Data and Disease Occurence

2 vital concerns of epidemiology

• The quality of data available for describing the health of populations

• whether these data are being applied in an appropriate manner

Big data

• Vast electronic storehouses of information

  • Internet search transactions

  • Social media activities

  • Data from health insurance programs

  • Electronic medical records

Why do we need big data in epidemiology?

• may cover entire populations/large number of people

• can be analyzed to discover patterns of variables (distribution and determinants) associated with health outcomes

• Allows for insights into determinants of morbidity and mortality

3 V’s of Big Data

• Volume

• Velocity

• Variety

Volume

• Big data requires flexible and easily expandable data storage and management solutions

• ~2.5 quintillion bytes of data are created each day

Velocity

• Big data infrastructure makes it possible to manage data more flexibility and quickly than before

• Velocity measures how fast the data are coming in

• Some data arrive in real-time, others in batches

• Not all platforms will experience the incoming data at the same place

Variety

• There are many different formats of health care data

  • Structured and free-text data are captured in the electronic health records

  • Diagnostic imaging

  • Data streaming from social media/mobile apps

  • Video, text, pdf, graphics, wearable devices

• Due to the wide variety of big data, we often need to synthesize the many different formats of data

Data linkage

• Used to join data elements contained in a database by tying them together with a common identifier

  • ex: medical record number, social security number, any kind of unique identifier

Data mining

• Who?

  • Data brokers

• What?

  • Gathering and exploring large troves of data

• Why?

  • find patterns and associations in the data which are unrecognized previously

• How?

  • Google/Facebook/Apple/Uber/retailers who track user activity

  • Google Flu Trends (GFT) in the fall of 2008: aimed to provide earning for the flu before the CDC released any information, but it ended up overestimating the prevalence of influenza

  • Fitbit/Apple watches

Online sources of retrieval

• Extensive resources are available for online retrieval of epidemiologic information, and the number of websites seems to be growing exponentially

  • Google Scholar

  • CDC

  • MEDLINE

  • Websites of organizations and publications related to epidemiology

  • State Health department

factors that affect the quality of data

• quality is determined by:

  • sources used to obtain the data

  • how completely the data cover the reference population

• quality of data affects:

  • permissible applications of the data

  • types of the statistical analyses that may be performed

Four questions should be asked about quality of data:

1. What is the nature of the data, including sources and content?

2. How available are the data?

3. How complete is the population coverage? (Completeness of popualtion coverage refers to the degree to which the dta reflect a population of interest to a researcher)

4. What are the appropriate uses of the data

What is the nature of the data?

• Where do the data come from

• It’s important to have a reliable source

• Examples: vital statistics (birth/death records), surveillance data, reportable disease statistics, case registries, health care records, etc

How available are the data?

• The investigator must have access to the data

• Release of personally identifiable information is prohibited in the US

  • Health Insurance Portability and Accountability Act (HIPAA)

  • Identifiable data must not be released without the person’s consent

• Data banks may release unidentifiable data

The Health Insurance Portability and Accountability Act of 1996 (HIPAA)

• Protects personal information contained in health records

• Data banks that collect information from surveys may release epidemiologic data as long as individuals cannot be identified

How complete is the population coverage?

• Completeness of population coverage refers to the degree to which the data reflect a population of interest to a researcher

  • Remember our sampling strategies—we want a data sample that will accurately represent the population of interest

• Representativeness (aka external validity) refers to the generalizability of the finding to the population from which the data have been taken

What are the appropriate uses of the data?

• the quality of your data may define what study design you must use for your study

  • ex. if your data cannot be used to estimate incidence of disease, you may have to conduct a cross-sectional or case-control study

U.S. Census Bureau

• A census of the population is taken every 10 years (1980, 1990, 2000) calculate the population size and other characteristics

• The Census Bureau also calculates the estimates of population size in the years in between the decennial

Vital events registration- Death records

Limitations to Death data

Birth statistics

Limitations to birth data

Surveillance data

Syndromic surveillance

CDC’s Nationally Notifiable diseases