session 94 panel discussion: beyond risk identification ... · adapted from gartner’s data...

Session 94PD, Beyond Risk Identification: Predictive Analytics in Health

Presenters: Elena V. Black, FSA, EA, MAAA, FCA

Yi-Ling Lin, FSA, MAAA, FCA Michael Y. Xiao, FSA, CERA, MAAA

SOA Antitrust Disclaimer SOA Presentation Disclaimer

https://www.soa.org/legal/antitrust-disclaimer/

https://www.soa.org/legal/presentation-disclaimer/

2018 SOA Health MeetingYI-LING LIN, FSA, MAAA, FCASession 94 – Beyond Risk Identification: Predictive Analytics in HealthJune 26, 2018

SOCIETY OF ACTUARIESAntitrust Compliance Guidelines

Active participation in the Society of Actuaries is an im portant aspect of m em bership. W hile the positive contributions of professional societies and associations are well-recognized and encouraged, association activities are vulnerable to close antitrust scrutiny. By their very nature, associations bring together industry com petitors

and other m arket participants.

The United States antitrust laws aim to protect consum ers by preserving the free econom y and prohibiting anti-com petitive business practices; they prom ote

com petition. There are both state and federal antitrust laws, although state antitrust laws closely follow federal law. The Sherm an Act, is the prim ary U.S. antitrust law pertaining to association activities. The Sherm an Act prohibits every contract, com bination or conspiracy that places an unreasonable restraint on trade. There are, however, som e activities that are illegal under all circum stances, such as price fixing, m arket allocation and collusive bidding.

There is no safe harbor under the antitrust law for professional association activities. Therefore, association m eeting participants should refrain from discussing any activity that could potentially be construed as having an anti-com petitive effect. Discussions relating to product or service pricing, m arket allocations, m em bership restrictions, product standardization or other conditions on trade could arguably be perceived as a restraint on trade and m ay expose the SOA and its m em bers to antitrust enforcem ent procedures.

W hile participating in all SOA in person m eetings, webinars, teleconferences or side discussions, you should avoid discussing com petitively sensitive inform ation with com petitors and follow these guidelines:

• Do not discuss prices for services or products or anything else that m ight affect prices

• Do not discuss what you or other entities plan to do in a particular geographic or product m arkets or with particular custom ers.• Do not speak on behalf of the SOA or any of its com m ittees unless specifically authorized to do so.

• Do leave a m eeting where any anticom petitive pricing or m arket allocation discussion occurs.• Do alert SOA staff and/or legal counsel to any concerning discussions

• Do consult with legal counsel before raising any m atter or m aking a statem ent that m ay involve com petitively sensitive inform ation.

Adherence to these guidelines involves not only avoidance of antitrust violations, but avoidance of behavior which m ight be so construed. These guidelines only provide an overview of prohibited activities. SOA legal counsel reviews m eeting agenda and m aterials as deem ed appropriate and any discussion that departs from the

form al agenda should be scrutinized carefully. Antitrust com pliance is everyone’s responsibility; however, please seek legal counsel if you have any questions or concerns.

2

Presentation Disclaimer

Presentations are intended for educational purposes only and do not replace independent professional judgment. Statements of fact and opinions expressed are those of the participants individually and, unless expressly stated to the contrary, are not the opinion or position of the Society of Actuaries, its cosponsors or its committees. The Society of Actuaries does not endorse or approve, and assumes no responsibility for, the content, accuracy or completeness of the information presented. Attendees should note that the sessions are audio-recorded and may be published in various media, including print, audio and video formats without further notice.

3

Female

Male

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

16.0%

18.0%

20.0%

22.0%

24.0%

AverageTermination%

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

16.0%

18.0%

20.0%

22.0%

24.0%

AverageTermination%

TerminationbyAgeBand

Data Analytics in Health

4

Rate Month 1 Month 2 Month 3 Month 4

ANTICIPATED SALES TOTAL $(000) 750 200 500 1,500PERSONNEL (% OF TOTAL SALES) 110% 110% 110% 110%

Human Resources - Headcount 5 5 5 5 5

Human Resources - Cost 25.00 25.00 25.00 25.00

Commission 0.10% 0.75 0.20 0.50 1.50Personnel Total $(000) 25.75 25.20 25.50 26.50

DIRECT MARKETING (% OF TOTAL SALES) 100% 100% 75% 40%

Telemarketing (% of Direct Sales) 100% 50% 50% 50%

Human Resources - Headcount 3 3 1.5 1.5 1.5

Infrastructure Support 25 10 25 10

Commission 0.10% 0.75 0.10 0.19 0.30

Training 25 10 25 10Telemarketing Total $(000) 53.75 21.60 51.69 21.80

Internet Marketing (% of Direct Sales) 25% 25% 25% 25%

Human Resources - Headcount 1 0.25 0.25 0.25 0.25

Website Development (one-time cost) 500

Hosting 10 10 10 10

Support & Maintenance 25Internet Marketing Total $(000) 535.25 10.25 10.25 10.25

Direct Mail (% of Direct Sales)

Human Resources - Cost

Material 1000 1000 1000 1000

Postage 250 250 250 250Direct Mail Total $(000) 1,250.00 1,250.00 1,250.00 1,250.00

Direct Marketing Total $(000) 1,839.00 1,281.85 1,311.94 1,282.05

What is currently being done?

What can be done?

• Pricing

• Claims reserving

• Plan design modeling

• Trend forecasting

• Risk scoring

• Care management

targeting/savings estimates

• Stress testing

• Data reporting

2009 2010 2011 2012 2013 2014 2015 2016

Year

9.2%

9.9%

9.9%

11.4%

8.8%

7.8%7.9%7.4%

6.6%

8.7%

7.2%

8.2%

AnnualTrend

Applying Data Analytics to Business Problems

5

PredictivePrescriptive

Spectrum of data analytics: hindsight to insight to foresight

Valu

e to

war

ds b

usin

ess

solu

tions

What happened?

Why did it happen?

What will happen?

0

1

2

3

4

5

6

7

8

A B C D

Diagnostic

Descriptive

How do I influence what will happen?

0

1

2

3

4

5

6

7

8

A B C D

0

4

8

12

16

0 4 8 12 16 20

Y 1

X1

0

4

8

12

16

0 4 8 12 16 20

Y 1

X1

Adapted from Gartner’s Data Analytics Maturity Model

Types of Problems/Models

6

! = #+%&Linear regression/logistical modeling• Risk adjustment

• Plan choice modeling

• Product conversion

Survival/Markov models• Disease progression

• Claims reserving

Classification/clustering• Provider referral patterns

• Targeted marketing

• Fraud identification

• High claimant identification

Time Series• Trend forecasting

• Stress testing

Case Study: Choice Modeling

7

Current Plan Options

Actuarial Value

Enrollment

HMO 0.90 10%

High Value 0.86 42%

Medium Value 0.81 45%

CDHP 0.76 2%

New Plan Options

Actuarial Value

Enrollment

Plan A 0.87

Plan B 0.81

Plan C 0.75

Plan D 0.68

An employer group wants to change the medical plans it offers to employees

Case study for illustrative purposes only

8

Subject Matter Expertise is CriticalID Medical Coverage Coverage Tier Zip Code2 Medium Value Option Employee + 1 Dependent 950666 Medium Value Option Employee Only 980537 Medium Value Option Employee Only 606308 HMO Plan Employee Only 951219 Medium Value Option Employee Only 33472

14 HMO Plan Employee + 1 Dependent 9461015 Medium Value Option Employee Only 9410916 Medium Value Option Employee Only 6047817 HMO Plan Employee + 1 Dependent 9460721 Medium Value Option Employee + 1 Dependent 1137724 Medium Value Option Employee Only 9458725 HMO Plan Employee Only 9410926 Medium Value Option Employee Only 9460627 Medium Value Option Employee Only 9413328 Medium Value Option Employee Only 6053231 Medium Value Option Employee + 2 or More Dependents 9563532 HMO Plan Employee + 1 Dependent 9120633 Medium Value Option Employee Only 0245834 Medium Value Option Employee Only 9410336 Medium Value Option Employee Only 9812139 Medium Value Option Employee Only 6061445 Medium Value Option Employee + 1 Dependent 1060448 HMO Plan Employee Only 9412153 HMO Plan Employee Only 9412254 HMO Plan Employee Only 9535655 Medium Value Option Employee Only 3302658 Medium Value Option Employee Only 9410759 Medium Value Option Employee Only 9412361 HMO Plan Employee Only 9450163 HMO Plan Employee + 2 or More Dependents 9459569 Medium Value Option Employee + 1 Dependent 3270372 Medium Value Option Employee Only 9004673 HMO Plan Employee + 2 or More Dependents 9136777 Medium Value Option Employee Only 9458781 Medium Value Option Employee Only 2817382 HMO Plan Employee Only 9569185 Medium Value Option Employee Only 1057389 Medium Value Option Employee Only 94115

CurrentMedicalPlan

HMOs

MediumValueOption


Importance of Data Visualizations

9

20 25 30 35 40 45 50 55 60 65 70 75 80

Age

-2

0

2

4

6

8

10

12

14

16

18

20

22

24

26

28

30

32

34

36

38

40

42

YearsofService

CurrentMedicalPlan

CDHPOption

HighValueOption

20K 40K 60K 80K 100K 120K 140K 160K 180K 200K

Salary

0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

SupplementalLifeDeduction

CurrentMedicalPlan

CDHPOption

HighValueOption


Feature Engineering

10

$0K $100K $200K $300K $400K $500K

Salary

$0

$50

$100

$150

$200

$250

$300

$350

$400

$450

$500

$550

$600

$650

$700

$750

$800

TotalMonthlyDeductions(noMed)

CurrentMedicalPlan

CDHPOption

HighValueOption

CDHPOption HighValueOption

$0K $50K $100K $150K $200K $250K

Salary

$0K $50K $100K $150K $200K $250K

Salary

$0

$50

$100

$150

$200

$250

$300

$350

$400

$450

$500

$550

$600

$650

$700

$750

$800

TotalMonthlyDeductions(noMed)


Process of Predicting and Evaluating Choice

11

Age / Stage in Life

Risk Tolerance

Premiums/ Contributions

($ and % of Pay)

Expected Claims

Plan Design

Individual Plan Election

Individual Annual Total

Claims

Total Employee Cost Sharing for the Individual

Plan Cost

Modeling Approach – Estimating Parameters

12

Age / Stage in Life

Risk Tolerance


($ and % of Pay)

Expected Claims

Plan Design


Heterogeneous Logit Model• i – individuals• j - plan options• k - # of attributes with weights !ik

• "ij – utility of plan option j to person i"ij = #j + $j !i + %ij

!ik = !0k + !1k &ik + 'k (ik

• Monte Carlo simulation and maximize log likelihood function

Modeling Approach – New Choices

13

Age / Stage in Life

Risk Tolerance


($ and % of Pay)

Expected Claims

Plan Design


Know preferences (!, " and #) and

now changing the attributes ($)

• More Monte Carlo to estimate probabilities

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

%ofTotalCountofMedicalCoverage

42%

46%

9%

2%

CurrentPlanEnrollmentCurrentMedicalPlan

HMOs

HighValueOption

MediumValueOption

CDHPOption

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

%ofTotalCountofSelectedNewPlan

57%

22%

17%

5%

NewPlanEnrollment SelectedNewPlan

PlanA

PlanB

PlanC

PlanD

Model Results

14


$70.0M $71.0M $72.0M $73.0M $74.0M $75.0M $76.0M $77.0M $78.0M $79.0M $80.0M $81.0M $82.0M

TotalAllowedCost

$10.0M

$10.1M

$10.2M

$10.3M

$10.4M

$10.5M

$10.6M

$10.7M

$10.8M

$10.9M

$11.0M

$11.1M

$11.2M

$11.3M

$11.4M

$11.5M

SelectedPlanCostShare

Examining the Range of Results

15

Probable Outcome?

Extreme Circumstances?


PlanA PlanB PlanC PlanD

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

LossRatio

LossRatiosbyPlan

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

LossRatio

TotalLossRatio

Interpreting Results for Business Intelligence

16


Now What?

17

Evaluate the Model on New Data

Refine the Model

Add New Features/Variables

Prescriptive Analytics

2018 SOA Health MeetingMICHAEL XIAO, FSA, CERA, MAAA

Session 94 – Beyond Risk Identification: Predictive Analytics in Health

June 26, 2018

2

SOCIAL NETWORKANALYSIS IN HEALTHCARE

3

What is social network analysis in healthcare and how do we define a relationship?

PHYSICIAN TOPHYSICIAN

PHYSICIAN TOFACILITY

TWO MAIN RELATIONSHIP TYPES: 1) PHYSICIANS THAT SHAREPATIENTS WITH OTHER PHYSICIANS; 2) PHYSICIANS THAT SHAREPATIENTS WITH FACILITIES.

4

What is social network analysis in healthcare and how do we define a relationship?

PHYSICIAN TOPHYSICIAN

PHYSICIAN TOFACILITY

TWO MAIN RELATIONSHIP TYPES: 1) PHYSICIANS THAT SHAREPATIENTS WITH OTHER PHYSICIANS; 2) PHYSICIANS THAT SHAREPATIENTS WITH FACILITIES.

5

At what level to do we define a “shared patient”?PATIENT LEVEL EPISODE OF CARE LEVEL

VS

60% OF PATIENTS THAT RECEIVE CARE EACH YEAR HAVE AT LEAST2 EPISODES OF CARE PER YEAR. THERE IS A SIGNIFICANTLYCLEARER RELATIONSHIP OF CARE AT THE EPISODE LEVEL.

6

Clinically related claims for a single patient are grouped together across a period of time.

SOURCE: Internal Data.

7

Why is social network analysis for episodes important?

Pareto Principle of Healthcare (80/20) Roughly Applies to Episodes

Patients with 3 or more episodes are 20% of the population and account for 60% of the cost.

Episodes with 2 or more physicians are 30% of the episodes and account for 70% of the cost.


8


$200 $50,000


9

How do we use this information?

Drive Better Specialist and Facility “Referrals”

Convince Stakeholdersof Value

Understand Patient Migration Patterns

Understand Geographic Patterns of Usage

Use Cases

10

Dallas – Individual Physician to Physician View

SOURCE: Internal data.

11

Dallas - Physician to Facility View


12

How are communities defined? Louvain Modularity

Greedy algorithm that maximizes the modularity within communities and minimizes the modularity between communities

Small changes can result in very different communities, but the trade-off is acceptable run-time

SOURCE: https://perso.uclouvain.be/vincent.blondel/research/louvain.html

13

Dallas – Physician to Physician Efficiency View (Minimum Shared Patient Threshold)


LEGEND

Node size = Total cost

Green = efficient physicianRed = inefficient physician

14

Dallas – Physician to Physician View Detail


LEGEND

Node size = Total cost

Green = efficient physicianRed = inefficient physicianBlack = insufficient data

15

Dallas – Physician to Physician Alternative (Bad) View (No Minimum Threshold)


16

Houston – Physician to Facility Efficiency Interactive View


17

Houston – Physician to Facility Community Interactive View


18

Open Source Technology Stack

Gephi: static visualizations (11, 13, 14, 15)

Python [bokeh + networkx]: interactive visualizations (16, 17)

19

Questions?

session 94 panel discussion: beyond risk identification ... · adapted from gartner’s data...

Documents