Global burden of disease (GBD) estimates help us understand how disease, injury and risk factors impact health at both the population and individual levels. Specifically, the GBD measures the prevalence and impact of fatal and non-fatal conditions at the country (and sometimes sub-national) level, as well as the underlying causes for these conditions. Researchers from the World Health Organization, Harvard University, and the World Bank started estimating the GBD in the 1990s, capturing 107 diseases and 10 risk factors. More recently, GBD estimates are built out by the Institute for Health Metrics and Evaluation (see the GDB 2010) The forthcoming GDB 2013 estimates will cover 303 diseases and injuries, 2,385 sequela, and 71 risk factors—resulting work of the contributions of over a thousand experts.
I recently attended the IHME GBD Technical Training Workshop to learn more about the GBD approach and the analytical methods used in the GBD studies. The amount of information that goes into the GBD is pretty amazing (I collected enough materials at the training to fill half a suitcase—and that was only on the methods). The result of IHME’s GBD work is a wealth of information on the composition, extent, and contributing factors to disease around the world (see the collection of papers IHME has published here).
I left with some observations, and remaining questions about the processes:
Data availability and quality is still a significant obstacle: Critical information on disease burden in many countries is non-existent or unavailable, making it difficult to know and understand health trends. Using IHME visualization tools, you can see the clear lack of available data regarding deaths caused by HIV in Zimbabwe, tuberculosis in Malawi, and diarrheal diseases in Sierra Leone—just a few of many examples. In the absence of sufficient or accurate data, disease burden is estimated using other variables such as regional trends or demographic and socioeconomic indicators. These modeling exercises attempt to counteract systematic biases or inaccurate reporting in country collected data. At times the adjusted differences can be significant— for example, see the differences in IHME mortality estimates, and Sudan and Pakistan country level data here.
Victoria Fan has written about some of the issues with uncertain and inaccurate raw data that underlies many health statistics here. This process reveals how little is known about mortality and causes of death in many countries. Will it serve as a driving force to encourage countries to produce more accurate and comprehensive health data? Limitations in the quality and coverage of administrative data is an issue we review in CGD’s Data for African Development Working Group report, due out on July 8.
Disability weights reveal some surprising preferences: Disability weights are used to quantify the effects of non-fatal health outcomes and compare the severity of different outcomes through the calculation of disability-adjusted life years (DALYs). IHME has generated their own data using survey tools to determine population preferences—quantifying how people feel about impact on quality of life of different diseases. At times this leads to counter intuitive results—for example the percentage of global burden of disability (in terms of years lost to disability) for acne is greater than that of each autism, chronic kidney disease, multiple sclerosis and HIV in developed countries. This is clearly a question of prevalence compared to severity. Another striking example is the relatively low ranking of cognitive disabilities based on population survey methods. For example, severe intellectual disability has a lower disability weight than distance vision blindness, moderate anxiety disorders, moderate alcohol use disorder, or a long term knee dislocation (with or without treatment), among others. See the full list of disability weighs from GBD 2010 here.
Incorporating new information can complicate policy making: As new data and modeling methods become available, IHME GBD estimates are revised, both for the current year and all previous years. These revisions are important but their effects can be complicated for policy making—for example, the World Bank’s recent release of new purchasing power parity numbers or recent GDP rebasings in sub-Saharan Africa. This raises the questions: how quickly should country priorities be adjusted in the case of a substantial revision in estimates? And could such data be used to monitor performance or for impact evaluation purposes?
The GBD method provides a tool for ‘crossing the road’: While jay-walking could be a potential risk factor included in the GBD—IHME uses what they call the ‘cross-walk’ tool to link data sources and pool study results that report on multiple related indicators. The tool helps address issues that arise when data exists in different forms by converting findings to comparable formats (i.e. a reference definition) using empirically observed relationships between the reference format and non-standard formats. It’s an pragmatic tool that might have applications outside the GBD analysis—for example, in systematic reviews of public health interventions or other development interventions where indicators are varied.
In whole, I learned that calculating the GBD is a complicated process and is certainly a feat of coordination and brainpower (for an indication of the complexity of the methods see the ‘summary flow chart’ here). The first installments of GBD 2013 have already been released (see here, here and here) and the remainder will be available by fall 2014.