Confidential #_x000D_
SDOH factors driving early vs late breast cancer diagnosis
Built teams from scratch
Working with marketing teams
Confidential #_x000D_
Analytical challenges
large number of variables
high multicollinearity
Confidential #_x000D_
Business challenges
Insistence on bivariate analysis
Have weekly calls with marketing teams
Confidential #_x000D_
Tell me about yourself
Aniket Bandhekar. Born and brought up in Mumbai. Pursued B.Tech. in Edible Oils Tech from I
Mumbai.
Got placed in Asian paints R&D lab in New Mumbai. My job was to create new paint products.
is here that I first realized my interest in analytical problem solving. I used to create a paint
using some recipe, conduct different test such as scratch test on the paint panel, hypothesize
what can be improved and how to improve it, tweak the paint recipe and try it out again.
After 1 year in Asian Paints, I joined IIM Lucknow for MBA. Here, I developed skills for using
analytics to inform strategy. Won lot of competitions - highlight being winning the HUL LIME.
Got placed in IBM. Worked in consulting domain for a few years. However, decided to pursue
analytics and made a switch to data science.
Post IBM, switched to ZS. Worked with the head of neurology portfolio for client and used
analytics to directly inform brand strategy. Covered nearly every aspect of commercial pharm
analytics like market landscape (TRx trends, NBRx trends for branded and generic), internal a
competitor performance (breadth and depth of writers, call volume analysis, growth and shar
as per LoT, patient segment, source of business and switches etc.), applied machine learning
techniques for use cases like patient identification for rare disease drugs, lookalike modelling
drivers analysis for market trends etc.
Got an opportunity to be a part of the global analytics capability center for Merck. Here, I
continue to work with global marketing teams to be their thought partner, understand busine
needs and use analytics to create insights and guide strategy. Besides, I have helped setup th
analytics center by hiring 40+ data scientists by conducting 100+ interviews within 6 months
have led the support for top MSD strategic priorties and multiple commercial teams. Currently
am leading a 10 member team to support brand strategy and performance insights for oncolo
and vaccines portfolio for Merck.
Confidential #_x000D_
Your work at Merck
Type of work
Built data science teams from scratch and enable them
Lead the brand strategy and performance insights team for oncology
and vaccines portfolio.
Work with marketing teams as thought partners, understand business
needs, translate business needs into analytical problem statements,
oversee the end-to-end analysis, create insights and guide business
decisions
Support top programs with analytics
Focus on building capabilities, products and achieving scale and
standardization
Confidential #_x000D_
Examples of work
Patient journey analytics
1. Patient demographics (Profile, symptoms, Dx at time of presentation,
comorbidities etc.)
2. Understanding Dx (HCP specialties who Dx and treat, screening tests before Dx,
who orders tests etc.)
3. Understanding treatment journey (Does the Dx HCP initiate treatment too, time
to treatment, adherence etc.)
Patient identification for rare disease
Physician lookalike modelling
Create test and control HCPs for a marketing campaign
Understand HCP Cx from using unstructured data
SHAP values based clustering
ZS work:
- Monitor market trends, competitor behavior, internal performance – identify areas
of concern and use analytics to solve them.
TRx and NBRx trends; Market growth (branded and generic); Growth and share as
per LoT, patient segment; HCP breadth and depth analysis; Call volume analysis,
SoB and switch analysis
- Redesign of patient drug affordability program
- Identified exclusion of drug for large insurance provider formulary
- Identified previously untapped patient opportunity
Late vs. early breast cancer Dx
dCEI
SDOH360
HEAT planning tool
KOL identification tool
Confidential #_x000D_
1
2
3
4
5
6
7
How do you prioritize your work?
Any challenges with this working model and how do you address it
Other challenges
Confidential #_x000D_
Learnings
Why looking for new job?
how do you manage team?
What are are looking for in new role?
Business problem statement/situation
task
analytical problem statements
Challenges
Steps taken
Confidential #_x000D_
Impact
assumptions of linear regression
Confidential #_x000D_
Workflow
Marketing teams -> Business engagement leads -> Us
Marketing teams come up with problem statement and communicate it to business engagem
We translate the business problem statements into analytical questions; design the analysis a
Coordinate with business engagement leads to present the findings to marketing teams
Work on any follow-up adhoc requests as needed
From a process perspective, all the projects are tracked through a JIRA dashboard
I understand the scope of work and assign resources as per expertise and availability
For each business request, we demand objective
Ask the business to assign priorities to the work
if some new work comes in and everyone is occupied, we are transparent and ask them what
Lots of things are lost in translation from marketing teams to business engagement leads
The overarching product strategy is not clear rather we are being passive recipients of requir
doing analytics in silo - This is again because we are not directly talking to marketing teams a
analytics are not done in house but rather done with vendors. So we dont know what all analy
done
All the workstreams are not talking to each other so there is lot of repetitive work
Lack of standardization, automation, dashboarding results in lot of time spent in doing repetit
Due to it being new team, work is not streamlined and hierarchy is not in place. Associates wo
with US teams with minimal guidance which results in QC issues
Business engagement leads being insistent on a particular methodology despite them not bei
Getting business buy in, not just getting it but sustaining it (Health equity)
Large sparse unstructured data for e.g., claims data. To address this, we use workarounds. Fo
Confidential #_x000D_
High vendor dependency. To address this, large scale initiatives for transition of repetitive wo
Got to work directly with global commercial teams head. In the previous roles, I have worked
Working in a highly ambiguous environement
Have come up with lot of proactive analysis - identify new patient opp, identify lack of insuran
We use innovative analytics techniques - using veeva link for content hyperpersonalization, te
Quickly picking up new areas - When I was in ZS, I was working on neurology brand analytics,
I have taken up lot of responsibilities here to grow quickly. I have helped setup the capability
supported multiple commercial teams, coordinated with global team leads to get work done i
ambiguous environment, and delivered consistently excellent results. I have now been entrus
responsibility to lead a 20 member team to deliver brand strategy and performance insights f
vaccines which constitutes 80% of merck business. However, I am still in a relatively juinior le
organizational hierarchy. I am expecting promotion in upcoming cycle but it will still be at a se
level. I am aspiring for a director level position and I think with my experience I am eligible fo
wont be able to reach there in the near future in Merck. Hence, I want to change job.
1. Have 1:1 sessions frequently to understand aspirations, issues faced, growth path, course
2. Enable them by encouraging to have trainings, externships, making presentations in front
and dealing with them directly
3. Inculcate critical thinking, asking the right questions and ensure that we are getting answe
4. Always talk about thinking about big picture
5. Keeping the growth path clear and ensure they are working towards it
Leading a large team of people driving analytics needs for a large section of an organization w
Ensuring that analytics team is not considered as a back end office but rather as equal counte
Ensuring that repetitive processes are automated and repetitive work is eliminated and thus t
Ensuring that we focus most of our time on next gen use cases - gen ai and llm (rep copilot),
Identify the sdoh factors driving late vs early diagnosis of breast cancer
We will do the analysis at a county level. For each county, we have the sdoh factors which wil
What is the patient cohort to be considered? - Time period, tumor status, look forward and loo
Dx codes for metastasis? Time gap between initial dx and metastasis diagnosis for it to be co
High multicollinearity
High # of variables (comorbidities (diabetes), demography (black pct), screening tests (mamm
Data cleaning (outlier removal, error value removal, imputation - with , column removal for hi
Feature removal was important as there were many features (>200), lot of features were rep
Confidential #_x000D_
5 techniques for iterative removal:
1. feature engg to combine features into 1 (comorbidity, screening related)
2. business sense based removal of features for which we don’t have a strong hypothesis or f
are non sensical
3. High correlation
4. VIF score
5. Lasso
Used multiple methods to determine important features - random forest, permutation feature
Output was used to do targeted interventions. Identify the geographies with the poorest late
linear relationship between y and x
error should be normally distributed which means that it is truly random and thus all the unde
no multicollinearity
homoscedasticity which means that the size of the error term should be same for all values o
Confidential #_x000D_
s (Business engagement leads are the demand planners and thought partners for us)
work priority is and should we close some other work and start this work
We are pushing for direct meetings with marketing
teams as much as possible however the working hours
are a roadblock here
I will be the SPOC for all requirements so that I can help
bring the synergies and realize scale and
standardization. I will also eventually oversee all
vendor delivery.
est and Control HCPs ZIP level analysis. Tracking sales at a zip level to analyze campaign impact. How
Confidential #_x000D_
vendor to inhouse/dashboard is conducted.
with VP level stakeholders and presented insights to team. In my new role, I will be co-ordinating direc
control HCPs zip level mapping, new techniques in breast cancer analysis
with digital, health equity, impact assessment and back to brand strategy with an expanded portfolio
contributing majorly to sales
n generating insights and driving strategy for the organization
efits of standardization and scale are realized.
g customer experience, hyperpersonalization etc.
x variables. Additionally, we will calculate the late diagnosis rate at a county level which will be the y v
period, should have a zip mapping etc.
a late diagnosis?
y), housing (# of rooms, # of people, housing problem pct., housing cost), transportation, education, i
ot were non sensical as well
Confidential #_x000D_
nce, drop column feature importance, gini importance based default feature importance, stepwise reg
and high patient population, check how it is performing on the top sdoh factors and determine the inter
ends have been captured
Confidential #_x000D_
mpaign impact. However, how to identify zips where the concerned HCPs are contributing to most sales
Confidential #_x000D_
co-ordinating directly with the Oncology and Vaccines global marketing teams.
expanded portfolio
which will be the y variable. We will use a regression model to identify the key sdoh factors.
tation, education, income (income inequality, median income) etc.)
Confidential #_x000D_
tance, stepwise regression etc. and all outputs were combined to determine final important features.
determine the interventions needed accordingly.
Confidential #_x000D_
tributing to most sales. Identified accounts which are contributing most to public sales of an analog an
Confidential #_x000D_
doh factors.
Confidential #_x000D_
al important features.
Confidential #_x000D_
c sales of an analog and used it as a proxy to identify top accounts. PIE analysis where we are learning
Confidential #_x000D_
where we are learning event predictors from 1 data and using it to predict events in another data.
Confidential #_x000D_
ts in another data.
Confidential #_x000D_
precision true positive / (true positive + false positive)
recall true positive / (true positive + false negative)
fpr false positive / (false positive + true negative)
ROC AUC curve is plotted TPR (recall) vs FPR (specificity)
Confidential #_x000D_
alse positive) out of all that were identified as true positive, how many were actually true posi
alse negative) out of all true positives, how many were correctly identified as positive
true negative) out of all negatives, how many were incorrectly identified as positive
(specificity)
Confidential #_x000D_
were actually true positive
ed as positive
as positive
