global burden of disease phe contribution to gbd project
TRANSCRIPT
Global Burden of Disease
PHE contribution to GBD project
GBD – what is it?
Presents estimates of all-cause mortality, deaths by cause, years of life lost, years lived with disability, and disability-adjusted life years by country, age, and sex
Originally conceived within WHO at much smaller scale – first report in 1993 Project taken to University of Washington with Chris Murray – creation of the Institute of
Health Metrics and Evaluation. GBD 2010 first output.
Over 90% of funding comes from Bill and Melinda Gates Foundation.
GBD 2013 expands upon methodology, datasets, and tools presented in GBD 2010
GBD 2013 includes 303 diseases and injuries estimated, 2,585 sequelae; attributed to 69 risk factors
most comprehensive effort to date to measure epidemiological levels and trends around the world (188 countries to date)…
Why is PHE interested?
Provides international benchmarks of health burden
•The Global Burden of Disease UK paper in the Lancet in 2013 very powerful for policy
•However:
• Only at UK level
• Previous data was not provided to IHME in any systematic way
•PHE facilitated sub-national estimate production at UK statistical region for 2014 release
•Future iterations looking to get more granular outputs
What are the main GBD outputs?
Main results are presented in terms of disability-adjusted life years (DALYs), a time-based measure that combines years of life lost due to premature mortality (YLLs) and years lived with a disability (YLDs), metrics that were specifically developed to assess the burden of disease.
One DALY can be thought of as one lost year of "healthy" life.
Analytical Principles
• Estimate all quantities of interest in all time periods. They believe an uncertain estimate even when data are sparse or not available is preferable to no estimate
• Synthesize all the appropriate data using statistical methods designed to handle both sampling and non-sampling error
• Method also allow the use of covariates to improve predictions for where data are sparse by borrowing strength across time or geography
• All estimates should be generated with 1000 (or more) draws from the posterior distribution of the quantity of interest
GBD 2013: flowchart of analytical components
Covariates
What is a covariate?• A variable that has a positive or negative relationship with a disease/
condition in the GBD – currently 192 used
• Other names: independent variables; predictors; explanatory variables
• GBD uses covariates to inform the estimation process in all models in the GBD Study
• For countries and conditions with lots of data, covariates play a minimal role in the estimation process
• For countries and conditions with little data, the role of covariates is very important
• Complete time series from 1980 to 2013 calculated for all covariates
Process for Covariates database
Mortality
12
Mortality: “The Envelope”
13
Mortality: “The Envelope”The importance of the all-cause mortality estimates:
Knowing the total number of deaths by age, sex, country, and year provides key information to policy makers and governments
The envelope is also important internally to the GBD:
• Mortality estimates are used as covariates
• Each of the causes of death are modelled independently and re-scaled to sum to the all-cause mortality envelope
14
Causes of Death
15
Constructed comprehensive database of 8730 site-years of data covering 188 countries from 1980 to 2013.
Vital registration – 4,133
Verbal Autopsy – 659
Surveillance Systems – 1,006
Survey/Census –60
Cancer Registries – 1,270
Sibling History Pregnancy Related Death – 1,572
Burial/Mortuary – 33
Police – 1,285
Hospital -- 42
Cause of Death database
Quality and Comparability• Mapping versions and adaptations of the data systems, such as ICD
• Garbage codes redistributed to plausible target codes using statistical methods, published studies and expert judgment
• Each data point examined and assessed for consistency with other sources for the same country, over time and how it fits with other estimates in the region
17
CODEm• Standard analytic tool for cause of death estimation- used for most causes
(some causes require custom methods)
• Develops a large range of plausible models for each cause and creates combinations ‘ensembles’ of the best performing models.
• Pulls directly from the COD database
• Displays results directly in the COD Visualisation Tool
18
Mixed Effects Linear Models
Space- time GPR models
Rate + +Cause fractions + +
Four possible family of models
CoDCorrect
Estimates for each age-sex-country-year for the 303 causes are constrained to equal the demographic estimate of all cause mortality for that age-sex-country-year.
This rescaling is undertaken at the 1,000 draw level to propagate the uncertainty in the estimates for each cause into the final results
19
Cause of Death Estimates
The importance of the cause of death estimates:
• Number of deaths by cause are key outputs of the GBD Study
• Age-standardised death rates are used as covariates in the non-fatal health outcome modeling
• The death estimates are the direct inputs for calculating YLLs (years of life lost)
20
Non-Fatal Health Outcomes
21
Details of Non-Fatal Health Outcomes Process
22
Non-Fatal Health Outcomes Approach• Create database on disease sequelae prevalence based on systematic
reviews of published studies, household examination and interview surveys, surveillance systems, notifiable diseases, cancer and other disease registries, hospital discharge data, primary care data.
• Flag and correct potential sources of bias
• For most diseases, use DisMod II, a Bayesian meta-regression tool to generate estimates.
23
Non-Fatal Health Outcomes
The importance of the non-fatal health outcomes:
•Prevalence is a direct input into the computation of YLDs (prevalence * disability weight = YLD)
•Prevalence and YLD estimates are key outputs of the GBD Project
•YLDs both capture morbidity associated with causes of death plus allow project to report a comparable measure of leading diseases that are not fatal
Disability Weights
25
Detail of Disability Weight database
26
Comorbidity Correction
• Assumption that one person cannot have disability>1 on the 0-1 disability weight scale if multiple conditions present
• To account for this GBD models comorbidity in a large micro-simulated population and use this to adjust disability weights in the final estimates
27
Risk Factors
Risk Factor Process
Calculating Risk Factor Burden1. Select risk-outcome pairs
2. Estimate exposure distributions to each risk factor
3. Estimate relative risk per unit of exposure for each risk-outcome pair;
4. Choose theoretical minimum risk exposure distribution (TMRED); and
5. Compute population attributable burden, including uncertainty.
Currently 69 risk factors estimated
Cause of death CVD
32
Cause of death CVD
33
Results visualisation
34
Results visualisation - YLD
35