Data Analysis - Overview

Please note that the GBD-2010 Injury expert group no longer exists.

- Please visit the website www.globalburdenofinjuries.org to find out more about other closely related collaborations of our group members.

- If you are interested in the ongoing work of IHME's GBD project, please visit their official website http://www.healthdata.org/gbd.

Last updated: March 17th 2009

Summary: Data sources for estimating the incidence of deaths and non-fatal injuries do not exist in many regions of the world. Thus data analysis in the GBD project relies heavily on extrapolating from regions where plenty of data is available to regions where information is scarce. The figures posted on this page provides a first sketch of what this process could look like. The application of these methods to estimate incidence in regions with incomplete data sources will introduce substantial uncertainty. This early sketch of the extrapolation process is provided here primarily to enable suggestions for improvements. Please contact Kavi and/or James with feedback.

Analytical Tools for Estimation

The extrapolation process proposed here relies on the use of three analytical tools:

• Injury pyramids: This is the ratio of deaths to non-fatal injuries. These will be age-sex specific.
• External cause to health state mappings (e-code => n-code mappings): This is the distribution of health states (TBI, long bone fracture, etc) for each external cause (pedestrian crash, fall, etc). Again age-sex specific
• Probability of admission: This allows mapping between incidence of injury hospitalization to incidence of injuries.

Data Availability Scenarios

Scenario 1 - death registration, hospital records, survey

Scenario 2 - death registration, hospital records

Scenario 3 - death registration, surveys

Scenario 4 - death registration, hospital records with unknown coverage

Scenario 5 - death registration

Scenario 6 - Hospital records

Scenario 7 - Survey

Scenario 8 - no data

Issues under further consideration

• In the presence of complete hospital data and a national health survey– which is better to use for incidence of non-fatal injuries – surveys or hospitals?
• Developing the tools
  • Characterizing probability of admission
    • Should be age-sex-injury (ncode) specific
    • Function of regional income?
  • Injury pyramid:
    • Is it better to build injury pyramids in terms of YLL and YLD intead of deaths-injuries?
    • How much does this vary by region, age and sex?
  • E=>N mappings
    • How much do these vary by region, age and sex?
• Predicting deaths in Scenario 7 (only survey data is available) – What is more reliable: CODMOD or estimating deaths from non-fatal injuries?
  • This should be checked in Scenario 1-4