A systematic review of health economic models of opioid agonist therapies in maintenance treatment of non-prescription opioid dependence

Physical and psychological dependence can occur with any opioid drug, but the non-prescription or ‘street’ use of heroin presents the greatest problems to society [1]. In 2010, the global prevalence of opioid use was estimated at 0.6–0.8% of the population aged 15–64 years (between 26.4 and 36.0 million opioid users), of which approximately half, or between 13 and 21 million, were using heroin [2].

Illicit non-prescription opioid dependence is associated with major medical, personal and social problems, including increased risk of infection with human immunodeficiency virus (HIV) or hepatitis C virus (HCV), increased risk of death due to suicide, overdose or violence, decreased quality of life, high rates of psychiatric co-morbidity, and involvement in criminal activity [3, 4].

It was estimated in 2008 that there are approximately 16 million injecting drug users (IDU) worldwide and that 3 million (18.9%) of them were living with HIV. Global prevalence of HCV infection among injecting drug users in 2010 was 46.7%, meaning that some 7.4 million injecting drug users worldwide are infected with HCV and 2.3 million injecting drug users are infected with hepatitis B [2].

In 2010, illicit drug use was found to be associated with between 99,000 and 253,000 deaths globally, with drug-related deaths accounting for between 0.5 and 1.3% of all-cause mortality among those aged 15–64 years [2].

Heroin addiction also has significant economic costs to society, resulting from the association between crime and opioid dependence. Many opioid-dependent individuals become involved in crime to support their drug use, but crime may also provide the money and the contacts to buy drugs [1]. Based on a review, the average heroin user is likely to engage in criminal activity for 40–60% of the time they are not incarcerated or not in treatment [5].

Other societal impacts result from the psychopharmacological effects of the drug, which may result in mistakes at work, lost productivity or unemployment [6]. Personal relationships may suffer or parental capacity may be hampered. Evidence from the United Kingdom (UK) shows neglect among children is correlated strongly with parental heroin use, and in the United States of America (US) parental problem drug use is one of the commonest reasons for children entering the care system [7].

The economic and social costs of Class A drug use were estimated to be £15.4 billion in 2003–2004 in England and Wales [6], with opioid and/or crack use accounting for 99%. Health and social care costs accounted for £557 million, implying that the majority of the costs are borne outside of health and social care provision.

Effective treatment for opioid dependence is available but is likely to be long-term or even life-long. Options include psychosocial/behavioural assistance, and pharmacological interventions including opioid agonist therapy. The most commonly used medications for opioid substitution include the opioid agonist methadone, and buprenorphine (with/without naloxone), a partial agonist/antagonist combination; less commonly used treatments include naltrexone, morphine sulphate, naloxone, diamorphine, and medical use of heroin [8].

Recent Cochrane reviews, as well as additional studies of maintenance treatment options, have consistently found opioid agonist therapy to be clinically effective and more cost-effective than no drug therapy in opiate-dependent users [1], and have found no major differences in rates of mortality or illicit drug use achieved with these treatments [9].

In financially constrained health systems with finite resources, increasing emphasis is being placed on the ability to demonstrate that healthcare interventions are not only effective, but also cost-effective. Economic evaluations, which use modelling techniques to consider the comparative clinical effects, patient values and cost of care of alternative options, are used by payers and healthcare policymakers to inform the decision-making process.

The emphasis when conducting an economic evaluation is on including all relevant evidence on costs and outcomes. Given the significant impacts on society, the economic framework in economic evaluations of non-prescription opioid dependence should include not only the direct medical costs associated with treatment and preventive interventions, but also costs borne by other areas of society such as social welfare services and the criminal justice system, and indirect costs associated with lost productivity [4].

In order to explore the ways in which previous modelling studies have approached this issue, we undertook a systematic review of published model-based economic evaluations of opioid agonist therapy in treating opioid dependence, including modelling of long-term costs and outcomes and budget impact modelling. The aim was to identify any ‘best practice’ methods for modelling approaches and the costs and outcomes considered, which could be used to guide and inform future health economic models in this area. To our knowledge, this is the first systematic review to examine the modelling approaches that have been applied in opioid agonist therapy for non-prescription opioid dependence.

Opioid dependence is typically a chronic condition with a dynamic and variable course. Individuals may have periods of illicit drug use interspersed with periods of treatment and periods of abstinence, and outcomes vary greatly from one individual to another. Opioid dependence also has complex effects on wider society, for example on the criminal justice system and victims of crime, so economic models should aim to capture these aspects in order to reflect the decision-making framework.

To our knowledge, this systematic review is the first to evaluate economic modelling studies conducted in opioid agonist therapy for non-prescription opioid dependence.

Several different modelling approaches have been reported, each with its own strengths and limitations. Decision-tree models typically accepted by HTA bodies are simple to construct and useful where short time horizons are appropriate and the estimation of outcomes is straightforward, but they do not easily capture time dependency or recurrent events in chronic or complex diseases such as heroin addiction. However, decision-tree models can be adapted to better reflect aspects of time dependency and/or longer term outcomes. An example of this is incorporation of Monte Carlo simulations, as was done in two of the HTA models reported [1, 23].

Other approaches that overcome the issues associated with modelling long-term or chronic conditions with discrete health states include cohort models or Markov models. In this type of model, patients are in one of a number of finite health states, which are mutually exclusive and represent clinically and economically important events. Movements between health states are determined by transition probabilities. Decision-tree models, cohort models and Markov models are favoured by HTA agencies for their simplicity of construction and analysis. However, the transition probability of moving from one disease state to another is independent of the patient’s disease history, and there is a fixed cycle time before a patient is eligible to move into a new state. These are potentially important limitations in opioid dependence, as transitions are affected by individual patient characteristics, previous behaviour and treatment history [22]. Ways to address these limitations within the Markov/cohort framework include using additional health states to account for disease history (as long as the number of states does not become too cumbersome), or the transition probabilities can be made time-dependent [1].

Patient-level simulation models such as dynamic models and Monte Carlo simulations can also be used to capture stochastic variation in outcomes between individuals, and can take account of heterogeneity in factors such as demographic characteristics and previous history. Their main deficiency is lack of transparency with regard to data inputs and how they are combined within the model, coupled with the prolonged time taken to run sensitivity analyses [1].

Dynamic models, used by three studies identified in this review [12, 20, 21] to evaluate population effects associated with needle sharing, allow internal feedback loops and time delays that permit the modelling of health changes across entire populations or systems. They are well established in the study of infectious disease transmission through populations [1].

Four studies identified in this review used Monte Carlo simulation [1, 22‐24], and a further study used it for the sensitivity analyses [14]. However, these types of models are known to produce different results from those of static models, and the direction of the results may be unpredictable compared with models such as decision trees or Markov models. Comparisons of results between dynamic models and static models cannot easily be made. As a result, dynamic models may not be accepted by HTA assessment groups, as noted in a systematic review and economic evaluation of methadone and buprenorphine for the management of opioid dependence in 2007 [1].

A limitation of our review is that it did not include health economic models based solely on data from the duration of a randomised controlled trial, i.e. those that did not extrapolate beyond the trial duration. Whilst this does represent another possible modelling approach, the short duration of such trials means it is unlikely to be suitable for modelling a chronic condition such as non-prescription opioid dependence.

Model parameters such as time horizons were reflected in the modelling approach. Time horizons in decision analytical models ranged from 6 months to 1 year while cohort/simulation models ranged from 5 years to a lifetime. Modelling guidelines [30] recommend that the time horizon of a model should be long enough to capture all important differences in costs or outcomes between two or more treatments or between treatment and no treatment. Both models used in NICE HTA appraisals were 1 year in length [1, 23]. As opioid dependence may require long-term or life-long treatment [4], longer time horizons are likely to be more appropriate for models of opioid dependence, so that a realistic picture of lifetime costs and benefits of treatment can be adequately presented.

The model inputs reflected the perspective taken. Costs included in the economic evaluations were mainly those relating to healthcare service costs associated with opioid agonist therapy, including costs to primary care services and hospital services. Wider societal costs, such as costs of drug-related crime, disorder and antisocial behaviour, and loss of productivity in the workplace, were included in some form in only a few of the models. Impacts on family and social networks have not been included in any model. Costs to the individual patient, such as out-of-pocket costs, the costs related to premature death, drug-related illness and the loss of earnings through criminality/imprisonment, sickness, temporary or permanent unemployment and reduced educational attainment are also not routinely included in economic models. This may be because these types of costs are not ordinarily included in economic evaluations of healthcare interventions, so comparability with other interventions would be hampered if they were included. Indeed, HTA groups around the world have varying guidelines regarding costs to be included in a reference case for economic evaluations. In Sweden and The Netherlands, all relevant direct and indirect costs and revenues for treatment and ill health, irrespective of the payee, should be considered [29, 32]; in Norway, unrelated medical and non-medical costs should not be included [31]; in the UK a broader perspective on costs (beyond the NHS and Personal Social Services) may only be considered in exceptional circumstances [30]; while in Australia, PBAC mainly considers the costs of providing health care resources but may also consider costs and cost offsets of non-health care resources, although these might not be as influential in decision making as health care resources [33]. In the treatment of drug addiction, consistent omission of wider costs and benefits in the reported models means that the impact of effective maintenance treatments will not be fully captured; this will under-estimate the benefit of maintenance treatment to society. For example, reducing the spread of HIV will benefit non-drug users who will avoid HIV infection [11, 12] and is an important societal impact of maintenance treatment. In the US, total lifetime treatment cost for HIV based on new diagnoses in 2009 was estimated to be $16.6 billion [34].

The most common form of evaluation was cost-utility analysis, with an outcome measure of the number of QALYs gained. This may reflect the preference of HTA agencies for this type of analysis as it incorporates the value placed on health effects by society [30, 33]. However, there are limited data sources available for utility weights for substance abuse. Including the impact on quality of life is important because substance abuse is associated with significant co-morbidities and personal and social impacts, all of which should be considered when estimating the benefit of maintenance treatment. Thus, QALY weights should take account of reductions in health due to the disease of addiction itself and the impact of treatment on quality of life. In this review, only one source was available where the standard gamble method was used with members of the general public to obtain QALY weights for health states in substance abuse [16]. Choice-based preference measures to capture the value of health-related quality of life impacts are recommended by HTA groups such as NICE in the UK [30], PBAC in Australia [33] and HTA groups in European countries such as Norway [31], The Netherlands [29] and Sweden [32]. However, with addiction therapies and services the question could be asked about whose values should be used. Members of the public valuing vignettes developed by patients may be skewed by moral judgements, so perhaps people who have recovered from drug addiction might be the best-informed group.

This systematic review identified 18 economic modelling studies in non-prescription opioid dependence reporting a range of modelling approaches. There appears to be no single standard emerging as the preferred approach and there are a number of advantages and disadvantages to the different modelling approaches, some of which can be overcome using advanced modelling techniques. These factors, together with acceptability to HTA agencies, need to be considered when selecting a model structure for economic evaluation of a new therapy indicated for the maintenance treatment of non-prescription opioid dependence.

The most common evaluation type was cost-utility analysis reporting the cost per QALY gained (11/18 models), reflecting the preference of HTA agencies for this form of evaluation, although the data on utility weights are limited. Typical outcomes include mortality and treatment retention. However, wider societal costs such as the cost of crime (expenditure by the criminal justice system in dealing with crimes committed, and cost consequences for the victims of crime), productivity, employment impacts, transmission of HIV and HCV and costs of treatment borne by patients (for example, transportation costs and value of time spent receiving treatment) are not typically included in health economic modelling. Given the wide ranging impact of illicit opioid abuse on the workplace, the healthcare system, and in the communities, including these non-typical costs may help present a more complete story of the economic burden.

This review identified some key elements that could form a standard framework and guide for future models, such as:

The range of modelling approaches found in the literature could not easily be compared in terms of quality but the evidence does indicate that there is currently no single standard emerging as the preferred approach. Indeed, a number of the methodological approaches taken in different modelling studies could be used together to begin to build a framework for future models in this disease area. Furthermore, some of the disadvantages of current model structures could be overcome by using more advanced modelling techniques or choosing a different model approach. We propose that to develop a future model for the pharmacoeconomic evaluation of opioid agonist interventions, researchers should first identify the most robust model currently available, replicate the model, and then extend that model in terms of the range of inputs described above. Such a model should be applicable to a range of scenarios, e.g. intervention versus no intervention, comparison between interventions, changes over time, societal impact. Providing access to the model and the associated model code to the health care community could stimulate future dialogue and make it easier to recognise the strengths and weakness of different modelling approaches.