Big data matters Maastricht University: Johan van Soest, Michel Dumontier CBDS: Bob van den Berg, Ole Mussmann

2. Sharing is caring
Maastricht University: Johan van Soest, Michel Dumontier
CBDS: Bob van den Berg, Ole Mussmann

3. Sharing is caring : Introduction
1. Why are we doing this and what is our goal?
2. What is our problem and how do we intend to solve it?
3. What is our approach and who are involved?
4. How does it technically work?
5. What have we learned so far and what’s next?

4. − The Maastricht Study is an extensive phenotyping effort to understand health in 10,000
individuals, with a focus on type 2 diabetes.
− Goal is to examine complications (e.g. cardiovascular disease), comorbidities (e.g.
depression, cognitive decline, and gastrointestinal diseases), and health utilization (e.g.
hospitalization).
− Research Questions:
– How does lifestyle (e.g. physical activity) affect morbidity, hospitalization, mortality?
– How do environmental factors impact lifestyle (e.g. physical activity) and health?
– How does the social-economic factors (e.g. neighbourhood income) affect individual
income, and in turn health?
Sharing is caring : The Maastricht Study

5. Sharing is caring : The Maastricht Study
To answer these questions, we need data
We need to combine
– deep clinical phenotyping data from the Maastricht Study
– regular health checkups and hospitalization via MUMC+
– and socio-economic and environmental data from CBS

6. Sharing is caring : The Maastricht Study
And:
We need to address social, legal, ethical concerns
– Legal: Are we allowed to combine data in this way?
– Ethical: Is it ethical to combine data in this way?
– Social: How do we encourage data sharing in a manner
that reduces the unproductive burden on researchers?

7. @micheldumontier::MedicalOncology:2017-09-
04
A set of 15 principles that apply
to any digital resource and their
metadata.
data, data repositories, software,
web services, scholarly
publications
7

8. Rapid Adoption of Principles
As of Sept 2017,
200+ citations since 2016
publication
Included in G20 communique,
EOSC, H2020, NIH, and more…
8
nature.com/articles/sdata201618

9. Sharing is caring : opportunity
Combining different data sources from
different owners dramatically
increases their value

10. Sharing is caring : challenge
Can we automate learning on
complementary data, while respecting
participants’ privacy?

11. Sharing is caring : problem definition
Clinical trials:
3-5% of hospital patients
95% of variables
Registries:
99% of hospital patients
10% of variables
Available data element
Missing data element
Data elements (variables)Patients

12. Sharing is caring : problem solution
Complementary
data analytics
Distributed
machine
learning
Probabilistic
record
linking
One-way
encryption,
hard matching
Development route
ValidateValidate
Without data
transport
With data
transport
Without patient
identifiers
With patient
identifiers
ValidateValidate

13. Sharing is caring : our approach
- Three interlocking work packages
- Technical
- Ethical, Legal and Society Issues (ELSI)
- Science
- Parties agree upfront on
- Data to be used and quality of the data
- Application of the FAIR principles
- Acceptance criteria for the outcome of the study
- Privacy by design: no access to the data for involved parties

14. Health, Ethics and Society
David Townend
+ ELSI Team
Maastricht Study
Annemarie Koster
+ MS team
CBDS
Marco Puts
Ole Mussmann
Bob van den Berg
Maastro Clinic
Andre Dekker
Johan van Soest
Sharing is caring : who are involved
Institute of Data Science
Michel Dumontier
Claudia van Oppen

15. Sharing is caring : how does it work?
Both UM and CBS have a
sensitive dataset

16. Sharing is caring : how does it work?
Some salt is added

17. Sharing is caring : how does it work?
The identifiers are being
hashed in the same way…

18. Sharing is caring : how does it work?
Both datasets are transferred
to a trusted third party

19. Sharing is caring : how does it work?
Data is combined, based on
the hashed identifiers

20. Sharing is caring : how does it work?
The result of the analysis is
shared with both parties

21. Sharing is caring : results
− Our technical proof of concept works,
– We can automate the process using proper security
measures
– Only the required data elements leave the institutes
− Our solution fits the constraints regarding patient consent
(UM) and compliance with the CBS Law (CBS).
− Groundwork for more advanced methods & algorithms

22. Sharing is caring : lessons so far
− Technology can not solve all trust issues
− Easy questions (e.g. plot age & income) can require complex
solutions
− A trusted third party is needed. For this project, CBS will also
perform that role

23. Sharing is caring : what’s next
− 3yr funding obtained as part of the National Science Agenda
(NWA) Work Package 5 - Accessibility & Interoperability
– Synergistic interactions with other WPs
− Keen to explore other opportunities including public-private
partnerships
– Learning from health and population data across
geopolitical regions
– Learning from public and private data (early drug
discovery and clinical trials)