and complications, and disease-specific mortality. Emphasis should be given to transparent extrapolation beyond trial data. Economic evaluations should be complete, include the societal perspective, and report—among other results—cost-utility ratios. One-way as well as multi-way deterministic or probabilistic sensitivity analyses should be conducted including those on the effect of heterogeneity of progression and heterogeneity of treatment. Internal model validation is mandatory and external validation should be attempted. Cross-validation should become standard, because several PD models have been published.

Approaches to model treatment effects varied and were either mediated via reduction of symptomatic progression and/or initial symptomatic improvement or via reduction of adverse effects. Considering the biology of PD and recommendations regarding good modeling practice, we believe that structural bias could be avoided and the substantial uncertainty in extrapolations could be reduced in future models, if they contain two separate components: 1) the underlying biological process, which can presently not be influenced by treatment, and 2) HY staging or clinical scales reflecting symptomatic treatment effects. To consider trade-offs in treatment effects on different outcomes, health states should be chosen in a way that they could be linked to outcomes such as Hoehn and Yahr stages, all complications and events of interest, UPDRS scores, utilities, and costs.

This study was supported by the German Federal Ministry of Education and Research, Competence Network Parkinson-Syndromes [BMBF No. 01GI9901/1]. The work of Dr Walbert was supported by the Fulbright Program.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References
  • 1
    Twelves D, Perkins KS, Counsell C. Systematic review of incidence studies of Parkinson's disease. Mov Disord 2003;18: 1931.
  • 2
    Rajput AH, Offord KP, Beard CM, Kurland LT. Epidemiology of parkinsonism: incidence, classification, and mortality. Ann Neurol 1984;16: 27882.
  • 3
    Sutcliffe RL, Meara JR. Parkinson's disease epidemiology in the Northampton District, England, 1992. Acta Neurol Scand 1995;92: 44350.
  • 4
    Mutch WJ, Dingwall-Fordyce I, Downie AW, et al. Parkinson's disease in a Scottish city. Brit Med J 1986;292: 5346.
  • 5
    Hoehn MM. The natural history of Parkinson's disease in the pre-levodopa and post-levodopa eras. Neurol Clin 1992;10: 3319.
  • 6
    Fahn S. The spectrum of levodopa-induced dyskinesias. Ann Neurol 2000;47(Suppl 1):S211.
  • 7
    Ahlskog JE, Muenter MD. Frequency of levodopa-related dyskinesias and motor fluctuations as estimated from the cumulative literature. Mov Disord 2001;16: 44858.
  • 8
    Schrag A, Jahanshahi M, Quinn N. What contributes to quality of life in patients with Parkinson's disease? J Neurol Neurosurg Psychiatry 2000;69: 30812.
  • 9
    Olanow CW, Watts RL, Koller WC. An algorithm (decision tree) for the management of Parkinson's disease (2001): treatment guidelines. Neurology 2001;56(Suppl 5):S188.
  • 10
    Oertel WH, Quinn NP. Parkinsonism. In: BrandtT, DienerHC, CaplanLR, KennardC, DichgansJ, eds., Neurological Disorders: Course and Treatment. San Diego: Academic Press, 2003.
  • 11
    Rascol O, Goetz C, Koller W, et al. Treatment interventions for Parkinson's disease: an evidence based assessment. Lancet 2002;359: 158998.
  • 12
    Honey C, Gross RE, Lozano AM. New developments in the surgery for Parkinson's disease. Can J Neurol Sci 1999;26(Suppl 2):S4552.
  • 13
    Volkmann J, Allert N, Voges J, et al. Safety and efficacy of pallidal or subthalamic nucleus stimulation in advanced PD. Neurology 2001;56: 54851.
  • 14
    Raiffa H. Decision Analysis: Introductory Lectures on Choices Under Uncertainty. New York: Random House, 1968.
  • 15
    Weinstein MC, Stason WB. Foundations of cost-effectiveness analysis for health and medical practices. N Engl J Med 1977;296: 71621.
  • 16
    Weinstein MC, Fineberg HV, Elstein AS, et al. Clinical Decision Analysis. Philadelphia: W.B. Saunders Company, 1980.
  • 17
    Drummond MF, O'Brien B, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes (2nd ed.). New York: Oxford University Press, 1997.
  • 18
    Hunink MG, Glasziou PP, Siegel JE, et al. Decision Making in Health and Medicine. Integrating Evidence and Values. Cambridge: Cambridge University Press, 2001.
  • 19
    Weinstein MC, O'Brien B, Hornberger J, et al. Principles of good practice for decision analytic modeling in health-care evaluation: report of the ISPOR Task Force on Good Research Practices-Modeling Studies. Value Health 2003;6: 917.
  • 20
    Kurlan R, Clark S, Shoulson I, et al. Economic impact of protective therapy for early Parkinson's disease. Ann Neurol 1988;24: 153.
  • 21
    Siderowf A, Holloway R, Mushlin A. Cost-effectiveness of pallidotomy and add-on medical therapy in advanced Parkinson's disease. Ann Neurol 1998;44: 517.
  • 22
    Smala AM, Berger K, Spottke EA, et al. Cost-effectiveness of cabergoline. Parkinsonism Relat Disord 2001;7(Suppl):S105.
  • 23
    Gold MR, Siegel JE, Russell LB, Weinstein MC. Cost-Effectiveness in Health and Medicine. New York: Oxford University Press, 1996.
  • 24
    Weinstein MC, Siegel JE, Gold MR, et al. Recommendations of the panel on cost-effectiveness in health and medicine. JAMA 1996;276: 12538.
  • 25
    Faglioni P, Botti C, Scarpa M, et al. Learning and forgetting processes in Parkinson's disease: a model-based approach to disentangling storage, retention and retrieval contributions. Neuropsychologia 1997;35: 76779.
  • 26
    Faglioni P, Saetti MC, Botti C. Verbal learning strategies in Parkinson's disease. Neuropsychology 2000;14: 45670.
  • 27
    Hsieh S, Hwang WJ, Tsai JJ, Tsai CY. Visuospatial orienting of attention in Parkinson's disease. Percept Mot Skills 1996;82: 130715.
  • 28
    Hsieh S, Hwang WJ, Tsai JJ, Tsai CY. Precued shifting of attention between cognitive sets in Parkinson patients. Psychol Rep 1996;78: 81523.
  • 29
    Sharpe MH. Patients with early Parkinson's disease are not impaired on spatial orientating of attention. Cortex 1990;26: 51524.
  • 30
    El-Deredy W, Branston NM, Samuel M, et al. Firing patterns of pallidal cells in Parkinsonian patients correlate with their pre-pallidotomy clinical scores. Neuroreport 2000;11: 341318.
  • 31
    Reynolds H, Wilson-Barnett J, Richardson G. Evaluation of the role of the Parkinson's disease nurse specialist. Int J Nurs Stud 2000;37: 33749.
  • 32
    Anonymous. A randomized controlled trial comparing pramipexole with levodopa in early Parkinson's disease: design and methods of the CALM-PD Study. Parkinson Study Group. Clin Neuropharmacol 2000;23: 3444.
  • 33
    Hempel AG, Wagner ML, Maaty MA, Sage JI. Pharmacoeconomic analysis of using Sinemet CR over standard Sinemet in parkinsonian patients with motor fluctuations. Ann Pharmacother 1998; 32: 87883.
  • 34
    LeWitt P, Oakes D, Cui L. The need for levodopa as an end point of Parkinson's disease progression in a clinical trial of selegiline and alpha-tocopherol. Parkinson Study Group. Mov Disord 1997;12: 1839.
  • 35
    Montgomery EB Jr, Lieberman A, Singh G, Fries JF. Patient education and health promotion can be effective in Parkinson's disease: a randomized controlled trial. PROPATH Advisory Board [Comment]. Am J Med 1994;97: 42935.
  • 36
    Martinez-Martin P. An introduction to the concept of ‘quality of life in Parkinson's disease’. J Neurol 1998;245(Suppl 1):S26.
  • 37
    Nuijten MJC. Bridging decision analytic modelling with a cross-sectional study – application to Parkinson's disease. Pharmacoeconomics 2000;17: 22736.
  • 38
    MacMahon DG. Parkinson's disease nurse specialists—an important role in disease management. Neurology 1999;52( Suppl 3):S215.
  • 39
    Spottke EA, Volkmann J, Lorenz D, et al. Evaluation of healthcare utilization and health status of patients with Parkinson's disease treated with deep brain stimulation of the subthalamic nucleus. J Neurol 2002;249: 75966.
  • 40
    Dewey RB, Rao SD, Holmburg SL, Victor RG. Ergotamine/caffeine treatment of orthostatic hypotension in parkinsonism with autonomic failure. Eur J Neurol 1998;5: 5939.
  • 41
    Dodel RC, Eggert K, Oertel WH. Economical research on Parkinson's disease: a review of the literature. Eur J Neurol 1999;6(Suppl 2):S315.
  • 42
    Dodel RC, Spottke EA, Oertel WH. Quality of life and costs of neurosurgical procedures in the treatment of Parkinson's disease. Aktuel Neurol 2000; 27(Suppl 1):S619.
  • 43
    Jost WH. Costs in the treatment of parkinsonism. J Neurol 2000;247(Suppl 4):S313.
  • 44
    Pogarell O, Oertel WH. New dopamine agonists for the treatment of parkinsons-disease. Aktuel Neurol 1998;25: 2029.
  • 45
    Siderowf AD, Holloway RG, Stern MB. Cost-effectiveness analysis in Parkinson's disease: determining the value of interventions. Mov Disord 2000;15: 43945.
  • 46
    Linna M, Taimela E, Apajasalo M, Martilla RJ. Probabilistic sensitivity analysis for evaluating cost-utility of entacapone for Parkinson's disease. Expert Rev Pharmacoeconomics Outcomes Res 2002;2: 917.
  • 47
    Palmer CS, Nuijten MJC, Schmier JK, et al. Cost effectiveness of treatment of Parkinson's disease with entacapone in the United States. Pharmacoeconomics 2002;20: 61728.
  • 48
    Iskedjian M, Einarson TR. Cost analysis of ropinirole versus levodopa in the treatment of Parkinson's disease. Pharmacoeconomics 2003;21: 11527.
  • 49
    Hoerger TJ, Bala MV, Rowland C, et al. Cost effectiveness of pramipexole in Parkinson's disease in the US. Pharmacoeconomics 1998;14: 54157.
  • 50
    Davey P, Rajan N, Lees M, Aristides M. Cost-effectiveness of pergolide compared to bromocriptine in the treatment of Parkinson's disease: a decision-analytic model. Value Health 2001;4: 30815.
  • 51
    Nuijten MJC, Van Iperen P, Palmer CS, et al. Cost-effectiveness analysis of entacapone in Parkinson's disease: a Markov process analysis. Value Health 2001;4: 31628.
  • 52
    Tomaszewski KJ, Holloway RG. Deep brain stimulation in the treatment of Parkinsons's disease. A cost-effectiveness analysis. Neurology 2001;57: 66371.
  • 53
    Shimbo T, Hira K, Takemura M, Fukui T. Cost-effectiveness analysis of dopamine agonists in the treatment of Parkinson's disease in Japan. Pharmacoeconomics 2001;19: 87586.
  • 54
    Hawthorne G, Richardson J. Measuring the value of program outcomes: a review of multiattribute utility measures. Expert Rev Pharmacoeconomics Outcomes Res 2001;1: 21528.
  • 55
    Mitchell SL, Harper DW, Lau A, Bhalla R. Patterns of outcome measurement in Parkinson's disease clinical trials. Neuroepidemiology 2000;19: 1008.
  • 56
    Fahn S, Elton RL. Members of the UPDRS Development Committee. Unified Parkinson's disease rating scale. In: FahnS, MarsdenCD, CalneDB, GoldsteinM, eds., Recent Developments in Parkinson's Disease II. New York: Macmillan, 1987.
  • 57
    Schrag A, Jahanshahi M, Quinn N. How does Parkinson's disease affect quality of life? A comparison with quality of life in the general population. Mov Disord 2000;15: 111218.
  • 58
    Dodel RC, Singer M, Köhne-Volland R, et al. The economic impact of Parkinson's disease. An estimation based on a 3-month prospective analysis. Pharmacoeconomics 1998;14: 299312.
  • 59
    LePen C, Wait S, Moutard-Martin F, et al. Cost of illness and disease severity in a cohort of French patients with Parkinson's disease. Pharmacoeconomics 1999;16: 5969.
  • 60
    Di Rocco A, Molinari SP, Kollmeier B, Yahr MD. Parkinson's disease: progression and mortality in the L-DOPA era. Adv Neurol 1996;69: 311.
  • 61
    Hoehn MM, Yahr MD. Parkinsonism: onset, progression, and mortality. Neurology 1967;17: 42742.
  • 62
    Riederer P, Sian J, Gerlach M. Is there neuroprotection in Parkinson syndrome? J Neurol 2000; 247(Suppl 4):S811.
  • 63
    Wermuth L, Stenager EN, Boldsen J. Mortality in patients with Parkinson's disease. Acta Neurol Scand 1995;92: 558.
  • 64
    Treves TA, Pedro-Cuesta JD. Parkinsonism mortality in the US, 1. Time and space distribution. Acta Neurol Scand 1991;84: 38997.
  • 65
    Foltynie T, Brayne C, Barker RA. The heterogeneity of idiopathic Parkinson's disease. J Neurol 2002;249: 13845.
  • 66
    Poewe W, Gerstenbrand F. Klinische Klassifikation des Parkinson-Syndroms-Subtypen und Übergänge zu Multi-System-Atrophien. In: SchnabertG, AuffE, eds., Das Parkinson-Syndrom. Frankfurt: Editiones Roche, 1986.
  • 67
    Kuntz KM, Goldie SJ. Assessing the sensitivity of decision-analytic results to unobserved markers of risk: defining the effects of heterogeneity bias. Med Decis Making 2002;22: 21827.
  • 68
    Siebert U, Goldie SJ, Kuntz KM. Pharmacogenomics-Bias. Simulationen zur Fehlbeurteilung von Arzneimitteleffekten bei Nichtberücksichtigung genetischer Heterogenität (Abstract). 47th Annual Meeting of the German Society for Medical Informatics, Biometry, and Epidemiology (GMDS). Berlin, Germany, 2002.
  • 69
    Doubilet P, Begg CB, Weinstein MC, et al. Probabilistic sensitivity analysis using Monte Carlo simulation. A practical approach. Med Decis Making 1985;5: 15777.
  • 70
    Weinstein MC, Toy EL, Sandberg EA, et al. Modeling for health care and other policy decisions: uses, roles, and validity. Value Health 2001;4: 34861.
  • 71
    Buxton MJ, Drummond MF, Van Hout BA, et al. Modelling in economic evaluation: an unavoidable fact of life. Health Econ 1997;6: 21727.
  • 72
    Anonymous. Decision analytic modelling in the economic evaluation of health technologies. A consensus statement. Consensus Conference on Guidelines on Economic Modelling in Health Technology Assessment. Pharmacoeconomics 2000; 17: 4434.
  • 73
    Siebert U. When should decision-analytic modeling be used in the economic evaluation of health care? Eur J Health Econ 2003;4: 14350.