Use of opioid substitution therapies in the treatment of opioid use disorder: results of a UK cost-effectiveness modelling study

Abstract

Aims: This study investigated the cost-effectiveness of buprenorphine maintenance treatment (BMT) and methadone maintenance treatment (MMT) vs no opioid substitution therapy (OST) for the treatment of opioid use disorder, from the UK National Health Service (NHS)/personal social services (PSS) and societal perspectives over 1 year.

Methods: Cost-effectiveness of OST vs no OST was evaluated by first replicating and then expanding an existing UK health technology assessment model. The expanded model included the impact of OST on infection rates of human immunodeficiency virus (HIV) and hepatitis C virus (HCV) infection.

Results: Versus no OST, incremental cost-effectiveness ratios (ICERs) for BMT and MMT were £13,923 and £14,206 per quality-adjusted life year (QALY), respectively, from a NHS/PSS perspective. When total costs (NHS/PSS and societal) are considered, there are substantial savings associated with adopting OST; these savings are in excess of £14,032 for BMT vs no OST and £17,174 for MMT vs no OST over 1 year. This is primarily driven by a reduction in victim costs. OST treatment also impacted other aspects of criminality and healthcare resource use.

Limitations: The model’s 1-year timeframe means long-term costs and benefits, and the influence of changes over time are not captured.

Conclusions: OST can be considered cost-effective vs no OST from the UK NHS/PSS perspective, with a cost per QALY well below the UK’s willingness-to-pay threshold. There were only small differences between BMT and MMT. The availability of two or more cost-effective options is beneficial to retaining patients in OST programs. From a societal perspective, OST is estimated to save over £14,032 and £17,174 per year for BMT and MMT vs no OST, respectively, due to savings in victim costs. Further work is required to fully quantify the clinical and health economic impacts of different OST formulations and their societal impact over the long-term.

Keywords:

• Opioid dependence
• opioid use disorder
• pharmacologic (drug) maintenance therapy
• economic evaluation
• societal costs

Introduction

Addiction can have a significant impact in some communities. There are an estimated 122,894 injecting drugs users (95% CI = 117,370–131,869), or 3.0 per 1,000 inhabitants aged 15–64 (95% CI = 2.87–3.22) in the UK¹ and 1.3 million in Europe². Effective treatment for opioid use disorder is likely to be long-term, and comprise psychological and behavioral assistance that can be supplemented with pharmacological interventions including opioid substitution therapy (OST) with methadone, buprenorphine (generic and Subutex), or buprenorphine/naloxone (Suboxone)^3,4.

Intravenous illicit heroin use is associated with a risk of overdose and of transmission of infectious diseases⁴. Therefore, drug policies are often aimed at reducing the morbidity and mortality associated with problematic drug consumption⁴, and effective OST programs can lead to reduced rates of problem heroin use^5,6. Additionally, there is compelling evidence that, among people infected with human immunodeficiency virus (HIV), OST can help recruitment onto, and adherence to, antiretroviral therapy⁷.

When appraising new healthcare interventions, there is often a requirement for evidence of their cost effectiveness to inform the resource allocation decision-making process⁸. This may require the development of health economic models. A recent systematic review of economic models that have been used in appraising OST revealed a limited number of models in the peer-reviewed literature, and no clear consensus to modeling approaches⁹. To explore this area further, we replicated an existing health technology assessment (HTA) model developed by Connock et al.¹⁰, and updated the model using 2016 costs. As studies have demonstrated that OST can reduce risk behaviors and HIV transmission⁴, we expanded the model to include the impact of HIV and/or hepatitis C virus (HCV) infection. Using the model, we investigated the cost-effectiveness of buprenorphine maintenance treatment (BMT) and methadone maintenance treatment (MMT) vs no OST, from UK National Health Service (NHS)/personal social services (PSS) and societal perspectives (the societal perspective included criminal justice costs as well as victim costs). Finally, we prepared an open source version of the model using R version 3.2.3, which is available to other investigators to review and further develop.

Methods

To study the cost-effectiveness of OST vs no OST, we began by replicating the cost-effectiveness model developed by Connock et al.¹⁰. This model was chosen because it was originally developed to assess the cost-effectiveness of BMT compared to MMT, as part of a UK HTA. It was also sufficiently well documented in terms of structure and data sources to allow us to use that information to construct our own model.

Once replicated, we updated the model to include 2016 prices for each cost parameter and expanded it to include the impact of infection with HIV and/or HCV. We then used the model to evaluate the cost-effectiveness of BMT and MMT (generics) vs no OST from both NHS/PSS and societal perspectives.

Replicating the original model

We replicated the original decision tree model outlined in Connock et al.¹⁰ (Figure 1). The model follows patients for 1 year. The main parameter of interest is retention in treatment. At 2, 6, 13, 25, and 52 weeks, patients are either retained in treatment or they drop out. At each of these time-points a utility value and cost are applied. Retention in treatment was derived from generating a survival curve for BMT, based on a Cochrane systematic review of flexible dose regimens comparing BMT and MMT, then applying the hazard ratio for the difference in retention between BMT and MMT to derive the MMT survival curve. Additional input parameters for level and nature of drug misuse and utility values are outlined in Table 1. Both NHS/PSS and societal (criminal justice system and victims of crime) resource use and costs (2004) were included. The model used UK information, where available, otherwise data from England and Wales were used.

Figure 1. Decision tree model structure (adapted from Connock et al.¹⁰).

Table 1. Level and nature of drug misuse and utility values (derived from Connock et al.¹⁰).

Parameter	Input
Proportions
OST and not drug misusing	0.3850
OST and drug misusers (injectors)	0.4400
OST and drug misusers (non-injectors)	0.5600
No OST and not drug misusing^a	0.0000
No OST and drug misusers (injectors)	0.6100
No OST and drug misusers (non-injectors)	0.3900
Weighted treatment specific utility values
OST: Buprenorphine
Week 1–2	0.7039
Week 3–52	0.7455
OST: Methadone
Week 1–2	0.7017
Week 3–52	0.7458
No OST	0.6231

^a Patients who dropped out of treatment were assumed to return to their previous drug misuse habits.

OST: opioid substitution therapy.

The model was replicated in Microsoft Excel, which was selected instead of the TreeAge software used in the original publication because it is more widely available and easier to validate. The National Institute for Health and Care Excellence (NICE) Decision Support Unit considers both packages to be suitable¹¹. The replicated model was then run using the same data inputs for resources and costs as published in the original report¹⁰. Where data inputs were not fully described in the original report, we tested a number of potential assumptions to identify the one that most closely replicated the outputs from the original model. The replicated model was run with the same 1-year time horizon as the original Connock et al.¹⁰ model.

The comparison of results of our replicated model with the original Connock et al.¹⁰ model is outlined in Table 2. There were a number of potential reasons for the small differences that are outlined in the discussion. However, we were confident that we had replicated the original model as closely as possible, with the difference in cost no more than £271 and in quality-adjusted life years (QALYs) of 0.0001.

Table 2. Comparison of Connock et al.¹⁰ model, with the replicated model results (2004 costs).

	QALYs		NHS/PSS costs		Societal costs
	Connock et al.^10 model	Replication	Connock et al.^10 model	Replication	Connock et al.^10 model	Replication
BMT	0.6774	0.6773	£2,491	£2,221	£30,992	£30,721
MMT	0.6990	0.6899	£1,971	£1,916	£28,345	£28,307
Difference of replication from Connock et al.^10 model
BMT	−0.0001		−£270		−£271
MMT	−0.0001		−£55		−£38

BMT: buprenorphine maintenance treatment; MMT: methadone maintenance treatment; NHS: National Health Service; PSS: personal social services; QALY: quality-adjusted life year.

Updating inputs

Cost parameters were updated to 2016, either by sourcing new data or uprating the original values using the relevant price index (Table 3). In terms of drug costs, the original Connock et al.¹⁰ model included the price for branded buprenorphine; however, as generic buprenorphine is now available, we used the generic price as the base case value.

Table 3. Resource use (derived from the Connock et al.¹⁰) and costs (2016 values).

Parameter	Resource use^10	Unit cost	Source
Drug costs (per mg)^a
Methadone		£0.0126	NHSBSA^26; 2016
Buprenorphine		£0.0803
Subutex^b		£0.3739	BNF^27; 2016
Suboxone^b		£0.3739
Dispensing fees (per treatment dispensed)
Buprenorphine, Subutex, Suboxone		£1.13	PSNC^28; 2016
Methadone		£1.13
Controlled drug fee (per treatment dispensed)
Buprenorphine, Subutex, Suboxone		£1.28
Methadone		£2.50
Supervised self-administration (per administration)
Buprenorphine, Subutex, Suboxone		£3.00	Thurrock Council^29; 2016
Methadone		£1.50
Counseling session per week	1.0	£10.91	Connock et al.^10; 2007^c
Urine test per week	0.5	£1.43	Connock et al.^10; 2007^c
GP visit per year
Successful health state	5.6	£36.00	Curtis^30, 2016
Unsuccessful health state	3.6
A&E visit per year
Successful health state	0.8	£163.24	NHS reference costs^31, 2015/16
Unsuccessful health state	0.7
Inpatient hospital stay per year
Successful health state	2.8	£470.21
Unsuccessful health state	1.75
Out-patient mental health visit per year
Successful health state	0.8	£101.46
Unsuccessful health state	1.3
Inpatient mental health visit per year
Successful health state	0.4	£429.00
Unsuccessful health state	1.5
Rate of drug arrests per year^e
Successful health state	0.80	£5,592.11	Connock et al.^10; 2007^d
Unsuccessful health state	0.30
Rate of acquisitive crime arrests per year^e
Successful health state	1.60	£2,119.68
Unsuccessful health state	1.35
Average time held in police custody per year (nights)^e
Successful health state	1.20	£108.66
Unsuccessful health state	2.00
Rate of court appearances per year^e
Successful health state	1.40	£1,100.78
Unsuccessful health state	2.20
Time spent in prison per year (days)^e
Successful health state	34.00	£108.44
Unsuccessful health state	36.00
Annual victim costs (per person on or off OST)^f
Successful health state		£14,004.69
Unsuccessful health state		£48,546.33

A&E: accident and emergency; BNF: British National Formulary; CPI: consumer price index; GP: general practitioner; HCHS: Hospital and Community Health Services; OST: opioid substitution therapy; PSNC: Pharmaceutical Services Negotiating Committee.

^a The model assumes a flexible dosing regimen and uses data on mean dose derived from Connock et al.¹⁰. Methadone cost/mg: Methadone Cost for 500 ml (1 mg/ml). Buprenorphine cost/mg: 30% at buprenorphine price for 7 tab (2 mg/tab) and 70% at buprenorphine price for 7 tab (8 mg/tab). Subutex cost/mg: 30% at Subutex price for 7 tab (2 mg/tab) and 70% at Subutex price for 7 tab (8 mg/tab). Suboxone cost/mg: 30% at Suboxone price for 7 tab (2 mg/tab) and 70% at Suboxone price for 7 tab (8 mg/tab).

^b Used in the sensitivity analysis only.

^c Inflated to 2016 using HCHS inflator³⁰.

^d Inflated to 2016 using CPI³².

^e Includes existing and new contacts with criminal justice services both on and off OST.

^f Includes costs incurred in anticipation of crime, e.g. installation of burglar alarms, as well as direct costs such as material or physical damage or loss.

Expanded model

In terms of inputs, the model was expanded to include HIV and/or HCV infection rates and associated costs and utilities, sourced from UK data, where available, otherwise data from England and Wales were used (Table 4). Patients enter the model with a prevalence of HIV and/or HCV. Once the patients enter the model, an incidence of HIV and/or HCV is applied to all patients without the disease based on drug misuse status and then added to the background prevalence. These figures are used to derive the proportion of patients with/without HIV and/or HCV in the OST and no OST arms. It is assumed that the prevalence/incidence rates of HIV among drug misusers who inject and drug misusers who don’t inject is the same based on several observational studies of drug misusers in treatment programs¹². Direct costs and utility decrements associated with the disease were applied (Table 4).

Table 4. Expanded model inputs.

Parameter	Input	Source
Proportions
Infection rate for patients who are on OST and not current drug misusers
HIV	0.0019	Calculated prevalence plus annual incidence calculated from Public Health England, Home office and Office of National Statistics 2016 data^33–37. Calculations shown below^a to^c
HCV	0.4961
HIV/HCV	0.0078
Infection rate among drug misusers (injecting)
HIV	0.0033
HCV	0.5396
HIV/HCV	0.0130
Infection rate among drug misusers (non-injecting)
HIV	0.0033
HCV	0.4908
HIV/HCV	0.0130
Costs
HIV infection (annual cost)
Direct cost: cost of treatment (triple therapy) and health service costs	£31,141.78	Mandalia et al.^38, 2010*
HCV infection (annual cost)
Direct cost of standard treatment (peginterferon alfa and ribavirin) and health service costs	£1,152.92	Calculated from NICE^39, 2009 (lifetime costs), Messina et al.^40, 2015 (genotype prevalence) and Campos et al.^41, 2007 (average duration of HCV)* Calculations shown below^d
Direct cost of new treatments** (interferon-free treatments) and health service costs	£44,852.00	Barclay et al.^42, 2016
Utilities†
OST and not drug misusing
No HIV/HCV	0.8673	Connock et al.^10, 2007
HIV	0.7806	Calculated using Zaric et al.^43, 2000‡
HCV	0.7806	Assumption§
OST and drug misusers (injectors)
No HIV/HCV	0.6332	Connock et al.^10, 2007
HIV	0.5699	Calculated using Zaric et al.^43, 2000‡
HCV	0.5699	Assumption§
OST and drug misusers (non-injectors)
No HIV/HCV	0.6834	Connock et al.^10, 2007
HIV	0.6151	Calculated using Zaric et al.^43, 2000‡
HCV	0.6151	Assumption§
No OST and not drug misusing
No HIV/HCV	0.8673	Connock et al.^10, 2007
HIV	0.7806	Calculated using Zaric et al.^43, 2000‡
HCV	0.7806	Assumption§
No OST and drug misusers (injectors)
No HIV/HCV	0.5880	Connock et al.^10, 2007
HIV	0.5292	Calculated using Zaric et al.^43, 2000‡
HCV	0.5292	Assumption§
No OST and drug misusers (non-injectors)
No HIV/HCV	0.6780	Connock et al.^10, 2007
HIV	0.6102	Calculated using Zaric et al.^43, 2000‡
HCV	0.6102	Assumption§

Calculations

^a Calculation of infection rate for patients who are on OST and not current drug misusers:

HIV: ((Prevalence HIV among persons who inject drugs in UK³³ × (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴) × UK Population 16 to 59 years in 2015³⁵) + ( Population with New HIV Diagnosis in 2015³⁶ × Proportion of 16-59 year olds who took Opioids in the last 3 months³⁴)) ÷ (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16 to 59 years in 2015³⁵).

HCV: ((Prevalence HCV among persons who inject drugs in UK³³ × (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴) × UK Population 16 to 59 years in 2015³⁵) ÷ (Proportion of 16-59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16 to 59 years in 2015³⁵)).

^b Calculation of infection rate among drug misusers (injecting):

HIV: ((Prevalence HIV among persons who inject drugs in UK³³ × (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16–59 years in 2015³⁵)) + (Population with New HIV Diagnosis who inject drugs³⁶) ÷ (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴) × UK Population 16–59 years in 2015³⁵).

HCV: (((Prevalence HCV among persons who inject drugs in UK³³ × (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16–59 years in 2015³⁵)) ÷ Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16–59 years in 2015³⁵ + Proportion of injectors who contract HCV)³⁷).

^c Calculation of infection rate among drug misusers (non-injecting):

HIV: ((Prevalence HIV among persons who inject drugs in UK³³ × (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16–9 years in 2015³⁵) + Population with New HIV Diagnosis who take non-inject drugs³⁶) ÷ (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴) × UK Population 16–59 years in 2015³⁵).

HCV: ((Prevalence of HCV among persons who inject drugs in UK³³ × (Proportion of 16-59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16 to 59 years in 2015³⁵)) ÷ (Proportion of 16–59 year olds who took Opioids in the last 3 months³⁴ × UK Population 16 to 59 years in 2015³⁵)).

^d Calculation of HCV annual cost = ((AVERAGE (Lifetime costs for Genotype 1 {Std treatment duration} in NICE 2009 (Lui study)³⁹, Lifetime costs Genotype 1 {Short treatment duration} in NICE 2009 (Lui study)³⁹, Lifetime costs Genotype 1 {Std treatment duration} in NICE 2009 (Yu study)³⁹, Lifetime costs Genotype 1{Short treatment duration} in NICE 2009 (Yu study)³⁹) × (Population with HCV Genotype 1[40]/(Population with HCV Genotype 1⁴⁰ + Population with HCV Genotype 2⁴⁰ + Population with HCV Genotype 3⁴⁰))) + (AVERAGE (Lifetime costs Genotype 2 & 3 {Std treatment duration} in NICE 2009 (Wagner study)³⁹, Lifetime costs Genotype 2 & 3 {Short treatment duration} in NICE 2009 (Wagner study)³⁹) × ((Population with HCV Genotype 2⁴⁰ + Population with HCV Genotype 3⁴⁰)/(Population with HCV Genotype 1⁴⁰ + Population with HCV Genotype 2⁴⁰ + Population with HCV Genotype 3⁴⁰)))) ÷ Average duration of HIV infection⁴¹.

HCV, hepatitis C virus; HCHS, Hospital and Community Health Services; HIV, human immunodeficiency virus; NHS, National Health Service; NICE, National Institute for Health and Care Excellence; ONS, Office for National Statistics; OST, opioid substitution therapy

* Inflated to 2016 using HCHS inflator³⁰.

** It was assumed a proportion of patients with HCV would receive the new interferon-free treatments that results in a cure. The proportion was estimated from the increase in HCV treatment on previous years reported by Public Health England in the 2016 report, Hepatitis C in the UK³⁷.

† HIV/HCV co-infection utilities were calculated by combining the utility values for HIV and HCV using methodology recommended by the NICE decision support unit⁴⁴.

‡ A quality adjustment multiplier of 0.90 was applied to the utility values of patients with no HIV/HCV to account for the reduced quality-of-life associated with HIV⁴³.

§ It was assumed that the adjustment in utility as a result of having HCV was the same as that for HIV, based on a study comparing the utility values across HIV and HCV⁴⁵.

In terms of outputs, the model included the impact of retention on a number of healthcare resources and non-clinical outcomes over a 1-year time horizon, which were drawn from the original work by Connock et al.¹⁰ and the self-reported National Treatment Outcome Research Study (NTORS) study^13–16. Outcomes reported included healthcare resource use (general practitioner [GP] visits, rates of accident and emergency [A&E] visits, inpatient hospital stays, out-patient mental health visits and inpatient mental health visits) and contacts with the criminal justice system (rates of drug arrests, acquisitive crime arrests, court appearances, and average time held in police custody (nights), and time spent in prison per year), as well as annual victim costs.

A model using R code was developed, and the outputs were validated against the Excel model to check for robustness. A publicly open source version of the model can be found in the supplementary files.

Sensitivity analyses

The deterministic sensitivity analysis tested the robustness of the model by varying certain parameters. We evaluated the impact of unit drug price on cost-effectiveness by determining the cost at which the incremental cost-effectiveness ratio (ICER) exceeded the threshold set by NICE (£30,000 per QALY). We also assessed the impact of branded buprenorphine (Subutex) and the branded combination of buprenorphine and naloxone (Suboxone) on the ICER. In the absence of long-term comparative clinical data demonstrating a difference between Subutex and Suboxone retention rates, we made the conservative assumption that they were the same, based on a short-term randomized controlled trial¹⁷. Additionally, we investigated the impact of removing victim costs from the overall economic societal costs. Probabilistic sensitivity analysis were conducted for BMT and MMT vs no OST. The results are presented in the form of cost-effectiveness planes and cost-effectiveness acceptability curves (CEACs) of incremental costs and outcomes to evaluate the uncertainty in results due to statistical variability around the parameter estimates. The cost-effectiveness plane graphically illustrates the ICER of multiple probabilistic simulations of a strategy. The CEAC curves use probabilistic simulations to demonstrate the likelihood a strategy is cost-effective at different threshold values of willingness-to-pay for an additional QALY.

Results

Impact of treatment vs no OST

We applied the replicated and expanded cost-effectiveness model to the use of either BMT or MMT vs no OST in the management of opioid use disorder. From a NHS/PSS perspective, vs no OST, OST is cost-effective, with ICERs of £13,923/QALY for BMT and £14,206/QALY for MMT (Table 5), i.e. OST offers incremental health benefits (QALY increase) over no OST at a modest expenditure on drugs.

Table 5. Results of the cost-effectiveness analysis.

	Cost (£)	Cost (£) difference	QALYs	QALY difference	ICER (£/QALY) vs no OST
NHS/PSS perspective
No OST	9,647		0.5883
BMT	10,362	716	0.6398	0.0514	13,923
MMT	10,541	895	0.6514	0.0630	14,206
Societal perspective
No OST	69,274		0.5883
BMT	55,243	–14,032	0.6398	0.0514	Dominant
MMT	52,100	–17,174	0.6514	0.0630	Dominant

BMT: buprenorphine maintenance treatment; ICER: incremental cost-effectiveness ratio; MMT: methadone maintenance treatment; NHS: National Health Service; OST: opioid substitution therapy; PSS: personal social services; QALY: quality-adjusted life year.

When total costs (NHS/PSS and societal) are considered, there are substantial savings associated with adopting OST, in excess of £14,032 for BMT vs no OST and £17,174 for MMT vs no OST (Table 5). For the two treatment arms, the total societal costs (which include criminal justice system and victim costs) are £55,243 and £52,100 for BMT and MMT, respectively, vs £69,274 for those not on treatment. Therefore, from a societal perspective, BMT and MMT offer both clinical improvement and substantial cost savings, i.e. they are dominant over the “no OST” scenario.

The results of our modeling exercise suggest that MMT may be associated with marginally better reductions in societal costs compared to BMT. This is based on minimal differences in the clinical efficacy in the input data used. A recent systematic review concluded that BMT and MMT may be equally effective from a clinical perspective¹⁸. Nonetheless, from both NHS/PSS and societal perspectives, the differences in costs between BMT and MMT are small compared to those between OST and no OST.

Impact of intervention on healthcare resources

Over the cycle of the model, OST results in a modest increase in GP and A&E visits, as well as inpatient hospitalizations (Table 6). On the other hand, there is a substantial decrease in the use of in- and out-patient mental health resources. It should be noted that these data only applied to 1 year of OST, and the model was not extended to determine the long-term impact of OST on healthcare resource use.

Table 6. Impact of OST on healthcare resource and non-clinical outcomes.

Outcome	BMT	MMT	No OST
Healthcare resource use (over 1 year)
GP visits	4.51	4.72	3.60
Rate of A&E visits	0.75	0.76	0.70
Rate of inpatient hospital stays	2.23	2.34	1.75
Rate of out-patient mental health visits	1.07	1.02	1.30
Rate of inpatient mental health visits	1.00	0.89	1.50
Criminal justice and victim costs (over 1 year)
Victim costs	£32,824	£29,282	£48,546
Rate of drug arrests	0.53	0.58	0.30
Rate of acquisitive crime arrests	1.46	1.49	1.35
Average time held in police custody (nights)	1.64	1.55	2.00
Rate of court appearances	1.84	1.75	2.20
Time spent in prison (days)	35.09	34.88	36.00

A&E: accident and emergency; BMT: buprenorphine maintenance treatment; MMT: methadone maintenance treatment; GP: General Practitioner; OST: opioid substitution therapy.

Impact of intervention on non-clinical outcomes

The impact of OST on societal cost drivers is outlined in Table 6. The greatest impact of OST is on the reduction in victim costs. The annual victim costs for a patient not in treatment is £48,546, compared to £32,824 for BMT and £29,282 for MMT. Using self-reported data from the NTORS^10,14, it appears that the rate of drug arrests and acquisitive crime arrests are higher during the first year of OST in individuals in OST programs, compared to rates for those not in treatment. Other aspects of involvement with the criminal justice system, including time in custody, court appearances, and time in prison, are slightly lower or unaffected by OST in the first 12 months of OST compared with no OST.

Sensitivity analysis

The deterministic sensitivity analysis evaluated the impact of unit drug price on cost effectiveness, as well as the impact of victim costs on overall economic societal burden.

Assuming a cost-effectiveness threshold of £30,000/QALY, we evaluated the cost point of OST at which it would no longer be considered cost-effective by NICE. With all other parameters unchanged, an ICER of £30,000/QALY would be associated with an increase of unit costs of 14-fold and 12-fold for BMT and MMT, respectively. Use of Subutex or Suboxone (price is the same), rather than generic buprenorphine, would increase the ICER to £20,963/QALY (NHS/PSS perspective).

Victim costs were by far the biggest driver of societal costs. When they are removed from the model, both BMT and MMT were no longer dominant over no OST, and had an ICER of £32,871 and £33,154 per QALY, respectively.

The probabilistic sensitivity analysis for BMT and MMT vs no OST are presented in the form of cost-effectiveness planes in Figures 2 and 3 and CEACs in Figures 4 and 5. The cost-effectiveness plane for BMT and MMT indicate that nearly all simulations were costlier than no OST, with the majority having a positive impact on QALYs. These results were consistent with the result of the base case. The CEAC curves indicate that there is a 64% probability that both BMT and MMT are cost-effective at a threshold of £30,000/QALY.

Figure 2. Incremental cost-effectiveness plane for BMT vs No OST. BMT, buprenorphine maintenance treatment; OST: opioid substitution therapy.

Figure 3. Incremental cost-effectiveness plane for MMT vs No OST. MMT, methadone maintenance treatment; OST: opioid substitution therapy.

Figure 4. Cost-effectiveness acceptability curve for BMT vs No OST. BMT, buprenorphine maintenance treatment; OST: opioid substitution therapy.

Figure 5. Cost-effectiveness acceptability curve for MMT vs No OST. MMT, methadone maintenance treatment; OST: opioid substitution therapy.

Discussion

We evaluated the cost-effectiveness of OST (MMT and BMT) vs no OST from both NHS/PSS and societal perspectives. From the NHS/PSS perspective, the ICERs for BMT (£13,923 per QALY vs no OST) and MMT (£14,206 per QALY vs no OST) were well below the UK NICE willingness-to-pay threshold (£30,000). From a non-NHS/PSS perspective, adoption of either form of OST would result in considerable savings due to a substantial reduction in victim-related costs (total cost savings of £14,032 and £17,174 per year for BMT and MMT vs no OST, respectively).

We were able to almost exactly replicate the Connock et al.¹⁰ results in our model. There were some slight differences, with most of the variance being accounted for by rounding effects, differences between TreeAge and Excel, half-cycle corrections and dosing information, uncertainty regarding doses per day, resource use was presented in months but cycle times were in weeks, and possible double counting of unsupervised dispensing over the first 6 months in the original model.

Demonstrating the cost-effectiveness of OST interventions will be important to aid healthcare decision-making in some countries. Using the replicated and expanded model, we demonstrated from the NHS/PSS perspective that both MMT and BMT are cost-effective, compared with no OST, within the ICER limits set by NICE in the UK¹⁹. The cost-effectiveness was similar for both forms of intervention, helping to support the observations elsewhere that availability of more than one intervention is important to help maximize retention in OST^3,18,20,21. Additionally, our evaluation of the role of unit drug cost on the ICER indicates that the cost/mg can increase by up to 14-times before OST is no longer cost-effective.

As observed by the World Health Organization³, methadone is often favored over buprenorphine. However, there may be times when buprenorphine is preferred as first-line OST, for example as a result of contraindications to methadone or patient preference due to adverse events^3,20,21. The Australian government identifies the following as drivers of preference for buprenorphine over methadone: individual differences in pharmacokinetic and pharmacodynamics, side-effects (in particular the sedative effects of methadone), and treatment flexibility due to milder withdrawal symptoms compared with those for methadone²¹. Slow-release oral morphine is another option for intervention that appears to be at least as effective as methadone, with positive effects on heroin use, mental symptoms, and patient-reported outcomes including heroin craving and treatment satisfaction^22–25.

The impact of opioid use disorder extends beyond the healthcare system^13,15. In the original Connock et al.¹⁰ model, victim-related costs were mainly drawn from the reported patterns of offences from the NTORS¹⁴, combined with estimates of the costs of crime from published literature to arrive at an estimate for the annual victim costs for drug users in treatment and for drug users not in treatment. These consisted of both costs incurred in anticipation of crime, e.g. installation of burglar alarms, as well as direct costs such as material or physical damage or loss. We found that there is a substantial economic benefit from the societal perspective that is greater than the investment needed in OST with methadone or buprenorphine. Recreating the analysis by Connock et al.¹⁰, omission of victim costs results in the loss of substantial cost savings to society, demonstrating that victim costs are the key driver of overall societal burden of heroin addiction. In this context, it is clear that a reduction in drug-related crime should be considered an important outcome of interventions¹³.

We also demonstrated an impact of OST on other non-clinical outcomes. In some cases there were reductions, whereas in others there was a small increase during the first year of OST, and there were some differences between MMT and BMT. Consequently, there is no clear interpretation of the impact of MMT and BMT on non-clinical outcomes. Given the scale of potential savings by adopting OST, there is a need to fully understand the societal impact of OST. This will likely require real-world evidence studies to inform future health economic models.

We consider the development of the model a step forward in providing researchers a methodology for testing the cost-effectiveness of OST formulations. However, further work is required to develop the model fully, principally due to the limited data in this area⁹. Additionally, future models should include a more comprehensive consideration of the economic and societal consequences (including victim costs) of heroin use⁹.

This study has several limitations. First, our model is limited to a 1-year time horizon; therefore, the beneficial impact of OST on longer-term outcomes (e.g. mortality) have not been included, and those that have been included (e.g. HIV/HCV infection rates, healthcare, and criminal justice resource use) are likely to demonstrate greater impacts over a lifetime. This would lead to even lower ICERs compared to no OST. To fully capture all the costs and benefits associated with OST (e.g. impact on risk behaviors, HIV transmission, overdoses, and mortality) requires a lifetime horizon. A lifetime model would require data on longer-term retention rates that could be extrapolated with a high degree of confidence and a better understanding of re-admission to treatment post-dropping out. It would also require the use of a different model structure (e.g. Markov Model) that supports analyses over a lifetime.

Second, the impact of OST on income loss (as a result of premature death) or productivity losses (employment status, absenteeism, presenteeism) have not been included in the model and would be important to include in a longer-term model. Third, adverse events were not included. Fourth, as our model (as with any model) is a simplification of the real world we accept that there is likely to be a degree of regional variability around some of the estimates (for example frequency, type and cost of dispensing, and supervision). Such variability was assessed in the sensitivity analysis, and could be further explored in future regional level models.

Finally, there is a requirement to generate health-state-specific utility values for HIV and HCV. This could be achieved either as part of a clinical study or using a panel such as the one used for the utility values in the original model^8,10.

Conclusions

Using the cost-effectiveness model we replicated and expanded, we have demonstrated that OST is cost-effective vs no OST, and offers substantial societal benefits in terms of reduced victim costs. The availability of multiple, cost-effective treatment options is important to help meet requirements and preferences for patients when selecting an intervention for OST to improve outcomes in line with expert guidance^3,18,21. Further work is required to understand the long-term cost-effectiveness of OST, including the generation of new data for resource use, criminality, and mortality. The availability of this model code in an open source format (R version 3.2.3) should make future collaborative research more straightforward, accurate, and comprehensive.

Transparency

Declaration of funding

Funding for this research was provided by Mundipharma International Limited, Cambridge, UK.

Declaration of financial/other interests

JK and WD are employees of Mundipharma International Limited, Cambridge, UK. AW, YY, JB, and AMM were employed by PHMR, which was contracted by Mundipharma International Limited to undertake this research. Peer reviewers on this manuscript have received an honorarium from JME for their review work, but have no other relevant financial relationships to disclose.

Abbreviations
A&E	=	accident and emergency
BMT	=	buprenorphine maintenance treatment
BNF	=	British National Formulary
CEAC	=	cost-effectiveness acceptability curve
CPI	=	consumer price index
GP	=	general practitioner
HCHS	=	Hospital and Community Health Services
HCV	=	hepatitis C virus
HIV	=	human immunodeficiency virus
HTA	=	health technology assessment
ICER	=	incremental cost-effectiveness ratio
MMT	=	methadone maintenance treatment
NHS	=	National Health Service
NICE	=	National Institute for Health and Care Excellence
NTORS	=	National Treatment Outcome Research Study
ONS	=	Office for National Statistics
OST	=	opioid substitution therapy
PSNC	=	Pharmaceutical Services Negotiating Committee
PSS	=	personal social services
QALY	=	quality-adjusted life year
UK	=	United Kingdom

Acknowledgments

The authors thank David Floyd and Ciara Wright for editorial assistance on behalf of PHMR Ltd. Preparation of the manuscript was funded by Mundipharma International Limited.