Surfacing the hidden demand for opioid dependent treatments for drug policy
makers
Mark Heffernan, Evans & Peck, Sydney Geoff McDonnell, Adaptive Care Systems, J enny
Chalmers and Alison Ritter, National Drug and Alcohol Research Centre, UNSW.
and UNSW Correspondence: mheffernan@ evanspeck.com
Abstract:
Illicit drug policy has been the subject of important SD studies addressing the interaction
between policing and medical treatment and estimating the prevalence of national
cocaine use. Here we modeled the impacts of policy changes associated with wider use of
newer opioid pharmacotherapies besides methadone. These newer drugs allow less
supervision of dosing and changes in the mix of prescribing and dispensing
arrangements. Key aspects of the model were estimation of potential demand for the
enhanced range of therapies and the cost and treatment impacts of changes in cycling on
and off treatments due to pricing and service configurations.
Here we describe the use of SD models to provide a logical consistent framework for
stimulating debate about incomplete and ambiguous data and clarifying the differences
in expectations and goals of treatment among broad groups of policy makers. Our
methodology included incorporating key concepts accepted from previous economic
equilibrium Markov models and control phase plots from previous modeling in the area.
Funded by the Australian National Council on Drugs www.ancd.org.au
This material is yet to be released.
Introduction
The early use of system dynamics in framing the interaction between medical treatments
and policing interventions resulted in publication of the classic book, The Persistent
Poppy by Levin, Roberts and Hirsch in 1975. This book outlined the delicate balance
between criminal and medical activities and the potential intolerable consequences of
extreme policies of “full prohibition” and “full legalization” mediated through feedback
effects via the price of heroin. The policy interventions described in this book include
educational effort, police effort, community education, re-entry programs, available
methadone treatments and counseling. An updated causal loop diagram, kindly supplied
by one of the original authors, Gary Hirsch, nicely illustrates the dynamic complexity of
illicit drug policy.
Community Definition of the
Problem (Criminal vs Medical)
Extemal Supply Community
Education
Police Effort——
Arrest Rate.
~ Treatment
Available
Educational \ J
Effort ——__ ADDICTS oN
Neaiddaiols THE STREET
ADDICTS IN
‘TREATMENT,
“ash vig
—— Effect , EX-ADDICTS
NON-USING we
POPULATION
Community Soca Successful Reentry
Economic Status
Fig 1 The Persistent Poppy, (Hirsch revised 2007)
Another important application of system dynamics to illicit drug policy was the National
Cocaine Prevalence Model by Jack Homer. This work included the integration of
multiple disparate data sets to reconcile the reported users of cocaine, the price, deaths
from overdose, arrests and the impact of introduction of crack. The results are
summarized in the following diagram.
Fig 2 Final Structure of the Cocaine Prevalence Model, (Homer 1996)
This paper describes the results of a current project which again uses system dynamics
modeling to develop and test the impacts of future government policy options in the
provision of medical treatments for illicit drug users.
Background
Methadone maintenance therapy has been the mainstay of medical treatment for opioid
dependence for many years. Newer oral drug substitutes for methadone are now
becoming widely available, particularly Buprenorphine (BuP), used alone or combined
with the narcotic antagonist, Naloxone. The Australian National Council on Drugs
(ANCD), a peak policy group, commissioned the Drug Policy Modeling Program
(DPMP) of the National Drug and Alcohol Research Centre to investigate the issues
related to new opioid dependent pharmacotherapies and advise on potential changes in
policy and practice.
# of Clients
30gcc a
2s000 a
1000
19000
soce Fa
1985 1900 1907 1998 1999 2000 2001 2002 2003 2004 2005 2006
Source: Australian Institute of Health and Welfare. Alcohol snd other drug treatment services in Austraiis.
Data Sets 2000-2006. Drug Treatment Series: Numbers 1 - 7. Canberra,
ed)
10 methadone is the only pharmacotherapy drug. From 2000 onwards
buprenorphine and ultimately buprenorphine-naloxone Is included. In 2006 there were 27, S88 methadone
patients; just over 70 per cent of all patients.
National Pharmacotherapy treatment patients from 1985 to 2006. Chalmers et al (2008)
This consultative project has produced a Pharmacotherapies Issues Paper (yet to be
released) and joint system dynamics modeling around the demand for services and costs
and benefits to the government and the community.
Currently the National and State Governments subsidises and provides a range of legal
prescribing and dispensing options for medications in addition to counseling and support
services. Medications are prescribed by public clinics, private doctors (primary care
practitioners’ offices or clinics or prisons services. The drugs are dispensed and
administered supervised at the public or private clinics, the community pharmacy or
prisons. In some cases “take-away” doses are available for partially supervised patients.
Regular reporting of patient numbers and medications dispensed is required by law, but
this data is not available through the life course of an individual patient. Therefore system
dynamics modeling was selected to assist the project to make sense of disparate datasets
and ‘triangulate’ estimates in order to gain consensus on the overall current state and the
consequences of future policy options.
Approach to Model Development
The DPMP was experienced in reviewing international and national literature and
synthesizing data from studies, surveys and reports. It had used a variety of economic,
stochastic, biostatistical and agent based models in the illicit drug policy area, but not
system dynamics. The most similar approach to system dynamics (that they were familiar
with) was a simple compartmental Markov model of illicit drug use. Parameters for this
model had been estimated using the usual Markov assumption of an equilibrium final
absorbing state.
The structure and behavior of this model was replicated using an ithink stock flow model
and the team then learnt that it was possible to relax the assumptions of the Markov
model and explore non-equilibrium conditions, including non-linear feedback
interactions.
We then proceeded to develop a stock-flow model of the flow of patients on opioid
dependent therapies, through various prescribing and dispensing locations.
Model structure
A simplified version of the model structure is illustrated in the following diagram.
Length of Stay
onM
'
Opioid to First First Pais] /
Dependent arnt | Treatment on ae
Users Never fe) Split Mto 0OT
new users Treated
=
(die(recover) new BuP
v
Patients Patients
on outof
Buphen | Treatment |
-orphine «| BuP toOOT “s. “ "7
wh he i
” d 3 3 remov
Los on ‘onBuP MS oor
OOT to BuP ae
mii
5
Time OOT
The detailed model includes splitting the patients on treatment into their various
prescribing and dispensing locations and allocating costs to the various payers (National
and State Governments and Users).
Within the methadone treatment sector there are a number of sub-sectors. To enter
treatment patients must be prescribed methadone by a medical practitioner, registered to
prescribe methadone. The model differentiates between three types of prescribing
medical practitioners, on the basis of who pays for the prescribing and the cost of that
prescribing; those employed by public treatment clinics, those working in private
practices (including those prescribing out of private clinics) and those employed to work
in the prison system. The Commonwealth government pays for prescribing in private
practices while the state government covers the cost of prescribing in prisons and public
clinics. The cost of prescribing in prisons and public clinics differs. Patients flow
between the three prescriber types, as well as flowing in and out of treatment. There is
also a dispensing sub-sector differentiating again between methadone dispensing
locations on the basis of who pays for dispensing and the cost of that dispensing.
Dispensing is undertaken under the control of a pharmacist. Prison patients are all
prescribed and dispensed in prison pharmacies. While the majority of patients prescribed
in a public clinic will be dispensed their methadone in that clinic some are dispensed
methadone by community pharmacists in the pharmacy. The pharmacy might be more
convenient; perhaps closer to home than the public clinic. All of the patients whose
prescriber is a medical practitioner in private practice are dispensed in a community
pharmacy. The State government pays for dispensing undertaken in public clinics and in
prisons while the patient pays for dispensing in community pharmacies. Hence there is a
patient flow from the prescribing sector to the dispensing sector and information flows
from both those sectors to the costs sector. Here the model calculates the costs borne by
the patient, State and Commonwealth Governments. Those costs only accrue when the
patient is in treatment, that is, when the patient is taking his/her methadone prescription.
Model Calibration
The various parameters and data sources are listed in the following table:
Variable Parameter Reference / Notes
Stocks at commencement
of simulation
Treatment naive opioid 12,000 =[3,500 x 4 yrs] - 1400 (10% outflow).
dependent population Consensus estimate.
Methadone treatment 27,346 2006 census data (unreleased).
Prescribers
Public = 7,853
GP =17,169
Prison = 2,324
Buprenorphine treatment 11,071 2006 census data (unreleased)
Between treatment 30,000 Calibrated from the model, based on length of
stay and steady state. At start of simulation.
Data in Dietze et al 2003: 63% ever in
treatment, 45% in treatment last 12 months;
26% in treatment on day of interview. Of the
current intx stock, 40% b/n tx is the lower limit;
142% is the upper limit. Currently set at 100%
Flows
Entrants to opioid
dependency
Flow from treatment naive
opioid dependent
population into treatment
for first time
Other
Allocation of inflow into
first treatment by drug
Allocation of inflow into
first treatment
* Public
* GP
¢ Prison
Length of stay
Methadone
+ Public
+ GP
+ Prison
- Between treatment
Buprenorphine
+ In treatment
+ Between treatment
Flow probabilities
between prescribers
From GP
From public
From between treatment
3,500 per annum
Average time to
treatment is 4 years
43%: buprenorphine
57%: methadone
25%
60%
15%
7 month
12 months
3 months
12 months
6 months
6 months
to public 10%
to prison 4.5%
to b/w tment 83.2%
death 0.8%
abstinence 1.5%
to GP 10%
to prison 5%
to b/w tment 82.7%
death 0.8%
abstinence 1.5%
to public 25.5%
to GP 51%
to prison 15%
death 2%
abstinence 1.5%
Unknown. Estimates of new users, 5% of total
IDU population (Razali et al., 2007, Caulkins et
al; Law etal).
5% of 69,346 = 3,500
This initialisation figure also accommodates our
recovery and death estimates.
Dietze et al (2003) median 3 yrs for methadone.
ATOS 4 yrs (State reports: av. age first
treatment 24-25 yrs, regular injector av. 20-21
yrs; 29%-40% meth 1* tx). This figure is
affected by the feedback loop (see below).
To equilibrate the model Based on census/state
data
National census (28%, 62%, 8%)
Bell et al. (2006) 31%, 56% and 9%. Back-
calculated from static proportions in each
allocation at any one time.
ATOS, Bell et al., State data
Feedback loop This figure depends on the ratio of no. in
treatment (methadone + buprenorphine) to no.
“between treatment’.
It is 4 when the ratio is less than 2, but falls ata
declining rate as the ratio increases from 2.
There is a limit on the years to entry of 2.
Death rate
* Pre treatment 5% per annum
+ in treatment 0.8% per annum Byrne, 2000, Caplehorn, 1996
+ between treatment 2% perannum
Abstinence rate (in and 1.5% per annum ATOS, Byme, cross-checked against
between treatment) international figures (NTORS, DATOS, Hser)
Pre-treatment abstinence 5% per annum Ravali et al., 2005; Caulkins et al., 2007
rate
Costs
Drug cost (per dose) $0.54 PBS $36 per litre; Img =.72c. Av meth dose
70mg
Costs - maintenance
+ public $14.58 per day NEPOD
+ GP $3.78 per day NEPOD
+ Prison $9.26 per day Warren & Viney, 2004
Costs - dispensing
+ Public $1.05 NEPOD
+ GP $5.00 From State surveys, averaged
+ Prison $1.05 Assumed same as public - no other data
Use of the Model
We set out to construct a model that could be used by policy makers to explore feasible
policy scenarios. We had no intention for the model to generate forecasts of the
implications of policy changes. Rather, we intended that the model communicate a
particular understanding of the system that could be used as a shared basis for debate on
policy issues. As well, the model needed to be able to simulate implications of policy
changes, given the current state of the system. Crucial to the calibration of the model was
discussion with policy makers to ensure that the model’s depiction of the system was
sufficiently realistic, without being cumbersome. In that process we learned, for example,
it was simpler to assume a system in equilibrium with constant numbers in treatment over
the life of the simulation in status quo, rather than being distracted by justifying a
constant upward or downward trend in the absence of data.
One example of ‘triangulating” estimates based on the structure of the model was the
ratio between Patients length of time in treatment and out of treatment. Published
estimates varied from 0.4 to 1.4. Based on the model structure and related estimates we
were able to infer that the figure was around 1.0
Policy Experiments
The key issues explored in this model were
1. Dispensing fees on patients
2. Increasing demand for treatment
3. Decreasing supply of treatment by retirements of prescribing primary care
physicians.
A simplified diagram of policy experiments is shown below.
External
Supply
+
Heroin M Service
Capacity
sen eens MUser
207" Payments
x
Length of Stay
onM
1
+ ee ee: 1
" seat reste Hirst new f Paiens | /
Dependent Treatment on
Users Raver o>] Salt TP Methadone - A AOU
newusers | Treated
yj Temioved
removed
(die/recover) new BuP
Patients Patients
on out of
Buphen Treatment |
i -orphine.-
Morbidity & iD -
Mortality Po "
ge removed ‘ removed
Ss y, ‘ oot
Los onB on BuP ey
OOT to BuP Gone
hf
Results of Policy Experiments
Commonwealth Government dispensing and prescribing Costs $A/month
'$ million
5.00 =<
=
4.00 , —e— Status Quo
3.00 —s=— Scenariol
—#— Scenario 2
2.0 $e rcs Scenario 3
1.00
0.00
0 12 24 36 48 #60 72 84 98 108 120
Months
Notes: Status Quo: Patient pays dispensing fees at pharmacies
Scenario 1: Commonwealth pays dispensing fees.
Scenario 2: In response the average length of stay in treatment for patients dispensed in
pharmacies increases by 50 per cent.
Scenario 3: A secondary response is that the time it takes for an opioid dependent person to enter
treatment for the first time is halved, on average, from 4 years to 2 years
Patients in treatment on a monthly basis before and after a 20 per cent reduction in the
time between treatment
Patients
WB000! [eras ctacmnaicinieensstcunmstontnie
40000
3500 = nee In treatment(1)
30000 f In treatment
25000 pee Methadone(1)
20000 —_ Methadone
15000
IAPR ARO SEE | sccorepey Buprenorphine(1)
10000 +——__t_—_____._._.—_. +
0 12 24 36 48 60 72 84 96 108 120 Buprenorphine
Months
Key
Work in Progress (for Presentation in J uly)
Weare planning some extensions to this work to further quantify the benefits of various
policies. Key indicators include the benefits in treatment, including reduction in crime
rate, policing and criminal justice costs and the morbidity and mortality avoided,
including heroin overdoses and HIV/AIDS reduction.
Further Policy Experiments (in Progress)
Phase Plots including the difference between abrupt and gradual changes in supply and
demand parameters
Expanding the scope to include related System Dynamics Work
Once the project team has successfully built and demonstrated simple models we are
exploring the possibility of extending the scope of the work to progressively include
additional feedback interactions. An example of some of the possibilities is shown below.
This addresses the perennial issues of interactions between criminal and health
interventions and the vexed question of relative direct and indirect contributions of
different intervention mixes to reduce crime and health risks on changing the rate of new
opioid dependent users.
External
Supply
*
Supply
l4 +
Heroin
Heroin
Demand
aa
<-> Policing —w
Arrest
Funds to
Policing
Wy
Perceived
Health Risk
+
Funds to
Treatment
MService
Capacity
Opioid
Dependent
to First
Treatment,
First
‘Treatment
MUser
+
Users Never
Treated
new users
o>
Split
if
Perceived
Health Risk
+ +
Morbidity &
Mortality
+
(die/recover)
* F
removed f
new BuP
LOS onB
Mto OOT
BuP to OOT
| Patients
out of
Treatment |
removed
on BuP
(OOT to BuP
Time OOT
Conclusion
This project demonstrates the successful development of a useful stock and flow model to
assist policy makers in considering the impacts of various policy experiments. It offers a
firm foundation of a simple well-calibrated model which has the capability to be
progressively expanded to challenge the current boundaries of analysis used in this area.
It has the potential to more successfully spread the understanding of feedback interactions
among health and policing policies by carefully building on earlier system dynamics
work in this area.
References
Australian Institute of Health and Welfare (AIHW) (2007) Alcohol and other drug
treatment services in Australia 2005-06: Report on the National Minimum Data Set Drug
Treatment Series no. 7. Cat. no. HSE 53 Canberra, AIHW.
Bell
Caulkins
Chalmers, J., Ritter A., Faes, C. with input from the expert advisory group (Nick
Lintzeris, Tamara Speed, Bob Batey and Alex Wodak) (2008) “Opioid Pharmacotherapy
Maintenance in Australia - A background issues paper”
Dietze
Hirsch GB 2007 Personal communication
HomerJB Why we iterate: scientific modeling in theory and practice
Sys.Dyn.Rev. 1996 12 1 p1-19
Levin G Roberts EB Hirsch GB The Persistent Poppy: A Computer-Aided Search for
Heroin Policy Ballinger Cambridge MA 1975 ISBN 0-88410-031-6
