Assessing premature and avoidable mortality Rafael Lozano, MD, on behalf of 189 authors of the paper

1. Global and regional mortality for
235 causes of death from 1990 to
2010
Assessing premature and avoidable mortality
Rafael Lozano, MD, on behalf of 189 authors of
the paper
“Global and regional mortality from 235 causes of death for 20
age groups in 1990 and 2010: a systematic analysis for the
Global Burden of Disease Study 2010”

2. Background
• Causes of death (CoD) is one of the most fundamental metrics
for population health.
• Trends in CoD provide an important summary of whether
society is or is not making progress in reducing burden of
premature mortality and especially avoidable mortality.
• Usually CoD assessments show success and failures of Health
Information Systems and provide directions of how to improve
them.
• GBD 1990 was the first comprehensive study to present the
global leading causes of death.
2

3. Global cause of death assessment: main
issues
• The universe of data
• Efforts to assess and enhance quality and comparability
of data
• The statistical modeling strategy
• Causes of death constrained to sum to all cause mortality
3

4. The universe of CoD data
We attempted to identify all Type Site years Countries
Vital Registration 2,798 130
available data on causes of
Verbal Autopsy 486 66
death for 187 countries from
Cancer registries 2,715 93
1980 to 2010
Police Reports 1,129 122
• We used 9 different sources Surveys/Census 1,564 82
of CoD data Maternal Mortality Surveillance 83 8
• We collected data on around Deaths in health Facilities 21 9
Burial and Mortuary 32 11
600 million deaths in the last
Country−years of vital registration, 1980−2010
30 years
• Data available varies by
disease:
o More on maternal, cancer,
injuries
o Less on NTD, diarrhea and
LRI pathogens
4

5. Assessment and enhancement of data
quality and comparability
1. Assessment of
completeness
Percent garbage from ICD vital registration
2. Causes of death
mapping
3. Redistribution of
misclassified causes of
death
4. Age and age-sex
splitting
5. Smoothing for
stochastic variation due
to small numbers
6. Outlier detection
5

6. Modeling causes of death
1. Causes of death ensemble modeling, CODEm (133 causes),
including all major causes except HIV. CODEm selects
models and ensembles of models based on out-of-sample
performance.
2. Negative binomial (12 causes).
3. Fixed proportion models (27 causes).
4. Disaggregation by pathogens or sub-causes (36 causes).
5. Natural history models (8 causes).
6. Mortality shock regressions (2 causes).
6

7. Combining results: CoDCorrect algorithm
• Because we developed single-cause models, it was imperative
as a final step to ensure that individual cause estimates
summed to the all-cause mortality estimate for every age-sex-
country-year group.
• This is one of the innovations of this study:
o Implemented taking into account uncertainty in every cause of
death model outcome
o We proportionately rescaled every cause such that the sum of the
cause-specific estimates equaled the number of deaths from all
causes generated from the demographic analysis (by country,
year, age, and sex).
o We applied CoDCorrect in a hierarchical way.
7

8. Shift of causes of death in the last 20 years
1990 2010
Injuries
Injuries 10%
9%
Comm/Mater
/Neonat/Nutr
Comm/Mater/ 25%
Neonat/Nutr
34%
Non Communicable Non Communicable
Diseases Diseases
57%
65%
46.5 million 52.7 million
8

9. Death decomposition analysis by changes in
population
75%
50%
25%
0%
-25%
-50%
-75%
All Causes Com/Mat NCD Injuries
Neo/Nut
% change 1990-2000 % change due to change in rates
% change due to pop ageing % change due to pop growth
9

10. Percentage of global deaths for female and
male individuals in 2010 by cause and age
Males Females

11. Rapid
shifts in
leading
causes of
global
death
11

12. Years of
life lost
puts more
emphasis
on
leading
causes in
children
12

13. Percentage of YLLs for all ages and both sexes
combined by cause and region in 1990 and 2010
1990 2010

14. YLL top 25 leading causes across 21 regions, 2010
14

15. Main findings
• The shifting pattern of the number of deaths by cause across
time, regions, and age groups is consistent with the three key
drivers of change.
• Our estimates of 235 causes of deaths are internally consistent
by age, sex, region, and year, but could be:
o Different from other publications (higher or lower),
o Similar to other estimates, or
o New data for global health.
• Causes of “millionaires” deaths
o HIV, LRI, diarrhea, malaria, TB (preterm, hepatitis)
o IHD, stroke, lung cancer, diabetes, cirrhosis (CKD)
o Road injuries (suicide).
15

16. Limitations
• Data on CoD even in settings with medical certification might not
always accurately capture the UCD.
• Redistribution of misclassified deaths could be improved with more
empirical data.
• Verbal autopsy data are more accurate for some causes and less for
others.
• For some cause models only a weak covariate selection was
possible.
• The use of negative binomial, fixed proportion models, and natural
history are in principle related to the lack of quality of data.
• UI are good for CODEm results but weak for other modeling
strategies.
• When expert opinion and data diverge, we tended to follow the
available data.
16

17. Final remarks
• GBD 2010 is the most comprehensive and systematic analysis
of causes of death undertaken to date.
• Adding time trends and quantifying the uncertainty differentiate
GBD 2010 from similar studies in the past.
• GBD 2010 is an asset for Global Health:
o More data then ever before
o New methods for improving the quality of data and the modeling.
• GBD 2010 has demonstrated that public health priorities
everywhere are changing, or soon will be.
17