1. 1
Big Data in Health:
The Global Burden of Disease Study
Peter Speyer
@peterspeyer
March 29, 2014

2. Big data in health
2
• Surveys
• Censuses
• Disease
registries
• Vital registration
• Verbal autopsy
• Mortuaries /
burial sites
• Police records
Variety Volume Velocity
• Hospital /
ambulatory /
primary care
records
• Claims data
• Surveillance
systems
• Administrative
data
• Literature
reviews
• Sensor data
• Social media
• Quantified self

3. IHME input data cataloged in the GHDx
3

4. From data to impact
4
Big data
Audiences
1. Data access
2. Data preparation
3. Data analysis
4. Data translation
Getting relevant data
in useful formats
to the right audiences

5. Example: Global Burden of Disease Study
• A systematic scientific effort
to quantify the comparative magnitude of
health loss due to diseases, injuries and risk factors
• GBD 2010 results published in
The Lancet in 2012
– 291 causes, 67 risk factors
– 187 countries
– 1990-2010
– By age and sex
• GBD 2013 Update in process
5

6. 1. Accessing the data
• Systematic identification of all relevant
data sources
– Data Indexers
– Lit reviews
• Challenges
– Data on paper, PDF, proprietary & obsolete
formats
– Patient / participant consent
– Confidentiality / de-identification
– Cost
6
Tristan Schmurr / Flickr
CC Chapman / Flickr

7. 2. Preparing data for analysis
• Data extraction
(databases, tables,
papers)
• Analysis of microdata
• Correction for bias
• Data quality issues,
e.g. garbage codes
• Cross-walks, e.g.
between ICDs
7

8. Demo: Tobacco Viz
8

9. 3. Analyzing data
• Use all available data
– Use covariates: indicators related to quantity of interest
• Test the modeling approach, e.g. predictive validity
testing (CODEm)
• Apply appropriate corrections, e.g. causes of death to
match all-cause mortality
• Quantify uncertainty
• Review: 1000+ experts, peer-reviewed publication

10. 4. Data translation
10
• Academic papers
• Policy reports
• Data search engine
• Data visualizations
– Input data
– Comprehensive results
– Key insights

11. Visualization demos
11

12. Parting thoughts
• Grab relevant data and get started
– Prep data with Data Wrangler, MS Power BI
– Visualize with Tableau Public, Google Fusion
– Learn to code: R, Python (analysis), JavaScript (viz)
• Check out IHME data sources and GHDx
Find resources & contact me at
speyer@uw.edu
@peterspeyer
http://healthdatainnovation.org
12