2nd Global Summit
Healthcare Fraud:
Prevention is better than cure
25-26 October 2012
Beaumont Estate, Old Windsor, UK
The Use of Advanced Analytics
to Prevent Healthcare Fraud
- Fraus omnia corrumpit –
Ted Doyle
Vice President, Fraud Analytics
UnitedHealth Group/OptumInsight
Julie Malida
Principal, Health Care Fraud Solutions
SAS Institute Inc.
2
Ted Doyle
Vice President, Fraud Analytics
UnitedHealth Group/OptumInsight
3
US Healthcare Fraud
National Trends
NHCAA Anti-Fraud
Management Survey
2011 Stats – NHCAA Management Survey
 Average Health Insurance company SIU realized combined fraud
recoveries, savings and prevented losses totaling over
$22.9M/year based upon average budget = $1.95M
 Average SIU staff = 20 FTEs
 Average recoveries = nearly $5.3M
 Average savings = more than $13.8M
 Average prevented losses = almost $7.7M
 Average number of open cases or investigations = 396
 Average number of cases handled by a US based SIU = 936
US Fraud Hot Spots
(Red Font with Star)
Brooklyn &
New York
City
Detroit
Chicago
Los Angeles
Dallas &
Houston
Baton
Rouge
Tampa &
Miami
US Top Ten Fraud Trends
Benefit Type
Fraud Scheme
Prescription Drug Services
“Drug Seeking” patients are doctor-shopping to obtain multiple medically
unnecessary prescriptions, causing benefit payments to increase but also
causing health risks for Payer member populations, which translates to
increased cost for medical care. Insurers lose between $8.6M and $857M
a year depending on plan size.
Ambulance Transportation
to Nowhere
Ambulance and Van services where no other office visit, ER or Inpatient
services provided at same time. In a 2006 OIG report, Medicare was found
to have improperly paid $402M for ambulance services that were not
rendered or medically necessary.
Infusion Therapy (IV
Therapy)
Medicare has identified over $2B in suspect payments for IV Therapy
associated with false AIDS diagnosis between 2002 and 2011. This
represents on average $222M a year.
Medical Identity Theft
With the proliferation of Medical Identity Theft, Payers need to identify
groups of patients who appear to be shared across multiple providers or
provider networks. Medicare identified over 100,000 member IDs
compromised (sold) and over $1B in savings for claims denied associated
with compromised member info between 2002 and 2011, representing on
average $111M a year.
Independent Diagnostic
Testing Facilities
Medicare alone allowed almost $1 billion for IDTF claims for 2.4 million
beneficiaries in 2010
US Top Ten Fraud Trends
Benefit Type
Fraud Scheme
Payments to excluded,
sanctioned or phantom
providers
Medicare allowed close to $41M for medical equipment and supply claims
with invalid, inactive or deceased referring physicians or for services ordered
by non-physicians.
Home Health Services
Medicare spending for Home Health Services has increased 81% since the
year 2000
Spike Billing
Payers need to ID spike billing over a rolling 12-month average but also
month-to-month spikes that don’t make sense based upon peer and/or
geographic trends. No prosecutorial case information or Regulatory reports
have been produced for this trend
Services while Inpatient
Public and Private sector Payers have seen an increase in
suspicious/fraudulent billing for outpatient services while the patient is in a
Facility setting. SIU/Analytic presentations at the annual training
conferences for the US National Health Care Anti-Fraud Association, United
Kingdom Health Insurance Counter Fraud Group and the European Union
Health Care Fraud and Corruption Network addressed this trends as a
significant concerns for new health care fraud.
Cosmetic Services – Dental,
Vision, Medical
Medically unnecessary cosmetic procedures, misrepresented (coded) as
medically necessary procedures. No prosecutorial case information or
Regulatory reports have been produced for this trend.
Comparative Billing Report
• CBR: A report developed in an effort to provide education to the
provider community by comparing billing practices across peer group
• Benefits: Education = behavior change = cost avoidance = savings
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
CBR 001 Physical Therapy Services with the KX Modifier
CBR 002 Chiropractic Services
CBR 003 Ambulance
CBR 004 Hospice
CBR 005 Podiatry
CBR 006 Sleep Study
CBR 007 Ordering Durable Medical Equipment: Spinal Orthotics
CBR 008 Outpatient Physical Therapy Services with the KX Modifier
CBR 009 Ordering Durable Medical Equipment: Diabetic Supplies
CBR 010 Chiropractic Services
CBR 011 Electrodiagnostic
CBR012 Ordering Durable Medical Equipment: Lower Limb Orthotics
CBR013 Advanced Diagnostic Imaging
CBR014 Pain Management Services
CBR015 Cardiology Services
CBR016 Evaluation and Management Services
•
For more info visit: http://www.safeguard-servicesllc.com/cbr/
Investigative Perspective
Why Advanced Analytics?
• Shift the focus from I’ll find the fraud to Show me the fraud.
• Free up SIU Investigators to create and investigate cases
• Discover unknown schemes and aberrant billing patterns
• Protect Payer resources at an enterprise level
• Predict trends and anomalies before to late
Challenge of Improper Claim
Detection
• This space represents the universe of claims
• Manual clinical review is impossible for entire space
• Goal: Stop as many reds (improper) for review as possible while
keeping the number of blues (proper) identified to a minimum
Advanced Analytics are Required
Using hybrid analytics for fraud detection
Enterprise
Data
Employer
Data
For known
patterns
For unknown
patterns
For complex
patterns
For unstructured
data
For associative
linking
Rules
Rules to surface known
fraud behaviors
Anomaly Detection
Algorithms to surface
unusual (out-of-band)
behaviors
Predictive Models
Identify attributes of
known fraud behavior
Text Mining
Leverage unstructured
data elements in
analytics
Network Analysis
Associative discovery
thru automated link
analysis
Examples:
• Inaccurate eligibility
information
• Unlicensed or
Suspended Provider
• Daily provider billing
exceeds possible
• CPT up-coding
• Value of charges for
procedure exceeds
threshold
Examples:
• Abnormal service
volume compared to
similar providers
• Ratio of $ / procedure
exceed norm
• # patients from
outside surrounding
area exceeds norm
Examples:
• Like patterns of
claims as confirmed
known fraud
• Provider behavior
similar to known
fraud cases
• Like provider/
network growth rate
(velocity)
Examples:
• Claim/call center
notes high-lighting
key fraud risks (e.g.,
policy questions)
• Static data elements
(e.g., address) used
for linking suspicious
activity
• Integration of rich
case file information
Examples:
• Provider/claimant
associated to known
fraud
• Linked members with
like suspicious
behaviors
• Suspicious referrals to
linked providers
• Collusive network of
providers & referrals
Medical
Procedure
Claims
Eligibility
Data
Provider /
Member
Referral
Known Bad
Lists
3rd Party
Data
New Phenomenon
Entrepreneurs in Healthcare Fraud
Entrepreneurs in Healthcare Fraud
• Who/what are these entrepreneurs?
• What are they doing?
• How do we identify them?
– Traditional scheme development
– A new analytic approach
• Wikipedia defines entrepreneur as:
– An entrepreneur is an owner or manager of a business enterprise who
makes money through risk and initiative.
– Is a term applied to a person who is willing to help launch a new venture
or enterprise and accept full responsibility for the outcome.
– “The entrepreneur shifts economic resources out of lower and into higher
productivity and greater yield."
Who/What Are These Entrepreneurs?
• A Healthcare entrepreneur can be:
– A business person who shows initiative to maximize income
– Middlemen who have inserted themselves into the healthcare delivery
system
– In both of the above cases, rules and regulations are usually
disregarded
Caution: Entrepreneur may or may NOT be a healthcare provider
•
Operations at OptumInsight regularly identifies “Entrepreneurs” in both
categories through ideation, referrals and tips from the healthcare
community
OptumInsight Anti-Fraud Analytics
• Our experience:
– The Provider will maximize revenue, but try to stay below
the radar by keeping dollar thresholds low enough
– Insert themselves into the healthcare delivery stream,
possibly as a “middleman” between legitimate providers
• Suppliers/vendors
• OBS facility charge
• HHA arranger
• IOM double billing
• “Concierge” services (delivery of drugs)
What Are They Doing?
• Submitting another bill when one or more providers have already
been paid in full
• Taking advantage of gaps
– In the claims processing system
– In Policies and Procedures
– Between UB forms and CMS forms
• Some take advantage of there being no applicable taxonomy code
for a non-healthcare type of provider
How Do We Identify Them?
• Old school methods: Look for them!
– Scheme ideation: Unlicensed entity (ASCs, FSED, OBS, Pharmacy,
HHA)
–
–
–
–
Duplicate claims
Data analysis
Inappropriate use of modifiers to prevent claim form hitting edits
More than one claim for single date of service (single or multiple
providers)
– Patient complaints
Example
Middleman company bills globally when the facility has
been paid the Technical component. Reading physician
gets the “interpretation” (modifier 26) split.
• According to policy, the payer will reimburse the interpreting
physician or healthcare professional only the professional
component because the facility is reimbursed for the technical
component of the service
• It would be inappropriate to bill the global code which
represents both the professional and technical component.
Example
Unlicensed Facility Entrepreneurs
Characteristics:
• Office Based Surgery entity at same
physical address as physician
• Physician paid global fee
• Corporate entity for office bills facility
charge
• Corporation is owned by the physician
• Classic double dip
Example
Concierge Service
Characteristics:
– Billing for self-administered or IV infusion drugs
– Arranges for the delivery of drugs directly to patient
or hospital outpatient department
– But is not a licensed pharmacy
– Has NO relationship with patients
– Doctors defending their use of the company
– Suspect kickbacks to Pharmacy and Doctors
– High Dollar claims and High Units billed
– Hold HHA license, but no employees
Analytics
Instead of scheme
analysis focused on a
single provider type on
a state by state
basis….
New data
analysis
techniques
examine multiple
indicators to
determine
aberrant billing
patterns
Detecting Healthcare
Entrepreneurs With Analytics
• Entrepreneurs are in the healthcare system to make money
rather than to deliver quality care so they will behave differently.
• For a healthcare provider to increase their reimbursement, they
must increase at least one of the following:
– Patients
– Units per Patient
– Cost per Unit
• Each of these can have various approaches. For example, to
increase patients a healthcare provider can:
– Open a new office location
– Aggressively market services
– Add patient “enticements” – waive copayments, spa coupons
– Engage in patient kick-backs – “rent-a-patient”
– Bill from a stolen patient list
23
Detecting Healthcare
Entrepreneurs With Analytics
• Our goal is to find Healthcare Entrepreneurs who are gaming
the payment system by inappropriately inflating patients, units
or unit cost
• Traditional outlier analysis often uses:
– Static benchmarks such as a provider who bills the highest code
in an upcoding group 5X more than their peers.
– Acceleration reports such as a provider billing a code 50% more
than last year.
24
Adjust For Known Changes
(Know Your Environment)
• New Member Groups
– Number of patients for healthcare provider increases 50%
 (more providers + static members = fraud)
• New Procedure Codes / New Use For Established Codes
– Healthcare provider’s use of a CPT dramatically increases, maybe a related
CPT was deleted and use shifted to this code?
= Need for provider Utilization Profiling & Morphing analysis
• Regional Practice Differences
– Medical school and practice partners greatly influence treatment patterns
• Epidemics, Seasonality and Disasters
– Fast rise of office visits in the Northeast, maybe a flu outbreak?
– Control at a state level
• Consider Provider Morphing
Healthcare Entrepreneurs
Conclusion
• Healthcare Entrepreneurs can range from licensed providers pushing
the envelope for revenue maximization to fictitious entities engaged in
fraudulent billing schemes
• Traditional scheme analysis is still an important part of an overall
detection strategy
– It yields savings
– Can help exclude inappropriate provider types
• Analytics can lead to the discovery of new schemes
– There are only three “levers” to increase reimbursement
• Patients, Units and Unit Cost
– Need both benchmarking and acceleration to accurately describe a
reimbursement pattern
– Many common occurrences will result in false patterns
• Remove known causes of variation such as new groups, regional differences, etc…
Healthcare Entrepreneurs Next Steps
• Clinical Prevalence
– Procedure codes can have multiple uses
– Linking them to diagnosis codes and peer literature on disease prevalence
will help us understand the complete treatment picture
– Should help weed out false positives
Use of Prevalence Data
in Fraud Detection
Epidemiology can be a useful tool
in Investigation
Examples:
•25.8 million children and adults in the United States—8.3% of the
population—have diabetes
•16.3% of the U.S. adult population—have high total cholesterol. The
level defined as high total cholesterol is 240 mg/dL and above
•These diseases are common and are evenly distributed amongst
Primary Care Physicians
Background
• Period Prevalence Rate is all cases whether old, new or
recurrent, arising over a defined period, either one or two
years. The denominator is the average population over
the period (or mid-point estimate)
– Specific rates permit rational and easy comparison of
disease patterns in different places and times for they can
be directly compared with each other
– There may be some regional variation but extremes are
worrisome either for fraud/waste/abuse or an epidemic that
merits public health investigation
• Prevalence values are additive as the population is the
common denominator
• ICD-10 has greater coding precision for many conditions
when compared with ICD-9 diagnosis coding
Example—Dx 357.81, Chronic
Inflammatory Demyelinating
Polyneuritis (CIDP)
• CBSA for zip code 33135 (Miami, Florida)
• Population-- 5,531,060
• 1695 Family Practice Physicians in CBSA*
• Prevalence of CIDP: 1 case per 10,000
• Expected CIDP cases in CBSA—553
• Insurer’s Florida Medicare membership 583,000
• Expected CIDP cases in Insurer’s population—583
• A single provider (Family Practice) was treating 9 individuals in
his practice for CIDP using high dose intravenous
immunoglobulin
`.
*
http://www.doh.state.fl.us/Workforce/Workforce/Annual_Reports/PhysicianWorkforce_Nov2010.pdf
Example—Dx 357.81, Chronic
Inflammatory Demyelinating
Polyneuritis (CIDP)--continued
• This doctor has 2% (9/553) of ALL CIDP cases in this entire
CBSA & 20% (9/58) of all expected CIDP cases in the insurer’s
Florida M&R market!
• Incidentally 4 other doctors (also Family Practitioners), had an
additional 3% (17/583) of ALL cases in the CBSA and 30%
(10/58) of the expected CIDP cases in the insurer’s Florida
M&R market!
• This concentration of patients with CIDP would not be expected
for a Family Practitioner
– All patients were receiving IVIG administered in high dosage
– The practices were clinics rather than specialists
• Review of records revealed diagnosis and treatment were both
fabricated with substantial recovery
IVIG can be used for
Hematologic Diagnoses
• Conditions such as Idiopathic Thrombocytopenic Purpura,
Monoclonal Gammopathy and Chronic Lymphocytic
Leukemia respond to IVIG
• Prevalence of different conditions with a common treatment
can be added together
• Number of patients with hematologic diagnoses with a
common treatment such as IVIG divided by population can
reveal areas where the prevalence is in excess of expected
• Single payer system and use of ICD-10 coding enhance
detection ability
• Review of records may yield misrepresentation of diagnoses
and/or treatment
Case Study - Infusion Therapy
• Scheme: Over-utilization of Infusion Therapy Drug (ITD) codes
– Excessive units
– Excessive frequency of service
• Costly services
• Small # of patients
$119 million
$8 million
$142 million
Infusion Therapy
• Scheme targets vulnerable populations
– Lower income/high unemployment
– HIV diagnoses
– Fraudsters buy member IDs
• From other fraudulent providers (list sharing)
• From patients themselves – offer kickbacks
– Cash
– Prescriptions, drugs, appliances, food
• Providers enlist physician complicity or bill using their provider
number without doctor knowing
– Often older physicians
– Pay kickbacks for referrals
• Providers enroll with Plan to get new provider number
IV Therapy: Detection Approaches
Detection
• Predictive Scoring Model (PSM)
• Peer Grouping
• Provider Profiling
• Link Analysis
• Static Code Analysis
• Prospective Rules
Provider Morphing: Definition
• Helps identify providers that may be abusing the payer system
• Compares distribution of billed procedure codes across two time
periods, weighted by paid dollars
• Large changes/swings may indicate the evolution of fraud schemes
Studying Provider behavior over time
• Patterns of provider behavior suggest fraud schemes
• Once a fraud scheme is identified, it can be stopped (or minimized)
• Fraudsters have financial incentive to adjust billing practices to
evade detection and maximize revenue
• For “Fraudster” group, when one “bad” behavior stops, new “bad”
behavior likely starts
• Traditional methods of fraud detection offer few clues as to what
that new bad behavior will look like
Provider Morphing: Sample Results
Provider
Procedure Procedure Description
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
Clinic XYZ
1992
1935
1810
1630
1480
1400
952
840
810
797
790
740
670
630
160
ANESTH, N BLOCK/INJ, PRONE
ANESTH, PERC IMG DX SP PROC
ANESTH, LOWER ARM SURGERY
ANESTH, SURGERY OF SHOULDER
ANESTH, LOWER LEG BONE SURG
ANESTH, KNEE JOINT SURGERY
ANESTH, HYSTEROSCOPE/GRAPH
ANESTH, SURG LOWER ABDOMEN
ANESTH, LOW INTESTINE SCOPE
ANESTH, SURGERY FOR OBESITY
ANESTH, SURG UPPER ABDOMEN
ANESTH, UPPER GI VISUALIZE
ANESTH, SPINE, CORD SURGERY
ANESTH, SPINE, CORD SURGERY
ANESTH, NOSE/SINUS SURGERY
Current Previous
Quarter Year
$36,876 $9,548
$26,317
$0
$13,650
$0
$20,648
$0
$16,745 $4,445
$20,283 $4,793
$23,806
$0
$21,681
$0
$28,955 $16,190
$10,367
$0
$35,330
$0
$34,504 $8,709
$8,253
$0
$8,362
$0
$23,299
$0
Data is grouped by provider to highlight changes in
behavior.
Current Quarter and Previous Year show the discrepancy
in paid amounts year-over-year for a given quarter.
The table above shows significant change in the type and
volume of procedures performed.
$500,000
$450,000
$400,000
$350,000
$300,000
$250,000
$200,000
$150,000
$100,000
$50,000
$0
Series1
1
2
$434,317
$83,670
Above, a suspicious Total Paid
discrepancy of over 5 times the previous
year’s paid amount.
Case Studies
Provider and CPT Paid Amount information
Total Paid Amount comparison
Provider
Procedure Procedure Description
Previous
Year
Current
Quarter
Provider
Provider
Provider
Provider
Provider
76942
64484
64483
64479
27096
$0
$0
$0
$0
$29
$5,625
$5,886
$14,398
$3,223
$3,173
A
A
A
A
A
ECHO GUIDE FOR BIOPSY
INJ FORAMEN EPIDURAL ADD-ON
INJ FORAMEN EPIDURAL L/S
INJ FORAMEN EPIDURAL C/T
INJECT SACROILIAC JOINT
• Physiatrist listed as Medical Director of a spa that recently went out of
business. DO and DC training
• 54% increase in patient volume, 78% increase in codes billed/pt visit
in current year with spike in echo guided biopsy and epidural/SI
injections
• Significant increase in submission of claims in current period
• Physical therapy and Chiropractor in practice with provider
Provider Morphing Summary
• Provider Morphing identifies potential aberrant behavior. It
has potential application as:
– Retrospective analysis tool to find suspicious behavior that
would require additional investigation
– Retrospective tool to ensure compliance for providers who have
agreed to modify certain billing behaviors
– Prospective flag that would allow the pending of claims subject
to further analysis of medical records
Ted’s Closing Thoughts
• Multiple Analytical Methods and Advanced Analytics are a Must
• Data, data and more data - use all data available to you
• Work together
– healthcare fraud is truly a global problem!
• Get creative!!
– “fraudster” will and does in order to maximize revenue
41
Julie Malida
Principal, Health Care Fraud Solutions
SAS Institute Inc.
42
THE NEED FOR MULTIPLE
ANALYTICAL METHODS
Analytic Approach: Business Rules
• Automates manual
processes
• Operationalize traditional
“red flags” or suspicious
loss indicators
• Effective regardless of
training or experience
level
• Catch suspicious claims
that would “fall through
the cracks”
Analytical Approach – Business Rules
Health Care Scenarios/Model Examples
1
Claims less than xx months of policy inception
2
Increase in coverage less than xx months of claim
3
Clinic/hospital distant from claimant’s home address for routine care
4
Bills are billed on holidays and weekends
5
Physician's diagnosis not consistent with treatment
6
Value of charges for procedure is excessive
7
Same drug prescribed for multiple family members
8
Doctors treatment always the same despite different injuries/accidents
9
Medication prescribed out of line with physician speciality
10
Doctor bills for emergency anaesthesia but hospital stay was non-emergency
Analytic Approach:
Unsupervised Methods
(Anomaly Detection)
• Use when no target exists
• Examine current behavior
to identify outliers and
abnormal transactions that
are somewhat different
from ordinary transactions
• Include univariate and
multivariate outlier
detection techniques, such
as peer group comparison,
clustering, trend analysis,
etc.
Avg. Number of PCS Services Submitted
Provider is not
only an outlier,
also shows
extreme variation
for average
number of
services submitted
per attending
provider
Analytic Approach:
Supervised Methods (Predictive Models)
• Use when a known target
Fraud Scores
(prior fraud) is available
• Use historical behavioral
information of known
fraud to identify suspicious
behaviors similar to
previous fraud patterns
• Include parametric and
nonparametric predictive
models, such as
generalized linear model,
decision tree, neural
networks, etc.
Predicted
Fraud Scores
Incomes
# of previous
investigations
Target Identification
Social Network (Link) Analysis
• Network scoring
– Rule and analytic-based
• Analytic measures of association
help users know where to look
in network
– Net-CHAID for local area of
interest (node) in the network
– Density, Beta-Index (network)
– Risk ranking with
hypergeometric distribution,
degree, closeness,
betweenness, eigenvector,
clustering coefficients (node)
• Modularity (sub-network)
Text Mining (Unstructured data)
 Up to 80% of insurer
data is unstructured
text
 Configurable parsing,
tagging, and extracting
of free text for use in
fraud analytics
 Combine quantitative
and qualitative data
with text analysis to
improve predictions
Text Mining (e.g., call center logs or doctor’s notes)
The Solution: A Hybrid Approach
50
Sample Client
Health Care Fraud Findings
Using a Hybrid Approach
Example 1 – Non-US Benefit Plan:
Targeting Doctor Shopping
• Problem statement:
– Plan previously relying on tips/law enforcement
– Drug diversion by pharmacies, drugs finding their way to the streets.
• Analytics Applied: Rules, Anomaly detection and link analysis
• Data Provided:
– One year of claims, provider and patient data (~5GB)
– Over 11k providers and 687k patients, $1.1B paid
• Findings:
– $192M in suspicious activity detected in top 10%, countrywide.
Example 1 – Non-US Benefit Plan:
Targeting Doctor Shopping
# SAS
Flagged
Providers
Total
Amount Paid
to SAS
flagged
Providers
Estimated
Fraud /
Waste /
Abuse
Score
Range
Percent
of
Providers
88-100
1%
110
$71,285,537
$3,187,105
$3,187,105
61-87
5%
403
$159,146,916
$7,115,299
$3,928,194
28-60
10%
512
$192,134,531
$8,590,143
$4,403,037
Estimated
Recovery
Example 1 – Non-US Benefit Plan:
Targeting Doctor Shopping
Validated Examples
Pharmacy A
Pharmacy B
• Example of two members with
•
very abusive behavior
•
•
•
Member 1 had 228 paid claims per
month (~7 claims daily) with a total
claim amount of $79K.
Member 2 had 110 paid claims per
month (~4 claims daily) with a total
claim amount of $20K.
Two prescribers responsible for ~99%
of these two members claims.
• $384K out of $963K(36%) of
claim dollars suspicious
•
Member had 642 paid claims per
month(~20 claims daily), and was
responsible for $63K of $67K(94%) of
the total amount paid to this pharmacy.
One physician prescribed $61K of the
claims.
Pharmacy C
•
•
Member had 129 paid claims per
month(~4 claims daily), and was
responsible for $40K of $116K(35%) of
the total amount paid to this pharmacy.
One physician responsible for $13K of
the claims.
Example 2 –
Commercial U.S. Health Plan
• Problem statement:
– Increasing number of patients exhibiting drug-seeking behavior for
Promethazine with Codeine (party cocktail) and Hydrocodone (pain
tablets).
• Analytics applied: Rules, anomaly detection and link analysis
• Data Provided:
– All claims, provider and member information for 1 year and 1 region
– ~ 414k claims, 116k members, $18.7M annually for these 2 drugs
• Findings:
– ~$1.5M in suspicious activity detected
– 1.4% of members taking these drugs flagged (1,587 patients)
Example 2 –
Commercial U.S. Health Plan
Based on SAS score
Hydrocodone
Promethazine with
Codeine
Total tablets
Total Cost
Total ML
Total Cost
Top 10 members
40,685
$1,999
89,834
$1,405
Top 50 members
198,170
$12,001
340,961
$5,326
Top 500 members
1,892,428
$106,726
1,956,103
$31,307
All members score>0
2,915,604
$163,077
2,398,391
$38,553
[1]
[1]
858 members had score > 0 for Hydrocodone
759 members had score > 0 for Promethazine
Example 2 Findings:
Top 10 Hydrocodone Members
Of all doctors visits ,
Member
inactive or 1+
Rx dispensed
outside
mem score
Member
dates
****4375
****0276
****0040
****8680
****3940
****8351
****5070
****4700
****3469
****0319
62
62
60
60
60
60
59
57
56
56
All 20 new
hydrocod Rx
had no prior doc
visit
Mem
Qty 12m
Inactive
1
1
1
0
1
0
0
0
1
0
4230
2784
3269
3765
5330
7260
3724
3863
2260
4200
Multiple
sources
only 3 did not
involve prescribing
this drug
# new
Rx
72
purchases of
other fraudprone drugs
% New
# Dr
#
#
#Rx
Rx of
Visit s Pharma Prescrib other
this
with No
ers
fraud
Drug
Rx of
prone
with No
this
drugs
Dr Visit Drug
20
16
23
71
28
48
45
57
28
14
100
94
100
97
86
88
98
95
54
100
1
0
0
24
4
3
1
10
12
0
2
3
2
5
1
3
6
6
5
1
3
3
5
10
2
1
7
8
10
2
Diff
Zipdist
8
0
0
18
36
72
0
14
33
15
Travels 35 miles more for hydrocod
compared to non-fraud-prone drugs
8
35
1
2
0
0
7
0
1
14
# Tablets (10mg Hydrocodone)
Detailed look at Member
1200
1000
• Monthly # tablets purchased
is excessive compared to
other hydrocodone users.
•All scripts came from 1
prescriber.
800
600
400
200
0
1
2
3
4
5
6
7
8
9
10
11
12
Month
Member
90th Percentile
patient buys more
frequently than when
supply runs out
Estimated Dose Per Day
180
160
140
120
100
80
60
40
20
0
0
100
200
300
Days Since First Purchase
400
Example 2 Findings:
Top 10 Promethazine Members
Member
score
95% of new Rx
Bought Prometh
for Prometh had
for 12 months.
no prior doctor
Not a seasonal
visit
Flagged Member Qty 12m # Months # New % New
user.
with
Spec
****7226
****5643
****5543
****0381
****5606
****3115
****4290
****6742
****0371
****3097
64
61
60
58
57
57
56
56
55
54
Inactive
0
0
1
0
1
0
0
0
1
0
(ML)
with Rx
for this
Drug
1 16,800
1 14,190
1 5,038
0 10,560
1 5,160
1 6,203
0 13,673
1 2,472
0 12,440
1 3,298
11
11
7
12
6
8
12
6
12
7
Rx
35
15
22
42
24
11
31
9
26
10
# Dr
Rx of
Visit s
this with No
Drug
Rx of
with No this
Dr Visit Drug
100
100
100
95
100
100
90
100
92
100
Traveled 4
miles more
for Prometh
compared to
non-fraud#
Diff
#Rx
prone
drugs
Pharma Prescrib other Zipdist
Went to 7
pharmacies
and 8
prescribers
for Prometh
#
0
0
0
22
1
0
3
0
1
1
fraud
prone
drugs
ers
1
2
4
7
4
2
6
2
4
2
2
2
7
8
9
7
4
9
3
8
31
14
15
44
3
4
22
9
16
11
Had 22 other doctor visits in 2009 that did not
involve the prescription of Prometh. Could this
be a cancer patient using Prometh as an antinausea drug?
0
0
0
4
0
0
1
0
6
0
Detailed look at Member
Dispensing Quantity
•Escalating usage.
•Heavy usage compared to 90%tile
of age group (age 47)
•Expected seasonal usage not
followed.
2000
1800
1600
1200
1000
800
600
400
200
Avg Dose pDay
0
1
2
3
4
5
6
7
8
9
10
11
12
130
Month
120
110
patient buys more
frequently than when
supply runs out
100
Est Dose Per Day
Total Quantity
1400
90
80
70
60
50
40
30
20
10
0
0
100
200
Days Since First Purchase
300
400
Drug Seeking Behavior Study:
Link Analysis
• Suspicious Network of Collusion:
• 2 members, both high scorers, same address
• 6 prescribers, 9 for the other, none in common
• Member 1 had activity from January-August 2009, and member 2
from June-August 2009.
• Activities of the 2nd member could have been stopped earlier if DSB
scoring and link analysis were performed regularly.
Example 3: Commercial U.S.
Health Plan - Targeting 5 Specialties
• Problem statement:
– Analyze professional claims payment activity in order to identify
patterns of fraud, waste and abuse in: Labs, DME, Pain
Management, Mental Health and Podiatry.
• Analytics applied: Rules, anomaly detection and link analysis
• Data Provided:
– All claims, provider and member information for 1 year and 1 region
– ~ 10k providers and $161M annually
• Findings:
– ~$16M in suspicious activity detected
– 623 providers flagged
Example 3: Commercial U.S.
Health Plan - Targeting 5 Specialties
Specialty
#
Providers
Amount paid
# Providers
Flagged **
Amount
Flagged
LAB
5,451
$75,820,727
382
$11,232,812
PAIN
2,158
$32,587,234
104
$2,466,119
DME
1,322
$42,059,927
86
$1,480,618
MENTAL
649
$7,897,088
45
$715,669
PODIATRY
493
$3,000,998
15
$93,086
10,073
$161,365,974
632
$15,988,304
All Five
** Includes all providers with at least 1 alert triggered
Example 3: Commercial U.S. Health Plan
Targeting 5 Specialties
• Goal: uncover potential fraud networks by linking flagged
providers based on name, address, phone number, tax id, Social
Security Number
Network 10884:
• 3 LAB providers  2
LAB companies
• total potential loss =
$81k
2 providers having same
address but other info
different. Medium scorer
colored green, while low
scorer colored white.
2 providers having same phone
number, tax id, similar names,
but different addresses  same
company, different branches?
High scorer (75).
Potential loss=$79k.
Example 4: U.S. Blue Cross Plan
Targeting Limited Benefit Abuse
• Problem statement:
– Abuse of benefits with inside limits (such as chiropractic and
acupuncture) by continuing the services using other family members’
ID.
• Analytics applied: Rules and anomaly detection
• Data Provided:
– All claims, provider and member information for 3 years (results on
most recent 12 months)
– ~ 8.57M claims, 750k members and $871M (most recent 12 months)
• Findings:
– ~$162k in suspicious activity detected
– 583 members flagged
Example 4: U.S. Blue Cross Plan
Targeting Limited Benefit Abuse
• When one family member’s benefit is used up, billing moves to another
family member with the same or similar diagnosis and same procedures:
– 483 members in 182 families with 120 Chiropractors
• Youngest member seen was 1 year old
• 3083 claims worth $111,110
– 13 members in 4 families with 8 acupuncture providers
• Youngest member seen was 11
• 104 claims worth $33,493
– 87 members in 32 families with 18 dermatologists
• Youngest member seen was 8
• 178 claims worth $17,525
Example 5: U.S. Public Agency
Targeting In-Home Support Services
• Problem statement:
• Desire to detect fraud earlier
• Improve the quality of their alerts
• Facilitate criminal investigations
• Analytics applied: Rules, anomaly detection, predictive modeling and Link
Analysis
• Data Provided:
– Beneficiary, provider, historical fraud referrals and vital records data for most recent 4
years
– ~170k beneficiaries and $1.9B paid in 2011
• Findings:
– 1020 alerts and $1.6M at risk
Example 5: U.S. Public Agency
Targeting In-Home Support Services
Results:
Anomaly Detection Rules
Cases Flagged
Distance and Geography
Provider Address out of State
39
Provider Address out of County
436
Distance Between Provider and Consumer
30 to 39 Miles
1749
40 to 69 Miles
2171
Over 70 Miles
567
Hours per Month
Severely Impaired cases (defined as eligible for up to 283 hrs. per
month but receiving more)
2292
Non-Severely Impaired (defined as eligible for up to 195 hrs. per month
but receiving more)
3289
Authorized for up to 300 hrs. per month but provider paid for more
756
Paid for Over 540 Hours
55
Paid for Over 720 Hours
12
Cross Participation
IHSS Provider is also Consumer of IHSS services
212
DMS Child Care Provider is also IHSS Consumer
289
Example 5: U.S. Public Agency
Targeting In-Home Support Services
More Results:
Anomaly Detection Rules
Cases Flagged
Payment After Death
Provider
Consumer
30 to 59 Days
41
545
60 to 119 Days
27
254
120 to 179 Days
19
102
180 to 364 Days
20
74
Over 365 Days
44
50
Person was only 1 link away from someone convicted or under investigation
259
71
Person was a suspect in another fraud case
558
564
Person was on the list of fleeing felons the state is pursuing
n/a
20
Convicted Drug Felon
n/a
27
Suspicious Activity
Suspicious Scenario (provider is not a relative and traveling >40 mi. to get
paid for small # of hours)
45
Closing Thoughts
• Multiple Methods are a Must
• Don’t just perpetuate EVERY linkage
• Re-score using other analytics after you build the links
• All models degrade over time:
– Need a continual feedback loop
– Need periodic testing of model performance
• Hoard the data, and use it all – even the unstructured
text data!
70
2nd Global Summit
Healthcare Fraud:
Prevention is better than cure
25-26 October 2012
Beaumont Estate, Old Windsor, UK