decision support algorithms with communimetrics
TRANSCRIPT
Decision Support Algorithms with Communimetrics
JOHN S. LYONS, PHD and APRIL D. FERNANDO, PHD
Center for Innovation in Population Health, University of Kentucky
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
CommunimetricsBackground and Philosophy
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
The Story of HelpingThings happen in people’s live and sometimes these events lead someone to believe that receiving care from a professional might be helpful.
Adults recognize that care might help. Individuals with limited means must rely on public systems for support.
This is where the story in the system begins …
Possible referral sources for adults:• Internal recognition of value• Family and Friends• Other Medical Professional
Possible referral sources for children:• Parents/Guardian• School staff or Teacher• Other Medical Professional
Understanding the Business of Helping: The Hierarchy of Offerings
Gilmore & Pine, 1997
Products
.
Produced for a retail market
Sushi
Transformations
Helping people change in some notable way.
Teaching people how to fish
Experiences
Purchasing a memory.
Sport FishingCommodities
Raw Materials
Fish
Services
Having someone apply a product for you.
Seafood Restaurant
Communimetrics: Understanding a Person’s Story
The individual shares their story.
The care provider listens to their
story.
The other story tellers share their
perspectives.
The stories are combined and a
single narrative is agreed upon.
01
02
03
04
Person-Centered Care Requires New Metrics
Clinimetric CommunimetricClassicalTest
ItemResponse
1 2 3 4 5 6 7 8
\
History of Measurement Theories
Communimetrics
§ Native Naturalism (Reality Theory) rather than British Empiricism
§ Non-arbitrary—every number has immediate meaning
§ Culturally and developmentally informed—the measure of a story
§ Based on qualitative approaches to synthesizing complex phenomenon—modified grounded theory
§ Post triangulation rather than pre-triangulation measurement
Scenario 1: Youth is distressed and the parent is minimizing the situation. With treatment the youth feels better and the parents come to realize the youth’s mental heath needs
0123456789
10
Catastrophizing Youth Minimizing Parent
AdmitTransition
Scenario 2. Parent is catastrophizing and youth is minimizing. With treatment the youth understand his her mental health needs better and the parent sees progress
0123456789
10
Minimizing Youth Catastrophizing parent
AdmitTransition
The problem with means of single perspectives—the average of two clinically successful treatment episodes equates to no effect
0
1
2
3
4
5
6
Youth Perspective Parent Perspective
AdmitTransition
01 Items are selected because they are relevant to service/case planning.
02 Each item uses a 4-item rating scale that translates into action.
03Rating should describe child/youth, not the child/youth in services.
6 Key Principles
04Consider culture and development before determining ratings.
05The ratings are agnostic as to etiology; it’s about the What, not the Why.
06Use a 30-day window in considering what is relevant to children, youth and their families.
RELEVANCE
ACTION LEVELS
CLIENT FOCUS
CULTURE & DEVELOPMENTTHE “WHAT”
30-DAY WINDOW
Action LevelsS t a t u s a n d I m p a c t
3
2
1
0No evidence of need. No need for action.
Significant history of need, or possible need that is not interfering with functioning.
Watchful waiting, additional assessment.
Need interferes with functioning. Action/intervention required.
Need is dangerous or disabling.Immediate/intensive action required.
Well developed centerpiece strength. Easily accessible by individual; essential for planning.
Useful strength. Evident and can be accessed by individual; useful for planning.
Strength identified. Requires building in order to be useful for individual or planning.
No strength identified. Considerable effort or building to create and develop strength.
Individuals & Family Program System
Decision SupportCare Planning
Effective PracticesEBPs
Outcomes Management
Quality Improvement
ROW 6
TCOM Grid of Tactics
EligibilityStep-Down
Resource ManagementRight-Sizing
Provider Transitions & Celebrations
ProgramEvaluation
Provider Profiles Performance/
Contracting
Case Management Integrated Care
Supervision
CQI/QAAccreditation
Program Redesign
TransformationBusiness Model
Design
Decision Support and CommunimetricsA Person-Centered Approach
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Decision Making and Decision Support
Decision Support Uses of CommunimetricMeasuresCommunimetrics is actively used in decision support within child welfare settings in the United States for each of the following key decisions. Many of these applications have been operations for years; some for more than a decade.
o Placement type and intensity
o Level of care
o Case management intensity
o Case rates
o Service Packages
o Evidence-based Treatments
o Safety and Risk
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATIONTo date, not all child serving systems using the CANS have implemented decision support algorithms.
Person-Centered Decision Support• Some approaches to level of care use service receipt or history to inform
level of care.• These approaches make the assumption that all such decisions are
perfectly indicated clinically.
• We know that, in reality, all sorts of factors actually influence service receipt in reality including availability, racial and cultural factors, etc.
• Thus using service receipt or history in decision support simply institutionalizes these biases into ongoing decisions.
• Compliance is simply an indirect indicator of prior service receipt.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
https://zoom.us/rec/play/-qGYol1YAYAawVU_ANCXsOFsOZjZfz2MtsvezTeKZLCl4w4FTH4m7-ogh2vJdIy7KYWM6IUEjNUo_jfE.--7iDI7T-OiNFlpi?continueMode=true&_x_zm_rtaid=G5g83OCeQ_SRinr0uFCL3g.1599071050241.5b24fc6ff52d727a8fda6a6e8b674c5e&_x_zm_rhtaid=757
Percent of hospital admissions that were low risk by racial group Adapted from Rawal, et al, 2003
0%5%
10%15%20%25%30%35%40%45%50%
1998 1999 2000 2001 2002
% of
Low
Risk
Adm
ission
s White
AfricanAmerican
Hispanic
Most existing measures use total scores with cut-offs for decision support applications. These cut-off approaches can be problematic at the margins because very small differences in a total score can lead to very different decisions.
In Larry P v. Riles (1984), for example, norm-based decision support using IQ in schools was deemed discriminatory in CA.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Communimetric algorithms and other measures’ approaches to decision support
Because of the item level reliability and action level format, patterns of actionable needs are used to generate algorithms using boolean (branching logic) models.
These models are far more intuitive clinically and therefore more defensible as they are easy to describe and the differences between individuals at different levels are always meaningful.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Communimetric algorithms and other measures’ approaches to decision support
Decision Making as a Consensus Process
Customization of Algorithms• Unlike most existing measures, the algorithms use a flexible TCOM framework to think
about decision support rather than an ‘off-the-shelf’ logic that applies the same standards everywhere.
• In reality different systems are different and to generate meaningful change you have to start where the system is currently functioning.
• For example, the original algorithm for residential treatment placement in Illinois was VERY low. This still resulted in a 30% decline in residential placements. Once the system was effectively evolved, then the algorithm could be adjusted to reflect the evolving system’s performance.
• As such, the approach allows for the customization of decision models by jurisdiction (county) to reflect different cultural contexts.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
The Utility of Communimetric Decision Support AlgorithmsEach of the following States have successfully used communimetricdecision support algorithms for LOC decisions in Child Welfare for more than a decade. There have been no problems and notable system improvement:
◦ Indiana
◦ New Jersey
◦ New York
◦ Wisconsin
◦ Tennessee
A number of other states and counties have used and continue to use communimetric decision support algorithms for shorter periods of time.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Statewide CANS Algorithm ExamplesNew Jersey, I l l inois and Tennessee
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
New Jersey’s System of CareThe f irst statewide implementation of a cross-systems
(comprehensive) version of the CANS
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
In New Jersey• Despite the number of children and youth served in their system of
care tripling over the past decade [Figure A]
• The actual number of youth placed in residential treatment has been reduced by over one third during the same period [Figure B]
• CANS algorithms were used to support this process and research using the CANS informed the design of the step down process
• At the same time nearly 1/3 of detention centers have been closed along with all state hospitals for children and youth.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Figure A. New Jersey’s Children’s System of Care Expansion: 2008-2016
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Figure B. New Jersey Children’s System of Care:
Number of Youth in Residential Care (2010-2016)
Illinois Department of Children and Family ServicesThe state with the f irst use of a CANS Algorithm applied within a
Team Decision Making process
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
In a series of published studies, Brian Chor and his colleagues studied the Illinois CANS decision support algorithm as used in their team decision making process. They found…
• Following the CANS level of care recommendation predicted greater clinical improvement as compared to not adhering to the recommendation. Chor, et al (2012) Children and Youth Services Review
• Serving children below the CANS recommended level of care resulted in less improvement in functioning, and serving them at a higher than recommended level of care resulted in reduced rates of improvement as compared to adhering to the CANS recommendation. Chor, et al (2014) Administration and Policy in Mental Health
• The CANS algorithms out-performed the placement decisions made by the child-family team Chor, et al (2013) Child Abuse and Neglect
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Communimetrics (Lyons, 2009)• Following the placement recommendations of the CANS decision
support algorithm within a Team Decision Making process was associated with more stable placements over time. [Figure C]
• Placing children below the CANS recommended level of care placement results in the second most stable placement history.
• Placing children in a higher than recommended level of care was associated with dramatically less stable placements over time.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Figure C. Survival analysis of time to placement disruption for children/youth whose placement matches CANS recommendations (Match= 0, green), those whose placed is at a lower intensity than recommended (match= -1, blue) and those whose placement is more intensive than recommended (match= 1, brown).
Tennessee Department of Children’s ServicesThe f irst state in which case workers completed the CANS
themselves
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Figure D. In an independent replication, Epstein, et al also found that following a CANS algorithm was associated with more stable placements than not following that algorithm on the initial placement in child welfare. Epstein, et al (2015) Residential Treatment for Children and Youth
0.00
0.25
0.50
0.75
1.00
0 100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
Time (Days)
Prop
ortio
n of
Und
isrup
ted
Reco
rds
Strata A
Strata B
4336 1848 944 580 381 253 176 124 81 57 40 22 13 6 2 1 1 0 0 0
9604 5132 2628 1683 1111 759 537 379 264 176 119 86 56 41 27 20 16 8 5 1Strata B
Strata A
Numbers at risk
Tennessee implemented the CANS as a result of a law suit
• The TN DCS implementation was the first to train case workers to complete the CANS. Training and support was provided to case workers to help them build the necessary skill set through Vanderbilt University’s Center for Excellence.
• TN State leadership credits the use of the CANS as part of the reason that they are now out from under this lawsuit as they went from being ranked as one of the least effective systems to one of the more effective systems during the decade after the CANS was fully implemented.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
In Summary• The CANS is widely used in decision support applications as described
above including Placement and Level of Care (and case rates/service packages).
• The CANS has been successfully used by a number of jurisdictions for more than a decade without incident and with sustained successful system change.
• Although many have been tried, no other cross-sector decision support approach has experienced this level of sustained success in child welfare and behavioral health.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Decision Support AlgorithmDevelopment Process
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Decision Support Algorithm Development• A description of the jurisdiction’s Level of Care structure and criteria is provided to
Praed.
• Identify youth in the current jurisdiction’s Level of Care structure and review their current CANS ratings (if available).
• Jurisdiction establishes the decision options – define each level and provide any child specific information about each.
• Items from CANS mapped to the jurisdiction’s defined levels and new algorithm is developed.
• Jurisdiction identifies a Panel of Experts to review the new algorithm recommendations. The panel of experts, along with Praed convene to discuss the initial algorithm recommendations.
TCOMCONVERSATIONS.ORG | @PRAEDFOUNDATION
Algorithm Variables Non-Algorithm Variables
Acuity # of Assessments
Low 7890
Med 2438
High 5093
Plots for Low, Med, High Acuity by Variable Type