Far too often analytics efforts have fallen short of making a tangible impact on outcomes because they haven’t been successfully implemented in real workflows. Predictive models remain at risk of becoming isolated in their use along the continuum of care where their integration may provide benefits larger than the sum of each silo. To combat this, UnityPoint Health (UPH) focused on integrating analytical models within the same readmission reduction strategy and coaching the care team to facilitate their adoption. Using this approach, one of UPH hospital’s risk-adjusted readmission indexes improved 40 percent over three years, surpassing internal system targets in performance and becoming the top performer in the health system. Learning Objectives: - Describe applicable predictive models useful in reducing 30-day readmissions. - Learn the elements of a successful readmissions reduction strategy in an integrated health system. - Understand common obstacles faced in the adoption of analytical tools and how to overcome them. View this webinar to gain knowledge of the analytics tools and methods UPH used, including innovative individualized risk heat-maps generated for each patient, strategies for analytics adoption, and lessons learned along the way.

1. Leveraging Predictive Models to
Reduce Readmissions
March 6, 2019
Rhiannon Harms
Executive Director, Strategic Analytics
UnityPoint Health
Ben Cleveland
Data Scientist
UnityPoint Health

2. © 2018
Health
Catalyst
What is your functional role?
1. Clinical — 16%
2. IT — 36%
3. Operational — 36%
4. Financial — 12%
Poll Question #1
2

3. Agenda
#4: The next time you have a melt down,
literally melt down. It is very impressive.
• Introduction
• Our Readmission Story
• Tools & Technologies
• Additional Opportunities
• Q&A

4. © 2018
Health
Catalyst
Learning Objectives
4
Describe applicable
predictive models useful
in reducing 30-day
readmissions.
Describe elements of a
successful readmissions
reduction strategy in an
integrated health system.
Describe common
obstacles faced in the
adoption of analytical tools
and how to overcome
them.

5. Agenda
#4: The next time you have a melt down,
literally melt down. It is very impressive.
• Introduction
• Our Readmission Story
• Tools & Technologies
• Additional Opportunities
• Q&A

6. © 2018
Health
Catalyst
6
UnityPoint Health in Nine Regions

7. © 2018
Health
Catalyst
Our Purpose:
To lead UPH’s quest to become a data-driven organization by:
• Embedding analytics into the strategic planning process,
• Enabling clinicians and business leaders to use data to make decisions,
• Identifying areas of opportunity to improve patient care, and
• Developing models used to predict population health and financial trends.
Analytics at UnityPoint Health (UPH)
7
We serve as problem-solving partners with a
focus on delivering the most valuable solution.

8. © 2018
Health
Catalyst
Our Team Capabilities
8
UPH Analytics teams provide three broad reporting and analytics capabilities:
• Key performance indicators
(KPIs)
• Standardized, on-going reports,
broadly accessed with self-
enabled interpretation
• Highly involved setup
• Advanced analytics
• Statistical methodologies
• Multivariate
• Highly intensive
• Highly consultative
• Answering unanticipated
questions
• Exploratory, not standard
reporting
• Majority of analytics team effort
• Not “data-dumping”
• Highly consultative
SCORECARDS &
DASHBOARDS
EXPLORATORY
ANALYSIS
PREDICTIVE
MODELING

9. © 2018
Health
Catalyst
Before Leveraging Predictive Solutions
Build on Current Scorecards and Reports
9
Day 1-30 Readmissions
by Service Line
Readmissions by Day of
Week & Time of Day
Readmissions by
Diagnosis/Product
• Daily retrospective reporting.
• Monthly risk-adjustments and
claims views
• By primary care provider
• By attending provider
• By discharge disposition
• By nursing unit
• Patient-level details as needed
Additional Views Include:

10. © 2018
Health
Catalyst
On a scale of 1-5, how effective is your organization in avoiding
preventable readmissions?
1. Not at all effective — 11%
2. Somewhat effective — 32%
3. Moderately effective — 46%
4. Very effective — 9%
5. Extremely effective —2%
Poll Question #2
10

11. © 2018
Health
Catalyst
Understanding What Leads to Readmission
11
What led to the readmission?
• Investigate caregiver concerns
• Check attendance at follow-up appointments
• Look for medication issues
• Ask the patient directly
Patient Readmitted
Daily
Readmissions
Report
Readmission
Interview with Patient
Readmission
Worksheet
Completed
Unit Daily Huddle
Daily
Readmissions
Meeting
Post Discharge
Planning Using New
Info
Patient
Discharged

12. © 2018
Health
Catalyst
Initiative Background
With 250,000 members in
value-based contracts, UPH
Analytics was commissioned
to build predictive analytics
tools that support the
transition to value-based care
and leverage the clinical
expertise from over 2,000
employed providers.
12

13. © 2018
Health
Catalyst
Project Scope
13
1. Improve model performance from
current industry standards like LACE
and HOSPITAL.
2. Predict not only which patients are at
risk of being readmitted, but also the
dates of highest risk.
3. Deliver the predictions in near real-time
in an automated, easy-to-understand,
cross-continuum tool.

14. © 2018
Health
Catalyst
Empowering the Readmissions Journey
14
2014
2015
2016
Re-established
readmissions focus.
Which patients are readmitting?
Why are patients readmitting?
Use data to identify groups
at risk for readmission.
Multi-disciplinary team involvement.
Proactive and routine use of data.
Moving from reactive to proactive readmissions prevention.
2013
Analytics and process
optimization.
2017

15. © 2018
Health
Catalyst
Continuing Deployment Across the
Care Continuum
15
16.1%
15.1%
14.6%
14.8%
15.0%
15.2%
15.4%
15.6%
15.8%
16.0%
16.2%
Apr-17 May-17 Jun-17 Jul-17 Aug-17 Sep-17 Oct-17 Nov-17 Dec-17 Jan-18 Feb-18 Mar-18 Apr-18 May-18
Home Care
Deployment Begins

16. © 2018
Health
Catalyst
Leveraging a Multi-Disciplinary Approach
16
Readmission Risk Tool
(RN Managers Identify High-Risk Patients)
Daily Huddle
(Unit Team)
Daily Readmissions Meeting
(Discuss Readmitted Patients)
Ongoing Data Analysis
(Readmissions Dashboard)
PCP, Home Health & SNF
(Data, Outreach, Partnerships)
Case Management & Chronic Disease Navigators
(Post-Discharge Planning & Follow-Up)

17. Agenda
#4: The next time you have a melt down,
literally melt down. It is very impressive.
• Introduction
• Our Readmission Story
• Tools & Technologies
• Additional Opportunities
• Q&A

18. © 2018
Health
Catalyst
Consider the Triggers for a New
Analytics Solution
18

19. © 2018
Health
Catalyst
The Big Three
19
Which patients do
we focus on?
What do we
do?
When do we do it?

20. © 2018
Health
Catalyst
Which patients do
we focus on?
What do
we do?
When do we do
it?
The Big Three
20

21. © 2018
Health
Catalyst
Solution
21

22. © 2018
Health
Catalyst
Cascading the Same Predictions to the
Post Acute Care Team
Patient Look-Up & Summary
Day-by-Day Risk View
Scheduled Appointment View
22

23. © 2018
Health
Catalyst
Readmission Risk Variables Span Four
Key Domains
23
Healthcare
Utilization
Diagnosis
History
Visit Specifics
(Vitals, Labs,
Meds)
Social
Determinants
of Health

24. © 2018
Health
Catalyst
UPH Readmission Risk Variables
24
Healthcare Utilization
• # of Appointments*
• Number of Late Appointment
Arrivals*
• # of Providers
• % of Appointment No Shows
• # of ED Visits
• # of Inpatient Visits
Social Determinants
of Health
• Age*
• Insurance Type*
• BMI*
• Marital Status
• Sexually Active
• Hispanic
• Tobacco Use
• Illegal Drug Use
• Female Partner
• Male Partner
Diagnosis History
• # of Chronic Conditions
• Anxiety
• Cardiovascular Disease
• Chronic Kidney Disease
• Dementia
• Depression
• Diabetes
• Heart Failure
• Hypertension
• Mood Disorder
• Obesity
• Pain
• Permanent Mental Disability
* High Relative Variable Importance

25. © 2018
Health
Catalyst
UPH Visit Specific Readmission Risk Variables
25
Visit Specific:
Vitals
• Diastolic BP*
• Systolic BP*
• Pulse*
• Respirations*
• Weight*
• Height.
• Temperature.
• Patient Pain Score.
Visit Specific:
Meds
• Anesthetics*
• Biologicals*
• Cardio Agents*
• Gastro Meds*
• Nutrition Meds*
• Topical Meds*
• Endocrine Meds.
• GU Meds.
• Hematology Meds.
• Infective Meds.
• Misc. Meds.
• Neoplastic Meds.
• Neuro Meds.
• Resp Meds.
Visit Specific:
Labs
• Calcium*
• Creatinine*
• Glucose*
• Hemoglobin*
• Potassium*
• Sodium*
• ABG.
• WBC Count.
Visit Specific:
Other
• Admit Type*
• LACE Score*
• LOS*
• Surgery*
• Chest Tube.
• Ventilator.
• Weekday.
• Month.
* High Relative Variable Importance

26. © 2018
Health
Catalyst
On a scale of 1-5, how effective is your organization in addressing the
social determinants of health of the patients and communities you
serve?
1. Not at all effective — 22%
2. Somewhat effective — 46%
3. Moderately effective — 27%
4. Very effective — 3%
5. Extremely effective — 1%
Poll Question #3
26

27. © 2018
Health
Catalyst
Experience with Social Determinants
27
Our internal findings are consistent with recent literature regarding the predictive
value of social determinants of health:
When considered along with clinical data (diagnoses, meds, labs, utilization hx),
SDoH in their current form are not significant predictors of readmission risk.
Logue, Everett, William Smucker, and Christine Regan. "Admission Data Predict High Hospital Readmission Risk." Journal of the American Board of Family Medicine 29 (2016): 50-59.
Jamei M, Nisnevich A, Wetchler E, Sudat S, Liu E (2017) Predicting all-cause risk of 30-day hospital readmission using artificial neural networks. PLoS ONE 12(7): e0181173.

28. © 2018
Health
Catalyst
Readmission Model Performance
Regional Performance:
• Area under the curve (AUC) ranges from
0.75-0.81
• Brier Score: 0.06-0.09
Performance in Literature:
• AUC: 0.7-0.82 (Kansagara, et al. 2011)
• Brier Score: 0.05-0.1
Kansagara, D., Englander, H., Salanitro, A., Kagen, D., Theobald, C., Freeman, M., & Kripalani, S. (2011, October 19). Risk Prediction
Models for Hospital Readmission: A Systematic Review. JAMA, 306(15), 1688- 1698.28

29. © 2018
Health
Catalyst
Heat Map Validation
29

30. © 2018
Health
Catalyst
System-Wide Utilization Continues to Grow
0
1000
2000
3000
4000
Feb
2017
Mar
2017
Apr
2017
May
2017
Jun
2017
Jul
2017
Aug
2017
Sep
2017
Oct
2017
Nov
2017
Dec
2017
Jan
2018
Feb
2018
Mar
2018
Apr
2018
May
2018
Over the past year, log-ins by care
team members to the Readmission
Risk Tool have doubled every three
months and are now in excess of
130 log-ins per day.
4,100 log-ins by
222 unique team
members in May.
30

31. Agenda
#4: The next time you have a melt down,
literally melt down. It is very impressive.
• Introduction
• Our Readmission Story
• Tools & Technologies
• Additional Opportunities
• Q&A

32. © 2018
Health
Catalyst
Integration with Other Care Coordination Tools
32

33. © 2018
Health
Catalyst
Testing After Deployment
• If the care team acts on the prediction, your new training data will be biased.
• If we successfully prevent high-risk patients from readmitting consistently, theoretically,
patients with similar profiles would not be high-risk in future models.
Development Data Test Data Biased!
2011 2015 2016 2017
UPH Inpatient Encounters
33

34. © 2018
Health
Catalyst
Lessons Learned
Risk scores trigger work
lists, risk visualizations
trigger conversations.
Conditions for adoption
include: the right use
case, staff involvement
in build, ease of
workflow, coaching on
interpretation, and peer
success.
Be mindful of silo
development efforts
involving pop health ‘risk’
– these can be
contradictory and may
create confusion as they
flow downstream to care.
managers.
When identifying
machine learning use
cases, focus on
uncertain “decision
points.”
34

35. Q&A

36. Thank You!