Fractal Regeneron Hackathon

Regeneron
Hackathon
Next Best Patient Type
17-12-2021

2
© 2021 Fractal Analytics Inc. All rights reserved | Confidential
Index
1. Problem Understanding
2. Solution Architecture
3. Exploratory Data Analysis
4. Feature List
5. Analysis
1. Clustering
2. Classification
6. Patient Profiles
7. HCP Target List
8. Impact

3
Problem Understanding
Eylea is an anti-VEGF drug approved by the U.S. FDA to treat wAMD, a common cause of blindness in the elderly
It’s observed that patients with wAMD continue the medication they started with as first line usage throughout their clinical
journey
BACKGROUND
Regeneron wants to identify the patient profiles who may be suitable for Eylea as the 1st line treatment and also identify target
HCPs to help enable the growth of Eylea as a brand of choice for 1st line treatment
OBJECTIVE
Areas of opportunity based on next best patient profile for Eylea which will result in growth in Rx and market share for
Regeneron
POTENTIAL BENEFITS

4
Frameworks
Solution Architecture
AI/ML
Framework
Features & Master Data Outcome
Patient
Profiles
HCP List
IQVIA
Data Plan Data
Providers Procedure
Codes
POS Code
Patients
Symphony
Data
Patients
Product Data
Cost Data
Outlets Data
Transactions
The proposed solution architect is an amalgamation of descriptive and predictive analytics to generate patient profiles based on their
demographic and clinical history and target HCPs based on their prescription pattern and network
Master
Data
with 50+
features
Raw Data

5
Total patients and claims distribution
Total Patients
Total
16.57 Mn
wAMD
0.90 Mn
Eylea (1L)
0.26 Mn
Total
161.50 Mn
wAMD
28.99 Mn
Eylea (1L)
2.81 Mn
Total Claims

6
Eylea is the 2nd most prescribed drug; has slightly high proportion in
Medicare vs other payer types
Only 581,678 patients out of 904,923 unique patients have NDC codes available for anti-VEGF
drugs. A particular person may have used multiple drugs
Most of the patients are covered under Medicare. A person may have multiple payers during their
treatment.

7
The gender and age distribution for patients using Eylea as 1st line
treatment is similar to the patients affected with wAMD

8
We created 50+ derived features using patient demographics &
claims history
Medical Features
Provider tier data
Claims Features General Features
Claims count of last 6-12 months
Claims amounts from last 6-12
months
Claims for wAMD and Eylea
patients
Medical conditions of patients (e.g.
Diabetes)
Records of other diseases except
wAMD
Patient demographics
Plan Type (Medicare or Medicaid)
Medical history (Injection counts,
Medication, etc.)

9
Cluster patients based on their first line of treatment
Multiple approaches explored for building profiles
Clustering
Classification models to predict patients suitable for
EYLEA first-line usage
Classification
Feature 1
Feature 2
Feature 3
…
Input
Prediction 1
Prediction 2
…
Output
Clustering could not segregate patient profiles i.e. EYLEA vs
Other drugs
Chosen Approach

10
Framework to create profiles for Eylea first line usage
Base Population:
Patients with pre-existing condition of wAMD
Model predicts EYLEA first line
usage for a patient
2018 & 2019 - training window (225306 patients)
Shaply Plots give us important features which helps
create member profiles
Plan Data
Providers
Patients Data
Claims Data
Diagnosis Codes
2020 - evaluation window (84103 patients)

11
Shap Plot explaining feature significance
• wAMD patient cost of Eylea first line users is
~1.3 times the wAMD patient cost for
members on other drugs
• Total count of wAMD claims for patients with
Eylea first line usage is 18% higher than
claims count for patients on other drugs
• Eylea is more common with members aged
75 and above
Patient Profiles
Cost incurred for wAMD treatment
Dosage amount
Total Medical cost for a patient
Count of wAMD claims
Indicator for Medicare members
Medclaims cost last 12 months
Tier 3 Flag
Preliminary Results

12
Prioritized HCP LIST
To be updated shortly

13
Impact
To be updated shortly

14
Appendix

15
Claims Made Year-Wise
Claims Data was Available from Sept. 2017 – Aug. 2021
There is no significant drop in the claims data even during the covid 19 period.

16
Column name Description Method of generating
Age Age of the patient Subtracting the Patients birth year from a
from a reference year
Gender Gender of a patient …
Eylea_firstline Patients who got treated with Eylea as First
as First line
Sorting on ‘service_date’ and then filtering for
filtering for ‘Eylea’ ndc code and then
dropping the duplicates based on patient_id
Other_Drug_firstline Patients who got treated with other drug as
drug as First line
Sorting on ‘service_date’ and then filtering for
filtering for ‘drugs other than Eylea’ ndc code
Eylea_flag Patients who got treated with EYLEA Filtering for ‘Eylea’ using ndc code and then
and then dropping the duplicates based on
based on patient_id
Features Creation

17
Column name Description Method of forming
wAMD_flag Patients who got diagnosed with w-AMD Filtering using provided icd codes for wAMD
wAMD on diag. codes column and then
then dropping the duplicates based on
on patient_id
Tier_1, Tier_2, Tier_3 Number of times a patient got treated with
with Tier 1, 2 and 3 providers respectively
Joining claims and provider data over
‘provider_id’ and ‘month_id’ and doing
doing groupBy on ‘patient_id’ while
aggregating for each tier counts
Medicare, Medicaid, Cash, Medicare_Part_B, Flags for the patient_ids who got treated
treated under any of these payers: Line of
Line of business
Joining claims and plan_type tables over
over ‘plantrak_id’ and filtering for each of
each of these payer-kinds and then dropping
dropping the duplicates based on patient_id.
Svc_chrg_amnt Total amount claimed by the patient for all
for all the claims filed by the patient
Doing groupBy on ‘patient_id’ while
aggregating for sum of svc_chrg
Contd…

18
wamd_svc_chrg Total amount claimed by the patient under
under wAMD condition
Filtering using provided icd codes for wAMD
wAMD on diag. codes column and doing
doing groupBy on ‘patient_id’ while
aggregating for sum of svc_chrg
Eylea_svc_chrg Total amount claimed by the patient for
for EYLEA medication for wAMD condition
Filtering for ‘Eylea’ using ndc code and doing
and doing groupBy on ‘patient_id’ while
while aggregating for sum of svc_chrg
Total_claims, wamd_claims
Wamd_eylea_claims
Total number of claims filed by the patient,
patient, under wamd condition and for eylea
for eylea medication respectively
Doing groupBy on ‘patient_id’ while
aggregating for count of Distinct claims. For
claims. For wamd, filtering on diag_grp_desc
diag_grp_desc and for eylea filtering on ndc
on ndc codes and proc_grp_desc
Conditions Flags (20 Conditions) Flags associated for a particular condition Condition flags are generated by matching
matching the ICD codes of those particular
particular conditions with 4 diagnosis code
code columns in the claims data
Contd…

19
Claims_count_0_x, Claims_count_0_x_eylea,
Claims_count_0_x_eylea,
Claims_count_0_x_wamd
Total claims count of the patients for the last 6
the last 6 -12 months, for the patients with
with wamd condition and for patients with
with eylea medication
Filtering on the service from date for the last
the last 6 and 12 months and then doing the
doing the necessary aggregations to convert
convert it from claim level to patient level.
Claims_cost_0_x, Claims_cost_0_x_eylea,
Claims_cost_0_x_eylea,
Claims_cost_0_x_wamd
Total claims cost filed by the patients for the
for the last 6 -12 months, for the patients
patients with wamd condition and for patients
patients with eylea medication
Filtering on the service from date for the last
the last 6 and 12 months and then doing the
doing the necessary aggregations to convert
convert it from claim level to patient level.
Injection count Totals injections taken by Patient Aggregating the injection counts at claims
claims level to member level
Have_ndc Wamd patients who have either ndc code
code available or proc_grp_desc in claims
claims data
Combining both ndc code and proc_grp_desc
proc_grp_desc column and then filtering
filtering patients with Non null values in the
in the new column.
Contd…

20
Clustering Fallbacks
Clustering was not leading to perfect segregation of the Eylea First Line
users into one cluster and other drugs users to different cluster
Cluster Cluster Size Eylea First
Line (Green)
%age
1 55,672 28,628 51.42%
2 538,548 133,406 24.77%
Distribution of Eylea First line users by Cluster
Cluster 1
Cluster 2

21
Eylea Other Drugs
Patients with wAMD treated with:

Fractal Regeneron Hackathon

More Related Content

What's hot (20)

Similar to Fractal Regeneron Hackathon (20)

Recently uploaded (20)

Fractal Regeneron Hackathon

Editor's Notes