Oct 2008 Washington DC Meeting

Meeting Minutes

When and Where: Oct 21^st 2008, World Bank, Washington DC. This meeting was an opportunistic gathering of expert group members who were already attending the ICE on Injury meeting in Washington DC. Our meeting was hosted by the World Bank Global Road Safety Facility.

Attended by: Jerry Abraham, David Bartels, Kidist Bartolomeos, Kavi Bhalla, Lois Fingerhut, Belinda Gabbe, Juanita Haagsma, James Harrison, Francisco Lopez-Valdes, Ronan Lyons, Maria Segui-Gomez (via Skype), Saeid Shahraz, Margaret Warner

Meeting Presentations

Data Hunt

ISSUES DISCUSSED

1. Group publications:

• Draft authorship guidelines for articles flowing out of the joint work of the expert group are now posted on the website. These were briefly discussed and there was general agreement with the principles outlined.
• In the interest of transparency, discussion topic leaders are encouraged to develop their papers online to the extent possible. The potential problem of journals not accepting materials that had been previously posted online was briefly discussed. James suggested asking a couple of journal editors whether this was a problem. Volunteers are needed to investigate.

2. Progress on data hunt (Discussion Topic 2):

• Kavi described the ongoing environmental scan of all relevant data sources that can inform GBD estimates. This search is showing that there are many untapped data sources that exist even in regions that are traditionally believed to be information poor (e.g. Sub-Saharan Africa).
  • Access to the data sources identified in the environmental scan is the big road block currently facing this work. We currently only have a very tiny fraction of the datasets that we know exist. How to remedy this:
    • All group members should use their connections to help improve data access for such work. At the meeting: Belinda mentioned their connections in Hong Kong, India, and Mexico. Lois mentioned the connections of ICE with Nicaragua, El Salvador, Colombia and Taiwan. Kidist mentioned work that WHO has done in many developing countries. Appropriate access to DSS/INDEPTH data is needed.
    • Call for contributions should be circulated on various relevant newsgroups, listservs, discussion forums, etc. Kidist and Lois have volunteered to put together a message for posting on the ICE Listserv. This message could serve as the template for messages distributed more widely.
    • We are working on establishing relationships with WHO and the World Bank so that we can request access through their country offices.

3. Progress on Definitions:

- Discussion Topic 1: Case Definition:

This topic is led by John Langley and Ronan Lyons, and the work was represented at the meeting by Ronan. He summarized progress and outcomes as follows. A series of e-mail exchanges has resulted in a proposed approach to definition which adheres to the principles for defining injury that have emerged from discussions in the Injury ICE and are stated in papers authored by John and others (e.g. Cryer & Langley Injury Prevention 2008;14:74-77). In outline, the approach limits attention to physical injury and is intended to include all cases incident in a population (i.e. it is designed to be immune from selection bias related to service access or availability). An inclusion criterion proposed by John was any condition of a type codable to the injury chapter of the ICD which resulted in one day or longer of reduction of usual activities. This threshold was the subject of discussion at the meeting. Issues raised included: the rationale for having any threshold; and ways in which the definition chosen might interact with characteristics of the main data sources on which the project will rely (esp hospital inpatient data and population surveys). Kavi reported that Saeid Shahraz has now inspected dozens of population-based surveys. While case definitions vary, most often they are framed quite loosely, and without any reference to a threshold (e.g. “were you injured…?” [due to some event in some time period]. On this basis, a case can be made to frame the study case definition in a similar way, since the advantage of greater specificity inherent in inclusion of a threshold would be moot if the definition must be applied to data that are defined in a less specific way.

We face a choice between having a theoretically sound definition (e.g. threshold of injury reducing normal activity for at least 1 day) but almost no data sources where this threshold has been made explicit to respondents (particularly in less affluent countries) or accepting the existing survey data and seeing whether it is possible to deduce whether it does/does not meet such a threshold. This could be based on other information within the survey, for example, distribution of ‘lay diagnoses’; remembered sprains or breaks will always exceed this threshold. Also it might be possible to compare information between surveys which have, or do not have, additional information on which to make such judgments. We need to use reasoned argument as well as the existence of specific data on threshold to judge whether the threshold is or is not met. A summary table of the questions lists of lay diagnoses (if present) and quoted incidence for all the surveys would be very helpful in coming to these decisions.

Injury "Sequelae" Definitions

Data Analysis - 1

- Discussion Topic 3: Injury state (“sequelae”) definitions:

James gave a presentation in which he outlined the nature of this task and the progress made (see web-site for a copy of the presentation). Discussion followed on several injury states concerning which substantial changes are proposed, in comparison with the approach taken in earlier GBD project (n.b. spinal injury, traumatic brain injury and severe burns), and on three elaborations of the criteria used to distinguish health states. The elaborations are (1) greater use of distinctions based on severity; (2) introduction of a set of priority health states, for which a data extraction algorithm will be used that selects cases on the basis of diagnosis codes in first or subsequent code positions (3) introduction of a treated/untreated distinction for certain conditions, which are believed to have strong treatment effects. James will provide further information on the process and outcomes of this aspect of the project via the Web-site.

Data Analysis - 2

4. Progress on data analysis:

- Survey data analysis:

• The analysis of survey data has emerged as a priority focus area for the expert group. Saeid Shahraz has been conducting a systematic review of all household surveys that included questions on injuries. He described results from his analysis of the World Health Surveys (60+ developing countries). The group discussed various technical issues related with recall bias, telescoping.
• We envision that the incidence of non-fatal injuries is the primary survey output of interest for our group especially since hospital and ER data with population wide coverage is not available for most regions. Where both (surveys + hospital data) are available, we need to do comparisons of incidence derived from both. It was decided that such comparative work will be done by Margy (using USA data), Fran and Maria (using Spanish and European data) and Saeid (using data from several developing countries).

- Hospital data analysis:

• GD external cause definitions are finalized (i.e. Discussion Topic 12 External Cause List is now closed for discussion)
• GBD “sequelae” definitions are almost finalized (see Progress on Definitions, above).
• Data extraction scripts are now being developed to extract data for the GBD project from hospital administrative data. These scripts aim at producing tabulations of GBD sequelae by GBD external cause groups.
  • A few key issues relating with the mechanics of data extraction were discussed. The scripts should:
    • Case inclusion/exclusion criteria:
      • Eliminate fatal cases from hospital tabulations
      • We need to eliminate multiple admissions for the same injury. Thus we limit tabulations to patients presenting at first point of contact with the healthcare system (e.g. "emergencias" for Mexico)
      • Include case if any of first three diagnosis codes are S&T (i.e. do not limit to first diagnosis).
    • Dealing with multiple “sequelae”: For each case, only one GBD “sequelae” will be picked based on a priority ranking developed along with the GBD sequelae definitions (above).
    • New version of GBD mapping will eliminate codes T79 and greater because these codes are sequelae codes themselves
    • If nature of injury and external cause codes are not linked, tabulate separately and generate external cause for obvious nature of injury codes (e.g. burns, poisoning, drowning, etc.)
    • If data from 2005 is >10,000 cases, use only 2005 data. Otherwise, generate tabulations for additional years (2002, 2003, 2004, 2006).

5. Progress on Discussion Topic 5 Dealing with multiple injuries: This topic is being led by Belinda Gabbe. Belinda Gabbe and Margy Warner provided an update to the group. A draft of the discussion document is under preparation and will be circulated to GBD members with a known interest in the topic shortly. Some of the key points discussed were:

· The approach to dealing with multiple injuries may differ for mortality and morbidity.

· Limitations of death certificate recording of multiple injuries

· Reliance on, and methods for identifying the, principal diagnosis in hospital morbidity datasets may not identify the injury with greatest risk of disability

· Some evidence in the literature that the worst injury is most important for mortality (though consensus on this issue has not been reached) but there is a growing body of literature to suggest that multiple injuries result in worse functional and health-related quality of life outcomes

· If the GBD project is to take account of multiple injuries then we need to:

o Establish key injury combinations that result in significantly greater disability or risk of death than individual injuries for inclusion as injury sequelae

o Establish the methods for allocating disability weights where there are multiple injuries using the full range of ICD diagnosis codes

§ Multiplicative methods assume independence between the diagnosis codes but this is unlikely given the patterns of injury associated with particular mechanisms

§ More complex methods assuming dependence of codes have been used for comorbidity in the Australian Burden of Disease Study 2003.

Outcome: Belinda will post a summary of the discussion document on the website and send the full draft to all interested contributors. We need to identify datasets with disability data and all ICD diagnoses to explore the issues discussed above.

6. Progress on Discussion Topic 7 dealing with unspecifieds: Summary recommendations for handling unspecifieds are posted on the website.

http://sites.google.com/site/gbdinjuryexpertgroup/Home/discussion-7-dealing-with-unspecifieds This document recommends age-sex proportional redistribution of unspecified categories over the corresponding specified categories. The group agreed that this is a good starting point for handling unspecifieds but further investigations are needed. As a first step, analysis of the use of injury-specific dump codes in the WHO Mortality database is being undertaken. This work will mimic the 2005 paper by Mathers et al. (Counting the Dead) in the WHO Bulletin but with a focus on injury dump codes.

In depth investigations are needed to see what causes may be classified to X59-Unspecified accidents. Lois and Margy will investigate with US data.

Dealing with Unspecifieds