Artificial Intelligence Enables Clinical Transformation
Artificial Intelligence Enables Clinical Transformation
Technology Has Reached a Point Where Change Is Possible
What’s happened?
What’s going to happen?
What should we do about it?
Big Data and Machine Learning in the Cloud
Teradata
Care Management
Cloud IngestionMetadata Catalog Security
AI & Machine Learning
Data-Driven Innovation
Integrated data sources for analysis (claims, pharmacy benefits, labs, social determinants of health, EMR, real-time ADT)
Data-driven Population Health strategies(using advanced datasets Blue Cross has deployed clinical programs to chronic condition management)
Predictive models (risk of hospitalization, readmission, emergency department visit)
High-touch clinical engagement (driven by artificial intelligence and advanced data sets)
Rigorous outcomes evaluation(what works, what to scale)
AI
Clinical Analytics Challenge
Framework for Assessing Value
Population Segmentation Model
Select segment(s)to address
Hypothesize root cause(s)
Playbook of known successes
New interventions (internal and external) to address root cause
Decide on “Plays” to run
Run play and monitor results
Playbook
Select lead, population, timing, success metrics
and control group
Success metric achieved
Success metric not achieved
Bring to scale
Add to Playbook
Pull from playbook as
needed
Identify problems to solve
• Long-Term Effect of Disease Management Program
• Diabetes Prevention Program — Prediabetes Intervention
• Quantitative Financial Analysis of Ochsner Digital Hypertension Program
• Office of Group Benefits Obesity Prevalence and Healthcare Utilization Pattern
• Zero Dollar Copay 2.1 Program Evaluation Expansion
• High-Cost Intervention Program Evaluation
• Pharmacy Carve-in/Carve-out Analysis
• Impact of High-Deductible Benefits on Member Long Term Health Outcomes
• Relationship of PCP Visit Cost-Sharing and Future Healthcare Expenditure
• Low-Value (Wasteful) Service Utilizer Impact on Member Future Healthcare Spend
• Bariatric Surgery Impact on Patient Future Healthcare Cost and Health Outcome
• Quality Blue Value Partnership Program Effectiveness on Lowering Member Total Cost of Care
• Quality Blue Primary Care Program Evaluation
• “Soundbites” Nurse Effectiveness on Improving Patient Engagement
• JIVA Nurse Notes – Using Natural Language Processing
• Near Real-Time Readmission Predictive Model –Using Authorization and ADT data feed
• Machine Learning of Medicare Advantage Medication Adherence Compliance Profile and Prediction
• First Trimester High-Risk Pregnancy Identification and Prediction
Quality, Cost Initiatives
Reduce Financial BarriersZero Dollar Copay, an incentive-based medication program
• Includes members within Blue Cross Disease Management program, with Blue Cross pharmacy benefit manager and have copay plan
• 110+ drugs from 34 therapeutic classes covering: DiabetesHeart diseaseHypertensionLung conditions (e.g. asthma)Mental health conditions
• Offers $0 copays for generic drugs
12
Increasing Medication AdherencePortion Days Covered (PDC) of Top Prescribed Medications
13
Zero Dollar Copay Disease Management ProgramReducing Healthcare Use and Costs
Healthcare Utilization Per 1,000 Members
Per Member Per Month (PMPM) Expenditures
• Members in the program had improved adherence rate compared to control group
• Members in low-income class improved more in medication adherence
• Scaled to 40,000 members; up to 72,000 in July
Net savings: -$72 + $41 = -$31 total PMPM-$372 total PMPY
Rx PMPM: +$41
Medical PMPM: -$72
Zero Dollar Copay: Program Results
Clinical Data Exchange Framework
Member EligibilityChecking
MSG or RecordStructure Edit
CDA
ADT MSG
Vendor Files
FHIR MSG
Data Normalization
Enterprise Master Patient Index
Enterprise Data Governance, Privacy & Security
Eligibility Product
Longitudinal Member Record
Claims
Field Edit
EDW_CDR
Data Lake
Enterprise Data Warehouse
Bulk Message Storage
ProviderNotification
Member & ProviderPortals
Analytics & Reporting
Provider
Terminology Mapping
ADT LAB
Blue Advantage
Commercial Clinical
Lab Files
DuplicateChecking In
tern
al C
onsu
mpt
ion
Data Formats
HL7 ORU
HL7 ADT
HL7 CCDA
HL7 FHIR
Custom
External Partnerships
Data Quality
Clinical Partnerships
Utilization MGMT
Advanced AnalyticsAP
I
Care MGMT
Keys to Success Data quality
Data diversity
Low latency
Call to action
Business integration
Lots of variables
Outcomes evaluation
Diverse AI algorithms
Scalable infra-
structure
Process redesign
Predictive Models Driving Clinical OutcomesRisk of
Hospitalization (ROH)
Risk of Emergency
Department Visit (ROE)
Risk of Readmission
Prediction of High-Cost Claimants
Customer Service
Complaints
Description Hospitalization in the next 6 months
Emergency department visit in the next 6 months
Readmission in the next 1 month
Future high-cost pool ($50,000+) in the next 12 months
Complaint to Customer Servicein next 1 month
Results
1.5x more likely to predict an unplanned admission than commercial model in high-risk group
1.35x more likely to predict emergency department visit than commercial model in high-risk group
Matching a daily authorizations-derived hospital census file with riskscore to notifyproviders
Top 1% generate 17% of cost
Between 200 and 350 members identified as high risk each week
PPV in High Risk Group(Top 1,000)
48% 75% 45% 75% 70%
Blue Cross’ Risk of Hospitalization (ROH) Outperforms Best Commercial Models
Important Predictors• Inpatient hospitalizations in last 2 years• Radiology services in recent 6 months• Ambulance services in recent 6 months• Cardiovascular procedures in recent year• Diabetes episode in last 2 years• # of members in family unit
High-Risk Segment Profiles• Rural parishes: East Carroll, Union,
Vernon, Washington• Low-income blocks
Average Cost perAdmission$15,000
# Members w/ Admit900
ROH PPV45%
Referred Annually2,000
X
Potential Impactable Cost
$13.5 Million=
=
X
Risk of Readmission• Event-level risk score runs Monday through Friday• Key predictors: admit diagnosis, discharge diagnosis, length of stay, service type, discharge
disposition, general risk score
Model re-calibrated for better performance
High Risk3.8% of total dischargesPPV 29.5 %
Medium Risk30.4% of total
dischargesPPV 15.6%
Low Risk65.8% of total discharges
PPV 5.2%
30-day readmission rate was 9.25% during study period
Blue Cross’ Customer Service Model Results
• Intention to Treat analysis was performed with logistic regression (Odds ratio 0.36, p value<0.05)
• 4,500 members identified as high risk
0Complaints after two-week mark when members receive outreach
64% Members who
are reached are 64% less likely to
complain than control
3xRate that control
group complained more than
treatment group
Two predictive models used to manage clinical needs of large commercial groupAn aggressive campaign using artificial intelligence and innovative clinical approaches to better support and manage the group’s members with three chronic conditions was deployed in late 2017
CAD
CHF
Diabetes
• Historically, this group has 19% of its members enrolled in Care Management programs related to these conditions
• Models were used to identify members with the highest opportunity for change; they focused on a much more refined 0.4% of the population
Predictive Models in Action: A Case Study
About 1% of the population referred to Care Management through Risk of Hospitalizationand Risk of EmergencyDepartment Visit models
18-month referral period
Engagement rate 4x better
Successfully Reduced Adverse Events
1,013
983
Pre 12 months Post 12 months
Number of AdmissionsEngaged Members
30
Calculation: (Admits/member month)*12 months*Members
288
246
Pre 6 months Post 6 months
Number of ED VisitsEngaged Members
42
Calculation: (ER Visits/member month)*6 months*Members
Real-Time Actionable Data
ADT (Admit, Discharge, Transmissions) and Re-Admission feeds are transmitting critical patient demographic information in real time.
• Helps providers stay on top of daily hospital utilizations • Helps providers see which patients are admitted/discharged from a facility• Occurs within 24 hours of the event
Blue Cross Care Management teams now can reach out to these patients via phone to monitor post-discharge behaviors and make sure members schedule follow-up office visits as needed
Enhanced Disease Management• Blue Cross offers state-of-the-art baseline Disease Management services,
plus enhanced services
• Programs focus on members with or at risk for certain chronic diseases
o Focuses on improving health outcomes for members with chronic conditionso Covers heart disease, chronic kidney disease, diabetes, respiratory (COPD and asthma)o Facilitated by Blue Cross nurses via telephone and community-based
• All programs have validated clinical outcomes and cost savings
Program Results: Financial Impacts
-$76
$74
-$150In Program Not in Program DiD
Average Medical PMPM Trends*
*Based on 36-month study of Blue Cross Disease Management program
In Program Not in Program DiD
*Based on 36-month study of Blue Cross Disease Management program
Enhanced Disease Management ResultsProgram Effects on Healthcare Utilization
Members in Blue Cross program:
3.32% fewer emergency department visits
1.53% fewer inpatient admissions
2.13% shorter length of inpatient stays
1.19% more prescriptions
Controlled HbA1c and blood pressure
FewerED visits
Fewerinpatient
admissions/admit days
ControlledHbA1c and
blood pressure
Moreprescriptions
• Medical PMPM trend differently depending on participation in program• Timing of separation points of the trend may reflect general acuity of
condition and its responsiveness to engagement.
Shorter time of response
Significant deterioration
CHD/Hypertension
Longer time of response
Diabetes
Diabetes and CHD/Hypertension Programs
• HbA1c showed increasing trend for those not in diabetes program
• CHD/Hypertension program participants had slightly decreasing trend in Systolic BP values
Clinical Outcomes by Participation
HbA1c
Systolic BP
Our In-House Clinical Team• 200+ clinical professionals:
doctors, nurses, dietitians, social workers, pharmacists
• Provide health coaching and support to members dealing with serious illnesses, chronic conditions or acute injuries
• Identify and close gaps in care, coordinate services
• Guide care transformation with providers in our Quality Blue programs
Community Nurse Program • Successful pilot over past two years• Works with designated healthcare
providers in a regional market• Engages members face-to-face using
AI and real-time data • Helps meet high-risk members’
needs, coordinate care, lower risks of readmission
• Improves health outcomes and helps to lower costs
Community Nurse Program: Results
Better engagement, better health outcomes
3.26 times higher engagement with community nurses vs. telephonic nurses only
Model now expanding to other areas in state
14.7%
62.6%
47.9%
Telephonic Nurse Community Nurse Difference
Member Engagement Rate by Outreach
Impactability Over RiskEvaluating which members will benefit the most from which interventions
Member A
Enroll in Care Management and offer six weeks of meal assistance
Engage via mail or email
Risk score
Impactability scorePlan action
High
Low
High
• Chronic kidney disease
• Low credit score
• Has called plan call center• Four family members• Age <80• 4 gaps in care
• On dialysis• In hospice care• Age >80• 0 gaps in careMember B
Key Takeaways• We discover, validate and scale interventions to improve and maintain the health
of those who would otherwise not seek care but who are at risk for deteriorating health or adverse events.
• Drive effective coordination, based on actionable data insights, internally between the Blue Cross care team and clinical team and externally with primary care physicians, specialists and facilities to enhance patient clinical outcomes and lower the total cost of care.
• To create effective programs, AI/ML-based targeting must be combined with robust outcomes evaluation.
• Compare models to industry benchmarks and standards to continually improve performance.
Research• Journal of Medical Economics• American Journal of Managed Care• Value in Health (ISPOR 2018 and 2019)• CDAO Exchange, August 2019• AHIP Institute and Expo, June 2019• Blue Cross and Blue Shield Association
National Summit, May 2019
Somesh Nigam, Ph.D.Chief Analytics and Data Officer