The summary examines using agent-based modeling to analyze buprenorphine treatment capacity and policy options. The model simulates physicians, patients, and treatment relationships over time. Initial analyses explore the impact of changing physician caps and reducing barriers for physicians not currently prescribing. Reducing barriers significantly improved access, though did not help rural patients, as new physicians located where others practice in cities. Further research will add patient heterogeneity, model costs and diversion, and validate the model using actual regional data. The goal is to explore policy options like expanding prescribing to nurse practitioners/physician assistants.
Racial Disparties in Treatment of Beahvioral Health Disorders in phila 10.27.09
Buprenorphine Treatment Capacity: Agent-Based Modeling for Policy Analysis
1. Buprenorphine Treatment
Capacity: Agent-Based Modeling
for Policy Analysis
Alexandra Nielsen, MS
Portland State University
American Association for the Treatment of Opioid Dependence
Atlanta, April 1
2. The Investigator Team
Alexandra Nielsen, MS
• Doctoral student*
Wayne Wakeland, PhD
• Professor and chair* of Systems Science Program
Systems Science Graduate Program, Portland State
University, Portland Oregon
The authors thank SAMHSA for funding this work,
and have no other relationships to disclose.
3. Agenda
Research questions
Review of research into treatment capacity
Introduction to Agent Based Models (ABM)
Research plan: ABM for policy exploration
Initial analyses: Geography and patient caps
Next steps: Nuanced policy exploration
4. Research Questions
Is current treatment capacity and accessibility
sufficient to meet patient demand? Considering:
• Physician patient caps
• Physician barriers to offering treatment
• Geography
• Patient willingness to travel
• State/payer policy barriers
• APN/PA exclusion from DATA 2000
5. Research Questions: Policy Exploration
What policies might improve treatment
capacity and accessibility?
• Changing physician patient caps
• Reducing barriers to offering treatment
–Widespread adoption of demonstrated care
delivery models, such as:
» VT hub and spoke, MA NCM, NM ECHO
• APN/PA prescribing authority under DATA
2000
6. Very Brief Review of Research
[1] Stein, B. D., Gordon, A. J., Dick, A. W., Burns, R. M., Pacula, R. L., Farmer, C. M., …
Sorbero, M. (2015). Supply of buprenorphine waivered physicians: The influence
of state policies. Journal of Substance Abuse Treatment, 48(1), 104–111.
Supply of BUP waivered physicians [1]
• 43% of US counties have no waivered physicians
• 7% had 20 or more physicians
• Medicaid funding, opioid overdose deaths, specific
state guidance ++ waivered physicians
7. Very Brief Review of Research
[2] Rosenblatt, R. A., Andrilla, C. H. A., Catlin, M., & Larson, E. H. (2015). Geographic and
Specialty Distribution of US Physicians Trained to Treat Opioid Use Disorder. The
Annals of Family Medicine, 13(1), 23–26.
Geographic and specialty distribution of waivered
physicians (2012) [2]
• 16% of psychiatrists waivered, mostly urban
practice
• 3% of primary care physicians waivered
• “most” counties no waivered physician
8. Review of APN/PA BUP Research
APN/PA interest in prescribing BUP
• 48% PAs/NPs at HIV conferences in 2006 [3]
• 29% of Kansas PAs surveyed, 65% report
turning away a patient seeking BUP
treatment [4]
[3] Roose, R. J., Kunins, H. V., Sohler, N. L., Elam, R. T., & Cunningham, C. O. (2008). Nurse
practitioner and physician assistant interest in prescribing buprenorphine. Journal of
Substance Abuse Treatment, 34(4), 456–459.
[4] Spiser, V., & Dumolt, S. (2011). Physician Assistants’ Interest in Prescribing Buprenorphine for
Opioid Addiction. In Proceedings of 7th Annual GRASP Symposium. Wichita State
University.
9. BUP Treatment Capacity in the Aggregate
Aggregate numbers don’t tell the whole story.
• If doctors x cap levels >patients seeking treatment,
then capacity is sufficient.
Does not ring true
• Most prescribers << capped levels
• Other prescribers have waitlists and cannot
serve all who seek treatment.
• Many counties have no prescribers
10. Introduction to ABM
Take real world data and real lived experience
Create an artificial world—like SimCity or Mine Craft
Fill it with artificial people—doctors, patients who
interact and make choices
Change things, “what if…”
See what happens in our artificial world
Reflect on model results with experts
11. What’s Different about ABM?
ABM models individuals not aggregate variables
• Doctor agents interact with patient agents in a
local environment
• Agents are all different
Location matters
Findings arise from local dynamics, and individual
decisions
16. Data: Where are Treatment Seekers?
NSDUH “people meeting dependence criteria”:
• 49.9% large MSA > 1 million (city and suburb)
• 35.5% small MSA < 1 million (city and suburb)
• 14.5% rural areas
– 48% live in small rural towns, 52% isolated rural [5]
[5] Kvamme, E., Catlin, M., Banta-Green, C., Roll, J., & Rosenblatt, R. (2013). Who
prescribes buprenorphine for rural patients? The impact of specialty, location and
practice type in Washington State. Journal of Substance Abuse Treatment, 44(3),
355–360.
18. Data: Treatment Relationships
Patients differ in how far they travel for treatment [6]
More rural, longer distance [7]
[6] Beardsley, K., Wish, E. D., Fitzelle, D. B., O’Grady, K., & Arria, A. M. (2003). Distance traveled to outpatient drug
treatment and client retention. Journal of Substance Abuse Treatment, 25(4), 279–285
[7] unpublished data from https://www.naabt.org/ treatment locator, 2006-2014.
19. Assumptions: Treatment Relationships
Patients will try to establish with the closest provider
If the provider is not accepting new patients, this
provider will refer the patient to other providers
nearby
Patients will stay with the established provider
Providers have a personal limit on the number of
patients they would consider treating by specialty
20. Data: Treatment Relationships
44% of waivered providers will not prescribe (due to
barriers)
Providers’ personal limits are modified by patient
caps
22. Simulation through Time
Many things happen at once—people die, start
seeking treatment, stop seeking treatment, enter
treatment, exit treatment.
Treatment retention modeled to fit closed cohort
study [8]
New doctors enter
Doctors increase
cap after 1 year
[8] Bell, J., Trinh, L., Butler, B., Randall, D., & Rubin, G. (2009). Comparing retention in treatment and
mortality in people after initial entry to methadone and buprenorphine treatment. Addiction,
104(7), 1193–1200
23. Simulation: Social Influence
Number of new treatment seekers based on NSDUH
“feel I need treatment” response
Location of new treatment seekers is geographically
near current treatment recipients
Simulates clusters of opiate dependent people
Simulates social diffusion of treatment in
communities of people dependent on opiates [9]
[9] Fox, A. D., Shah, P. A., Sohler, N. L., Lopez, C. M., Starrels, J. L., & Cunningham, C. O. (2014). I
Heard About It from a Friend: Assessing Interest in Buprenorphine Treatment.
Substance Abuse, 35(1), 74–79.
24. Demonstration of Model Application
Two policies considered
• Changing prescribing caps
– 100 - 1000 (high)
– 30 - 500 (low)
• Removing barriers for waivered, not prescribing
providers
– Percentage not prescribing 44%, 40%, 30%, 20%, 10%, 0
25. Simulation: Outcome Metrics
Preliminary outcomes presented (4 years)
• Number of buprenorphine recipients
• Number patients who sought but didn’t get
treatment
• Mortality of those waiting for treatment
• Percentage of patients who received treatment
when seeking
• Treatment seekers too far from a provider
• Treatment seekers ever waitlisted
27. Research Questions: Policy Exploration
What policies might improve treatment
capacity and accessibility?
• Changing physician patient caps
• Reducing barriers to offering treatment
–Widespread adoption of demonstrated care
delivery models, such as:
» VT hub and spoke, MA NCM, NM ECHO
• APN/PA prescribing authority under DATA
2000
30. Discussion: Changing Caps
No statistically significant change in outcome metrics
Due to model assumptions
• Most providers are not willing to treat many
patients, even specialists
• A small number of specialist providers are willing
to treat many patients
May have local impact on specific communities not
detected by average metrics
38. Discussion: Reduce Non-prescribing
Statistically significant improvement in access
No significant impact on rural patients
• New providers practice where others do, cities.
Do other providers serve patients in HPSAs?
39. Research Plan: Nuanced Model
Add patient heterogeneity and impact on treatment
outcomes
Model diversion and public health outcomes
• Evaluate policies on most good/least harm
Model costs of treatment vs no-treatment
Model barriers to care
Model validation against empirical data, expert face
validity
40. Research Plan: Nuanced Policy Exploration
Under what assumptions could APN/PA prescribing
affect treatment capacity and address equitable
access?
How might good care delivery models address
provider barriers?
How are costs assessed and to whom to they
accrue? (Can we make the business case?)
What other public health concerns arise—diversion,
accidental exposure?
41. Research Plan: Validation of Abstract
Geography with Actual Geography
Using real geographic region, census data
Using representative survey data to make a synthetic
population of potential patients
Using actual data on prescriber locations and
specialties
Compare specific regional model with general
abstract model