Role of Data in the Decade of Action Road Safety- 2011-2020
1. Role of Data in the Decade of Action
Road Safety: 2011-2020
Kavi Bhalla, PhD
Research Scientist
Department of Global Health and Population
Harvard School of Public Health
2. Why Road Traffic Injuries?
0
50
100
150
200
250
300
350
400
1940 1950 1960 1970 1980 1990 2000 2010
Year
DeathsperMillionPeople
AUS AUT BEL
CAN CHE DEU
DNK FIN FRA
UK IRL ITA
JPN NLD NOR
SWE USA
Roadinjurydeathspermillionpeople
3. Why Road Traffic Injuries?
0
50
100
150
200
250
300
350
400
1940 1950 1960 1970 1980 1990 2000 2010
Year
DeathsperMillionPeople
AUS AUT BEL
CAN CHE DEU
DNK FIN FRA
UK IRL ITA
JPN NLD NOR
SWE USA
Roadinjurydeathspermillionpeople
4. Goal: A framework for injury metrics
• Funding: World Bank Global Road Safety Facility
Two grants (since 2006)
• Build and implement a framework for estimating
national burden of road traffic injuries in 18 countries
• Adapt methods to Africa (ongoing)
• Research Team
– Kavi Bhalla, Saeid Shahraz, Jerry Abraham, David Bartels,
Nicole DeSantis, and Pon-Hsiu Yeh
• External collaborators: GBD-Injury Expert Group
5. Talk Overview
Estimating the burden of road traffic injuries in:
1.Information-rich settings
– E.g.: Mexico, Iran, Sri Lanka, Colombia ….
2.Information-poor settings: Africa
– E.g.: Mozambique: Triangulating to national estimates
from multiple data sources
– Extending methods to other African countries
3.Global estimates of burden of road traffic injuries
– GBD 2010 study
6. Information Rich Settings: Mexico
National burden of road injuries*
DEATHS
HOSPITAL ADMISSIONS
EMERGENCY
ROOM VISITS
HOME CARE
Envelope from survey : further breakdown
Using hospital registry (selected provinces)
Surveys:World Health Survey,ENSANUT
Envelope from survey : further breakdown
Using Ministry of Health and IMSS Hospitals
Broken down by
• age and sex groups
• urban/rural
• institutional care received
• injury severity
• victim mode (pedestrian,
motorcycle, car occup, etc)
• impacting vehicle
• injuries (head, limb, etc)
• time of day
• type of road
death registration
* International Journal of Injury Control and Safety Promotion, Aug 2010
9. Talk Overview
Estimating the burden of road traffic injuries in:
1. Information-rich settings
– E.g.: Mexico, Iran, Sri Lanka, Colombia
2. Information-poor settings: Africa
– E.g.: Mozambique: Triangulating to national
estimates from multiple data sources
– Extending methods to other African countries
3. Global estimates: Burden of road traffic injuries
– GBD 2010 study
10. Mozambique: Data Sources
DEATHS
NON-FATAL INJ
2003 DHS – Trauma Module;
Maputo City Hospital Records
1. Mortuary data from Maputo city
– Urban; Medico-legal deaths (injuries)
– Retrospectively collected data; 10 years
2. Demographic Surveillance Site; Manhica
– Rural; Causes of death from verbal autopsy
3. Post-census mortality survey (INCAM)
– Nationally representative (~18000 deaths)
– Verbal autopsy – “injury” is a cause
11. Triangulating to National Estimates
Estimating Injury Mortality
–Urban Injury Mortality
• National Verbal Autopsy => Mortality envelope
• Maputo Mortuary => Disaggregate envelope
–Rural Injury Mortality
• National Verbal Autopsy => Mortality envelope
• Manhica DSS => Disaggregate envelope
12. Inputs: Estimating Injury Deaths
National Verbal Autopsy Study Urban Mortuary (Maputo City)
Road injury
Fall
Drowning
Fire
Poisoning
Suicide
Homicide
Other
Unspecified
0% 10% 20% 30% 40% 50%
% of all Injuries
0
2
4
6
8
Urban Rural Total
%injurydeaths
Rural DSS Verbal Autopsy (Manhica)
0% 10% 20% 30% 40% 50%
Road Injury
Fall
Drowning
Burn
Cut
Firearm
Blunt Object
Explosives
Poisoning
Hanging
Intoxication
Strangling
Other
Unknown
% of all injuries
Unintentional
Suicide
Homicide
Unknown
“Injury Envelopes”
18. Data Sources:Non-fatal Injury- Surveys
Sudan Ethiopia Ghana Burkina-
Faso
Nigeria Uganda
NATIONAL
- Sudan
Household
Health Survey
2010
NATIONAL
• World Health
Survey
• Socio-
Economic
Survey of
Disabled
Population
• Health &
Nutrition
Survey
• Welfare
Monitoring
Survey
COMMUNITY
• Jimma Injury
Survey
NATIONAL
• World
Health Survey
• DHS
Maternal
Mortality
• Core
Welfare
Indicators
Questionnaire
• Living
Standards
Measurement
Survey
• Child Labour
Survey
COMMUNITY
• Kumasi &
Brong-Ahafo
Injury Survey
• Accra Injury
Survey
NATIONAL
• World Health
Survey
• Enquete
Burkinabé Sur
les Conditions
de Vie des
Mengages
(CWIQ)
NATIONAL
• Nigeria
Injury Survey
• Core Welfare
Indicators
Questionnaire
• Living
Standards
Survey
• General
Household
Survey
COMMUNITY
• Lagos
Household
Survey
NATIONAL
• National
Household
Survey
• Northern
Uganda
Baseline
Survey
COMMUNITY
• Kawempe &
Mukono
Community-
based Injury
Survey
19. Data Sources-Non-fatal Injury-Hospital
Sudan Ethiopia Ghana Burkina-
Faso
Nigeria Uganda
• Health
Managemen
t
Information
System
(HMIS)
• Black Lion
hospital-
based injury
surveillance
• District
Health
Information
System
(DHIS)
• Health
Managemen
t
Information
System
(HMIS)
• Hospital
morbidity
tabulations
• Ministry of
Health
Hospital
Statistics
• WHO
Hospital-
based Injury
Surveillance
• ICCU
Hospital-
based Injury
Surveillance
20. Themes: Census Data
• Household mortality questions are common
• Often ask if death was from injury
=> Can provide injury totals
• Face validity tested in S. Africa
22. Themes: Urban Mortuaries
• Exist in most urban centers
• Issues: Cause coding and catchment areas.
0%
20%
40%
60%
80%
%ofallinjurydeaths
Ethiopia Ghana Mozambique
Nigeria Sudan Uganda
Zambia
23. Themes: Rural DSS Sites
• Common in Africa; Causes of death via VA
• Issues: External cause coding
0%
2%
4%
6%
8%
10%
12%
(Injurydeaths)/(Totaldeaths)
DSS Sites, 1999-2001, Injury death fraction
24. Talk Overview
Estimating the burden of road traffic injuries in:
1.Information-rich settings
– E.g.: Mexico, Iran, …
2.Information-poor settings: Africa
– E.g.: Mozambique: Triangulating to national
estimates from multiple data sources
– Extending methods to other African countries
3.Global estimates: Burden of road traffic injuries
– GBD 2010 study
25. Global
Mortality
Model Sensible global
health priorities
TheoryRegional
estimates
Theoretical Input
Discussion Papers
Case Definition
GBD “Sequelae” defs
Multiple Injuries
Handling unspecifieds
Recurrent injuries
…
…
Real World Data
Death registers
Hospital records
Verbal autopsy data
Health surveys
Mortuaries
Literature Reviews
GBD sequelae
Model: Burden
of non-fatal Injuries
GBD INJURY EXPERT GROUP
www.globalburdenofinjuries.org
Mortality data
26. Architecture of Global Injury Data
Non-fatal Injury Data Sources
• Surveys
– Strength: Population incidence of road injuries
– Shortcoming: Poor measurement of sequelae
• Hospital Records
– Strength: Precise (ICD) sequelae descriptions
– Shortcoming: Data not available from all regions;
Denominator population is often not known
29. Road Injuries Incidence:
HOSPITALIZED
(by region, age, sex)
Sequelae Incidence:
HOSPITALIZED
(by sequelae, region, age, sex)
Sequelae Incidence:
NOT HOSPITALIZED
lack of medical care
(by sequelae, region, age, sex)
Sequelae Incidence:
NOT HOSPITALIZED
do not need admission
(by sequelae, region, age, sex)
Surveys DISMOD
Covariates
Mapping:
Road Injury Sequelae
Hospital data
Access to Care
Surveys
Probability of admission
in a high access to care setting
Model
NZ: Hospital data
30. Seq. Incidence:
HOSPITALIZED
Seq. Incidence:
NOT HOSPITALIZED
lack of medical care
Seq. Incidence:
NOT HOSPITALIZED
do not need admission
Sequelae durations
• % life long; excess mortality
• duration of short term
Disability Weights
(for three types of sequelae)
Australian Hospital
Registry
Model (contd)
GBD field studies
Burden of non-fatal of road injuries
(YLDs)
31. Talk Overview
Estimating the burden of road traffic injuries in:
1.Information-rich countries
– E.g.: Mexico, Iran, …
2.Information-poor settings: Africa
– E.g.: Mozambique: Triangulating to national
estimates from multiple data sources
– Extending methods to other African countries
3.Global estimates: Burden of road traffic injuries
– GBD 2005 study
32. Conclusions
• Lots of Existing Data: even in Africa: HDSS,
mortuaries, surveys, hospital, censuses, etc.
• Analysts Wanted: to develop methods for
eliminating bias, triangulating to policy relevant
statistics
• Emerging Research Field: Global Health
Metrics: with unique methods, research
community, and political stakeholders.
33. Thank You!
Acknowledgements
• Funding: World Bank Global Road Safety Facility
– Two grants over six years
• External collaborators: GBD-Injury Expert Group
• Research Team
– Saeid Shahraz, Jerry Abraham, David Bartels, Nicole
DeSantis, and Pon-Hsiu Yeh
Find out more
–www.globalburdenofinjuries.org
–email: kavi_bhalla@harvard.edu
41. Data Sources Availability
1. Global Data Sources
a) Mortality
b) Health/Injury Survey
c) Hospital records
2. Data Sources in Africa
Detailed Information: www.GlobalBurdenofInjuries.org
42. Background
• Closely associated with ongoing GBD-2010 study
– Earlier GBD studies used few African data sources
– GBD-Injury expert group
• approximately 170 members
• www.globalburdenofinjuries.org
• Funder: World Bank Global Road Safety Facility
• History
– Original Study: National Road Traffic Injury Estimates
– Vision
• Should construct best estimates with all existing data sources
– 18 Focus countries
43. LATIN AMERICA EAST ASIA & PACIFIC
Brazil Mauritius
Colombia
Ecuador SOUTH ASIA
Mexico* Sri Lanka*
Argentina*
Uruguay MID. EAST & NORTH AFRICA
Iran*
EAST EUROPE & CENTRAL ASIA
Croatia SUB-SAHARAN AFRICA
Czech Republic Mozambique
Hungary
Kazakhstan HIGH INCOME COUNTRIES
Latvia Spain
Slovenia USA
Original Study: 18 Focus Countries
45. Triangulating to National Estimates
• Original Method:
–Deaths: using national death registration data
• adjust for completeness, redistribute unspecifieds
–Non-fatal injuries
• Incidence from population surveys
• hospital data to estimate “sequelae” => convert to
burden
• But, what to do about Mozambique?
–There is no death registration in Africa!
46. Data Sources in Mozambique - Deaths
1. INCAM-2007: National Verbal Autopsy Study
– ‘Total Injury’ death rates
2. One Urban Mortuary (Maputo City)
– Medico-legal autopsies for unnatural deaths
– 10 years of retrospective data: 1993-2003
– age, sex, cause (intent and mechanism)
3. One Rural Demographic Surv. Site (Manhica)
– Verbal Autopsy: 1999-2003
– age, sex, cause (intent and mechanism)
47. Constructing a national estimate
• Two Mozambique(s):
–Urban Mozambique
• ‘Total Injury’ death rate (by age-sex) from INCAM
• Further breakdown using Maputo city mortuary data
– Rural Mozambique
• ‘Total Injury’ death rate from INCAM
• Further breakdown using Manhica HDSS data
–National = Urban + Rural
49. Overview
• Background
• Data sources in one country – Mozambique
• Data architecture in Africa
– Censuses
– Mortuaries
– HDSS Sites
– Hospital data
⇒ Conclusion: Plenty of data: Analysts Wanted!
• Data Collection Process
51. POPULATION CENSUS DATA
• Household mortality questions are common in
African censuses.
• Usually intended for estimating maternal mortality
• Sometimes they ask if death was from injury
• Our Goal: Use census data to estimate total injury
incidence by age, sex, urban/rural
56. Censuses with Injury questions
Analyzed this far:
• South Africa – 2001
• South Africa – 2007 (large community survey)
• Sudan: South and North
• Lesotho 2006
• Malawi 2008
Hope to analyze:
• Ghana 2010 (In-field starting Sept 26)
• Mozambique INCAM survey (report available)
57. Validation: Are the census results
sensible?
Using South African death registration data
58. Fraction of total deaths that are from injuries?
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0
1-4
5-9
10-14
15-19
20-24
25-29
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
75-79
80-84
85+
Total
(Injurydeaths)/(Totaldeaths)
South Africa 2001: Injury death fraction, male
Census
VR
64. Overview
• Background
• Data sources in one country – Mozambique
• Data architecture in Africa
– Censuses
– Mortuaries
– HDSS Sites
– Hospital data
⇒ Conclusion: Plenty of data: Analysts Wanted!
• Data Collection Process
65. Urban Mortuary Data
• Mortuary Data is very commonly available
• 7 countries => 7 mortuaries
66. Mortuary Data – 7 countries
COUNTRY PERIOD No. of Cases
Sudan
• Khartoum Mortuary
• Omdurman Mortuary
• 6 years (2004-2009)
• 4 months (in 2010)
~15,000
255
Uganda
(Kampala)
6 months (1 July -31 Dec 2007) 757
Ethiopia
(AddisAbaba)
1 yr (1 Jul, 2006 -30 Jun 2007) 2114
Zambia
(Lusaka)
13 months (Nov 2007-Dec
2008)
594
Ghana
(Kumasi)
2 yrs (2005 - 2006) 1545
Nigeria
(Ibadan)
3 yrs (2007-2009) 1045
Mozambique
(Maputo city)
10 yrs (1994-2003) 12354
67. How complete is mortuary data?
• Medico-legal requirements vary dramatically across
Africa
– Nigeria: Family has a right to opt-out of medico legal
investigations.
– Zambia (Lusaka): Burial registration is strictly enforced
• Can we quantify completeness and quality?
68. Completeness Test for Sudan –
Khartoum Mortuary
Compare
1. Number of injury deaths in Khartoum mortuary;
&
2. Number of injury deaths in Khartoum from 2008
Census
Completeness = (#1) / (#2)
69. MORTUARY CENSUS COMPLETENESS
MALE 1222 1347 91%
FEMALE 208 307 68%
OVERALL 1430 1654 86%
Completeness: Khartoum Mortuary Data
DEATHS BETWEEN AGES 15-49 YEARS
70. Quality of cause-of-death coding
Sudan : Khartoum Mortuary
Mixed coding is a serious issue in existing mortuary records
74. Typical HDSS site
• Example: Navrongo,
Ghana
• Established: 1993
• Community size:
– 144,000 people
• Fairly rural
• Demographic and health monitoring: includes verbal
autopsy on all deaths and regular morbidity surveys
(including injuries)
79. Injury Surveys: Measurement Issues
• Injury Involvement Questions
– Our Injury Definition: “resulting in one day disability”
• very rare, e.g. TASC
– What is usually measured:
– Common: without threshold (were you injured in the last
month?)
– Common: Hospitalized e.g. WHS, LSMS, national health surveys
– Relatively rare: injury resulting in one day loss of school/work
• Recall period
– 2 weeks, 1 month (very common), 3 month, 4 months, 5 months,
6 months, 7 months, 1 year (common)
80. Measurement Issues: Recall Biases
InjuryRate,per1000
All road injuries (hospitalized and non-hospitalized)
Source: WHS
Last
Month
2 to 3
months
ago
4 to 6
months
ago
6 to 12
months
ago
Source: World Health Surveys: Aggregated data from 53 countries
month
incidence,per1000
81. Measurement Issues: Recall Biases
InjuryRate
All road injuries (hospitalized and non-hospitalized)
Source: WHS
Hospitalized Road Inj
Substantial recall effects for the non-hospitalized cases
Last
Month
2 to 3
months
ago
4 to 6
months
ago
6 to 12
months
ago
incidence,per1000
month
82. Non-fatal road injury incidence (1 /2)
82
0 10 20 30 40 50 60 70
KAZ
Asia, Central-GEO
CHN
Asia, East-CHN
IND
BGD
IND
IND
NPL
PAK
PAK
IND
Asia, South-PAK
KHM
VNM
VNM
LKA
MYS
MYS
VNM
PHL
VNM
LAO
VNM
LKA
MUS
MMR
Asia, Southeast-THA
Australsia-NZL
Caribbean-DOM
SVK
SVN
CZE
HUN
BIH
Europe, Central-HRV
RUS
LVA
UKR
EST
Europe, Eastern-RUS
ESP
Europe, Western-NLD
Road injuries per 1000 persons
83. Non-fatal road injury incidence (2 /2)
830 10 20 30 40 50 60 70
Latin America, Andean-ECU
MEX
COL
GTM
Latin America, Central-MEX
URY
BRA
PRY
IRN
IRN
ARE
MAR
TUN
TUR
North Africa / Middle East-SYR
Sub-Saharan Africa, Central-COG
ETH
MOZ
UGA
MWI
KEN
COM
ZMB
TZA
Sub-Saharan Africa, East-ETH
ZAF
ZAF
KEN
ZAF
NAM
SWZ
Sub-Saharan Africa, Southern-ZWE
BFA
GHA
NGA
GHA
GHA
NGA
NGA
NGA
NGA
CIV
GHA
GHA
SEN
NGA
TCD
MLI
MRT
Sub-Saharan Africa, West-BFA
CAN
North America, High Income-USA
Road injuries per 1000 persons
Sub Saharan Africa - West
Sub Saharan Africa - East
Sub Saharan Africa - South
Sub Saharan Africa - Central
84. Conclusions about Surveys
84
• Approximately 2% of the population has a non-
fatal road injury every year
• Survey-based measurements are readily available
• Analytical corrections required for comparability
(ongoing)
86. Overview
• Background
• Data sources in one country – Mozambique
• Data architecture in Africa
– Censuses
– Mortuaries
– HDSS Sites
– Hospital data
⇒ Conclusion: Plenty of data: Analysts Wanted!
• Data Collection Process
87. Data Collection Process
• Environmental Scan to Identify Sources
– Google Searches
– Lots of emails: GBD Injury Exp Gp has 160+ members
– Call for data in PLoS Medicine
– Systematic Literature Review
• Enabling data access: 3 levels of data sharing
– Micro data (individual level records)
– Data tabulated to our specifications
– Published reports with most detailed tabulations
88. Country Visits
• Prior to country visit
– Conduct data source scan
• During country visit
– National Statistics Office
– University researchers
– Mortuaries and ED at main hospital
– Police (Traffic and Crime)
• What helps during visit
– Being poor
– Having a country collaborator
– Having a plan that suggests high likelihood of getting
to every other data source in the country
89. Thank you
89
• Find out more:
– www.globalburdenofinjuries.org
– email: kavi_bhalla@harvard.edu
• Acknowledgements:
– This research has been funded by a grant from the
World Bank Global Road Safety Facility
– The data sources analyzed in this project have
been collected by other researchers and agencies.
92. Injury Expert Group: Operations
• Publications:
– strongly encouraged
– Authorship: should reflect principles commonly
used in the academic community for large multi-
center collaborative studies
– Example:
93. Authorship:
- lists all those who contributed + “on behalf of the Global Burden of Disease
Injury Expert group”
101. Other Key Themes
• Household Surveys: Key Issues
– Recall biases, defining injury thresholds
• Hospital Data
– only source for injury diagnosis
– Key Issues:
• burden in the presence of multiple sequelae