Published by Oxford University Press on behalf of the International Epidemiological Association
ß The Author 2008; all rights reserved.
International Journal of Epidemiology 2008;37:891–901
doi:10.1093/ije/dyn121
Evaluation of cause-of-death statistics
for Brazil, 2002–2004
Elisabeth França,1 Daisy Xavier de Abreu,1 Chalapati Rao2* and Alan D Lopez2
Accepted
27 May 2008
Background Mortality statistics systems with reliable cause-of-death data constitute a major resource for effective health planning; however, many
developing countries lack such information systems. Brazil has a long
history of registering deaths, and a critical assessment of the quality of
current cause-of-death statistics in its five different regions is crucial
to identify strengths and weaknesses in the data, and present options
for improvement.
Methods
Quality of cause-of-death data from 2002 to 2004 was evaluated using
an assessment framework based on four main attributes: generalizability, reliability, validity and policy relevance. A set of nine criteria:
coverage, completeness, consistency of cause patterns with general
mortality levels, consistency of cause specific mortality proportions
over time, content validity, proportion of ill-defined causes and nonspecific codes, incorrect or improbable age or sex patterns, timeliness,
and geographical disaggregation were used to assess the four
attributes of data quality.
Results
Completeness of death registration varies from 72 to 80% in the
northeast regions, compared with 85–90% in the Southeast and
Centre-West regions, and 94–97% in the wealthier South region.
The proportion of ill-defined deaths is an important problem in
reported causes of death from almost all regions. Lack of adequate
evidence limits the assessment of content validity of registered
causes of death. Coverage, consistency of causes with general level
of mortality, consistency over time, age and sex patterns, timeliness
and usability of statistics for subnational purposes were judged to
be reasonable and increase confidence in using the statistics.
Conclusions There is considerable heterogeneity in the quality of cause-of-death
statistics across Brazilian regions, especially for criteria such as
completeness and ill-defined causes. These factors can influence
generalizability and validity of reported causes of death, and must be
considered in the interpretation and use of data for secondary
descriptive analyses such as burden of disease estimation at regional
level, with suitable adjustments to account for bias. The differences
identified in this study could be a useful guide for defining measures
and investments needed to improve data quality in Brazil.
Keywords
1
2
Mortality, cause of death, vital statistics, evaluation, Brazil
Programa de Pós-graduação em Saúde Pública, UFMG, Belo
Horizonte, MG, Brazil.
School of Population Health, University of Queensland,
Brisbane, Australia.
* Corresponding author. School of Population Health,
University of Queensland, 288 Herston Road, Herston QLD
4006, Australia. E-mail: [email protected]
891
892
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Introduction
Analyses of causes of death are fundamental for
monitoring the health situation of populations and for
planning suitable interventions. However, one of the
major limitations of such analyses in most developing
countries is the quality of vital statistics which
encompass biases that are not easily measured. So,
systematic evaluation of national cause-of-death
statistics is vital for decision-makers to identify gaps
and present options for improvement.
In Brazil, despite the fact that the first law making
it mandatory for the State to register deaths dates
back to 1888,1,2 compilation of mortality statistics was
initially quite fragmented throughout most of the
country, except for some seats of State Government.3
In the 1970s, two important initiatives were introduced to improve the situation with respect to vital statistics. In 1973, the Brazilian Institute of Geography
and Statistics (IBGE) was entrusted with the collection, compilation and analysis of data from the Civil
Register Office in each municipal council. Later, the
Ministry of Health created the Mortality Information
System (MIS) in 1975, to complement the work of the
IBGE in vital statistics, with responsibility for compiling cause of death data.3,4 Since then, a number of
steps have been taken by the Ministry of Health to
improve the quality of mortality data, e.g. the implementation of active recording of deaths in hospitals
and close collaboration with civil registrars to register
deaths which occur outside hospitals.
Despite continuous improvements in the MIS, a
recent international data quality assessment categorized Brazil as producing medium-quality death registration data,5 based on estimated national levels of
completeness of registration and proportions of deaths
classified to non-specific causes of death. However,
this assessment did not explore likely differentials in
completeness and quality at subnational level, despite
its obvious importance for detailed mortality analyses
for health policy and monitoring mortality change
within the country.
In this article we have applied a comprehensive
evaluation framework6 to recent cause-specific
mortality data at national and regional levels, to
identify differences in the quality of cause-of-death
data reported by Brazilian regions. The application of
this framework in an entirely different data collection
and availability setting, and for the first time in a
regional context for a country may well help other
countries to decide on the generalizability of these
methods for cause-of-death quality assessment.
Methods
Selection of analytical units
Brazil, with an estimated total population in 2004 of
about 180 million, has around 5500 municipalities
which are grouped into five regions—North, Northeast,
Southeast, South and Centre-West. Economic and
population features differ widely among these regions.
The Northeast (28% of Brazil’s total population) and the
North (8% of the national total) have living standards
below the national average. On the other hand, the
South and Southeast regions, comprising 102.6 million
or 57% of the total national population have the highest
living standards. The Centre-West region (7% of the
population) has living standards intermediate between
these two regional groups (Table 1).
Data sources
Data on deaths were provided by the Brazilian MIS. In
this system, when a death occurs in a hospital or health
facility, the attending physician issues a death certificate according to the international form recommended
by the International Classification of Diseases 6th Revision (please see Appendix 1). For deaths that occur at
home, the cause of death is certified in one of four ways:
Table 1 Socioeconomic characteristics of regions in Brazil, 2003
North Northeast Southeast
South Center-West
Brazil
14 064 278 49 862 741 76 333 625 26 315 184
12 532 306 179 108 134
Populationa
Percentage of population (region/country)a
Percentage of urban population
b
Average literacy rate (15 years þ)
c
Low family income per capita (%)
7.9
27.8
42.6
14.7
7.0
100.0
69.9
69.1
90.5
80.9
86.7
81.2
89.9
76.8
93.2
93.6
90.5
88.4
32.6
40.4
13.6
12.2
19.0
20.7
4.5
34.7
80.8
30.5
38.4
55.3
Water supply system at home (%)
57.5
83.3
95.5
93.9
85.5
89.6
Own a refrigerator (%)
83.9
79.4
96.5
96.0
93.1
91.7
Own a telephone at home (%)
37.4
37.4
66.0
67.1
59.1
57.8
Access to sanitation (%)
Source: adapted from Brazilian Institute of Geography and Statistics—IBGE, 2005.
a
2004.
b
2000.
c
< half the minimum Brazilian wage.
EVALUATION OF CAUSE-OF-DEATH STATISTICS
(i) by an attending physician if present; or
(ii) by a physician from the official Death Investigation Service where available; or another
physician from the municipality; or
(iii) by the civil registrar in which case the cause of
death is usually not recorded. However, even in
the first two instances, an ill-defined cause
could be assigned; or
(iv) by a coroner where there are suspicious circumstances surrounding the death.
The Death Investigation Service [Serviço de Verificação
de Óbitos (SVO)] is an official institution linked to the
Health Service which performs autopsy in natural
deaths with an unknown cause (home or hospital
deaths). In 2003 the SVO operated in 13-seat State
Governments and also in some other cities. Since 2006
an initiative has been implemented by the Ministry of
Health to create SVO in the majority of larger cities in
the country. However, current statistics do not reflect
the influence of these developments as yet.
A standardized electronic program called SCBSistema de Seleção de Causa Básica (‘Underlying
Cause of Death system for Microcomputers’)7 is used
to select and code the underlying cause of death, using
codes from the Tenth Revision of ICD since 1996. Data
cleaning and compilation is done at the municipal,
provincial and state level, and an electronic data file is
transferred to the national office every 3 months.8
MIS provides annual reports of deaths by cause, age,
sex, place of residence and other key variables on the
Internet and on CD ROM. Raw data are also available to
download from the Internet (www.datasus.gov.br)
along with a specific software named ‘Tabwin’, developed by the Brazilian Ministry of Health, which enables
large volumes of data from MIS to be processed and
analysed at different levels of aggregation by municipality, state or region. Population data were provided by
the Brazilian Institute of Geography and Statistics
(IBGE) and made available on the site www.
datasus.gov.br.
Assessing cause-of-death data quality
We have applied the evaluation framework formulated
by Rao et al.,6 based on four main attributes, which
encompass nine criteria: (i) generalizability (with the
criteria coverage and completeness); (ii) validity (use of
non-specific codes, content validity, incorrect or improbable age or sex patterns); (iii) reliability (two criteria:
general level of mortality and consistency of causespecific mortality proportions over time); and (iv) policy
relevance (timeliness and geographical disaggregation).
Generalizability can be assessed by posing two questions: (i) does the mortality information system cover
all the target population, i.e. the whole country (coverage)? (ii) Are all deaths in the population covered by the
mortality information system registered (completeness)? Completeness for each region was assessed by
a two-step process. First, population counts from the
1991 and 2000 Census rounds were adjusted for
893
completeness using the ‘Generalised Growth Balance’
(GGB) method.9 This method assesses the relative
completeness of two successive census counts for a
given population by comparing the age distributions
observed in them. The method assumes the population
is closed to migration, but not necessarily stable (i.e.
constant birth and death rates). However, a limitation
of this method is that it is very sensitive to age
misstatement. Subsequently, these adjusted population
data and annual intercensal deaths were analysed using
the ‘Extinct Generation Method’ (EGM)10 to arrive at
an overall estimate of the completeness of death
registration for the intercensal period (1991–2000).
The EGM Method is based on the premise that there is a
relationship between the number of current deaths in a
population, and the number of people alive in the
population. Thus, in a closed population with perfect
recording of deaths, the size of the population age a at
time t can be estimated by accumulating the deaths to
that cohort after time t until the cohort was extinct. If
current deaths are under-reported, the ratio between
the estimated number of persons at age a and the
observed number of people at age a will be less than one,
which represents a measure of the completeness of
death registration in adults, assuming that death
understatement is constant across all adult ages. Like
the GGB, the EGM method is also sensitive to age
misstatement. Completeness estimates were used to
adjust observed mortality rates at ages 5 years and above
in 2003, assuming an average annual rate of improvement of completeness of 0.03%. This rate was calculated
using the completeness estimated by the EGM method
for the 1980–1991 and 1991–2000 intercensal periods to
derive the annual average rate of improvement in
completeness over these periods, which was found to be
0.03%. The adjusted mortality rates were used to
construct sex-specific life tables for each region, in
combination with levels of child mortality estimated
from the National Household Sample Survey conducted
in 2003.11
Validity of cause-of-death data is another key important element of the framework, as death registry may be
nearly complete, but it may contain a high percentage of
incorrect diagnoses. We extensively reviewed the
epidemiological literature on studies conducted to
investigate the ‘true’ registered causes of death over
the past three decades in Brazil, and discuss these
findings in the context of the content validity criterion.
When evaluating the proportion of ill-defined deaths
among all causes of death, we considered not only
deaths coded to the ICD codes for ‘symptoms, signs and
ill-defined conditions’ (ICD-9 codes 780–799 and ICD10 codes R00–R99), but also other codes which are not
informative as underlying causes of death from a policy
perspective.5 These include non-specific cardiovascular
causes such as cardiac arrest and unspecified atherosclerosis, among others; cancer without mention of primary site; and injuries of undetermined intent. In
addition, we examined sex and age patterns of some
894
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
causes of death that are sex-specific (e.g. prostate
cancer) or strongly age-dependent (e.g. perinatal
causes) to identify departures from the expected
patterns indicating low validity of the data.
To assess consistency of cause of death patterns with
the general level of mortality, we compared observed
proportionate mortality for each region by age and sex,
according to the three broad cause groups (group I:
communicable, maternal, perinatal and nutritional conditions; group II: non-communicable diseases; group
III: injuries) defined by the Global Burden of Disease
Study (GBD)12 with similar proportions derived from
compositional cause of death models.13 These models
are based on the premise that there is a predictable
relationship between broad cause group proportions
and the overall level of mortality and development in a
population. Previous empirical work suggests that these
models adequately represent mortality change, except
in populations such as in Southern Africa, severely
affected by HIV/AIDS mortality.12,13 We used the
estimated age-specific death rates by sex from the
regional life tables estimated from our methods, and
regional estimates of income (GDP) as input variables
for the cause of death models, and calculated for each
region, the number of standard deviations by which
observed cause-specific mortality proportions differ
from mean model predicted proportions. The GDP for
Brazil and states was derived from the 2000 census.14
The GDP for each region was calculated as a weighted
average of GDPs for its constituent states and adjusted
for international US dollar value.
Consistency of cause-specific mortality proportions
over time was assessed by examining the time trend
from leading causes less susceptible to real temporal
fluctuations in males and females during 2002–2004.
Significant fluctuation was evaluated by applying
Poisson regression methods.
The relevance of the data for policy-makers was
assessed using a maximum time lag of about 2 years
for the release of final tabulations (timeliness) and
availability of detailed information on health differentials in populations within health service jurisdictions at population level (geographical disaggregation).
Results
Generalizability
To our knowledge, there was no information published as to whether any municipalities had failed to
comply with reporting protocols, leading to exclusion
of their mortality data from numerators. If coverage is
indeed 100%, then we estimate that there are varying
levels of completeness of death registration in Brazil
(Table 2), although at national level, completeness of
adult death registration is around 90%.
The observed gradients across regions are consistent
with expectations that completeness of registration
systems as well as estimates of life expectancy at birth
are related to general economic development
Table 2 Estimated levels of completeness of death
registration in regions of Brazil, 1991–2000
Region
North
Male
%
80.6
95% CI
79.6–81.5
Female
%
95% CI
79.7
78.7–80.6
Northeast
78.3
74.9–81.6
72.1
68.4–75.7
Southeast
90.0
88.8–91.1
87.4
86.3–88.6
South
96.9
96.1–97.6
93.6
93.0–94.3
Centre-West
87.0
85.9–88.1
85.4
84.0–86.8
Brazil
91.2
90.4–92.0
88.9
88.2–89.5
(Tables 1–3). Interestingly, the life-table analysis
suggests that estimated levels of adult mortality for
males in the North and Northeastern regions are
lower than in the more developed Southeast region,
despite having the highest estimated levels of child
mortality (Table 3).
Validity
The proportion of ill-defined deaths is important in
reported causes of death from almost all regions,
particularly in the North and Northeast. However,
there is only one published study that explores the
potential misclassification patterns of these causes in
registration data from these regions.15 Non-specific
codes of cardiovascular disease account for 3.3% of all
causes and the variability across regions is smaller than
for ill-defined causes (Table 4). Nonetheless, on face
value at least, there has been an improvement in this
dimension of data quality from all regions over time,
both in terms of magnitude and the reduction of
differentials across regions (Figure 1). From another
perspective, the majority of ill-defined causes originate
from home deaths, ranging from 46% in the Southeast
to 82% in the Northeast (Table 5).
There have been a large number of epidemiological
studies conducted in Brazil to measure the accuracy of
the defined causes of death in the MIS by comparing the
cause reported on death certificates with clinical records
and, occasionally, autopsy reports15–17(see Appendix 2
for detailed list of articles). Some studies have also
addressed the problem of coding agreement.7,18 The
studies are generally designed to investigate validity of
one or several related causes within a cause group, or,
different measures are used to infer validity, such as
sensitivity and specificity, or kappa scores of agreement.
Moreover, there are no standard definitions for reference diagnoses used for comparison. While results of
studies on maternal deaths, injuries and some specific
disease categories provide evidence of under reporting,15,19 there is more confidence in the reliability of
mortality statistics on chronic diseases.20
Cause-of-death statistics from each region were
screened for plausibility of causes with age and sex
dependency, and we have not found any incorrect or
improbable age or sex patterns for major causes of death
EVALUATION OF CAUSE-OF-DEATH STATISTICS
895
Table 3 Risks of child and adult mortality, and life expectancy at birth by region, Brazil 2003
5q0
Observed
Estimateda
Observed
Adjustedb
eo
Observed
Adjustedc
North
0.025
0.041
0.193
0.235
69.2
66.0
Northeast
0.026
0.050
0.208
0.259
68.3
64.2
Southeast
0.019
0.023
0.252
0.276
67.1
65.7
Gender/Region
Males
45q15
South
0.017
0.024
0.222
0.228
68.5
67.8
Center-West
0.020
0.023
0.225
0.254
68.2
66.2
Brazil
0.022
0.030
0.231
0.250
67.8
66.2
North
0.021
0.034
0.104
0.129
76.1
72.8
Northeast
0.021
0.040
0.108
0.148
75.7
71.2
Southeast
0.016
0.015
0.117
0.133
75.9
75.1
South
0.014
0.017
0.109
0.116
76.8
76.4
Center-West
0.017
0.020
0.116
0.135
75.9
74.3
Brazil
0.018
0.025
0.113
0.126
75.8
74.2
Females
a
NHSS-National Household Sample Survey, IBGE, 2004.
Extinct Generations Method (EGM), adjusted to 2003 data.
c
Based on 5q0 estimated and 45q15 adjusted.
b
Table 4 Percentage of total deaths assigned ill-defined and non-specific codes in different regions of Brasil, 2002–2004
Non-specific code group
Symptomsa
North
21.2
Northeast
25.4
Southeast
8.8
South
6.4
Centre-West
6.0
Brazil
13.1
0.5
1.3
1.5
0.6
0.4
1.2
Cancer
0.6
0.6
0.8
0.9
0.8
0.8
Cardiovascular diseaseb
2.7
3.1
3.1
3.8
3.9
3.3
25.0
30.5
14.3
11.7
11.1
18.3
Injurya
a
Total ill-defined
a
Codes ICD-10 R00-R99 (Symptoms); Y10–Y34 and Y87.2 (Injury); C76, C80, C97 (Cancer).
Codes ICD-10 I47.2,I49.0,I46,I50,I51.4,I51.5,I51.6,I51.9,I70.9 (Cardiovascular disease).
b
60
50
%
40
30
20
10
0
1979
1981
1983
1985
North
1987
1989
Northeast
1991 1993
Year
Southest
1995
South
1997
1999
2001
2003
CentreWest
Figure 1 Regional trends in proportions of deaths assigned ill-defined and non-specific ICD codes in Brazil, 1979–2004.
same ICD-10 codes as used in Table 4 for the period 1996–2004, with corresponding ICD-9 codes for the period 1979–1995
896
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 5 Regional variation in proportions of home deaths that were assigned ill-defined causes during 2002–2004
Deaths during 2002–2004
Total deaths (all causes)
North
157 391
Northeast
758 613
Southeast
143 74 48
South
475 682
Centre-West
180 086
Brazil
300 92 20
Home deaths (%)
27.9
35.7
16.2
23.0
20.6
23.0
Ill-defined among home deaths (%)
56.1
58.4
25.5
17.4
17.4
38.6
Total ill-defined deaths (%)
21.2
25.4
8.8
6.4
6.0
13.1
Home ill-defined out of total ill-defined (%)
74.0
82.0
46.6
62.7
59.7
67.8
in Brazil. Figure 2 shows patterns for malnutrition and
road traffic accidents to illustrate this point. There were
very few deaths due to road traffic accidents in the
North. While the age patterns for road traffic accidents
were similar across regions, this is not so in the case of
malnutrition, with the Northeast and Southeast regions
tending to code larger numbers of deaths among the
elderly to this cause. Such variations should be
examined in more detail through focussed research
to identify whether these variations are real, or a result
of certification and coding practices.
Reliability
Comparisons between observed and predicted proportionate mortality by age, sex and cause at regional level
reveal several important findings. For instance, significantly higher than expected (42 SDs) proportionate
mortality from injuries occurs among males at adult
ages in all regions, particularly in the North and CentreWest, and was also markedly high for infants in the
South. Communicable diseases account for higher
proportions in males (35–59 years) and females
(50–69 years) in the Southeast, and also among the
elderly in both sexes in the northern regions (Figure 3).
Trends in cause-specific mortality over time for the
leading causes of death in males and females in each
region are consistent, and there are no fluctuations to
suggest problems with data quality arising from
changes in certification or coding practices.
Policy relevance
Assessment of regional mortality data suggests that
data availability is both timely (about 18 months time
lag) as well as adequate for subnational analyses.
Tabulations according to almost all the variables on
the medical certificate (approximately 60) are available
for municipalities and other subnational geographic
divisions, and include some socioeconomic variables
such as occupation and ethnicity, although data on
these variables are considered less accurate.17,21
Discussion
To our knowledge this is the first ever systematic
evaluation of the quality of mortality and cause of death
data for Brazil, according to a framework and specific
criteria developed from similar studies elsewhere.6,22,23
In addition to using a comprehensive framework, this
article also demonstrates the application of findings
from one criterion i.e. completeness of death registration; to adjust reported mortality statistics to estimate
region specific mortality patterns in Brazil. The results
have two potential uses: first, they could be used to
adjust reported statistics to develop cause-of-death
estimates by region (the subject of a forthcoming
article); and secondly, they inform local health managers about the quality of data generated by their
mortality information systems, and may well result in
measures to improve data quality. The evaluation
results suggest that Brazilian regional mortality statistics are of reasonable quality for the majority of criteria,
although there remains considerable heterogeneity
among regions for criteria such as completeness and
proportions of ill-defined causes.
In terms of coverage of data we have assumed that
compliance is 100%, given that there is an existing
provision for government health funding which permits
fund disbursement only upon satisfactory submission
of mortality data for a specified reporting period.4,24
Further detailed investigation of this aspect would not
only lead to more accurate assessments of completeness
of death registration, but also identify specific issues
that impede efficient compilation and reporting of
mortality statistics.
Measures of completeness are essential for accurate
estimation of summary measures of mortality such as
life expectancy at birth, and to assess potential
reporting bias from deaths not captured in cause-ofdeath statistics. Application of indirect demographic
methods suggests that levels of completeness in Brazil
and regions are generally satisfactory, more so in the
South and Southeast. Death registration is much less
complete in the Northeast and North, a finding
consistent with other studies using different methodologies.3,25,26 Registration is more complete for
males than females, confirming the pattern that has
been found in other countries.5
According to our assessment, the proportion of illdefined and vague causes used to code deaths in
Brazil remains a key issue. Although the proportion
has been declining, it is still high (13.1% of all deaths
were assigned an ill-defined cause in Brazil in
2002–2004), and accounts for nearly a quarter of all
deaths in the North and Northeast. For this reason,
recent Brazilian analyses of causes of death have been
EVALUATION OF CAUSE-OF-DEATH STATISTICS
Road traffic accidents - North- 2003
Number of deaths
3000
2500
600
2000
400
1500
1000
200
500
0
<01
Males
Road traffic accidents - Northeast - 2003
3000
Number of deaths
Malnutrition - North- 2003
800
0
Malnutrition - Northeast - 2003
800
2500
600
2000
1500
400
1000
200
500
<01
0
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
<01
Males
Road traffic accidents - Southeast- 2003
Number of deaths
3000
2500
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Females
Malnutrition - Southeast- 2003
800
600
2000
400
1500
1000
200
500
0
<01
Road traffic accidents - South- 2003
3000
0
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Males
Number of deaths
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Females
0
2500
<01
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Females
Malnutrition - South- 2003
800
600
2000
1500
400
1000
200
500
0
<01
0
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Males
Road traffic accidents - Center West - 2003
3000
Number of deaths
897
2500
<01
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Females
Malnutrition - Center West - 2003
800
600
2000
400
1500
1000
200
500
0
<01
0
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Males
<01
01−04 05−09 10−14 15−19 20−29 30−39 40−49 50−59 60−69 70−79 80 e+
Females
Figure 2 Age patterns by sex for deaths from road traffic accidents and malnutrition, regions of Brazil, 2003
carried out only in capitals or selected municipalities
which are considered to have better data quality.27
Strengthening of the Death Investigation Service
could improve this situation.
For defined causes, although the medically certified
causes of death may well be inaccurate for individuals,
possible compensating error patterns at the population
level may mean that misdiagnoses may not be that
important from a public health perspective20 especially
for broad cause groupings.12 It is difficult to assess how
valid this claim may be across Brazil since the majority
of published studies have concentrated on the South
and Southeast and may well not be representative of
the country as a whole (see Appendix 2). An overall
assessment of content validity of reported causes of
death requires a full-scale epidemiological study on a
national sample of deaths,28 applying standard data
collection methods, uniform case definitions, and using
defined statistical measures of validity. Such a study
needs to be conducted in Brazil.
4
4
4
2
2
2
0
0
0
−2
−2
−2
−4
−4
−4
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
−6
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
0
South, Males
6
6
6
4
4
4
2
2
2
0
0
0
−2
−2
−2
−4
−4
−4
−6
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
0
1
5
Brazil, Females
6
−6
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
North, Females
4
4
2
2
0
−4
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Southeast, Females
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
CentreWest, Males
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Northeast, Females
4
2
0
−2
1
6
6
S.D.
S.D.
Southeast, Males
S.D.
Northeast, Males
6
0
−2
−2
−4
−4
−6
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
85
0
1
5
South, Females
6
6
4
4
2
2
2
0
0
0
−2
−2
−2
−4
−4
−4
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
CentreWest, Females
6
S.D.
4
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Age
Group 1
Group 2
−6
0
1
5
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Group 3
* Group 1: Communicable, maternal, perinatal and nutritional conditions; Group 2: Noncommunicable diseases; Group 3: Injuries
Figure 3 Number of standard deviations by which observed cause-specific mortality proportions differ from mean predicted proportions, Brazil and regions,
2002–2004
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
S.D.
North, Males
6
898
Brazil, Males
6
EVALUATION OF CAUSE-OF-DEATH STATISTICS
It is likely that the completeness and validity of data
have improved in all regions in recent years, owing to
the efforts which have been made mainly by the
Brazilian Center for Classification of Diseases and the
Brazilian government to strengthen the vital registration system.4,24 Recent efforts of the Brazilian government have focussed on helping states and
municipalities from the North and Northeast regions
strengthen expertise for household interviews and
health service investigations targeting a reduction in
ill-defined causes of death.29 However, these changes
in data quality parameters over time, although
positive, can lead to substantial changes in time
trends in cause-specific mortality and have to be
taken into account in analysis.
Interestingly, the relationship between the level of
mortality from all causes and the relative composition
of causes in Brazil and regions follows quite closely
the expected patterns from application of the model
proposed by Salomon and Murray (2002),13 even for
the Northeast where injury deaths are relatively
consistent with predictions. Thus, even though the
levels of ill-defined causes are high, and with a
different age distribution to defined causes,15 it is
possible that the ill-defined causes have not substantially changed the broad proportional distribution of
causes of death in the Northeast. On the other hand,
in the North and Centre-West the observed proportions of deaths from injuries among adult males were
markedly higher than the predicted values, as for
infants in the South. More complete registration of
adult deaths from injury could possibly explain these
findings in the North. In the Centre-West, however,
the mortality rate for road traffic accidents is
particularly high among adult males.
These differences as well the higher proportion of
communicable diseases in males in the Southeast could
be real or artefactual, since the current mortality
patterns arising from violence and HIV/AIDS are not
adequately represented in the historical cause of death
models used to predict the comparison distributions for
this assessment. However, these differences appear
epidemiologically plausible, given the reported high
proportions of deaths due to violence in Brazil30 and the
possibility of HIV/AIDS-related deaths in the Southern
region of Brazil. On the other hand, the observed higher
than expected proportions of deaths due to communicable diseases among the elderly in both sexes in the
Northern regions and in females (50–69 years) in the
Southeast are possibly an artefact, potentially arising
from a tendency to attribute deaths to infectious
complications (pneumonia, septicaemia), rather than
underlying non-communicable diseases (cancers,
stroke, dementia and other nervous system disorders).
For infants, the observed mortality rate for injuries
(71.4/100 000 live births) in the South was 63% higher
than in the Southeast (data not presented). This was
largely the result of comparatively high mortality from
aspiration related codes (ICD-10 W75, W78, W79 and
899
W84) which collectively accounted for 78% of infant
male deaths from injuries in the South. These findings
suggest that the observed differences are likely to be real
as the South has better quality cause-of-death information than other regions, although it is also possible that
some of these codes were SIDS (Sudden Infant Death
Syndrome) which could be underreported in the
South.31 It has also been observed that the clinical
diagnosis of aspiration (W78-79) is not always confirmed on autopsy.32 All these findings warrant more
in-depth investigation, to ascertain the exact nature of
the differences between observed and predicted proportionate cause-specific mortality.
From the perspective of timeliness and geographical
disaggregation, the MIS could be considered highly
adequate, as the final output is published within 2
years of the reference period and the system produces
statistics which could be used to evaluate the health
situation of the population at all administrative levels,
including municipalities. The MIS therefore has the
potential to be an important tool for monitoring the
effects of health and development programs within
different geographic areas.
Although the application of the cause-of-death
evaluation framework at the sub national level in
Brazil has important findings, there are certain
findings from the application of the cause-of-death
evaluation framework at the sub national level in
Brazil should be viewed with some caution. First, in
the assessment of completeness of death registration,
there is uncertainty in the use of indirect demographic methods such as the EGM, arising
from assumptions regarding absence of migration,
and constancy of incompleteness across all ages.
Although these assumptions might not be supported
by empirical data for Brazil, it should be borne in
mind that there is no adequate method for all
situations. Migration does occur in Brazil, and is
generally concentrated at young adult ages, which
potentially affects our measures of completeness. In
our study, we have followed recommended guidelines
for assessing plausibility of completeness estimates,33
and while it is difficult to assess their accuracy, we
judge them to be reasonable based on the
relative pattern of completeness across regions.
Secondly, regarding cause-of-death attribution, apart
from the issue of ill-defined causes, there is uncertainty
regarding the validity of registration diagnoses for
specified causes of death. We assume that any errors in
diagnoses that result in misclassification are likely to
cancel out in the final tally, observed in several
international settings, but this assumption requires
periodic verification from studies on content validity
mentioned earlier. Finally, there is uncertainty in the
applicability of the cause-of-death models used to test
consistency between total and cause-specific mortality
in Brazil.
Despite these limitations, we believe that the
framework provides adequate insight into the quality
900
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
of data in Brazil, and indicates that the South,
Southeast and Centre-West regions have satisfactory
cause-of-death data from the MIS. For the Northern
and Northeast regions, data on both levels of
mortality and causes of death need to be corrected
for biases reported from this evaluation, prior to their
use for their intended purposes. Further applications
of the evaluation framework at sub national level will
reveal additional nuances in the methodology, which
could be taken into account in future iterations.
Innovative and sustained efforts will be required to
achieve the goal of complete death registration and
high-quality cause-of-death certification and coding
practices, especially in the Northern regions. These
regions with poor data quality are also those with the
lowest socioeconomic levels, with the greatest health
inequalities, and the greatest need for health investments. Closing the gap between regions in terms of
mortality data quality will also serve to strengthen the
evidence base for programs to reduce mortality
differentials in Brazil.
Conclusions
Problems with quality of cause-of-death statistics
across Brazilian regions, especially for criteria such
as completeness and ill-defined causes can influence
generalizability and validity of cause-of-death reports,
and must be considered in the interpretation and use
of data to minimize bias.
Measures and investments needed to improve data
quality are most urgently needed with respect to
death registration and certification practices in the
North and Northeast regions. However, certification
and coding practices in other regions of the country
also need to be strengthened in order to increase the
utility of cause-of-death data for health policy.
Supplementary data
Supplementary data (appendices) are available at IJE
online.
Acknowledgments
E França and DX Abreu were sponsored by the
Brazilian Coordination of Improvement of Higher
Education Personnel (CAPES), Grants No. 124206-7
and 233206-3.
Conflict of interest: None declared.
KEY MESSAGES
Evaluation of cause-of-death statistics based on a comprehensive framework is essential to assess
usability of data for health policy and research.
Subnational cause-of-death statistics for regions in Brazil demonstrates considerable heterogeneity in
terms of completeness of death registration and proportions of ill-defined causes of death, but are
satisfactory for other dimensions of data quality.
Well-designed epidemiological studies are required to assess content validity of reported causes of
death.
Gaps between regions in terms of mortality data quality should be reduced in order to strengthen the
evidence base for health policy and research in Brazil.
References
1
2
3
4
5
Silveira MH, Laurenti R. Vital events: aspects of Brazilian
vital registration and the inter-relationship between the
present legislation and vital statistics. Rev Saúde Pública
1973;7:37–50.
Oliveira LAP, Simões CCS. IBGE and population research.
R Bras Est Pop 2005;22:291–302.
Vasconcelos AMN. Quality of mortality statistics in Brazil.
R Bras Est Pop 1998;15:116–24.
Mello-Jorge MHP, Laurenti R, Gotlieb SL. Quality
analysis of Brazilian vital statistics: the experience of
implementing the SIM and SINASC systems. Cieˆncia Saúde
Coletiva 2007;12:643–54.
Mathers CD, Mafat D, Inoue M, Rao C, Lopez AD.
Counting the dead and what they died from: an
6
7
8
assessment of the global status of cause of death data.
Bul World Health Organ 2005;83:617–25.
Rao C, Lopez AD, Yang G, Begg S, Ma J. Evaluating
national cause-of-death statistics: principles and application to the case of China. Bull World Health Organ
2005;83:617–25.
Santo AH, Pinheiro CE, Rodrigues EM. Comparative
evaluation of underlying causes of death processed by
the Automated Classification of Medical Entities (Acme)
and the Underlying Cause of Death Selection (Scb)
Systems. Rev. Saude Publica 1998;32:1–6.
Brazilian Ministry of Health. Procedures from the
Mortality Information System Manual. Brası́lia:
Fundação Nacional de Saúde, 2003. Available from:
http//portal.saude.gov.br/portal/arquivos/pdf/sis-mortalid
ade.pdf.
EVALUATION OF CAUSE-OF-DEATH STATISTICS
9
10
11
12
13
14
15
16
17
18
19
20
21
Hill K. Estimating census and death registration completeness. Asian Pac Popul Forum 1987;1:8–13, 23, 24.
Bennett NG, Horiuchi S. Mortality estimation from
registered deaths in less developed countries.
Demography 1984;21:217–34.
IBGE. National Household Sample Survey 2004. Available
from: http//www.ibge.gov.br/home/estatistica/populacao/
trabalhoerendimento/pnad2004/defaulttab-hist.shtm.
Murray CJL, Lopez AD. Estimating causes of death: new
methods and global and regional applications for 1990.
In: Murray CJL, Lopez AD (eds). The global burden of
disease. Boston: Harvard School of Public Health, 1996.
pp. 118–200.
Salomon JA, Murray CJL. The epidemiologic transition
revisited: compositional models for causes of death by age
and sex. Population Dev Rev 2002;28:205–28.
Brazil Gross Domestic Product (GDP/PIB) per Capita by
State:2000. database on the Internet. Demographia. [cited
2006 Nov 20]. Available from: http//www.demographia.
com/db-brazil-gdpreg.htm.
Mello-Jorge MHP, Gotlieb SL, Laurenti R. The national
mortality information system: problems and proposals for
solving them. II – deaths due to natural causes. Rev Bras
Epidemiol 2002;5:197–211.
Fonseca LAM, Laurenti R. The quality of medical
certification for the cause of death in São Paulo, Brazil.
Rev Saúde Pública 1974;8:21–29.
Cordeiro R, Peñaloza ERO, Cardoso CF et al. Validity of
information on occupation and principal cause on death
certificates in Botucatu, São Paulo. Cad Saúde Pública
1999;15:719–28.
Santo AH. Quality Evaluation of the Coding of Causes of
Death in the State of São Paulo, Brasil. Informe
epidemiolo´gico do SUS 2000;9:189–98.
Laurenti R, Mello Jorge MHP, Gotlieb SL. Maternal
mortality in Brazilian State Capitals: some characteristics
and estimates for an adjustment factor. Rev Bras Epidemiol
2004;7:449–60.
Laurenti R, Mello Jorge MHP, Gotlieb SL. The accuracy
of official mortality and morbidity statistics related
to chronic non-communicable diseases. Cieˆncia Saúde
Coletiva 2004;9:909–21.
Cardoso AM, Santos RV, Coimbra CEA Jr. Infant
mortality according to race/color in Brazil: what do the
national databases say? Cad Saude Publica 2005;21:1602–8.
22
23
24
25
26
27
28
29
30
31
32
33
901
Ruzicka L, Lopez AD. The use of cause-of-death statistics
for health situation assessment: national and international experiences. World Health Stat Quart 1990;43:249–57.
Mahapatra P, Chalapati Rao PV. Cause of death reporting
systems in India: a performance analysis. Natl Med J India
2001;14:154–62.
Brazilian Ministry of Health. Decree no.1882/GM of
18th Dec 1998. Available at: http/www.mp.se.gov.br/
CAD/Arquivos/port1882.doc.
Andrade CLT, Szwarcwald CL. Socio-spatial inequalities
in the adequacy of Ministry of Health data on births and
deaths at the municipal level in Brazil, 2000–2002.
Cad Saúde Pública 2007;23:1207–16.
Paes NA. Quality of death statistics by unknown
causes in Brazilian states. Rev. Saúde Pública 2007;
41:436–45.
Ishitani L, Franco GC, Perpétuo IHP, França E. Socioeconomic inequalities and premature mortality due to
cardiovascular diseases in Brazil. Rev Saúde Pública 2006;
40:684–91.
Gittelsohn A, Senning J. Studies on the reliability of vital
and health records: I. Comparison of cause of death and
hospital record diagnoses. Am J Public Health 1979;
69:680–89.
Brazilian Ministry of Health. Reorganization and qualification of health information systems. In: Ministry of
Health Health Surveillance in the Brazilian Unified
System (SUS): strengthening the response capacity to
old and new challenges. Brası́lia: MH, 2006. pp. 43–47.
Available at: http//portal.saude.gov.br/portal/arquivos/pdf/
livro-nova-vigilancia-web.pdf.
Brazilian Ministry of Health. Impact of violence on the
health of Brazilians. Brası́lia: Secretaria de Vigilancia em
Saúde. 2005.
Victora CG, Nobre LC, Lombardi C et al. Epidemiology of
sudden infant deaths in cities of Rio Grande do Sul,
Brazil: A comparative study of cases and controls. Rev
Saude Publica 1987;21:490–96.
Peres
LC.
Sudden
unexpected
infant
death
syndrome in Ribeirão Preto, Brazil. Rev Paul Med
1998;116:1803–7.
Hill K, Choi Y. The adult mortality in developing
countries project: substantive findings. Adult Mortality in
Developing Countries Workshop. California: The Marconi
Center, 2004. pp. 1–18.