ifm analytics

22
© 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation's express consent. IFM Analytics What’s New? Supriti Singh Senior Scientist, Insurance Fraud Manager FICO FICO® Insurance Fraud Manager User Group: San Diego, CA | May 7--8, 2014

Upload: vartan

Post on 24-Feb-2016

84 views

Category:

Documents


0 download

DESCRIPTION

FICO® Insurance Fraud Manager User Group :. San Diego, CA | May 7--8, 2014 . IFM Analytics. What’s New?. Supriti Singh Senior Scientist, Insurance Fraud Manager FICO. FICO® Insurance Fraud Manager User Group :. San Diego, CA | May 7--8, 2014 . IFM Analytics: What’s New?. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation's express consent.

IFM AnalyticsWhat’s New?

Supriti SinghSenior Scientist, Insurance Fraud ManagerFICO

FICO® Insurance Fraud Manager User Group:San Diego, CA | May 7--8, 2014

Page 2: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation's express consent.

IFM Analytics: What’s New?Speakers

Nitin BasantSenior ScientistFICO

Jeremy Greene PhDLead ScientistFICO

RobinScientist IIFICO

FICO® Insurance Fraud Manager User Group:San Diego, CA | May 7--8, 2014

Page 3: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.

Agenda

3

► Network Analytics For Insurance (IRE)-Nitin

► Member Centric Analytics-Nitin► Focused Analytics-Robin► IFM Claims Model Analytics: Peak

and Valley Detection-Jeremy► Discussion- Supriti

Page 4: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.4

Network Analytics for Insurance (IRE)

Page 5: IFM Analytics

5 © 2014 Fair Isaac Corporation. Confidential.

The future of Analytics

Learn

LearnLearn

Claims

FICO Fraud

Trained Models

FICO Outlier

Detection Models

FICO Link

AnalysisLearn Known Patterns to Filter• E.g. combinations of specific

procedure codes and modifiers

Detect Networks• Detect provider/ pharmacy/

facility networks to profile and score

Learn Additional Characteristics• E.g. # number of providers in

the network

Learn New Fraud Tags• Including from Outlier Detection Models

and Link Analysis

Page 6: IFM Analytics

6 © 2014 Fair Isaac Corporation. Confidential.

Name Carole Brand

Address 12395 Cedar Park Rd., Suite 225

City ClevelandState OHZip 44102

Inve

stig

atio

nal

Rev

iew

Suspect(s) Matches and Visualization

Search Criteria

Proa

ctiv

e D

etec

tion

High Scoring Claims, Providers, pharmacies, etc.

Proactive BatchNetwork Scoring

Prioritized Suspicious Network LeadsSearch/Match

960

650

560

960

CasesPharmacies

Members3rd Party Private3rd Party Public

Scoring

Engine

Search/Match

CasesPharmaciesPrescribers

Members3rd Party Private3rd Party Public

Network Analytics

All Claims, providers,pharmacies, etc.

Prescribers

Claims

Claims

Available in IFM

Page 7: IFM Analytics

7 © 2014 Fair Isaac Corporation. Confidential.

► Focused Analytics► Networks with high concentration of suspicious claims/ providers► Seemingly unrelated providers who share a lot of members► Providers with multiple NPIs► Etc

► Networks Model► Identify all the networks in the data► Develop a model to score the networks as an entity

Network Analytics

Page 8: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.8

Member-centric ModelsUsing medical, pharmacy and facility data together

Page 9: IFM Analytics

9 © 2014 Fair Isaac Corporation. Confidential.

► A high priority for IFM

► Follow the member using member ID across different claim types like medical, pharmacy, facility, etc.

► Consistent and reliable data across claim types improve the scope and quality of analytics► Example – Member ID, Rendering Provider ID

► Model development using medical and facility claims underway

Member-centric models

Page 10: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.10

Focused Analytics

Page 11: IFM Analytics

11 © 2014 Fair Isaac Corporation. Confidential.

Focused AnalyticsHigh Units

Example 1 Example 2Diagnosis Unspecified general medical examination Pain in joint, lower leg

Procedure93000 - Electrocardiogram, routine ecg with at least 12

leads73560 - Radiologic examination, knee; one or two views

Claim Begin Date 2/26/2013 11/16/2012Claim End Date 2/26/2013 11/16/2012Allowed Amount $8,548 $2,168Paid Amount $8,458 $2,168Units 402 50

Identifies claim lines which have unusually high number of units as compared to the norm.

Example 1- 402 is a lot of EKG’s especially when they are done for an “unspecified general medical exam”.

Example 2- 50 x-rays of the knee for knee joint  at a cost of $2169.00 paid is too High. One or two would be more appropriate.

Page 12: IFM Analytics

12 © 2014 Fair Isaac Corporation. Confidential.

Focused AnalyticsProcedure Upcoding

Identifies providers who have unusually high tendency to perform/ bill for more expensive procedures.

Procedure groups created using data driven cluster on textual description of procedure codes and diagnosis codes.

Provider Specialty: Orthopedic Surgery

Procedure Code Average Allowed Provider Activity Peer Activity73580 - RADIOLOGIC EXAMINATION, KNEE, ARTHROGRAPHY, RADIOLOGICAL SUPERVISION AND INTERPRETATION

$104.92 75.36% 0.19%

73564 - RADIOLOGIC EXAMINATION, KNEE; COMPLETE, FOUR OR MORE VIEWS $42.83 19.81% 13.75%

73562 - RADIOLOGIC EXAMINATION, KNEE; THREE VIEWS $36.40 2.90% 24.42%

73610 - RADIOLOGIC EXAMINATION, ANKLE; COMPLETE, MINIMUM OF THREE VIEWS

$35.19 0.00% 13.67%

73630 - RADIOLOGIC EXAMINATION, FOOT; COMPLETE, MINIMUM OF THREE VIEWS

$34.77 0.00% 14.97%

73560 - RADIOLOGIC EXAMINATION, KNEE; ONE OR TWO VIEWS $29.74 1.93% 18.57%

Page 13: IFM Analytics

13 © 2014 Fair Isaac Corporation. Confidential.

Focused AnalyticsHigh RVU

The analysis aims to identify providers with unusually high Relative Value Units (RVU) sum on a particular day.

RVU examples - 32854 - Lung transplant with cardiopulmonary bypass 90.00 RVU99201 - Office/outpatient visit new patient 0.48 RVU

Provider Specialty: Chiropractic 1.1 Million claim lines in a year 22 Million dollars paid in the same year

Service Date 11/19/2012 12/17/2012 RVU Sum 2,977.22 2,828.64 Claim Lines 6,227 6,039 Members 2,997 2,911

Page 14: IFM Analytics

14 © 2014 Fair Isaac Corporation. Confidential.

Focused AnalyticsImpossible Day

Identifies providers who billed services adding up to an excessive number of hours in a single day. Uses time-based procedure codes .

Timed Procedure examples - 96102 - Psychological testing by technician 60 minutes96150 - Health and Behavior Assessment 15 minutes

Provider Specialty: Pediatrics 280,875 minutes from 248 days averaging 19 hours a day

Service Date 2/8/2011 1/19/2011Total Minutes 2,000 (~33 hours) 1,965Total Paid 4,498.02 4,408.98Claim Lines 192 204Members 112 116

Page 15: IFM Analytics

15 © 2014 Fair Isaac Corporation. Confidential.

► Additional analytics available as a service► Models currently available only for professional claims► Flexible input file format► Currently packaged/ready for deployment► Accelerated innovation-to-value for clients► More to come

► Geospatial – members visiting providers geographically distant► Procedure Unbundling

Focused Analytics

Page 16: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.16

IFM Claims Model Analytics: Peak and Valley Detection

Page 17: IFM Analytics

17 © 2014 Fair Isaac Corporation. Confidential.

► The IFM Medical Claims model currently has 6 analytics: High Dollar Procedure, High Dollar Day, Procedure Rate, Procedure Repetition, Unusual Modifier, and Missing Modifier.

► All 6 analytics use your data.

► However, the two dollar analytics currently make some statistical assumptions.► A very common example of a statistical assumption is to assume that data follows a “normal

distribution” (see below) since normal distributions have nice mathematical properties.

Current IFM Approach

Page 18: IFM Analytics

18 © 2014 Fair Isaac Corporation. Confidential.

► Points that fall inside the purple circle will currently receive a high score.► Example: The “normal distribution” assumption is in red.

► However, sometimes these are just contracted fee schedules for a subset of providers (i.e., false positives).

Current IFM Approach: High Dollar Procedure

$ for procedure code0

20

40

60

80

100

120

140

Average $ for procedure code

# claim lines

Page 19: IFM Analytics

19 © 2014 Fair Isaac Corporation. Confidential.

► NO STATISTICAL ASSUMPTIONS

► We make a smooth curve out of the histogram.

► We identify peaks and valleys and flag the high dollar valleys as outliers.

New Solution: “Peak and Valley” Detection

$ for procedure code

# claim lines

20 30 40 50 60 70 80 90 100

Average $ for procedure code

Valleys flagged as outliers

Valley not flagged as outliers

Page 20: IFM Analytics

20 © 2014 Fair Isaac Corporation. Confidential.

► We have tested this new algorithm on the High Dollar Procedure analytic.

► 28% more fraud detected

► 17.5% increase in savings

Results

Page 21: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential.21

Our Analytics Team in San Diego and Bangalore

Nitin Basant Jeremy Greene Jessy Su Robin Snehal KatreSupriti Singh

Himanshu Jain Vivek Bhardwaj

Page 22: IFM Analytics

© 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation's express consent.

Thank YouSupriti [email protected]

FICO® Insurance Fraud Manager User Group:San Diego, CA | May 7--8, 2014