A presentation by SMART Infrastructure Facility Research Director Dr Pascal Perez to the 11th International Multidisciplinary Modeling and Simulation Multiconference (I3M), Bordeaux, September 2014.
A Lifetime Individual Sampling Model for Heroin Use and Treatment Evaluation in Australia
1. A Lifetime Individual Sampling Model
for Heroin Use and Treatment
Evaluation in Australia
Nagesh Shukla
Van Hoang
Marian Shahanan
Alison Ritter
Vu Lam Cao
Pascal Perez
September 2014
2. Introduction
• Australian federal and state governments spend about AUD 1.7b pa in prevention,
treatment, harm reduction and law enforcement to combat illicit drugs.
• There is an increasing pressure from both the government and the public to know
– whether the current spending is optimal; and/or
– what needs to change to increase the benefits of spending.
• This is particularly important for complicated policies where there are many
external costs and benefits, and as such; there are diverse views about the value
of the projects.
• The aim of this study is to
– assess the net social benefit of current heroin treatment strategies, and
– compare different combinations of treatment alternatives through modelled scenarios
2
3. Conceptual model
3
• Initial Population with over 97,000 heroin
users and heroin abstainers.
• Each individual is transitioned from one
health state to others using predefined
(individual based) state transition
probabilities.
• Time step is defined as the length of stay in
each state, individually driven.
• Population is evolved and added a sub-population
of new drug initiators each year.
• Net Social Benefit is computed based on the
outcomes of the simulation model.
• Main data sources:
– Australia Treatment Outcome Study (ATO) Dataset
– MIX Study Dataset
– National Opioid Pharmacotherapy Statistic Annual
Data (NOPSAD)
– Alcohol and Other Drug Treatment Services National
Minimum Data Set (AODTS-NMDS treatment data)
4. Model Components
4
Initial
Population
States
Transition
Time
State
Transitions
Costs and
Outcomes
Net Social
Benefit
5. Model Components (cont.)
Initial Population
• Is estimated as the current NSW heroin using population.
• Over 97,000 heroin users and heroin abstainers.
• Each year, a sub-population of new initiators is added to include new drug users.
• Each individual in the initial population has the characteristics as:
– Age: starting with 18 to 60 years spread
– Gender: male or female
– State: current state
– Opioid use history
– Incarceration history
– Treatment history
• The initial population is evolved over the lifetime
6. Model Components (cont.)
States
• Model has a set of mutually exclusive states which
are:
– large enough to capture the complexity of the treatment
process and
– low enough to ensure the resulting model is tractable
and does not overburden the model with very detailed
and specific data requirements.
• There are 2 main types of states:
– Drug use state: S1, S2, S3
– Treatment states: S4, S5, S6
• The model also considers 3 important locations
(stages) in the drug using individual’s trajectory:
– In Community: S1 to S6
– In Prison: S8, S9, S10
– Death Stage: S11, S12
• There is only 1 treatment state in the prison stage
due to the insufficient in-prison treatment data.
State Name Stage
Abstinence (S1) COMMUNITY
Irregular Use (S2) COMMUNITY
No Treatment & Use (S3) COMMUNITY
Withdrawal (S4) COMMUNITY
Residential Rehabilitation (S5) COMMUNITY
Pharmacotherapy (OTP) (S6) COMMUNITY
Counselling Only (S7) COMMUNITY
Abstinence (S8) PRISON
No Treatment & Use (S9) PRISON
Treatment (S10) PRISON
Drug Related Death or 60+ Years
DEATH
Old (S11)
Non-Drug Related Death (S12) DEATH
7. Model Components (cont.)
Transition Time
• Is heterogeneous ‘time to transition’ for each individual in the model based on
his/her attributes such as age, sex, treatment history, and state.
• Is defined as the length of stay (LOS) in each state, individually driven, stratified by
age, sex, history.
• Free from traditional fixed time steps for individual movements across states as
using continuous function for individual’s length of stay determination
8. Model Components (cont.)
State Transition
• After finishing assigned LOS in a state, individuals transition to other states based
on transition probability functions dependent upon the individuals’ attributes.
• There are 2 types of transition functions in the model:
– An equation: empirically derived, specifies the probability based on individual’s characteristics and
history of the transition.
– A probability distribution of the likelihood of transition: empirically derived from summary data,
based on a known distribution of an event
• Once a distribution function is established, Monte Carlo sampling is used to
choose transition probabilities.
9. Model Components (cont.)
Cost & Outcomes
• During running through cycles, the model will accrue costs and outcomes (also
referred to as rewards) within each cycle.
• Main categories of costs in the model:
– Treatment costs: per days and transition
– Crime costs: including social costs, penalty, and police costs
– Life-years: saved, or lost
– Other health care utilization (i.e. hospital, emergency department visits, and treatment for specific
diseases such as Hepatitis B and C)
– Economic impact on family burden event
• Main categories of benefits in the model:
– Earnings due to returning to work after successful treatments
– Cost-savings to the government and society due to successful treatments
(e.g. reduction of crime and health care utilization).
10. Model Components (cont.)
Net Social Benefit
• Once the costs and benefits have been calculated, the criterion for assessing the
overall efficiency of an intervention is the Net Social Benefit (NSB).
푁푆퐵 =
푇
푡=1
퐵푡 − 퐶푡
1 + 푟 푡−1
– 퐵푡 are benefits in year t,
– 퐶푡 are costs in year t,
– r is the discount rate, and
– T is the duration in years under consideration.
• The NSB is the sum of the present value of all benefits minus the sum of the
present value of all costs.
• A policy is potentially worthwhile if NSB is > 0.
11. Model Architecture
Java (Eclipse)
PostgreSQL
(Intermediate Data)
11
State
Transition
Algorithm
Cost/Benefit
Estimation
Population
Generator
PostgreSQL
(Output Data)
PostgreSQL
(Source Data)
Java Swing
JDBC
Graphic User Interface
12. Work In Progress Results
• An initial prototype simulation model has built, that creates the initial
heroin using population, new heroin initiators, and transitions to different
states.
• Developing user interface to support users to interact with the model to
design and run different scenarios.
• In the process of feeding the model with validated transition functions,
per unit/event costs, and benefits.
• The final step in the modelling will be to validate whether the model is
consistent with heroin user career trajectory.
12