Eric Carlson June 8, 2018 Actuaries Club of the SouthWest Milliman IntelliScript Underwriting with Rx Based Models
Eric Carlson
June 8, 2018Actuaries Club of the SouthWest
Milliman IntelliScript
Underwriting with Rx Based Models
Agenda
Rules Engine and Mortality
Simplified Issue Findings
Predictive Model and Mortality
Prescription Data and Mortality
Combining Predictive Models and Clinical U/W
Proprietary and Confidential
The Future of Underwriting …
Electronic requirements (Rx, Medical Data, MIB, MVR, Credit …)
Decision engines driven by data
Predictive Models
Automation
Increasing
APS, Labs
Cycle times
Costs
Decreasing
Better Customer Experience
Proprietary and Confidential
Multiple Levels of Data
Obtain the authorization
Submit the query
Review the results
Clients
Health Plan
PBM
Clearing
House
Retail
Pharmacy
Data Sources
Proprietary and Confidential
Prescription Histories
1PrescriptionBrand and generic name | Dosage and quantity | Date of fill
2PhysicianSpecialty | Contact information
3PharmacyContact information
4Dates of eligibilityWith or without prescriptions
5Underwriting significance indicatorRed, yellow, green
Proprietary and Confidential
Mortality Study Timeline
2009
Milliman / RGA study
1M exposure years
2,500 deaths
2012
Milliman study
21M exposure years
45,000 deaths
Began to validate
and expand Irix
2015
Milliman study
53M exposure years
231,000 deaths
Created Risk Score
2017
Milliman study
104M exposure years
469,000 deaths
Update Risk Score
model
Proprietary and Confidential
Rx: Relative Mortality by Maximum Drug Priority
0%
5%
10%
15%
20%
25%
30%
35%
40%
20%
40%
60%
80%
100%
120%
140%
160%
180%
No Hit Eligibility Only Green Yellow Red
Exposure
Rela
tive M
ort
alit
y
What is a rules engine?
Irix™UW Guidance
Conditions
Severity
Decisions
Rule Variables
Indication / Therapeutic class
Drug combinations
Fill timing (date or duration ranges)
Physician specialty / count
Gender / Age
Diagnosis / Procedure combinations
Drug / Diagnosis combinations
Other variables
Data Input
Rx
Medical Data
Application Data
Other (MIB, MVR …)
Proprietary and Confidential
Irix – Insulin Matters
Diabetes
165% relative mortality
Non insulin dependent
128%
Insulin Dependent
269%
Proprietary and Confidential
Irix – Context Matters
Zofran (ondansetron)
200% relative mortality (RM)
Pregnancy
OB/GYN
58%
1 isolated fill
1 Fill
160%
CINV
Oral Steroid
751%
Proprietary and Confidential
Irix – Drug Combinations Matter
Spironolactone
243% relative mortality
Thiazide Diuretics (177%)
Ace / Angio II (ARBS) (119%)
Beta Blocker (137%)
With 2 out of 3 of:
338%
Thiazide Diuretics (177%)
Ace / Angio II (ARBS) (119%)
Beta Blocker (137%)
Without 2 out of 3 of:
169%
Proprietary and Confidential
Irix – Morphine Equivalence Matters
* MED = Morphine equivalent dosage
Opioids
113% relative mortality
Low MED*
109%
High MED*
346%
Proprietary and Confidential
What is an Rx-based predictive model?
Holistic Multi-variate Rx Model of Mortality Risk
Predicts Relative Mortality of a Life or Group of Lives
Delivered via a Rules Engine
Proprietary and Confidential
0%
100%
200%
300%
400%
500%
600%
700%
800%
0
500,000
1,000,000
1,500,000
2,000,000
2,500,000
3,000,000
3,500,000
4,000,000
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0+
Rel
ativ
e M
ort
alit
y
Live
s
Relative Mortality and Lives
Risk Score 2.0 - Lives Risk Score 2.0 - RM
Models can predict relative mortality accurately.
142017 Milliman Mortality Study: 25M lives, 15M Rx hits, 469K deaths Proprietary and Confidential
Predictive Model Benefits
Evidence based and data driven
Stratify risk within a given medical condition
Detect unintuitive patterns
Quickly and consistently interpret large amounts of data
Easy to test, implement, use, and update
Proprietary and Confidential
Predictive Model Challenges
Limitations of data or lack of data
How to reflect low-incidence risks?
Operational challenges
Change in underwriting
Field underwriting more difficult
Client / agent communication challenges
Proprietary and Confidential
Risk Score stratifies platelet inhibitor risk.
0%
50%
100%
150%
200%
250%
300%
350%
400%
0
15,000
30,000
45,000
60,000
75,000
90,000
105,000
120,000
x < 1 1 ≤ x < 1.5 1.5 ≤ x < 2 2 ≤ x
Risk Score Range
Very Serious Platelet Inhibitor (Plavix)
Lives
Relative Mortality
Proprietary and Confidential
Risk Score stratifies risk within conditions.
19
0%
50%
100%
150%
200%
250%
300%
350%
400%
450%
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
0.00 < x < 0.50 0.50 < x < 1.00 1.00 ≤ x < 1.50 1.50 ≤ x < 2.00 2.00 ≤ x < ∞
Rela
tive
Mo
rta
lity
Liv
es
Diabetes Third Line with Insulin
Risk Score 2.0 Lives Risk Score 2.0 RM
Proprietary and Confidential
SI Case Study Background
Mostly auto-decision via Irix
Risk Score as of time of underwriting
Have deaths on issued and declined cases
Proprietary and Confidential
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f Li
ves
Risk Score
Risk Score Distribution by UW DecisionSI Case Study - Hits Only
Lives - Issue Lives - Decline
SI Case Study – Distribution of Lives
Issue
Decline
Average Score (Hits Only)
Issue 0.96
Decline 1.52
Proprietary and Confidential
0%
100%
200%
300%
400%
500%
600%
0.00 ≤ x < 1.00 1.00 ≤ x < 1.50 1.50 ≤ x < 2.00 2.00 ≤ x < 3.00 3.00 ≤ x
Rel
ativ
e M
ort
alit
y
Risk Score Range
Relative Mortality by Risk Score and UW DecisionSI Case Study - (Hits Only)
Issue Decline
SI Case Study – Relative Mortality
Decline
Issue
Proprietary and Confidential
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f Li
ves
Risk Score
Lives - Issue Lives - Decline
Thresholds can be adjusted to achieve desired business results.
Low ThresholdHigh Threshold
Proprietary and Confidential
Some issued premium now gets declined
Equal amount of declined premium now gets issued
Set Risk Score threshold to issue the same amount of business.
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
New Issues New Declines
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
New Issues New Declines
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
Lives - Issue Lives - Decline
Threshold
83%
Before Risk Score
75%
After Risk Score
Issued Cases Relative A/E
Same amount of business issued
$4 Million increase in profit
9% Mortality improvement
Proprietary and Confidential
Set Risk Score threshold to maintain the same mortality A/E.
Some issued premium now gets declined
More declined premium now gets issued
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
New Issues New Declines
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
New Issues New Declines
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 > 4.0
# o
f A
pps
Lives - Issue Lives - Decline
Threshold
$56.1 M
Before Risk Score
$66.0 M
After Risk Score
Premium Issued
Same mortality A/E
$2.9 Million increase in profit
18% More issued business
Proprietary and Confidential
Optimal solution combines clinical underwriting with the power of predictive modeling.
Proprietary and Confidential
Paradigms
Clinical Underwriting
Condition based
Univariate
Uses clinical expertise
Predictive Model
Statistical basis
Multivariate analysis
Single risk metric for each case
Underwriting decision