economic evaluation in opioid modeling: systematic review · 2020. 10. 27. · conclusion: the...

- Contents lists available at sciencedirect.comJournal homepage: www.elsevier.com/locate/jval

Economic Evaluation in Opioid Modeling: Systematic Review

Elizabeth Beaulieu, MA, Catherine DiGennaro, BA, Erin Stringfellow, PhD, MSW, Ava Connolly, Ava Hamilton, BA,Ayaz Hyder, PhD, Magdalena Cerdá, DrPH, MPH, Katherine M. Keyes, PhD, MPH, Mohammad S. Jalali, PhD, MSc*

* AddreBoston,1098-30

A B S T R A C T

Objectives: The rapid increase in opioid overdose and opioid use disorder (OUD) over the past 20 years is a complex problemassociated with significant economic costs for healthcare systems and society. Simulation models have been developed tocapture and identify ways to manage this complexity and to evaluate the potential costs of different strategies to reduceoverdoses and OUD. A review of simulation-based economic evaluations is warranted to fully characterize this set of literature.

Methods: A systematic review of simulation-based economic evaluation (SBEE) studies in opioid research was initiated bysearches in PubMed, EMBASE, and EbscoHOST. Extraction of a predefined set of items and a quality assessment wereperformed for each study.

Results: The screening process resulted in 23 SBEE studies ranging by year of publication from 1999 to 2019. Methodologicalquality of the cost analyses was moderately high. The most frequently evaluated strategies were methadone and bupre-norphine maintenance treatments; the only harm reduction strategy explored was naloxone distribution. These strategieswere consistently found to be cost-effective, especially naloxone distribution and methadone maintenance. Preventionstrategies were limited to abuse-deterrent opioid formulations. Less than half (39%) of analyses adopted a societalperspective in their estimation of costs and effects from an opioid-related intervention. Prevention strategies and studies’accounting for patient and physician preference, changing costs, or result stratification were largely ignored in these SBEEs.

Conclusion: The review shows consistently favorable cost analysis findings for naloxone distribution strategies and opioidagonist treatments and identifies major gaps for future research.

Keywords: economic evaluation, opioid overdose, opioid use disorder, simulation models, systematic review.

VALUE HEALTH. 2020; -(-):-–-

Introduction

Overdose deaths involving opioids have increased fourfold inthe United States between 1999 and 2017, and opioid-relatedoverdose deaths and hospitalizations have risen sharply in theUnited States and worldwide.1

Death due to opioids is a leading cause of unintentional deathfor Americans and has contributed to a decrease in US life ex-pectancy.2 In 2018, a substantial number of deaths in the UnitedStates (estimated 128 people per day)3 were attributed to anoverdose involving opioids.1,2,4,5 The opioid crisis is characterizedby multilayered dimensionality, with many moving inter-connected parts ranging from the individual to societal level.6

Opioid-related illness has significant economic costs for health-care systems and society. Costs to the US economy of the opioidepidemic—including healthcare, mortality, criminal justice ac-tivities, family assistance, and productivity loss—were estimatedat $631 billon total over the 4-year interval 2015 to 2018.7 Severalmedications are effective in improving opioid-related healthoutcomes, including naloxone to reverse an opioid overdose8 and

ss correspondence to: Mohammad S. Jalali, PhD, MSc, MGH Institute for TecMA 02114. Email: [email protected]/$36.00 - see front matter Copyright ª 2020, ISPOR–The Professional So

methadone, buprenorphine and buprenorphine-naloxone, andinjectable naltrexone to treat opioid use disorder.9

Resource allocation in public health decision making is oftenguided by guesswork, leading to misappropriated resources. Tobetter guide allocations, it is critical to assess the cost-effectiveness and efficiency of alternatives when consideringstrategic solutions. Evaluations include assessment of in-terventions directly targeted at individual patients and decisionsabout broader implementation strategies within a greater sys-tem.10 Economic evaluations provide an analysis given currentobservable conditions. Yet conditions constantly change in a dy-namic world, so an estimate of future costs given a set of as-sumptions about future conditions could prove even more useful.Simulation-based economic evaluations (SBEEs) (eg, Markov,agent-based, system dynamics, and compartmental models)represent a set of such tools that can assess allocative efficiency inopioid policies while accounting for weaknesses in standardeconomic evaluations.

Specifically, SBEEs extrapolate beyond short timeframes andcan offer projections of future cost-effectiveness and utility.11 In

hnology Assessment, Harvard Medical School, 101 Merrimac St., Suite 1010,

ciety for Health Economics and Outcomes Research. Published by Elsevier Inc.

www.sciencedirect.comwww.elsevier.com/locate/jvalmailto:[email protected]

2 VALUE IN HEALTH - 2020

addition to the use of existing real-world data, their ability toaccommodate gaps in data allows for the simulation of pop-ulations for more accurate estimates of population-level effects.12

SBEEs are especially useful to model phenomena with rapidlychangeable landscapes, such as the prescription and illicit opioidsupply channels characteristic of the opioid crisis. Simulationmodels are increasingly used in public health to account for thesystemic complexity of the opioid crisis,13–17 but SBEEs representonly a fraction of a wider pool of economic evaluations, and theirusefulness in answering public health policy questions deservesgreater attention. SBEEs have the potential to aid decision makersin implementing strategies to ameliorate the impacts of the opioidcrisis. Strategies that might be modeled using SBEEs include pol-icies or interventions that alter the flow of people from thehealthy population into disordered states—for example, usingnonopioids to treat pain or abuse-deterrent formulations of opioidpain medications; treatment programs that move people withsubstance use disorders into remission; or strategies that reduceharm to people once they have developed substance use disorderssuch as reducing likelihood of fatal overdose. See Sharareh et al17

for information about the benefits and weaknesses of differentmodeling methods in opioids research and Jalali et al18 for thetype of policy questions they can answer.

The current systematic review provides insight into SBEEresearch pertaining to interventions to reduce opioid use andrelated harm. The primary goal is to identify what has beenstudied in existing SBEE research and to review and evaluatecharacteristics of the published studies. A secondary objective is tosynthesize the results when possible. This review will highlightweaknesses in the existing body of SBEEs and gaps in the scopeand quality of these economic evaluations. Reviews of economicevaluations have been published in the area of opioid-related in-terventions before many relevant simulation models addressingthe opioid crisis were developed19 or which have a narrow scopefocused on economic evaluations of a specific subtype of opioidmisuse or use disorder treatment strategy.17,20,21, For example, a2016 review21 assessed a broad scope of any type of economicevaluation for opioid use disorder interventions, yet cut off theinclusion time window to post-2007 studies, which excludes somehigh-quality intervention efforts. A 2017 review20 examined thelimited scope of opioid agonist treatments among people withnonprescription opioid dependence; this criterion excludes ele-ments included in the current review such as opioid antagonist(ie, naloxone and extended-release naltrexone) treatments andpeople with strictly prescription opioid dependencies who havenot escalated to heroin use. A 2019 scoping review17 includedsimulation and conceptual models for policies aimed to resolvethe opioid epidemic but did not focus on economic evaluation. Nopublication exists to date that reviews modeling-based economicevaluations for opioid crisis interventions. The current reviewencompasses SBEEs that provide a cost-based analysis for anintervention aimed at addressing the opioid crisis, and the scope isnot limited to any subtype of intervention strategy.

The rest of this article is organized as follows: we describe theprocess of the review in the Methods section; present thedescriptive and technical characteristics in the Results section;and summarize and synthesize the results when possible in theDiscussion section.

Method

Search Strategy and Selection Criteria

A multiphase search strategy was applied to collect economicevaluations of interventions to reduce opioid use and related

harms based on simulation models. PubMed was first searched fornon-animal research journal articles published in English from theinception of PubMed (1966) to September 2019. The followingMeSH terms were used in the PubMed search: (analgesics, opioid,heroin, naloxone, methadone, opioid-related disorders, prescrip-tion drug misuse, OR fentanyl) AND (computer simulation/[edu-cation; economics; methods; statistics and numerical data; supplyand distribution], Markov chains, OR systems analysis). The fiveMeSH term categories of “computer simulation/[ ]” ensured theinclusion of a wide range of simulation modeling approaches(agent-based, system dynamics, compartmental, and micro-simulation, among others); however, decision tree and regressionanalyses were excluded. To complement the search and to ensurethat all relevant journal articles were included in the initial sam-ple, results from EBSCOhost and Embase were also included.

Cost Screening

Two independent reviewers conducted a general cost-relatedscreen to filter out articles not conducting cost analyses. The re-viewers excluded articles that did not analyze an interventionstrategy with an outcome around opioid use or did not report aquantitative, cost-related result. Analyses of cost or price dy-namics within drug markets were also excluded in this stage—forexample, an article on the heroin epidemic in Baltimore22 wasexcluded because its only cost analysis was related to prices ofdrugs in the market rather than the cost of an intervention toreduce opioid overdose deaths and use disorder.

For each of these screens, following best review practices,23 a10-article pilot test was performed to ascertain mutual under-standing of the inclusion criteria among the reviewers. After eachscreen, the reviewers compared findings, recorded the percentagreement, and discussed discrepancies to arrive at a mutualagreement. In cases where the inclusion decision could not bemade based on title and abstract, the article was escalated to full-text review. The full-text reviews were then conducted to ensurethat the actual content of the studies aligned with the inclusioncriteria. For any step of the screening, extraction, and abstractionprocesses in which a consensus between the 2 reviewers was notreached, a third reviewer was available to resolve discrepancies.

Data Extraction and Abstraction

For each study that met all inclusion criteria, 2 researchersindependently read the full text and supplemental materialswhere applicable, and they abstracted and entered informationinto predefined 17-item extraction sheets adapted from system-atic review data extraction tools.24 The information items wereselected to characterize the core characteristics of the analyses byobserving trends of analyzed interventions over time; economicevaluation techniques and specifications; simulation modelingapproaches underlying the economic evaluations; the core finding(cost-effectiveness or cost utility); and differences across strate-gies that drove the cost-related results. Each extraction item isdescribed briefly in the Supplementary Materials found at https://doi.org/10.1016/j.jval.2020.07.013.

Quality Assessment

Each of the economic evaluations was appraised using theDrummond 10-point checklist.25 The Drummond checklist, whichhas also been used by other systematic reviews of economicevaluations,26-28 informs a quality appraisal of economic evalua-tions with consideration of a variety of methodological attributesand was selected for its ease of communication to a wider audi-ence (Fig 1). This checklist has the advantage that it allows iden-tification of the minimum acceptable standard for reporting of the

https://doi.org/10.1016/j.jval.2020.07.013https://doi.org/10.1016/j.jval.2020.07.013

Figure 1. Quality assessment of economic evaluations.

PubMed, EBSCOhost, Embase Full-text journal articles

891 papers

Duplicate papers (244) removed20 papers added manually

667 papers for initial review of title/abstract

Title and abstract cost screen76 papers

Second screen and full-text review39 papers

Exclusion:

Exclusion:

- Non-simulation study design- No opioid-related outcomes- Additional duplicates591 papers excluded

- No cost analysis- Only dynamics are within drug market37 papers excluded

-- 3

general methods and results, and it allows direct comparisonacross economic evaluations and their results.28 The tool enablesreporting of each economic evaluation’s fulfillment of 10 criteriaitems ranging from defining the research question, establishmentof appropriate alternatives, effectiveness and costs, analyticalapproach, accounting for uncertainty, and presentation of results.One researcher identified whether each article met each criterion.In cases of ambiguity, the designation was escalated to a secondresearcher for further evaluation. It was permissible for a givenitem within the checklist to remain marked as partially met afterthe second review; this is reported in the results.

Results

Study Selection

Figure 2 represents the search strategy and results. The initialsearch produced 76 articles for simulation models of the opioidcrisis, and they were uploaded into an online screening tool

Figure 2. PRISMA flow diagram.

(Abstrackr29) to be screened for whether the title or abstractcontained any mention of cost analysis or economic evaluation.There was 79% agreement in the designations given by the 2 re-viewers. Following discussion between the reviewers, theconsensus reached brought 39 studies into passage to inclusionfrom this screening.

The second screen assessed whether each of the 39 studiesreported a cost-related result quantitatively and included anintervention strategy that was examined for an effect or outcomearound opioid use. The 2 reviewers had an initial 85% agreementbefore reaching full consensus. Studies were excluded if the articledid not report a cost-related finding quantitatively, did not modelits effectiveness outcome in a manner directly related to opioids,or focused only on the transmission of human immunodeficiencyvirus (HIV) or hepatitis C virus (HCV) as the mechanism of effec-tiveness for the intervention. These exclusion criteria wereadopted to avoid evaluation of articles targeting evaluation ofstrategies that were not primarily surrounding opioids-relatedinterventions. For example, we included a study30 that modeled

Table 1. Extracted items from 23 studies.

Study Study characteristics Cost andeffectivenessestimation

Results

Descriptive characteristics Technicalcharacteristics

Asche 201536 � Perspective: Medicaid� Intervention and compar-

ator: sublingual film formula-tion vs tablet formulation ofbup/nx combination

� Simulated population:Medicaid patients with opioiddependence

� Treatment setting:office-based

� Country/Currency (adj. year):US; USD (2010-12)

� Cost source: MarketScanMedicaid claims database

� Industry funding: yes (ReckittBenckiser Pharmaceuticals)

� Modelingapproach: Markovmodel

� Economic evalua-tion (EE) method:budget impactanalysis

� Sensitivity anal-ysis: deterministic

� Time horizon: 5years

� Discounting: 3%cost

� Cost: emergencycare, medication,inpatient care

� Effectiveness:impact on budget;transition probabil-ities between initia-tion, maintenance,discontinuation

� Results: sublingual bup/nx has 100%market share; cost = $6.4B sublingualfilm is progressively replaced bygeneric tablet; cost = $6.464B

� Most sensitive parameter: The ratioof probabilities of nonpsychiatric hos-pitalization in the “off-treatment” statebetween film and tablet, and the pricerebate for tablet

� Authors’ conclusion: Using the sub-lingual film formulation for more pa-tients treated with bup/nx is predictedto increase outpatient care costs, butit would also generate savings inemergency care and hospitalizations.Total direct medical costs for Medicaidwould be lower for sublingual-film-treated patients at current drug prices

� % EE assessment items fully satis-fied: 80%

Barnett 199937 � Perspective: healthcare payer� Intervention and compar-

ator: MMT vs “drug-free”treatment

� Simulated population: 25-year-old people who use heroin

� Treatment setting: clinic-based

� Country/Currency (adj. year):US; USD (1996)

� Cost source: literature� Industry funding: no


� EE method: cost-effectivenessanalysis


� Time horizon:lifetime

� Discounting: 3%cost and effects

� Cost: emergencycare, medication,inpatient care

� Effectiveness: Life-years gained;mortality rates

� Results: ICER = $5915 per LY gained� Most sensitive parameter: do not

identify specifically; ICER found to be, $10k/life-year over wide range ofmodeling assumptions

� Author’s conclusion: giving opiateaddicts access to methadone mainte-nance has an ICER of $5915 per life-year gained

� % EE assessment items fullysatisfied: 70%

Barnett 200138 � Perspective: healthcare payer� Intervention and compar-

ator: BMT vs MMT� Simulated population: adults

in opiate dependence mainte-nance treatment

� Treatment setting: clinic, of-fice, and hospital


� Cost source: literature, Red-Book, French national reporting

� Industry funding: no

� Modelingapproach:Compartmentalmodel†

� EE method: cost-utility analysis




� Cost: medication;urinalysis, physicianevaluation, psycho-social interventions;costs associatedwith HIV/AIDS

� Effectiveness:QALYs; number ofinjection drug usersin maintenance

� Results: ICER = $14 000-$84 700 per QALY gained (low HIVprevalence);$10 800-$66 700 per QALY gained(high HIV prevalence). ICERS givenacross range of buprenorphine cost/dose, expansion strategy, and metha-done status.

� Most sensitive parameter: do notidentify specifically; Tables 3 and 4display how ICERs change in responseto differentquality of life adjustment assumptions

� Authors’ conclusion: Cost-effectiveness of buprenorphinemaintenance depends on its price perdose. At $5 or less per dose,buprenorphine maintenance is cost-effective; at $15/dose it is cost-effective if its adoption does not leadto a net decline in methadone use or ifa medium to high value is used toyears of life for PWID and those inmaintenance therapy; at $30/dose, itis cost-effective only under the mostoptimistic assumptions.


continued on next page


Table 1. Continued


Results


Carter 201739 � Perspective: societal� Intervention and compar-

ator: subdermal implantable vssublingual buprenorphine

� Simulated population: peopleseeking treatment for OUD

� Treatment setting: officebased


� Cost source: literature, insur-ance claims reports

� Industry funding: yes (Brae-burn Pharmaceuticals)



� Sensitivity anal-ysis: univariate andprobabilistic

� Time horizon: 12months

� Discounting: no

� Cost: direct medicalcosts and non-medical costs;clinical and societalpenalties of relapseand illicit opioid use(criminal justice,work lost)

� Effectiveness:QALYs; number ofabstinent patients,retained patients

� Results: subdermal implantablebuprenorphine costs $4386 less andgives 0.031 more QALYs than sublin-gual buprenorphine

� Most sensitive parameter: relativemonthly probability of relapse whileon treatment

� Authors’ conclusion: subdermalinjectable buprenorphine preferredover sublingual buprenorphine fromhealth-economic perspective fortreatment


Chalmers 201240 � Perspective: healthcarepayer(s) and patient

� Intervention and compar-ator: subsidized vs unsubsi-dized MMT

� Simulated population:patients enrolled in MMTtherapy

� Treatment setting: clinicbased

� Country/Currency (adj. year):Australia; AUS $ (2006/2007)

� Cost source: NEPOD database,survey data


� Modelingapproach: Systemdynamics model

� EE method: othercost analysis (esti-mating cost burdenif subsidy policyadopted)


� Time horizon:cyclical model withmonthly andannual estimates

� Discounting: no

� Cost: medication,prescribing anddispensing; andproportionatebearers of the cost

� Effectiveness: Costestimation ofmethadone main-tenance treatment;behavioral effectsof entry and reten-tion to treatment

� Results: Annual cost to subsidizeMMT = $94M to $175.8M

� Most sensitive parameter: do notidentify specifically; cost estimationdepends on behavior effects assump-tions of how patients stay longer intreatment and treatment-naivepatients enter treatment sooner

� Authors’ conclusion: if Australiangovernment(s) were to providedispensing fee subsidies for metha-done maintenance patients, it wouldbe costly, but these additional costsare offset by the social and healthgains achieved from the programs


Cipriano 201851 � Perspective: societal� Intervention and compar-

ator: distributing naloxone kitsin secondary schools vs nonaloxone distribution

� Simulated population: sec-ondary school students at riskof overdose

� Treatment setting: secondaryschools

� Country/Currency (adj. year):Canada; Can $ (2017)

� Cost source: literature, Tor-onto School Board data, Cana-dian Red Cross data, CanadianInstitute for Health Information


� Modelingapproach:Decision-analyticmodel


� Sensitivity anal-ysis: deterministicand probabilistic


� Discounting: 1.5%cost and effects

� Cost: naloxone kits,program setup,training, adminis-tration, mainte-nance; emergencymedical care, long-term healthcarecosts

� Effectiveness:QALYs; mortality inevent of overdose

� Results: Scenarios in which the pro-gram would be cost-effective

� Most sensitive parameter: mostsensitive to cost and intensity of stafftraining, lifetime costs of person whooverdoses, and intensity of naloxonetraining program

� Authors’ conclusion: makingnaloxone available in schools is cost-effective if there are at least 2overdoses/year in the school


Coffin 2013 (1)52 � Perspective: societal� Intervention and compar-

ator: naloxone distribution forlay administration vs nonaloxone distribution

� Simulated population: peoplewho use heroin averaging 28-38years of age

� Treatment setting: commu-nity based

� Country/Currency (adj. year):Russia; USD, converted fromRussian currency (2010)

� Cost source: Russian Federa-tion data, program reports,media







� Cost: naloxone kits;emergency andhospital care;transport costs;heroin user cost tosociety


� Results: ICER = $71 USD per QALYgained

� Most sensitive parameter: efficacyof lay-administered naloxone atpreventing overdose death and thecost of naloxone

� Authors’ conclusion: naloxone dis-tribution to heroin users for lay over-dose reversal is highly likely to reduceoverdose deaths in target commu-nities and is robustly cost-effective



-- 5

Table 1. Continued


Results


Coffin 2013 (2)53 � Perspective: societal� Intervention and compar-

ator: naloxone distribution forlay administration vs nonaloxone distribution

� Simulated population: 21-year-old new US people whouse heroin (base model) to ages31 and 41



� Cost source: literature, CDCdata







� Cost: naloxone kits;EMS visits andtransport to hospi-tal; emergencydepartment care;heroin user cost tosociety

� Effectiveness:QALYs and life-years gained;absolute andrelative overdosedeath rates

� Results: ICER = $438 per QALY gained� Most sensitive parameter: cost

effectiveness result was robust torange of inputs; most sensitive to ef-ficacy of lay-administered naloxoneand cost of naloxone.

� Authors’ conclusion: the interventionto distribute naloxone to heroin userswould likely reduce overdose deaths,increase QALYs, and be highly cost-effective


Jackson 201541 � Perspective: state-leveladdiction treatment payers

� Intervention and compar-ator: injectable XR-NTX vs MMTvs BMT

� Simulated population: adult18-65-year-old males agesenrolled in treatment for opioiddependence in the UnitedStates

� Treatment setting: officebased and clinic based

� Country/Currency (adj. year):US; USD (2015*)

� Cost source: literature, CMSdata





� Time horizon: 168days (approx. 0.5year)

� Discounting: no

� Cost: medication,counseling, medi-cation manage-ment and oversight

� Effectiveness:opioid-free days;transitionprobabilitiesbetweenabstinence, opioiduse, retention intreatment

� Results: ICER for XR-NTX relative toMMT = $72.42 per opioid-free daygained; BMT is dominated

� Most sensitive parameter: cost-effectiveness of XR-NTX compared toMMT is sensitive to effectivenessinputs and may be altered byuncertainty in relative costs

� Authors’ conclusion: XR-NTX is acost-effective medication for treatingopioid dependence if state addictiontreatment payers are willing to pay atleast $72 per opioid-free day


King 201642 � Perspective: third-party payersin the US

� Intervention and compar-ator: office-based BMT vs clinic-based MMT

� Simulated population: 1,000adult, opioid-dependentpatients with no history oftreatment within 30 days

� Treatment setting: officebased and clinic based


� Cost source: expert input,Redbook, CMS data





� Time horizon: 1year

� Discounting: no

� Cost: medication,counseling, medi-cation manage-ment and oversight

� Effectiveness:Number of drug-free weeks andadditional patientin treatment;probabilities ofretention

� Results: ICER for MMT vs BMT is $10437 per additional patient in treat-ment gained and $8515 per additionalopioid abuse-free week gained

� Most sensitive parameter: weeklycost of MMT had the largest impact oncost-effectiveness when retention intreatment was the outcome

� Authors’ conclusion: MMT is a cost-effective alternative to BMT for newlyinitiated opioid-dependent adults foropioid maintenance treatment in theUnited States from the perspective ofa third-party payer


Krebs 201843 � Perspective: societal� Intervention and compar-

ator: OAT for all treatment re-cipients vs observed standardof care (54% treatment initia-tion with medically managedwithdrawal)

� Simulated population: patientspresenting with OUD; basemodel patient is 35 years old

� Treatment setting: publiclyfunded opioid use disordertreatment facilities


� Cost source: literature, legisla-tive reporting


� Modelingapproach: Semi-Markov cohortmodel


� Sensitivity anal-ysis: probabilistic



� Cost: healthcareresource use andcriminal activity

� Effectiveness:QALYs; survivalprobability, HIVincidence, rate ofincarceration

� Results: Initial OAT strategy giveshigher QALYs (12.93 vs 12.52) at alower cost (946,804 vs 1,025,061)

� Most sensitive parameter: OAThealthcare costs while not directly intreatment

� Authors’ conclusion: OAT deliveredto all patients presenting for treat-ment provides greater health benefitsand cost savings than the observedstandard of care. This strategy maxi-mizes the value of publicly fundedtreatment of opioid use disorder inCalifornia




Table 1. Continued


Results


Langham 201854 � Perspective: public health sys-tem; societal perspective insensitivity analysis

� Intervention and compar-ator: distribution of naloxonefor use by nonmedical re-sponders vs no naloxonedistribution

� Simulated population: Adultsat risk of heroin overdose


� Country/Currency (adj. year):United Kingdom; Pound sterling£ (2016)

� Cost source: Royal Pharma-ceutical Society, National HealthService

� Industry funding: yes (Mundi-pharma International Ltd.)





� Discounting: 3.5%cost and effects

� Cost: medication,training/education,kit distribution,emergency care


� Results: ICER = 899 pounds per QALYgained

� Most sensitive parameter: rate offirst overdose; proportion of wit-nessed overdoses; efficacy ofnaloxone; proportion of witnessedoverdoses when naloxone is available;social network modifier

� Authors’ conclusion: Distribution oftake-home naloxone decreased ODdeaths by 6.6% and was cost-effective.ICER is 899 pounds per QALY gained,well below the 20 000 set by UKdecision makers


Masson 200444 � Perspective: long-rangeperspective of the healthcaresystem

� Intervention and compar-ator: MMT and 1 hour/weekpsychosocial therapy duringfirst 6 months vs 180-daydetoxification, 3 hours/week ofpsychosocial therapy, and 14education sessions during thefirst 6 months

� Simulated population: 196adults with diagnosed opioiddependence

� Treatment setting: researchclinic in an established drugtreatment program

� Country/Currency (adj. year):US; USD (2004*)

� Cost source: public health sys-tem administrative database





� Time horizon: 16months with pro-jection to 10-20years


� Cost: health serviceutilization and psy-chosocial therapy/education

� Effectiveness:QALYs; years inopioid use healthstate

� Results: ICER = $16 967 per LY gained;ICER , $20 000 per QALY gained

� Most sensitive parameter: do notidentify specifically; table shows how$/LY and $/QALY change in responseto a set of 1-way inputs

� Authors’ conclusion: methadonemaintenance is a cost-effectivetreatment relative to 180-day-longmethadone detoxification


Morozova 201945 � Perspective: payer (Ministry ofHealth and municipalauthorities)

� Intervention and compar-ator: scale-up strategies forOAT for OUD vs currentcapacity

� Simulated population: peopleat risk of and with OUD


� Country/Currency (adj. year):Ukrainian cities; USD, convertedfrom Ukrainian currency (2016)

� Cost source: private and na-tional reports


� Modelingapproach:Compartmentalmodel





� Cost: costs associ-ated with increasedopioid agonist ther-apy capacities

� Effectiveness:QALYs; access, initi-ation, retention intreatment

� Results: increased OAT capacity (12.2-, 2.4-, and 13.4-fold) would be cost-effective at WTP for QALYs gained ofone GDP/capita. ICER given acrossstrategies for each city

� Most sensitive parameter: do notidentify specifically

� Authors’ conclusion: A substantialincrease in opioid agonist treatment(OAT) capacity in 3 Ukrainian citieswould be cost-effective for a widerange of willingness-to-pay thresholds



-- 7

Table 1. Continued


Results


Nosyk 201246 � Perspective: societal; Ministryof Health and third-party payerin sensitivity analysis

� Intervention and compar-ator: diacetylmorphine (heroin)vs methadone maintenance

� Simulated population: peoplewith chronic opioid depen-dence refractory to treatment


� Country/Currency (adj. year):Canada; Can$ (2009)

� Cost source: North AmericanOpiate Medical Initiative, courtrecords





� Time horizon: 1-,5-, 10-year andlifetime

� Discounting: 5%effects

� Cost: medication,human resources,overhead, drugtreatments for HIVand HCV infection,nonmedical costs:criminal activity(charges, justicesystem,victimization)

� Effectiveness:QALYs; transitionprobabilities be-tween treatment,abstinence, relapse

� Results: ICER for diacetylmorphineversus MMT in Can$ per QALYs gainedis cost saving and more effectiveacross all time horizons

� Most sensitive parameter: the onlyscenarios where diacetylmorphine isnot strictly cost saving and they reportICER numerically is the Ministry ofHealth perspective and when theyapply mortality estimates from Gron-bladh et al

� Authors’ conclusion: A treatmentstrategy featuring diacetylmorphinemay be more effective and less costlythan methadone maintenance treat-ment among people with chronicopioid dependence refractory totreatment. Societal costs woulddecrease via reduction in crime, andboth duration and quality of life of thetreatment recipients would increase.


Ritter 201647 � Perspective: societal� Intervention and compar-

ator: comparing residentialrehabilitation, opioid substitu-tion treatment, counseling only,in-prison treatment strategiesto one another

� Simulated population: peoplewho use heroin between agesof 18 and 60 each with variousgenders, HIV and HCV statuses,and treatment histories

� Treatment setting: various(clinic based, residential treat-ment, prison)

� Country/Currency (adj. year):Australia; AUS$ (2012)

� Cost source: literature, DOHdata, Medical Benefits data,NSW committee data,experimental


� Modelingapproach: Micro-simulation model

� EE method: othercost analysis(microsimulationmodel building andvalidation)

� Sensitivity anal-ysis: no


� Discounting: 3%effects

� Cost: treatmentprovision, health-care services, crim-inal activity, life-years lost, familybenefit oftreatment, HIV/HCVtreatment

� Effectiveness: life-years saved;transitionprobabilitiesbetweenabstinence,irregular andregular use,withdrawal,treatment statesand mortality rates

� Results: estimated costs over heroinuse careers; reported itemized costsby heroin use state, crime event, etc.

� Most sensitive parameter: n/a� Authors’ conclusion: authors were

able to build a stable, tractable modeland verified all parameters. Validationagainst external data sources revealedhigh validity. While there are limita-tions associated with any model, theheroin career model now has the po-tential to be used for simulations ofalternate policy scenarios.


Schackman 201248 � Perspective: healthcare payerand patient

� Intervention and compar-ator: long-term office-basedbuprenorphine-naloxone (bup/nx) vs no treatment for clinicallystable opioid-dependentpatients

� Simulated population: clini-cally stable patients with opioiddependence who alreadycompleted 6 months of office-based bup/nx treatment

� Treatment setting: officebased


� Cost source: literature, USDOLdata, CMS data, state data (CT)







� Cost: medication,treatment delivery

� Effectiveness:QALYs; transitionprobabilities be-tween in/off treat-ment and on/offdrugs states

� Results: ICER = $35,100 per QALYgained (bup/nx, compared to notreatment)

� Most sensitive parameter: quality-of-life weight assumptions

� Authors’ conclusion: office-basedbup/nx for this set of patients can be acost-effective alternative to notreatment at an accepted threshold of$100 000/QALY




Table 1. Continued


Results


Townsend 201955 � Perspective: societal andhealth sector

� Intervention and compar-ator: comparing 8 naloxonedistribution strategies to oneanother

� Simulated population: peopleat risk of opioid-relatedoverdose death



� Cost source: literature, phar-maceutical data, SAHMSA data,DOJ data, USDOL data, USGAOdata





� Time horizon: life-time with 5-yearsocial cost


� Cost: naloxone kits;training; time costsof naloxonetraining; ambu-lance and emer-gency departmentvisits; productivitycosts of OUD andoverdose; costs tocriminal justice sys-temEffectiveness:QALYs; mortalityrates and QOL dueto less hypoxia andreduction in misuse

� Results: high layperson / high policeand fire / high EMS distribution strat-egy is dominant from the societalperspective with an ICER of $15 950per QALY gained

� Most sensitive parameter: probabil-ity that police and fire arrive beforeEMS

� Authors’ conclusion: evidence sup-ports increased naloxone distributionto laypeople, police/fire, and EMS;under resource constraints laypeopleand EMS should be the priority


Uyei 201756 � Perspective: healthcare sector� Intervention and compar-

ator: comparing no treatment;naloxone distribution alone;with linkage to addiction treat-ment; with PrEP; and with link-age to addiction treatment andPrEP strategies to one another

� Simulated population: peopleat risk of opioid-relatedoverdose death via injection.Starting cohort is 22 years ofage and HIV-negative

� Treatment setting: commu-nity and hospital based


� Cost source: literature, stateCMS data (CT)


� Modelingapproach: Decisionanalytical Markovmodel



� Time horizon: 5,10, and 20 years

� Discounting: 3%cost

� Cost: cost ofnaloxone distribu-tion, addictiontreatment, emer-gency care, HIVcare

� Effectiveness:QALYS; overdosesaverted, survival,life expectancy,HIV-related deathsaverted

� Results: naloxone distribution plusPrEP and linkage to addiction treat-ment has an ICER at 20 years of $95337 per QALY gained relative tonaloxone distribution plus linkage toaddiction treatment. The other 3strategies were dominated

� Most sensitive parameter: ICERs forthe fifth (all-inclusive) strategy weremost sensitive to variation

� Authors’ conclusion: naloxone dis-tribution through syringe service pro-grams is cost-effective compared withsyringe distribution alone, but whencombined with linkage to addictiontreatment is cost saving comparedwith no additional services. A strategythat combines naloxone distribution,PrEP, and linkage to addictiontreatment results in greater healthbenefits in people who inject drugsand is also cost-effective at the $100000 per QALY gained willingness topay threshold.


Yenikomshian201735

� Perspective: healthcare payerand physician

� Intervention and compar-ator: ER ADOs vs ER non-ADOs

� Simulated population: 10 000adult chronic pain patients

� Treatment setting: physiciandecision in primary care


� Cost source: literature, CMSdata

� Industry funding: yes (PfizerInc.)

� Modelingapproach: Markovprocess model

� EE method: othercost analysis (im-pacts of prescribingdecision non-incrementally)


� Time horizon: 12months

� Discounting: no

� Cost: prescribing,medication, costassociated withabuse or misuseevents

� Effectiveness:misuse and/orabuse-relatedevents and NNH;reduced misuse orabuse events

� Results: patients prescribed ER ADOshad 87 to 417 fewer misuse- and/orabuse-related events than patientsprescribed ER non-ADOs with asavings of $8 to $35 per patient. NNHranged from 185 to 40.

� Most sensitive parameter: resultsrange by population (commercial,Medicaid, Medicare, VA) and weresensitive to decreases in the proba-bility of misuse and/or abuse-relatedevents but showed reductions in mostscenarios

� Authors’ conclusion: a physician’sdecision to prescribe ER ADOs ratherthan ER non-ADOs could lead to largereductions in misuse and/or abuse-related events and associated costsacross many patient populations



-- 9

Table 1. Continued


Results


Zaric 2000 (1)30 � Perspective: healthcare payer� Intervention and compar-

ator: expanded MMT vs currentlevel of MMT

� Simulated population: IDUsand non-IDUs with varying HIVstatus, 18 to 44 years of age



� Cost source: literature, censusdata, national report







� Cost: health carecosts with andwithout MMTexpansion for bothHIV- and non-HIV-related medicalneeds

� Effectiveness:QALYs; treatmentavailability, reducedrisky behavior andspread of HIV,mortality

� Results: ICER = $8200 per QALYgained (high HIV prevalence) and $10900 per QALY gained (low HIV preva-lence). Under various other assump-tions the ICER ranges from $10 000 to$38 300 per QALY gained (high HIVprevalence) and from $15 200 to $36100 per QALY gained (low HIVprevalence)

� Most sensitive parameter: do notidentify specifically; the expandedMMT capacity remains cost-effectiveeven if it is twice as expensive and halfas effective as current MMT slots

� Authors’ conclusion: expansion ofMMT is cost-effective on the basis ofcommonly accepted criteria formedical interventions


Zaric 2000 (2)49 � Perspective: healthcare payer� Intervention and compar-

ator: expanded MMT vs currentlevel of MMT

� Simulated population: IDUsand non-IDUs with varying HIVstatus, 18 to 44 years of age



� Cost source: literature, censusdata, national report





� Time horizon: 26years with 10-yearforward projection


� Cost: non-HIVhealthcare, HIVcare, methadonemaintenance byHIV and IDU status

� Effectiveness: LYsand QALYs; treat-ment availabilityand HIV infectionsaverted

� Results: ICER of 10% MMT expansionis $10 900/$8400/$6300/$8200 perQALY gained for 5%/10%/20%/40%HIV-prevalence communities,respectively

� Most sensitive parameter: do notidentify specifically; incremental MMTslots are likely to be cost-effectiveeven if they cost twice as much andare half as effective in reducing riskybehavior as current MMT programs

� Authors’ conclusion: expandingexisting MMT programs is a cost-effective healthcare intervention thatcan play an important role in slowingthe spread of HIV and improving thelength and quality of life for IDUs;expansion is cost-effective even inpopulations with low HIV prevalenceamong IDUs


Zarkin 200550 � Perspective: societal� Intervention and compar-

ator: methadone treatment vsno intervention

� Simulated population: heroinusers aged 18-60 in generalpopulation


� Country/Currency (adj. year):U.S.; USD (2001)

� Cost source: literature, na-tional report, FBI data, USDOJdata, USDHHS data, censusdata


� Modelingapproach: MonteCarlo simulation

� EE method: cost-benefit analysis

� Sensitivity anal-ysis: yes;deterministic

� Time horizon:approximating life-time by modelingages 18-60 (42years)

� Discounting: 3%cost and benefits

� Cost: heroin use,treatment, criminalbehavior, employ-ment, and health-care use; crime,incarceration, andloss of employment

� Effectiveness: out-comes converted tomonetary terms;mortality, crime,employment loss

� Results: lifetime benefit-to-costratio = 37.73 (dynamic model) and4.86 (static model)

� Most sensitive parameter: lifetimecrime costs and mean per-individualeconomic benefit depends highly onprobability of committing a crime.Life-years, number of years usingheroin, and mean lifetime crime costsdepend highly on the mortality rate ofheroin users

� Authors’ conclusion: increasing ac-cess to treatment significantlyincreased treatment benefits andcosts and dominates the alternativestrategy of improving treatment pro-cess by lengthening stay in treatment


Note. Cost-benefit analysis: comparison of interventions and their consequences where both are expressed in monetary terms. Cost-effectiveness analysis: economicanalysis that compares the relative costs and outcomes of 2 or more alternatives. Budget impact analysis: economic assessment that estimates the financialconsequencesof adoptingan intervention.Cost-utility analysis: economicanalysis that compares the relative costs andquality-adjustedoutcomesof2ormorealternatives.


adj., year indicates year to which costs were adjusted; ADO, abuse-deterrent opioid; BMT, buprenorphine maintenance treatment; bup/nx, buprenorphine-naloxone; EE,economic evaluations; ER, extended release; ICER, incremental cost-effectiveness ratio; IDU, injection drug user; LY, life-year; MMT, methadone maintenance treatment;NNH, number needed to harm; NTX, naltrexone; OAT, opioid agonist treatment; OUD, opioid use disorder; QALY, quality-adjusted life-year.*Year of currency adjustment not explicitly stated in article; year of publication shown instead for context in comparing costs across studies.†Model details in prior work.51

-- 11

interventions as explicitly affecting non-HIV costs and utilities butexcluded a study31 that had a primary focus on HIV transmission.Inclusion of studies that focus primarily on HIV would furtherincrease the heterogeneity of study design, research methods, andfindings of the final collection of studies. Following the full-textreview and conflict resolution discussions after each screening,23 studies were finally included. Ten studies were published insubstance abuse journals, 7 in health economic and managementjournals, and 6 in general medicine and public health journals.

Descriptive Characteristics of Studies

Seven major descriptive characteristics were extracted, dis-cussed below, and presented in detail in Table 1.

PerspectiveThe perspective of an economic evaluation depends on the

research question and which costs and effectiveness measures theanalysis considers. We divided these primary perspectives into 3general categories: healthcare sector, payers, and societal.Healthcare sector perspectives account for “formal health caresector (medical) costs borne by third-party payers or paid for out-of-pocket by patients,” including “current and future health costs,related and unrelated to the condition under consideration.”32

Payer perspectives account for the subset of healthcare sectorcosts covered by the payer. The societal perspective accounts formedical-related costs and factors such as an intervention’s effectson reduced criminal activity or worker productivity and wageslost.32,33 The Second Panel on Cost Effectiveness in Health andMedicine34 recommends use of the healthcare sector and societalperspectives as reference case analyses. Less than half (9 articles;39%) of the studies adopted a societal perspective in their primaryanalysis. The others (14 articles; 61%) were limited to a healthcaresector or specific payer perspective for healthcare costs.

Intervention and comparatorInterventions and accompanying comparators of interest

included one prevention strategy (abuse-deterrent opioids),35

treatment strategies (opioid agonist, partial agonist, or antagonistmaintenance therapies),30,36–50 and strategies involving naloxonedistribution (distribution to laypeople; emergencymedical services(EMS), police and fire workers; and secondary schools).51–55 Onestudy considered multiple sets of strategy combinations.56 Thefocusof economic evaluationshas shiftedover timeas the incidenceof overdose has escalated. Analyses of opioid agonist or partialagonist therapies (methadone and buprenorphine maintenance)were the predominant strategy evaluated between 1999 and 2005,and these continue to be relevant through the present day. Evalu-ations of naloxone distribution strategies increased beginning in2013,51–56 coinciding with the rapid rise in overdose deaths andincreasing fentanyl contamination of the heroin supply.

As shown in Figure 3, methadone maintenance treatment(MMT) is covered with the highest frequency, followed bybuprenorphine maintenance treatment (BMT) and naloxone (theonly explicit harm reduction strategy explored). Injectablenaltrexone,41 prescription injectable diacetylmorphine46 (ie, theactive ingredient in heroin) for opioid use disorder, and abuse-deterrent opioid formulations35 were each evaluated only once.

Simulated populationThe simulated populations varied in age, and some models had

initial conditions where nonusers could transition into opioid-related states. Modeled individuals were commonly defined bydrug use status (7 articles; 30%—eg, individuals who actively useheroin), disorder status (8 articles; 35%—eg, chronic pain patients,people with opioid use disorder), treatment status (4 articles;17%); or event risk status (4 articles; 17%—eg, people at risk ofopioid-related overdose death).

Treatment settingTreatment or strategy setting of models underlying the eco-

nomic evaluations were clinic-based (10 articles; 43%), office-based (4 articles; 17%), both clinic- and office-based compara-tively (3 articles; 13%), or community-based (6 articles; 26%).

Country and currencyEighteen studies (78%) were based on North American (US or

Canada) models, so costs for most of the economic evaluations areexpressed in US dollars (USD). The remaining 5 (22%) were basedon models for Australia (2), the UK (1), Ukraine (1), and Russia (1);these studies’ costs are expressed in the given country’s currencyor converted to USD.

Cost sourceEighteen studies (78%) used cost data from national or state

data, often supplemented by literature and private data; theremaining 5 used only literature or industry sources to determinecosts.

Industry fundingFour studies (17%) were funded by the pharmaceutical in-

dustry; the rest were unfunded or funded by NIH Institutes, theDepartment of Veteran Affairs, the Connecticut Department ofPublic Health, or philanthropic organizations.

Technical Characteristics of Studies

Five major technical characteristics are extracted, discussedbelow, and presented in detail in Table 1.

Modeling approachThe sample of studies represents a wide variety of modeling

approaches. Thirteen (57%) were Markov-based models (Markov,semi-Markov, Markov process, or decision-analytical Markovmodel). The next most common modeling approaches were othercompartmental (4 articles; 17%) and decision-analytic models (3articles; 13%). Finally, the studies included 1 (4%) each of a systemdynamics, microsimulation, and Monte Carlo model.

Economic evaluation (EE) methodEE methods were mostly cost-utility (15 articles; 65%) or cost-

effectiveness analyses (3 articles; 13%); there was 1 (4%) budgetimpact analysis and 1 (4%) cost-benefit analysis, and 3 (13%) costanalyses that do not fit a predefined economic evaluation frame-work (“other cost analysis” in Table 1). We labeled economicevaluations that report a cost-effectiveness result incrementallyusing a quality-adjusted outcome (eg, QALY) as a “cost-utility”


analysis, even in cases where “cost-effectiveness” was applied inthe article title.30,38,39,43–46,49,48,51,52–56 This designation does notrepresent an error or shortcoming on the part of the originalstudies because the terminology has become common in healtheconomic literature. Quality adjustments are done by accountingfor the quality of life associated with various health states usingutilities ranging from 0 (death) to 1 (perfect health). Economicevaluation approaches reflected their respective aims, which mostoften (18 articles; 78%) were to contrast the cost-effectiveness ofmultiple interventions. As shown in Appendix Figure 1 in Sup-plemental Materials found at https://doi.org/10.1016/j.jval.2020.07.013, cost-utility analyses were employed most frequently.

Sensitivity analysisTwenty-two studies (96%) performed sensitivity analyses. Ten

(43%) performed a deterministic sensitivity analysis only, and 12(52%) performed a probabilistic sensitivity analysis, often inaddition to a deterministic sensitivity analysis. More recent pub-lications were more likely to be probabilistic (median publicationyear 2017, vs 2005 for deterministic approaches). A singular mostsensitive parameter was not usually identified, but cost-effectiveness results often varied depending on the inputs. Themagnitude of the dependence was not consistently reported. Au-thors’ conclusions were typically qualitative interpretations of themain cost-related result.

Time horizonTime horizons specified by the model ranged from 1 year or

less (5 articles; 22%) to lifetime or approximating lifetime (10 ar-ticles; 43%).

DiscountingThe most common discount rate chosen was 3% (14 articles;

61%) but ranged from 1.5% to 5%. All studies applied discountingexcept in cases with 1 year or smaller time horizons, where itwould not apply.

These characteristics inform the context of conclusions deci-sion makers may draw from the results of the economic evalua-tions. This gauging of technical attributes of SBEEs may expediteidentification of appropriate areas of future research. If, forexample, a decision maker is particularly interested in learningabout interventions’ impact on budget, this section shows that

Figure 3. Distribution of evaluated interventions. Studies could haveactive ingredient in heroin.

current SBEE literature is limited to only one budget impactanalysis (ie, Asche 201536), and research investments might bemade in that area. Additionally, data collection efforts need to bealigned with research investments in order to develop bettermodels for better decisions. Our study offers insights into whattypes of health and cost data are still needed across the diverseinterventions for OUD.

Cost and Effectiveness Estimation

Input costs for all studies typically included healthcare costsassociated with the intervention directly such as cost of medica-tion, administrative costs, and treatment delivery costs. Seven(30%) studies accounted for costs associated with adverse eventssuch as emergency care in the event of overdose. Some input costcomponents appeared only within the context of specific types ofinterventions. For example, pharmacotherapy maintenance38

usually entails urinalysis testing, and training and education costsoften incurred within naloxone distribution strategies.51,54 Costsfor comorbid conditions such as HIV and HCV were accounted forin 7 (30%) studies. When a perspective beyond the healthcaresystem is adopted, as in the case for societal perspectives, inputcosts such as the costs associated with criminal activity (8 articles;35%) and worker productivity (6 articles; 26%) were alsoconsidered.

The effectiveness measures extraction item gives the summaryoutcomes unit and notes the driving factor for differences ineffectiveness between intervention strategies. Summary effec-tiveness units were usually life-years (3 articles; 13%) or quality-adjusted life-years (QALYs) (15 articles; 65%) in accordance withmost economic evaluation approaches being cost-effectiveness orcost-utility analyses. The important distinction—in economicevaluation generally and in the context of SBEEs specifically—be-tween cost-utility and cost-effectiveness analysis is ultimately thepresence of quality adjustment; cost-utility studies account for thedifference in quantified estimates of quality of life gained, not justsimply unadjusted count of life-years gained.57

Incremental Cost-Effectiveness Ratios and Other CostAnalysis Findings

Although the heterogeneity of the studies precluded a meta-analysis, some cost-effectiveness results may be summarized in

more than one evaluated intervention. Diacetylmorphine is the


-- 13

cases where economic evaluations analyzed common alternativesor framed their findings in similar ways. Two studies55,56 analyzedmultiple mixed strategies and found an all-inclusive strategy ofnaloxone distribution, pre-exposure prophylaxis (PrEP), andaddiction treatment was the cost-effective alternative to sub-combinations of naloxone distribution, PrEP, and addiction treat-ment or each approach alone, at an incremental cost-effectivenessratio (ICER) of $95 337 per QALY gained.56 A particular naloxonedistribution strategy combination (high levels of distribution tolaypeople, police, fire workers, and EMS) was cost-effectiveagainst the other combinations with an ICER of $15 950 perQALY gained.55 The intersection of sensitivity analysis results andintervention type point to how interventions might perform amidchanging conditions. A trend that emerged for the most sensitiveparameters across multiple studies51,52,53,54,58 relates to the un-certainty around lay naloxone distribution: these results tended tobe sensitive to cost and intensity of training, efficacy, and pro-portion of witnessed overdoses.

Incremental cost-effectiveness ratio (ICER) results are displayedunder the results column in Table 1. Economic evaluations showedfavorable cost-effectiveness results under a commonly acceptedwillingness to pay threshold of $50 000 per QALY for naloxonedistribution strategies against the alternative of no naloxone dis-tribution52,53,54 and for opioid antagonist therapies (MMT or BMT)against an alternative of no treatment37,44,48 or against a currentstandard of care.43 The economic evaluations contained strongevidence that MMT is cost-effective against BMT41,42 or no treat-ment37 with one mixed result38 that is discussed herein.

In the budget impact analysis,37 the total cost incurred byscaling up a health intervention in terms of market share wascompared to the alternative of smaller or 0% market shares. Ageneral cost analysis40 investigated government payer or patientshare of cost burden under MMT subsidy policy implementationand found subsidizing methadone would reduce the cost burdenon patients and costs are offset by social and health gains. Otherstudies analyzed health and cost impacts of the abuse-deterrentopioid prescribing decision separately rather than incremen-tally35 or estimated costs over heroin use.47

Only 1 study41 examined injectable naltrexone and comparedit with MMT or BMT, and it was considered cost-effective againstMMT if policymakers were willing to pay an additional $72 perpatient per opioid-free day. One study46 examined prescribeddiacetylmorphine (ie, heroin), and it was found to be cost-effectiverelative to MMT. Additionally, cost findings were in favor of thesublingual over the tablet formulation of buprenorphine-naloxonefrom a public payer perspective36 and the subdermal implantableover the sublingual version of buprenorphine. Three studies thatlooked specifically at scale-up of interventions for buprenorphineor methadone maintenance therapies30,45,49 showed evidence infavor of cost-effectiveness.

Quality Assessment

Eleven (48%) of the 23 studies met the full 10 checklist criteria(see the overall score (%) in Table 1 and details in Appendix Table 1in Supplemental Materials found at https://doi.org/10.1016/j.jval.2020.07.013). Figure 1 shows the proportion of items that were fullyor partially satisfied.

Discussion

Summary of Findings: Quality, Scope, and Gaps

We identified 23 studies that conducted simulation-basedeconomic evaluations (SBEEs), analyzing the cost of various

interventions to address opioid overdose and use disorder. Thequality assessment indicated that the economic evaluations wereof moderate to high quality, with most studies fully satisfying atleast 90% of the quality assessment items. It appears particularlydifficult for researchers to identify the important and relevantcosts and consequences for each alternative, which could be dueto the challenges of data availability and complexity. This chal-lenge is especially present when attempting to model a popula-tion, such as people who use heroin, whose behaviors areparameterized with high uncertainty over a long time horizon.37,47

Other cost-related challenges appeared, for instance, in measuringthe quantities of drug-free therapy and detoxification servicesreceived,37 differences in cost estimation methods betweenmethadone and buprenorphine treatments,42 and missingsocietal-relevant costs such as lost wages in the societalperspective.47 When performed, sensitivity analyses help resolvethese concerns in part by showing that varying the cost inputs didnot alter the results drastically37—although interpretation iscomplicated by the lack of a standard, generally acceptablethreshold of willingness to pay for some outcomes such as addi-tional patients retained in treatment.42

The topical scope of the studies covered 6 main analyses ofcost-effectiveness: methadone expansion; comparisons ofdifferent medications for opioid use disorder (methadone,buprenorphine, injectable naltrexone, and, in one study, medicallyprescribed diacetylmorphine [ie, the active ingredient in heroin]);comparisons of modes of medication delivery (eg, sublingual vssubdermal buprenorphine-naloxone); comparisons of medica-tions to treatment without medications; abuse-deterrent versusnon–abuse-deterrent opioid prescribing; and comparisons ofdifferent naloxone distribution strategies. The most commoncomparisons involved opioid agonist treatments, particularlymethadone, a full agonist, and buprenorphine, a partial agonist.

Economic evaluation results depend on many specifications—including geographies represented, choice of input parameters,model structures, assumptions inherent to the underlying models,and various reported outcomes—that make it difficult to arrive atblanket conclusions across a given extraction category. Because ofhigh heterogeneity among the studies, additional synthesis in theform of a quantitative aggregation of results was ruled out.Nonetheless, results were consistent enough to give a reliableframework for discussion of overall findings.

The only harm reduction strategy evaluated across studies wasnaloxone distribution; other harm reduction strategies not treatedin this collection of SBEEs—which may be an area for futureresearch—include syringe exchange, safe consumption sites, fen-tanyl test strips, and education programs that teach how to usemore safely.59-61 The inclusion and exclusion criteria were set upsuch that they would have captured SBEEs on harm reductionstrategies other than naloxone distribution. Economic evaluationsthat compared a naloxone distribution strategy against nonaloxone distribution consistently arrived at a favorable cost-effective result for naloxone distribution with ICERs under awillingness-to-pay threshold of $50 000 per QALY.52-54 Thestudy55 that addressed the question of which persons should betargeted to carry naloxone recommends prioritization of distri-bution to laypeople likely to experience or witness an overdoseand first responders.

Economic evaluations for opioid antagonist therapies (MMT orBMT) against an alternative of no treatment37,44,48 or against acurrent standard of care43were also found tobe cost-effective at the$50 000 threshold. The cost-effectiveness comparison ofMMTwithBMT was mixed: in a model38 with HIV effects, one extreme sce-nario found BMT to be cost-effective relative to MMT at awillingness-to-pay threshold of $100 000 (ICERs ranged from $84



700 to $10 800 perQALY). This study38was published in 2001, at theearliest part of the time window for economic evaluations repre-sented in the review, and contained high uncertainty because thetreatmentwasnot introduced in theUSmarket until 2002.62Amorerecent analysis42 foundMMTcost-effective over BMT if the decisionmaker is willing to pay $10 437 per additional patient in treatmentor $8515 per additional opioid-freeweek gained. It is not surprisingthat MMT was found more effective given its superior clinical re-sults in patient retention relative to buprenorphine ornaltrexone63,64; its cost-effectiveness depends on the relevant de-cision maker’s willingness to pay for the given outcomes. Notably,the one study evaluating prescription diacetylmorphine found itless costly and more effective than MMT, which is consistent withclinical studies finding superior outcomes in patients prescribeddiacetylmorphine (often in addition to methadone).65-67

The one study41 examining injectable naltrexone should beconsidered in light of additional research that has been conductedsince 2015. First, studies in the United States have expanded,finding significant induction failure (ie, very early relapse68) andgreater rates of discontinuation,69 albeit equivalent success rates ifpeople can be successfully inducted. Moreover, the FDA has issueda warning letter to Alkermes, the manufacturer of Vivitrol, forwithholding information about the opioid overdose risk associ-ated with its drug.70 Thus, the limitation pointed out by Jacksonet al41 (that there is a “relative lack of evidence on the effective-ness”) is less true now, and additional economic evaluationsshould be conducted taking into account these recent findings.The results of the review demonstrate a dearth of economicevaluations of prevention strategies for the opioid crisis, as only 1study35 addressed this angle.

To the end of using the results of this collection of economicsevaluations to guide allocative efficiency, we can arrive at a broadconclusion that naloxone distribution and methadone-assistedtreatment are options with consistently favorable cost-effectiveness results. Each study should be considered within itsown context of relevant alternatives. Responsible interpretation ofthe economic evaluations’ results will take into account eachstudy’s sensitivity analysis approach and input ranges andwhether the inputs are varied in a manner that reflects thepractical context for the decision makers’ application. The SBEEstudies reviewed did not typically take account of patient andprescriber preferences or acceptability of an intervention. Only 2studies compared different formulations of buprenorphine,finding that the relative value of sublingual or tablet form wasdependent on contextual factors that have since changed. Specif-ically, sublingual formulations were estimated to have lower totaldirect costs than tablets at drug prices as of the study36 publishedin 2015, which was before the FDA’s approval in 2018 of cheaper,generic formulations of sublingual buprenorphine-naloxone.71

Implantable buprenorphine, a relatively new form of the medi-cation that was only approved in 2016,72 was found cost-effectiverelative to the sublingual formulation, but recent studies find thatpatients still generally prefer sublingual and other short-actingforms over injectable or implantable buprenorphine.73-75 Patientpreferences could preclude realization of any projected cost sav-ings if patients are reluctant to use the product. This challenge ofaccounting for patient preference has been an active area ofimproving cost-effectiveness research.76

Besides making an intervention potentially irrelevant, lowacceptability can mask costs of rapid scale-up for interventions thatgo underutilized. Feedback dynamics are largely unaccounted for,and, simply, preferences can change. Increasingly, urgent policyquestions surround the rising costs of medications due to manu-facturers’ increased perception of demand and willingness to pay.Failing to account for rising prices can dramatically reduce cost-

effectiveness and limit access to life-saving medications, particu-larly for treatments requiring prolonged use of medication, whichis often the case in treatments for opioid use disorder. Cost inputswere cited as among the most sensitive parameters in studiesacross a variety of intervention types: opioid agonist therapies (ie,methadone and buprenorphine maintenance),42,43,50,77 extended-release naltrexone,41 and naloxone distribution strategies;51-53

changing costs has the potential to present a shift in cost-effectiveness across a multitude of strategies.

Seven of the 23 studies cite cost as among the most sensitiveinputs to the cost-effectiveness of treatment; rising prices of acentral input to treatment costs may play an important role inusing modeling insights to generate effective policy strategies.

Finally, only 2 studies37,50 stratified results by age and gender;stratification by other factors such as socioeconomic status,ethnicity, or urban versus rural locality was not performed by anyof the reviewed study. Stratification importantly provides insightinto patient heterogeneity, reflecting how results for the so-calledaverage patient may not represent patients in a given subgroup,thus informing the degree of appropriate generalizability of eachstudy’s conclusions.

Limitations and Future Research

This review was limited to journal articles published in English.Economic evaluations such as those conducted alongside a clinicaltrial or based on retrospective data without a simulation extensionwere excluded.

The quality assessment of economic evaluations was limited tothe Drummond checklist. This checklist is one of multiple eco-nomic tools that exist, each with advantages and disadvantages,designed to aid quality assessment of economic evaluations. Theprimary advantage in this application is its offering of familiarity,ease of understanding, and accessibility to a general audience;however, the checklist cannot identify weaknesses in the under-lying simulation model. The assessment informs how well thestudy has included the minimum quality aspects for economicevaluation. If a simulation model was of poor methodologicalquality, relied on weak or invalid assumptions, or was not cali-brated, the economic evaluation could still attain a perfect qualityscore as long as the components of the economic evaluation weremet. The Quality of Health Economic Studies (QHES) instrument isan example of a more specific quality assessment instrument thatmight be employed for future research assessing the quality andrigor of the simulation modeling methods included in this re-view.78 A potential area for future research can also includereviewing the models retrospectively to assess their accuracy inpredicting cost-effectiveness based on now observable historicaltrends. Areas that merit further attention as a topic for futureresearch include understanding the tradeoff between makingopioids available for pain versus restricting opioids and facingpossible effects of undertreated pain; syringe exchange programs;and supervised consumption sites.

Conclusions

This systematic review identified 23 simulation-based eco-nomic evaluations (SBEEs) in the opioid literature. The charac-teristics of this set of studies reflect the heterogeneity andcomplexity behind this public health crisis and the large set ofdecisions required in attempting to model it. Despite slight de-clines in fatal opioid overdoses in 2018 and 2019,79 Americans arestill dying at unprecedented rates from opioids. In addition to thetremendous human toll that these deaths take, there are signifi-cant societal and healthcare system costs. Many of these

-- 15

healthcare costs will remain, if not rise, as people with opioid usedisorder access treatment and initiate medication maintenance.By making the best available use of limited data, SBEEs can helpaccount for the chronic nature of the opioid crisis and the effectsof various interventions within a greater dynamic system.

Supplemental Material

Supplementary data associated with this article can be found in theonline version at https://doi.org/10.1016/j.jval.2020.07.013.

Article and Author Information

Accepted for Publication: July 25, 2020

Published Online: Month xx, xxxx

doi: https://doi.org/10.1016/j.jval.2020.07.013

Author Affiliations: MGH Institute for Technology Assessment, HarvardMedical School, Boston, MA, USA (Beaulieu, DiGennaro, Stringfellow, Con-nolly, Jalali); Department of Epidemiology, Mailman School of Public Health,Columbia University, New York, NY, USA (Hamilton, Keyes); Division ofEnvironmental Health Sciences, College of Public Health, The Ohio StateUniversity, Columbus, OH, USA (Hyder); Center for Opioid Epidemiologyand Policy, Department of Population Health, New York University Schoolof Medicine, New York, NY, USA (Cerdá); Sloan School of Management,Massachusetts Institute of Technology, Cambridge, MA, USA (Jalali).

Author Contributions: Concept and design: Beaulieu, DiGennaro, JalaliAcquisition of data: Beaulieu, Hamilton, Cerda, KeyesAnalysis and interpretations of data: Beaulieu, Stringfellow, Connolly,

Hamilton, Hyder, Cerda, Keyes, JalaliDrafting of the manuscript: Beaulieu, DiGennaro, Stringfellow, Connolly,

Hyder, Cerda, Keyes, JalaliCritical revision of the paper for important intellectual content: Beaulieu,

DiGennaro, Stringfellow, Connolly, Hyder, Cerda, Keyes, JalaliSupervision: Jalali.Conflict of Interest Disclosures: Dr. Keyes reports personal fees

related for consultation and testimony in product litigation. No other dis-closures were reported.

Funding/Support: The research team was supported in part by theU.S. Food and Drug Administration (U01FD006868-01) and the NationalInstitute on Drug Abuse (UM1-DA049415).

Role of the Funder/Sponsor: The funder had no role in the design andconduct of the study; collection, management, analysis, and interpretationof the data; preparation, review, or approval of the manuscript; or decisionto submit the manuscript for publication.

REFERENCES

1. Centers for Disease Control. CDC’s efforts to prevent opioid overdoses andother opioid-related harms. https://www.cdc.gov/opioids/pdf/Strategic-Framework-Factsheet_Jan2019_508.pdf. Accessed February 7, 2020.

2. CDC Director’s Media Statement on U.S. Life Expectancy [press release]; 2002.https://www.cdc.gov/media/releases/2018/s1129-US-life-expectancy.html.Accessed February 5, 2020.

3. Wilson N, Kariisa M, Seth P, Smith H IV, Davis NL. Drug and Opioid-InvolvedOverdose Deaths — United States, 2017–2018. MMWR Morb Mortal Wkly Rep.2020;69:290–297.

4. Scholl L, Seth P, Kariisa M, Wilson N, Baldwin G. Drug and opioid-involvedoverdose deaths—United States, 2013–2017. MMWR Morb Mortal Wkly Rep.2018;67:1419–1427.

5. Center for Behavioral Health Statistics and Quality. Results from the 2017National Survey on Drug Use and Health: Detailed Tables. US Health andHuman Services. 2018.

6. Jalali MS, Botticelli M, Hwang RC, Koh HK, McHugh RK. The opioid crisis: acontextual, social-ecological framework. Health Res Policy Syst. 2020;18:87.

7. Opioid Epidemic Cost the U.S. Economy at Least $631 Billion Over Four Years:Society of Actuaries’ Analysis [press release]; 2019.

8. McDonald R, Campbell N, Strang J. Twenty years of take-home naloxone forthe prevention of overdose deaths from heroin and other opioids—concep-tion and maturation. Drug Alcohol Depend. 2017;178:176–187.

9. The National Academies of Sciences Engineering and Medicine. Medicationsfor Opioid Use Disorder Save Lives. Washington, DC: The National AcademiesPress; 2019.

10. Hoomans T, Severens JL. Economic evaluation of implementation strategiesin health care. Implementation Sci. 2014;9(168).

11. Hawkins N, Grieve R. Extrapolation of survival data in cost-effectivenessanalyses: the need for causal clarity. Med Decis Making. 2017;37(4):337–339.

12. Goldie SJ. Public health policy and cost-effectiveness analysis. JNCI Mono-graphs. 2003;2003(31):102–110.

13. Wakeland W, Nielsen A, Schmidt TD, et al. Modeling the impact of simulatededucational interventions on the use and abuse of pharmaceutical opioids inthe United States: a report on initial efforts. Health Educ Behav. 2013;40(1suppl):74S–86S.

14. Wakeland W, Nielsen A, Geissert P. Dynamic model of nonmedical opioid usetrajectories and potential policy interventions. Am J Drug Alcohol Abuse.2015;41(6):508–518.

15. Chen Q, Larochelle MR, Weaver DT, et al. Prevention of prescription opioidmisuse and projected overdose deaths in the United States. JAMA NetworkOpen. 2019;2(2).

16. Pitt AL, Humphreys K, Brandeau ML. Modeling health benefits and harms ofpublic policy responses to the US opioid epidemic. Am J Public Health.2018;108:1394–1400.

17. Sharareh N, Sabounchi SS, McFarland M, Hess R. Evidence of modelingimpact in development of policies for controlling the opioid epidemic andimproving public health: a scoping review. Subst Abuse. 2019;13:1178221819866211.

18. Jalali MS, Botticelli M, Hwang R, Koh HK, McHugh RK. The opioid crisis: needfor systems science research. Health Res Policy Syst. 2020;18:88.

19. Doran CM. Economic evaluation of interventions to treat opiate dependence:a review of the evidence. Pharmacoeconomics. 2008;26(5):371–393.

20. Chetty M, Kenworthy JJ, Langham S, et al. A systematic review of healtheconomic models of opioid agonist therapies in maintenance treatment ofnon-prescription opioid dependence. Addict Sci Clin Pract. 2017;12(1):6.

21. Murphy SM, Polsky D. Economic evaluations of opioid use disorder in-terventions. Pharmacoeconomics. 2016;34(9):863–887.

22. Agar MH, Wilson D. Drugmart: heroin epidemics as complex adaptive sys-tems. Complexity. 2002;7(5):44–52.

23. Siddaway A, Wood A, Hedges L. How to do a systematic review: a bestpractice guide for conducting and reporting narrative reviews, meta-analyses, and meta-syntheses. Annu Rev Psychol. 2019;70:747–770.

24. Joanna Briggs Institute Reviewers’ Manual: 2014 edition. The Joanna BriggsInstitute. 2014.

25. Drummond MF, O’Brien B, Stoddart GL, Torrance GW. Methods for the Eco-nomic Evaluation of Health Care Programmes. Oxford: Oxford University Press;1987.

26. Nickel F, Barth J, Kolominsky-Rabas P. Health economic evaluations of non-pharmacological interventions for persons with dementia and theirinformal caregivers: a systematic review. BMC Geriatr. 2018;18(1).

27. Gibson E, Begum N, Sigmundsson B, et al. Economic evaluation of pediatricinfluenza immunization program compared with other pediatric immuni-zation programs: a systematic review. Hum Vaccin Immunother.2016;12(5):1202–1216.

28. Gonzalez-Perez JG. Developing a scoring system to quality assess economicevaluations. Eur J Health Econ. 2002;3(2):131–136.

29. Wallace BC, Small K, Brodley CE, et al. Deploying an interactive machinelearning system in an evidence-based practice center: abstrackr. Paper pre-sented at: Proceedings of the ACM International Health Informatics Sym-posium (IHI); 2012.

30. Zaric GS, Barnett P, Brandeau M. HIV transmission and the cost effectivenessof methadone maintenance. Am J Public Health. 2000;90(7):1100–1111.

31. Alistar SS, Owens DK, Brandeau ML. Effectiveness and cost effectiveness ofexpanding harm reduction and antiretroviral therapy in a mixed HIVepidemic: a modeling analysis for Ukraine. PLoS Med. 2011;8(3).

32. Neumann PJ, Sanders GD, Russell LB, et al. Cost-Effectiveness in Health andMedicine. 2nd ed. New York, NY: Oxford University Press; 2016.

33. Garrison LP Jr, Pauley MV, Willke RJ, Neumann PJ. An overview of value,perspective, and decision context—a health economics approach: an ISPORspecial task force report [2]. Value in Health. 2018;21(2):124–130.

34. Sanders G, Neumann P, Basu A. Recommendations for conduct, methodo-logical practices, and reporting of cost-effectiveness analyses: second panelon cost-effectiveness in health and medicine. JAMA. 2016;316(10):1093–1103.

35. Yenikomshian MA,White AG, Carson ME, et al. Projecting the cost, utilization,and patient care impact of prescribing extended release non-abuse-deterrentopioids to chronic pain patients. J Opioid Manag. 2017;13(5):291–301.

36. Asche CV, Clay E, Kharitonova E, et al. Budgetary impact of the utilization ofbuprenorphine/naloxone sublingual film and tablet for Medicaid in theUnited States. J Med Econ. 2015;18(8):600–611.

37. Barnett PG. The cost-effectiveness of methadone maintenance as a healthcare intervention. Addiction. 1999;94(4):479–488.

38. Barnett PG, Zaric GS, Brandeau ML. The cost-effectiveness of buprenorphinemaintenance therapy for opiate addiction in the United States. Addiction.2001;96(9):1267–1278.

39. Carter JA, Dammerman R, Frost M. Cost-effectiveness of subdermalimplantable buprenorphine versus sublingual buprenorphine to treat opioiduse disorder. J Med Econ. 2017;20(8):893–901.

https://doi.org/10.1016/j.jval.2020.07.013http://doi.org/https://doi.org/10.1016/j.jval.2020.07.013https://www.cdc.gov/opioids/pdf/Strategic-Framework-Factsheet_Jan2019_508.pdfhttps://www.cdc.gov/opioids/pdf/Strategic-Framework-Factsheet_Jan2019_508.pdfhttps://www.cdc.gov/media/releases/2018/s1129-US-life-expectancy.htmlhttp://refhub.elsevier.com/S1098-3015(20)34399-0/sref3http://refhub.elsevier.com/S1098-3015(20)34399-0/sref3http://refhub.elsevier.com/S1098-3015(20)34399-0/sref3http://refhub.elsevier.com/S1098-3015(20)34399-0/sref4http://refhub.elsevier.com/S1098-3015(20)34399-0/sref4http://refhub.elsevier.com/S1098-3015(20)34399-0/sref4http://refhub.elsevier.com/S1098-3015(20)34399-0/sref5http://refhub.elsevier.com/S1098-3015(20)34399-0/sref5http://refhub.elsevier.com/S1098-3015(20)34399-0/sref5http://refhub.elsevier.com/S1098-3015(20)34399-0/sref6http://refhub.elsevier.com/S1098-3015(20)34399-0/sref6http://refhub.elsevier.com/S1098-3015(20)34399-0/sref7http://refhub.elsevier.com/S1098-3015(20)34399-0/sref7http://refhub.elsevier.com/S1098-3015(20)34399-0/sref8http://refhub.elsevier.com/S1098-3015(20)34399-0/sref8http://refhub.elsevier.com/S1098-3015(20)34399-0/sref8http://refhub.elsevier.com/S1098-3015(20)34399-0/sref9http://refhub.elsevier.com/S1098-3015(20)34399-0/sref9http://refhub.elsevier.com/S1098-3015(20)34399-0/sref9http://refhub.elsevier.com/S1098-3015(20)34399-0/sref10http://refhub.elsevier.com/S1098-3015(20)34399-0/sref10http://refhub.elsevier.com/S1098-3015(20)34399-0/sref11http://refhub.elsevier.com/S1098-3015(20)34399-0/sref11http://refhub.elsevier.com/S1098-3015(20)34399-0/sref12http://refhub.elsevier.com/S1098-3015(20)34399-0/sref12http://refhub.elsevier.com/S1098-3015(20)34399-0/sref13http://refhub.els

economic evaluation in opioid modeling: systematic review · 2020. 10. 27. · conclusion: the...

Documents