rm world 2014: datamet risk analytics engine
DESCRIPTION
TRANSCRIPT
DATAMET RISK ANALYTICS ENGINE
Kleber GallardoCEO, Alivia Technology
8/20/2014
• , 8/16/2014: “Fraud and systematic overcharging are estimated at roughly $60 billion, or 10 percent, of Medicare’s costs every year, but the administration recovered only about $4.3 billion last year. The Centers for Medicare and Medicaid Services, which is responsible for overseeing the effort, manually reviews just three million of the estimated 1.2 billion claims it receives each year. “
• Medicaid waste fraud and abuse estimated 60-80 Billon
• US Healthcare National expenditure was 2.8 Trillion or $8,915 per person in 2012 – US HHS
What’s the Business Problem?
Global Healthcare Costs
Why is it Difficult?
• New Fraud Patterns created all the time
• Identify groups and networks of risky providers and members in the medical claims arena.
• Help investigators to do risk based investigations and focus on the most likely providers and members of fraud, waste and abuse.
• System automatically and continually learns patterns of risk based on known data fraud cases as well as continually learn on new cases as they are identified.
• This enables continuous feedback and improvement of the system and automated detection of risk cases based on these patterns.
How does data mining help?
Implementation Process
Pilot and Prod
Build Analytics Engine
Data Warehouse
Infrastructure
Identify and Access Data
Sources
Hire and Train Team
Collaborative Process
Business and Technology Identify Need
DataMet Risk Analytics Process
Stand AloneData sources
Databases, flat files of
different types
Preprocess dataCollect, clean, Aggregate and
store
Revise/refine Analysis
Data Analyst reviews output
Take action based on findings
Integrated Data Warehouse
Results feedback
Investigatorinterprets
results
Data analysis findings
DataMet Analytics EngineMachine learning,
statistics, KDD, data mining, risk scoring, and
others
DataMet Risk Engine Components• Rapid Miner and R for the analytics
• RapidMiner server for the dashboards, user management, resource sharing and scheduler
• Ontologies – Hierarchies of concepts
– Data Source Management
– Connection between data attributes to concepts
• Rule templates and Process Templates– Connect data with groups or sequences of algorithms with data sources
• DataGrid and Process Grids– Enable high speed access and processing of the data from various sources
• Social Network Visualization to show relationships between entities
• Geo Mapping to show locations of concentrations
• Designed to support Multilingual and Multi Currency capability
DataMet Sample Algorithms
• Benford’s Laws• Outlier Detections• Decision Trees• SVM• Attribute Ranking• Correlation Analysis• Death Match• Fuzzy Matching• Network Relationship Graphs
Risk Matrix
LOW
LOW
LOW
MEDIUM
MEDIUM
MEDIUM
MEDIUM MEDIUM
HIGH HIGH
HIGH
HIGH
HIGH
CRITICAL CRITICAL
CRITICAL
CRITICAL
MEDIUMMEDIUM
HIGH
LOW
MEDIUM
MEDIUM
HIGH
MEDIUM
SevereMajorMarginal Minor Moderate
Almost Certain
Likely
Possible
Unlikely
Rare
FINANCIAL OR SYSTEMIC RISK LEVEL
EVEN
T LI
KEL
IHO
OD
Risk Analytics Process
Apply Algorithms
Calculate Risk Score
Predictive Models
Detail Analysis
Data Driven Risk Assessment
ETL1
2
4
3
6
5
Cleanse
Risk RankedInsights
Output
Data Sources
DataMet
DataMet 21st Century Risk Analysis Process
Data Sources
Captured Resources
Collect Analyze ReportTrack
Results
Data Driven Risk Assessment
Improved ProcessesCollaboration
Management Dashboards
Learning
Tracking And Follow up
Improved Algorithms
Informed Decisions
Pharmacy Claims Example
Filter Narcotics
Calculate Risk Score with Analytics Engine
Filter Paid > $250/Year
Select Members with Risk Score greater than 25
Claims Members Paid 170M 2M 5B
Filter for year 201225M 0.9M $577M
14M 0.7M $541
0.67M 0.16M $9.5M
114K 6K $7.4M
5K 130 $1.4M
1
2
4
3
6
5
Get Paid Claims
• Capture indicators of risk
– Aggregates
– Rates of change
– Ratios
– Daily/Weekly/Monthly/Quarterly/Yearly
Metrics and Granularity
• Homeland Security
• Tax Records
• Health Care Claims
• Bank Transactions,…
Details Analysis of Risk Cases
Risk Analytics and Business
• Risk analytics is not an intuitive process
• It is the exercise of applying rules, algorithms and domain knowledge to data
• The data model is imbedded in the business processes and in the data that it produces
• Subject area expertise is essential to understanding the data and success of the project
• The blend of subject area knowledge and analytic expertise is a powerful combination
• DataMet can effectively identify groups and networks of risky providers
• DataMet helps investigators to focus on the most likely providers and members of fraud, waste and abuse.
• DataMet automatically learns patterns of risk based on known data fraud cases as well as well as new cases .
• DataMet enables continuous feedback and improvement of the system and automated detection of risk cases based on these patterns.
• DataMet is aimed at reducing the global waste and abuse which is estimated to be $600 billion/year US and globally
Conclusions