|Modelling the treated course of schizophrenia: development of a discrete event simulation model
|Heeg B M, Buskens E, Knapp M, van Aalst G, Dries P J, de Haan L, van Hout B A
This is a critical abstract of an economic evaluation that meets the criteria for inclusion on NHS EED. Each abstract contains a brief summary of the methods, the results and conclusions followed by a detailed critical assessment on the reliability of the study and the conclusions drawn.
The study examined unspecified methods to increase compliance (by 20%) with therapies for the treatment of schizophrenia. Initially, conventional antipsychotic medication was administered. This was switched to atypical antipsychotic drugs and, subsequently, to clozapine if the administered treatment was ineffective or the side effects were intolerable.
Type of intervention
Secondary prevention and treatment.
Economic study type
The target population comprised patients experiencing their second or third psychotic episode, who deteriorated over time, for whom the decision about potentially long-term medical therapy had been undertaken. First-time episode patients were excluded.
The setting was both the community and secondary care. The economic study appears to have been carried out in the Netherlands and the UK.
Dates to which data relate
Some of the effectiveness and health care resource use data used in the model were derived from papers published between 1984 and 2003. The costs were reported in 2002 prices.
Source of effectiveness data
The effectiveness data were derived from a review of the literature, which appears to have been non-systematic. When data were unavailable, expert panels were used.
Discrete events simulation (DES) was used to model the costs and outcomes. The time horizon considered was 5 years. A Dutch expert panel was used to develop the concepts used in modelling, and the model was refined through group discussion and individual interviews. A hypothetical sample of 2,000 patients was randomly generated from the baseline distributions assigned to patient characteristics. The model was run 2,000 times for each simulation.
Outcomes assessed in the review
The outcomes assessed were the probabilities of experiencing side effects (i.e. extrapyramidal symptoms, tardive dyskinesia, sedation, weight gain and agranulocytosis) with conventional therapy, atypical therapy and clozapine.
Study designs and other criteria for inclusion in the review
Sources searched to identify primary studies
Criteria used to ensure the validity of primary studies
Methods used to judge relevance and validity, and for extracting data
Number of primary studies included
Four studies appear to have been reviewed in order to obtain the model parameters exclusively derived from the review.
Methods of combining primary studies
Investigation of differences between primary studies
Results of the review
The probabilities of side effects with conventional therapy were 0.176 for extrapyramidal symptoms, 0.027 for tardive dyskinesia, 0.25 for sedation, 0.1 for weight gain and 0 for agranulocytosis.
The probabilities of side effects with atypical therapy were 0.07 for extrapyramidal symptoms, 0.006 for tardive dyskinesia, 0.07 for sedation, 0.12 for weight gain and 0 for agranulocytosis.
The probabilities of side effects with clozapine were 0 for extrapyramidal symptoms, 0 for tardive dyskinesia, 0 for sedation, 0 for weight gain and 0.01 for agranulocytosis.
Methods used to derive estimates of effectiveness
A UK expert panel populated the model's parameter values based on their own personal experience and their knowledge of the literature. All experts were asked to comment on the model structure, its assumptions, and the estimates of epidemiological and treatment patterns.
Estimates of effectiveness and key assumptions
The following parameters were estimated through the expert panel.
The probability of being compliant between episodes when in community treatment (0.6 with conventional treatment, 0.65 with atypical treatment and 0.65 with clozapine),
The probability of being compliant between episodes when in intense community treatment was 0.75 with conventional treatment, 0.8 with atypical treatment and 0.8 with clozapine.
The probability of being compliant between episodes when in a staffed hostel was 0.65 with conventional treatment, 0.7 with atypical treatment and 0.7 with clozapine.
The probability of being compliant between episodes when in hospital was 0.8 with conventional treatment, 0.85 with atypical treatment and 0.85 with clozapine.
A further series of parameters were used to describe the probabilities of switching treatment due to adverse events (i.e. extrapyramidal symptoms, tardive dyskinesia, sedation, weight gain agranulocytosis and more than two relapses on treatment), and those of changing treatment when in community treatment, intense community treatment, staffed hostel or hospital for patients who presented risk, patients who could take care of themselves, patients who could moderately take care of themselves, and patients who could not take care of themselves. The full set of parameter values was provided in Table 1 of the paper.
Measure of benefits used in the economic analysis
The number of relapses was used as the measure of benefit. This was discounted at a rate of 1.5% over the 5-year time horizon. Some other model outcomes reported in the paper were the total duration of psychosis, the time between relapses, the average Positive and Negative Syndrome Scale for Schizophrenia (PANSS) scores during relapse and between relapses, and the time on first-line treatment.
The perspective adopted for the cost estimation was that of the UK health care services. The direct medical costs included medication, visits to a psychiatrist, and costs associated with residing in particular locations (i.e. community treatment, intense community treatment, staffed hostel or hospital). These were estimated at 2002 prices. The costs of medications were derived from the Drug Tariff 2002. The costs associated with residing in particular locations were derived from the Personal Social Services Research Unit Report. The unit costs were reported separately, although the quantities of resources used were generated by the model itself. The costs were modelled over a 5-year period and were discounted at a rate of 6%. The type of costs estimated were the average costs per patient.
Statistical analysis of costs
The examination of uncertainty was restricted to the sensitivity analysis. The unit costs were treated as point estimates.
The indirect costs were not estimated.
One-way sensitivity analyses addressed:
the effect that atypical medication had on symptoms,
the threshold level at which symptoms would result in harm,
the costs in the various locations,
the potential risk by gender to cause self-harm,
the probabilities that medications would be switched,
the time between relapses, and
the level of severity.
A multi-way, probabilistic sensitivity analysis was also performed to evaluate second-order uncertainty. It was applied jointly to the above parameters as well as to compliance, excluding severity. No explicit justification was provided for the choice of the values used in the sensitivity analysis. The 95% confidence intervals (CIs) were reported for some of the results of the probabilistic sensitivity analysis.
Estimated benefits used in the economic analysis
The average number of relapses would be 4.08 if compliance could be reduced by 20% versus 3.53 if compliance were not improved (resulting in a reduction of 0.55). If compliance could be improved by 20%, there would be a reduction of 0.44 years in the total duration of psychosis, the time between relapses would increase by 0.44 years, the symptom scores would be reduced both during relapses and in between relapses, and the time on first-line treatment would be increased by 2.93 months.
None of the values used in the univariate sensitivity analysis had a substantial impact on the reduction in the number of relapses. The highest value was 0.61 when severity was assumed to be "non-severe", while the lowest was 0.48 when severity was assumed to be "very severe". The most influential parameter was the potential risk of causing harm. The results from the multivariate sensitivity analysis showed that the 95% CI for the increased number of relapses when compliance was improved was 0.04 to 0.99.
If compliance could be increased by 20%, the average (discounted) cost per patient would be 99,073. For a situation with no improvement in compliance, the cost per patient would be 82,925 (the incremental discounted savings amounted to 16,147).
When univariate sensitivity analyses were carried out, the costs were most sensitive to the potential risk of causing harm by gender, the probability of switching medication, and changes in the location costs. However, none of these variations from the base-case materially altered the implications of the model. The results of the multivariate sensitivity analysis showed that the 95% CI for the discounted cost-savings when compliance was improved was -39,821 to 274.
Synthesis of costs and benefits
The benefits and costs were not combined.
The authors' conclusions related primarily to the modelling methodology (discrete events simulation, DES), which they argued offers an easier way of promoting the necessary dialogue with clinicians than Markov modelling (where the dialogue relates to the more alien notion of transition probabilities). In terms of the results of the model, the authors concluded that increasing compliance by 20% resulted in cost-savings that were not sensitive to the parameter variations that they had tested.
CRD COMMENTARY - Selection of comparators
The authors considered current practice in the treatment of schizophrenia as the comparator for the economic analysis. They modelled the impact of an improvement in the patient's compliance. However, there were no details on how this improvement in compliance would be achieved, which introduces uncertainty into the validity of the intervention considered at analysis.
Validity of estimate of measure of effectiveness
The manner in which the estimation of effectiveness parameters was developed involved a review of the literature, which appears to have been non-systematic, and the use of two expert panels. One panel was involved in developing the conceptual model, while the other was involved in estimating the parameter values, both on the basis of personal experience and knowledge of the literature. It was difficult to comment on the validity of the estimates obtained from the review of the literature, as little information on the methods used to identify, select and assess the primary studies was given. The authors did not fully describe the process by which parameters were estimated from the expert panels, nor did they report the number of experts involved in the process. However, it appears to have involved an attempt to achieve expert consensus. The parameter values were tested using a sensitivity analysis, although it was not reported how the extreme values were selected. A limitation reported by the authors was that information obtained from experts is often anecdotal and decisions about how to treat patients are often taken after the consideration of a broad range of issues. A further limitation reported was that the data around treatment effectiveness often related to a short-term period.
Validity of estimate of measure of benefit
The summary measure of health benefit was specific for the type of disease and treatments considered at analysis. It was obtained through the modelling exercise and appears to have been appropriate. However, it does not enable the study results to be compared with those obtained for different interventions. The authors reported that there were no data about quality of life in relation to treatments for schizophrenia.
Validity of estimate of costs
The cost perspective adopted was that of the UK health care service. Most of the important costs appear to have been included in the analysis. However, given that the intervention used to increase compliance was not identified in the study, the associated costs of increasing compliance do not appear to have been included in the costing, and this would bias the results of the cost estimation in favour of the situation of increased compliance. The unit costs (from published sources) and total costs (generated by the model and its parameters) were reported. Extensive sensitivity analyses of the costs were performed to examine variability in the model parameters, although the choice of the ranges used was not justified. The unit costs appear to have been treated as point estimates, with no sensitivity analysis provided.
The authors compared their cost estimates with those from another study. They concluded that the fact that their own estimate was almost twice that of the other study might be accounted for by inflation and differing case severity. The issue of the generalisability of the results to other settings was not addressed. The authors highlighted a number of other limitations in their mode. First, the modelling of compliance (in reality compliance can take place in a variety of ways). Second, psychiatrists may switch therapies for several reasons. Third, the correlation between various side effects. Finally, the fact that severity can be measured along a range of dimensions.
Implications of the study
The authors suggested that further research to quantify the issue of whether patients are able to take care of themselves, and how this would impact on the effectiveness of the treatments, could be worthwhile. The study presented several caveats that should be taken into consideration when interpreting the results.
Source of funding
Supported by Janssen Pharmaceutica, N.V., Belgium.
Heeg B M, Buskens E, Knapp M, van Aalst G, Dries P J, de Haan L, van Hout B A. Modelling the treated course of schizophrenia: development of a discrete event simulation model. PharmacoEconomics 2005; 23(Supplement 1): 17-33
Other publications of related interest
Csernansky JG, Mahmoud R, Brenner R, the Risperidone-USA-79 Study Group. A comparison of risperidone and haloperidol for the prevention of relapse in patients with schizophrenia. N Engl J Med 2002;346:16-22.
Geddes J, Freemantle N, Harrison P, et al. Atypical antipsychotics in the treatment of schizophrenia: a systematic overview and meta-regression analysis. BMJ 2000;321:1371-6.
Guest J, Cookson R. Cost of schizophrenia to UK society. An incident-based cost-of-illness model for the first 5 years following diagnosis. Pharmacoeconomics 1999;15:597-610.
Subject indexing assigned by NLM
Adult; Antipsychotic Agents /economics /therapeutic use; Decision Making; Female; Humans; Male; Models, Economic; Schizophrenia /drug therapy /economics /therapy; Self Care /classification
Date bibliographic record published
Date abstract record published