Transcript Document

Measuring cause specific
mortality: the use of verbal
autopsies
Alan Lopez
Chalapati Rao
Uses of cause of death data
• To study and explain levels, trends and differentials in age
specific mortality (Preston and disciples)
• To guide priorities for resource allocation for intervention
programs, biomedical and sociomedical research
• To monitor public health programs
• To provide clues for epidemiological research
Sources of national cause-specific mortality data
– vital registration systems – GOLD STANDARD
– sample registration systems
– household surveys
– population laboratories and surveillance systems
– epidemiological estimates
• For deaths registered in these systems, cause of death is
either
– certified by a medical practitioner
– based on "verbal autopsies"
– not given at all
VR: Data availability, around 2000
Useable data
Complete
enumeration
Total countries in
region
Africa
4
1
46
The Americas
32
14
35
Eastern Mediterranean
7
4
22
Europe
47
39
51
South-East Asia
4
0
11
Western Pacific
21
8
27
World
115
66
192
WHO Region
Data quality
Quality
Criterion used
Countries
High
ICD 9 or 10 coding, and
completeness >90% and illdefined codes <10%
Australia, Bahamas*, Canada, Chile*, Cuba, Estonia, Finland, Hungary,
Iceland*, Ireland, Japan, Latvia, Lithuania, Malta, Mexico, New Zealand,
Republic of Moldova, Romania, Russian Federation, Saint Vincent and the
Grenadines*, Singapore, Slovakia, Ukraine, United Kingdom, United States
of America, Uzbekistan, Venezuela
Medium
Completeness 70-90% OR
ill-defined codes 10-20% OR
non-ICD codes used
Antigua and Barbuda*, Austria, Azerbaijan, Barbados*, Belarus, Belgium*,
Belize*, Brazil, Brunei Darussalam, Bulgaria, Colombia*, Cook Islands,
Costa Rica, Croatia, Czech Republic, Denmark*, El Salvador*, France*,
Georgia, Germany, Grenada*, Guyana*, Israel*, Italy, Kazakhstan, Kuwait,
Kyrgyzstan, Luxembourg, Mauritius, Mongolia, Netherlands, Niue, Norway,
Panama, Philippines*, Republic of Korea, Saint Kitts and Nevis*, Saint
Lucia*, Seychelles, Slovenia, Spain, Sweden, Switzerland*, The former
Yugoslav Republic of Macedonia, Trinidad and Tobago*, Turkmenistan*,
Uruguay
Low
Completeness < 70% OR illdefined codes > 20%
Albania, Argentina, Armenia, Bahrain, Bosnia and Herzegovina*, Cyprus*,
Dominica*, Dominican Republic*, Ecuador, Egypt, Fiji, Greece*, Iran,
Jamaica*, Kiribati, Nicaragua, Paraguay, Peru, Poland, Portugal, Qatar, San
Marino, Serbia and Montenegro, South Africa*, Sri Lanka*, Suriname*,
Syrian Arab Republic, Tajikistan*, Thailand, Tonga*, Tuvalu
Strategies to improve COD data availability
• Accelerate development of civil registration
– SLOW, EXPENSIVE, LOW GOVERNMENT PRIORITY
• Introduce / improve physician certification, ICD coding, and
statistical processing of data
– CRITICAL, NEEDS BIG ‘WHO’ PUSH, NO CHAMPIONS
–
• Introduce and develop local applications of verbal autopsy
procedures for data acquisition through
• vital registration
• sample registration
• demographic surveillance systems
• household surveys
What is VA ?
• VERBAL AUTOPSY (VA), a two step procedure
– Data collection : interview of bereaved relatives to
collect information on symptoms experienced by
deceased before death, using some form of survey
instrument
– COD assignment : methods include
– physician review of VA data
– ICD certification, coding, and tabulation
– computerised algorithms for population fractions
Principles of VA
• Based on recall by relatives of symptoms \ illness prior to death,
sometimes difficult for adult deaths (similar symptoms)
• Requires identification of clearly distinguishable symptom complexes
for each cause of interest, not available for some adult causes
– lung cancer, TB, diabetes, different forms of liver disease etc
• So far, found useful and validated for infant and maternal deaths;
deaths due to injuries
• Recent experiences suggest utility of gathering information from
medical documents if available within household
Historical use of VA
• Measuring cause specific mortality in populations
– SCDR / SRS - India
– DSP / VR – China
– AMMP (Tanzania)
• Investigating COD in specific age-sex cause groups
– infant / child deaths
– maternal causes of death
– Injury related deaths
• Investigating outbreaks / epidemics
– Ebola fever epidemic in West Africa
– Diarrhoeal disease in Bangladesh
Historical use (contd)
• Assessing coverage and effectiveness of disease
specific interventions
– insecticide impregnated bed nets in Africa
– Pneumococcal vaccine trials in Bohol
– Home based neonatal care in India
• In developed countries
– confidential enquiries of maternal mortality
– Sudden infant death syndrome
– QOL / mental health status in terminal illness among elderly
VA in India
•Survey of causes of death – rural (SCDR) 1967 – 1998 (1400 PHCs, rep)
•Structured instrument, lay assignment of cause, ICD classification since 1996
•problems with incomplete coverage, inadequate investigator training /
physician verification of cause
•New system being implemented in SRS, covering 8 million
•New instruments, field protocols, training support
•Five year retrospective survey planned soon
•independent field studies by Ind C Med Res in 5 zones,  1 million pop each
•Need for integrated approach, involving civil registration systems
SCDR Results
Leading causes of death, 1996-1998 and GBD 2000 estimates for India
1996
Year
1997
1998
GBD 2000
Ill defined causes
20.6
19
18.4
0
Respiratory (incl TB)
17.8
16.4
17.2
15
Infectious (excl TB)
14.8
13.1
12
16
Cardiovascular (excl stroke)
10.2
12.1
12.5
19
External causes
8.4
8.9
9.6
10.3
Perinatal
7.9
8.7
7.9
8
Neuropsych (incl stroke)
6.2
6.5
6.6
6.6
Cancers
3.6
4.3
4.3
6.9
Anaemia
3.4
3.2
3.3
0.4
Digestive
2
2.1
2.1
3.3
Endocrine
1.2
1.1
1.2
1
Other causes
3.9
4.6
4.9
13.5
Total
100
100
100
100
43168
44409
40351
10109157
Cause group
Deaths
VA in China
• Two mortality statistics systems – NMS, DSP
• NMS (VR) – medical certification (urban) / lay reporting (rural) 120 m pop
• DSP –– 1981, now 145 points, nationally representative, 10 m pop
• medical certification / VA for household deaths (80%)
• Semi structured instrument, physician COD assignment, ICD classification
since 1987
• Household visit by township hospital staff
• Much reliance on free text response to ‘ What was the cause of death in this
person?’
• Use of supportive medical documentation available at home / from hospital
records
• Needs to be tested for reliability, and if possible, validity
• Research projects underway (UQ, Harvard)
DSP Results
5q0
DSP
17.3
45q15 Males
45q15 Females
122
75
GBD 2000
Leading causes of death in 2000
Cause
DSP 2000
GBD 2000
Stroke
18.7
17.7
COPD
12.9
13.7
8.2
2.6
6.3
7.5
Other cardiac diseases
IHD
Ill-defined diseases
5.0
Trachea, bronchus and lung cancers
3.9
3.5
Liver cancer
3.8
3.6
Hypertensive heart disease
3.6
2.3
Stomach cancer
3.5
4.5
Lower respiratory infections
3.0
3.3
Self-inflicted injuries
2.7
3
Road traffic accidents
2.3
2.6
Oesophagus cancer
2.1
2.4
Other malignant neoplasms
2.0
0.6
All other causes
Total deaths
22.1
45716
32.7
875873
VA in Tanzania
• Adult morbidity and mortality project in 3 districts
• 1992 onwards
• Instruments developed by LSHTM – AMMP
• Physician assignment of cause
• Non ICD mortality classification
• Recent introduction of ICD certification / coding (2003)
• Validation study underway (UQ, Harvard, LSHTM)
AMMP results - 2000
Cause
Percent
TB/AIDS
21.4
Acute febrile illnesses
20.4
Ill defined causes
11.2
Cardiovascular
8.3
Perinatal (incl stillbirths)
7.9
Others
7.4
Acute respiratory infections
6.1
External causes
5.2
Diarrhoea
5.0
Neoplasms
4.3
Digestive
2.8
Total
100.0
Deaths
3721
Important issues in VA
• Standard survey instruments including modules for
– Free text narrative
– Structured questions
– Recording household medical document information
• Interviewer
– Education background, training
• Choice of respondent
– Proximity to deceased, education, age and sex, cultural factors
• Recall periods
– Minimum and maximum intervals
• Cause of death assignment and ICD coding
– Physician / trained health professional review using standard protocols
– Computerized programs
– ICD mortality tabulation lists
• Validation studies
Validation studies
• To develop standard verbal autopsy instruments and
procedures that are applicable in different epidemiological
and cultural settings with minimal modifications
• To measure biases in community cause of death patterns
when using VA instruments validated in hospital based
studies
• To improve understanding of quality of cause of death
information for estimating global and regional mortality
patterns
Methods
• VA validation study in Tanzania
• VA validation and mortality statistics evaluation
study in China
Study design - Tanzania
For each death
3000 deaths
Medical record
DC/ underlying
cause
from reviewer 1
Verbal autopsy
DC/ underlying cause
from reviewer 2
Agreement
DC/ underlying cause
from reviewer 1
Disagreement
reconciled by
consensus
MR Underlying cause
Verbal autopsy validation
DC/ underlying cause
from reviewer 2
Agreement
VA Underlying cause
Study design - China
2700 deaths
Medical record
Death certificate
Routine system
validation – urban
areas
1900 deaths
Verbal autopsy
validation
Routine
system death
certificate
3500 deaths
Verbal autopsy
death
certificate
Routine system
reliability –
rural areas
Future research
• Gates proposal – Africa, Bangladesh, Philippines
• New UQ sites in Indonesia, Thailand
• Egypt, Syria, other EMRO countries ?
• Expression of interest by Indian SRS to adopt WHO
VA methodology ?
• Other opportunities in Africa through PEPFAR?
Conclusions
• Information on symptoms could be combined with available clinical
evidence for judging cause of death at individual level
• Scope for application of Bayesian principles in deriving population level
cause specific mortality fractions from data gathered in surveys
•
• For adult deaths, could be useful for understanding broad cause group
mortality at population level
•
• A measure of discriminatory power of individual questions / algorithms for
identifying specific causes of adult deaths would be useful
Conclusions (contd)
• Need to evaluate biases from VA instrument validation in hospital
studies, and cross cultural comparability of responses to specific
questionnaire items
• Need for standardization of protocols for both data collection and
cause of death assignment
• All VA implementation should contain elements of validation
• Examples of potential application
– Sentinel sites in northern Brazil
– Representative sites in Tanzania
– Data quality improvement in Thailand, China
• No other option to rapidly increase usability of cause of
death data from developing countries