Professional Documents
Culture Documents
EurJPop2012 PDF
EurJPop2012 PDF
net/publication/235937367
CITATIONS READS
16 965
7 authors, including:
Frova Luisa
Italian National Institute of Statistics
88 PUBLICATIONS 783 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Task Force on Revision of the European Standard Population 2013 View project
All content following this page was uploaded by Marilena Pappagallo on 29 August 2014.
Abstract Multiple cause of death (MCOD) data have been used to recalculate
mortality levels attributed to a given condition, and to determine the most frequent
associations of causes involving this condition. In this article, we begin with a
description of how the MCOD data is collected, and we discuss data quality. After
presenting the main indicators specifically developed to analyse these data, we
provide a concrete illustration of the method based on a comparison of cancer-
related mortality in Italy and France. The results for the two countries are strikingly
similar. The change in mortality levels is modest for most anatomic sites: cancer is
often selected as the underlying cause of death (UCD). The most notable exception
(cancer of the prostate) potentially exemplifies future trends, with more effective
treatments and increased incidence of this disease among elderly people due to
population ageing, cancers may more often play a contributing role in mortality. For
all anatomic sites, the reporting of a neoplasm as both underlying and contributing
cause of death is a feature of cancer-related mortality. We then categorize all other
associations into five patterns (‘degeneration of the contributing cause’, ‘risk factor
for the UCD’, ‘common cause’, ‘consequence/complication of the UCD’ and
‘symptom of the UCD’) that reflect current medical knowledge.
V. Egidi
Dipartimento di Scienze Statistiche, Sapienza Università di Roma (SUR),
Viale Regina Elena, 295, 00161 Rome, Italy
123
A. F. Désesquelles et al.
In an article published in the 60s, Dorn and Moriyama (1964) lent support to the
widely accepted idea that ‘a single cause, no matter how selected, no longer
adequately describes the morbid conditions responsible for a large proportion of
deaths’. About 25 years earlier, Theodore Janssen (1940) had been one of the first to
highlight the importance of tabulating and analysing multiple1 causes of death. His
recommendation was followed by Lillian Guralnick (1966) who proposed a number
of methods for tabulating the exhaustive information reported on death certificates.
Almost 50 years after these pioneering studies, mortality is still analysed primarily
in terms of the underlying cause of death (UCD). Yet, multiple cause of death
(MCOD) data are increasingly produced and disseminated by statistical offices, and
they have been used in several ways that can be broadly classified into three lines of
research.
First, demographers and statisticians have used these data to build multiple-cause
life tables (David 1974; Lee and Thompson 1974; Manton et al. 1976; Manton and
Poss 1979; Wong 1977). Here, the aim is to compute the probability of death when
1
Multiple causes include both underlying and contributing causes.
123
Analysing Multiple Causes of Death
one or more causes of death are eliminated. Standard methods that rely on the UCD
and on competing risks theory require assumptions about independence/dependence
among causes. These assumptions may or may not be appropriate, depending on the
morbid processes under study. MCOD statistics offer interesting options for
addressing this problem: morbid states may indeed be redefined in order to represent
patterns of diseases which are more nearly independent (Manton and Poss 1979).
The second line of research aims at recalculating mortality levels attributed to a
given condition. To put it differently, the aim is to calculate mortality indicators that
take into account all the conditions reported on death certificates. This approach has
been developed mainly to reassess the contribution to mortality of certain conditions
that are frequently reported as a contributing cause of death. Among others, diabetes
(Jougla et al. 1992; Goldacre et al. 2004c; Romon et al. 2008; Tsung-Hsueh et al.
2010), asthma (McCoy et al. 2005; Fuhrman et al. 2009), hypertension (Wing and
Manton 1981), alcohol and tobacco use (Speizer et al. 1977; Nizard and Munoz-
Pérez 1993; Nam et al. 1994; Goldacre et al. 2004d), hepatitis (Wise and Sorvillo
2005), and musculoskeletal disorders (Coste and Jougla 1994; Ziadé et al. 2010)
have received particular attention. But cancers (Wrigley and Nam 1987; Mannino
et al. 1998; Goldacre et al. 2004a), heart failure, and stroke (Goldacre et al. 2005;
Goldacre et al. 2008) as well as other respiratory diseases (Goldacre et al. 2004b;
Fuhrman et al. 2006) have been studied too. Some authors have used the MCOD
approach on a larger or even unrestricted list of conditions (Manton and Stallard
1982; Crews 1988; White et al. 1989; Steenland et al. 1992; Tardon et al. 1995;
Mackenbach et al. 1995; Goldacre et al. 2003; Frova et al. 2009; Désesquelles et al.
2010). Most analyses we found in our review are country-specific and some have
focused on deaths at old or very old ages (Kohn 1982; Manton 1986; Manton and
Myers 1987; Stallard 2002; Désesquelles and Meslé 2004). With the rise in life
expectancy, a growing share of the population reaches ages where comorbidity is
very frequent. As a consequence, the number of reported conditions on the death
certificates is expected to increase.
Finally, the third line of research focuses on cause combinations. ‘Statistics
showing combinations of causes come nearer the truth than do those based on the
single cause principle because the majority of deaths actually result from a
combination of causes’ (Janssen 1940). When Janssen wrote these words, the
population ageing process was not as advanced as it is now, but degenerative and
man-made diseases that characterize the third stage of the epidemiological transition
(Omran 1971) had started to represent a growing share of total morbidity. Following
Nam (1986), Crews (1990) suggests that the underlying-cause approach oversim-
plifies the changes occurring in the epidemiological profile, and he believes that the
MCOD approach is more appropriate to examine the complexity of degenerative
disease processes.
In this article, we begin with a description of how MCOD data are collected, and
we investigate data quality. Then, we present the various indicators that have been
developed to revaluate mortality levels and examine combination of causes. Finally,
we provide a concrete illustration of the analytical potential of this approach based
on the comparison of cancer-related mortality in Italy and France. As we shall see,
the French and the Italian cause of death data are not fully comparable but the data
123
A. F. Désesquelles et al.
collection methods and the health and mortality profiles of the two countries are
similar enough to ensure that the comparison is not meaningless. This being said, we
believe in the heuristic virtues of cross-country comparison. As we cannot be certain
about data quality, both similarities and dissimilarities in the results obtained in the
two countries must be interpreted with caution, but they are a potentially fruitful
source of information. In particular, they can help to identify failures of the coding
and/or certification practices as well as other drawbacks of the data gathering tools.
Data for the two countries are for the year 2003. Among all causes of death,
cancer ranks first in France and second in Italy. In a previous study (Désesquelles
et al. 2010), we showed that, compared to less life-threatening diseases, cancers
are more often selected as the underlying cause of death. However, this result may
vary across anatomic sites because survival chances are not the same for all
cancers. Neoplasms with the best survival profiles can be considered as chronic
conditions, and long-term cancer survivors may develop other health problems
that they may eventually die from. All in all, given the increased incidence of
cancers among elderly people due to population ageing, and the development of
more effective treatments, cases of comorbidity involving cancers are expected to
become more frequent (Eakin et al. 2006). So, from a public health perspective,
the study of the interactions between cancers and other diseases will be
increasingly relevant.
Though the idea that death often results from the conjunction of several diseases,
conditions or risk factors, was already present in the mind of statisticians who
implemented the first International Classification of Diseases at the end of the 19th
century, their main efforts at that time focused on the production of tables
attributing one cause to one death. Strategies for deciding about the cause to be
chosen have changed over time. First, people in charge of data collection drew up
rules to help choose the underlying cause from among all the information reported
on the death certificate. At that time, they did not trust certifying physicians, who
often considered the immediate cause of death as the underlying one. To help coders
make a decision, the US Bureau of Census issued the first ‘Manual of Joint Causes
of death’ that provided practical cases for which a choice had to be made, and rules
to be followed (Moriyama 1956). This manual was revised several times during the
interwar years. It was only after the Second World War that greater trust was placed
in the certifying physicians. This trend was reinforced by the introduction of new
death certificates on which certifiers had to determine and describe the most likely
sequence of events that led to death, starting from the immediate cause and going
back into the past to identify the first medical problem that initiated the whole
process.
The elaboration of very precise rules to select the UCD improved cross-country
comparability. For a long time, the concern to ensure UCD comparability diverted
123
Analysing Multiple Causes of Death
people in charge of cause of death statistics from taking into account all the
information reported on the death certificates. The ultimate aim was still to produce
consistent tables for the UCD. It is fair to say that before the computer age, it was
not a simple task to deal with tables linking several causes for one death. Until now,
not many countries have published data on multiple causes of death. The US, where
these tables have been produced in electronic format since 1968, remains an
exception. In France and Italy, the interest in multiple causes of death is much more
recent.
In France, the recording of multiple causes began in the early 50s with the
implementation of a new certificate on which the physician could report several
conditions that had contributed to the death (Aubenque 1975). This certificate was
based on the international model recommended by the World Heath Organization
(WHO). Physicians progressively became familiar with this new tool, and the
proportion of deaths with more than one coded condition grew from 25 % in 1955 to
45 % in 1967. After this period of regular increase, the proportion jumped suddenly
to 60 % in 1968. That year, the recording of causes of death was taken over by the
French National Institute for Health and Medical Research (Inserm), and the quality
of statistics immediately improved. Another jump occurred in 2000 with the
adoption of the ICD-10 and the implementation of an automatic coding system. In
the 2000s, the proportion of deaths with more than one coded condition reached
85 %. However, Inserm has only provided access to the complete information
reported on the death certificates since 2000.
In Italy, the collection, processing and dissemination of causes of death statistics
has been under the authority of the National Institute of Statistics (Istat) since 1926.
The Italian death certificate, also based on the WHO model, was introduced in 1951
with the adoption of the ICD-6. A first evaluation of co-morbidities over the period
1951–1954 was performed in 1958 (Istat 1958), but collection of MCOD data did
not begin until 1995, with the introduction of the automated coding system (Istat
2004). In 2008, 98 % of death certificates had more than one coded condition.
Codes for all reported conditions have been available in the databases since 1995,
but the first official release of MCOD data is for 2003, when the ICD-10 was
adopted.2 In Italy, data on contributing causes are not available for deaths attributed
to an external cause and for infant deaths. For the remainder of this study, all these
deaths have thus been excluded from the analysis.
The production of cause of death statistics relies on two steps that are both crucial
for the quality of the MCOD data. First, the certifying physician reports the chain of
causes leading to death on the certificate. Second, this information is coded. A
growing number of countries now use an automatic coding system. To our mind,
this innovation represents a major advance towards improved quality of MCOD
data. Human intervention is limited to problematic cases that cannot be processed
automatically and, at least in theory, the WHO coding rules can be applied
2
Detailed tables are available since 2003 on the Istat website.
123
A. F. Désesquelles et al.
3
Note that both the selection of the underlying cause and the reporting of the contributing causes of
death are potentially affected by inadequate/insufficient training.
4
The two certificates are displayed in the Appendix.
123
Analysing Multiple Causes of Death
3,5
0,5
0
1-9 10- 20- 30- 40- 50- 60- 70- 80- 90+
19 29 39 49 59 69 79 89
Fig. 1 Average number of causes reported on death certificates, all entries and entries on parts I and II—
deaths over the age of one, excluding deaths from external causes, France and Italy, 2003. Mechanisms of
death and ill-defined contributing causes are excluded. Data France: Inserm CépiDc mortality database/
Italy: Istat mortality database
13 lines in Italy (8 in part I, 5 in part II5) versus 6 in France (4 in part I, 2 in part II). This
is may explain why, on average, more causes are listed in Italy (4.0) than in France
(3.1). Some of the reported contributing conditions are ill-defined causes or
mechanisms of death that do not shed light on the mortality process. In a previous
study (Désesquelles et al. 2010), we found that these causes are more frequently
reported as contributing causes in Italy.6 For the remainder of this article, all ill-defined
causes or mechanisms of death as contributing cause entered on death certificates have
been excluded from the computation.7
Figure 1 displays the average number of causes entered on death certificates in
France and Italy by age, for all entries as well as for entries on parts I and II
separately. The average number of entries on part II is very similar in the two
countries. So the difference between the two countries is explained by the fact that
the Italian certifying physicians report more causes on the first part of the death
certificate. Figure 1 also shows that the average number of reported causes varies by
age. Both real differences in the morbid process and variations in certification
practices by age at death may explain this finding. The increase up to age 80
(Mackenbach et al. 1995; Désesquelles and Meslé 2004; Désesquelles et al. 2010) is
consistent with the increased frequency of comorbidity with age, but the subsequent
5
Part I is for the diseases or conditions that directly led to death. Part II is for any other significant
condition that unfavourably influenced the course of the morbid process but is not related to the condition
directly causing death. In many cases, it is difficult to decide whether a condition should be mentioned in
part I or part II. If not explicitly specified, all indicators presented in this article are computed taking into
account entries in parts I and II.
6
When excluding these causes from the computation, the difference between the two countries narrows
(2.4 in France vs. 3.0 in Italy).
7
Note that we considered as redundant information and removed from the databases any recorded cause
that, for a given death, is exactly identical (at the four-digit level of the ICD-10 classification) to another
already registered cause.
123
A. F. Désesquelles et al.
decline is more puzzling. We cannot rule out a selection effect, people with fewer
diseases are likely to live longer. But we suspect that it is an artefact resulting from a
less thorough description of the morbid process when death occurs at very old ages.
At this point it is worth noting that when comparing the two countries for some
specific underlying causes of death, the situation may be very different from the one
described above. Figures 2a–c display the average number of death certificate entries
by age in France and Italy for three underlying causes that exemplify the main patterns
emerging from the comparison of the two countries. In the first pattern (corresponding
to the leading causes of death: neoplasms, diseases of the circulatory system and
diseases of the respiratory system), more causes are mentioned in Italy than France at
all ages, and the gap increases with age. The second pattern (infectious and parasitic
diseases as well as diseases of the endocrine, nutritional and metabolic system,
diseases of the digestive system and diseases of the skin and the subcutaneous tissue)
differs from the first one insofar as the gap between the two countries remains stable
with age. For all other underlying causes of death, the average number of entries is very
similar in the two countries, and there is no decline with age of the average number of
entries on the death certificates. So for these underlying causes of death, the
differences in death certificate format have no impact.
Apart from the effect of age and UCD, several other characteristics of the
decedent may affect the number of causes reported on the death certificate. For
example, the information available to the certifying physician is likely to be more
precise when death occurs in a medical structure. Marital status may also play a
role: the spouse, if present, may provide information on the decedent’s medical
history. We use a Poisson model to examine simultaneously the effect of the
following decedent characteristics on the number of listed causes: age at death,
UCD, sex, marital status (married, never married, widowed and divorced) and place
of death (hospital, residence for elderly people, private residence, other place). The
results for the two countries (Table 1) are very similar. As models are performed on
large numbers (507,547 deaths in France and 550,835 in Italy), almost all effects
reach statistical significance. After controlling for all characteristics, the shape of
the number of reported causes by age (increase followed by a decline in the oldest
age groups) is confirmed. We find fewer reported causes for females and for never-
married decedents but corresponding relative risks are only slightly under one. The
effect of the place of death is clearer: more causes are reported when death occurred
in hospital and, for Italy, in a residence for elderly people. The lowest relative risk is
for deaths occurring in a private residence. All these results are very similar to those
obtained by Wall (2005) with US data, except for marital status. In her study, death
certificates of married people less frequently listed more than one reported
condition. Regarding the UCD, some causes (diseases of the musculoskeletal
system/connective tissue, diseases of the digestive system, endocrine, nutritional
and metabolic diseases, as well as infectious and parasitic diseases) are associated
with a higher number of reported conditions, and some others (diseases of the
nervous system, mental and behavioural disorders, diseases of the respiratory
system and, not surprisingly, ill-defined causes/mechanisms of death) with a lower
number. The two main causes of death (neoplasms and diseases of the circulatory
system) occupy an intermediate position.
123
Analysing Multiple Causes of Death
(a)
3,50
3,00
France
2,50
Italy
2,00
1,50
1,00
1-9 10- 20- 30- 40- 50- 60- 70- 80- 90+
19 29 39 49 59 69 79 89
(b)
Average number of causes
4,00
3,50
3,00
France
2,50
Italy
2,00
1,50
1,00
1-9 10- 20- 30- 40- 50- 60- 70- 80- 90+
19 29 39 49 59 69 79 89
(c)
4,00
Average number of causes
3,50
3,00
France
2,50
Italy
2,00
1,50
1,00
1-9 10- 20- 30- 40- 50- 60- 70- 80- 90+
19 29 39 49 59 69 79 89
Fig. 2 Average number of causes reported on death certificates by age for three underlying causes of
death. Deaths over the age of one, France and Italy, 2003. a Neoplasms as UCD. b Infectious and parasitic
diseases as UCD. c Diseases of the nervous system as UCD. Mechanisms of death and ill-defined
contributing causes are excluded. Data France: Inserm CépiDc mortality database/Italy: Istat mortality
database
2 Indicators
123
A. F. Désesquelles et al.
Table 1 Impact of various decedent characteristics on the number of causes reported on the death
certificate: relative risks resulting from a Poisson model
France Italy
Age at death
1–39 0.92 0.85
40–49 0.95 0.91
50–59 0.97 0.93
60–69 0.99 0.96
70–79 Ref Ref
80–89 0.98 1.00 ns
90? 0.93 0.94
Sex
Males 1.03 1.02
Females Ref Ref
UCD
Infectious and parasitic diseases 1.07 1.13
Neoplasms 1.00 1.02
Diseases of the blood and blood-forming organs 1.11 1.00 ns
Endocrine, nutritional and metabolic diseases 1.18 1.15
Mental and behavioural disorders 0.96 0.81
Diseases of the nervous system 0.93 0.87
Diseases of the circulatory system Ref Ref
Diseases of the respiratory system 0.96 0.93
Diseases of the digestive system 1.12 1.06
Diseases of the skin and subcutaneous tissue 1.01 ns 0.93
Diseases of the musculoskeletal system/connective tissue 1.25 1.10
Diseases of the genitourinary system 1.12 0.98
Other diseases 1.20 1.09
Symptoms, signs, abnormal findings, ill-defined causes 0.44 0.36
and Mechanisms of death
Marital status
Divorced 1.00 ns 1.00 ns
Widowed 1.00 ns 1.00 ns
Never married 0.99 0.98
Married Ref Ref
Place of death
Private residence 0.87 0.90
Hospital Ref Ref
Residence for elderly people 0.96 1.04
Other 0.89 0.96
Mechanisms of death and ill-defined contributing causes are excluded. Deaths over the age of one,
excluding deaths from external causes, France and Italy, 2003. Data: France: Inserm CépiDc mortality
database/Italy: Istat mortality database
ns not significant
123
Analysing Multiple Causes of Death
u dc;xis the number of deaths observed at age x with underlying cause u and con-
tributing cause c; u dx is the number of deaths observed at age x with cause u as
underlying cause; dx is the standard number of deaths at age x
Following a more investigative approach, several indicators have been elaborated
to determine the strength of any association of causes of death. Considering that the
reporting of a cause on the death certificate is a dichotomous event, the strength of a
given association can be estimated with odds ratios. The odds ratio for causes A and
B is defined as the ratio between the odds of cause B being reported on the
certificate, when cause A is also mentioned, divided by the odds of cause B being
reported, when cause A is not mentioned (Chamblee and Evans 1982; Smith and
Kliewer 1995; Wilkins et al. 1997; Redelings et al. 2007; Frova et al. 2009). The
ratio of observed to expected joint frequencies is another commonly used indicator.
In its general form, it is computed as the ratio between the actual number of deaths
involving a given combination of causes, and the expected number of deaths with
that combination, assuming that the combined causes are independent (Israel et al.
1986; Stallard 2002). A ratio over one suggests that causal relations between the
combined causes may exist (Australian Bureau of Statistics 2003). Mackenbach
et al. (1995) have proposed a more refined indicator. The ‘standardized prevalence
123
A. F. Désesquelles et al.
ratio’ is the result of an indirect standardization procedure: the age- and sex-specific
prevalence rates of a contributing cause among all deaths are applied to the age and
sex distribution of the deaths attributed to a given UCD to compute an expected
number of deaths. Having the structure of a standardized mortality ratio, this
indicator depends on the age and sex distribution of the UCD under study. As a
consequence, direct comparisons between underlying causes are not allowed. To
overcome this problem, we developed the cause of death association indicator
(CDAI) that can be used to compare various UCDs within a country, as well as
across various countries for a given UCD (Désesquelles et al. 2010). In the formula
of the CDAIs, both the observed and expected prevalence rates are age standardized.
The CDAI between a contributing cause c and an underlying cause u compares how
frequently (1) the cause c is contributing to deaths assigned to the underlying cause
u and (2) the cause c is contributing to all deaths of the country. It is thus the ratio
between an observed and an expected proportion of deaths with cause c as
contributing cause. It is worth noting that the CDAI may also be seen as the ratio
between the standardized prevalence at death of the combination u, c conditional to
the specific underlying cause u (SPc/u), and the unconditional standardized
prevalence of the same contributing cause (SPc)8:
P u dc;x P
P u dc;x
SPc=u x u dx d x x dx x dx
CDAIu;c ¼ ¼ P P 100 ¼ P du dc;xx 100
SPc dc;x
d x d x x d x
x dx x d x
u dc;xis the number of deaths observed at age x with underlying cause u and con-
tributing cause c; u dx is the number of deaths observed at age x with cause u as
underlying cause; dc;x is the total number of deaths observed at age x with cause c as
contributing cause (regardless of the underlying cause) dx is the total number of
deaths observed at age x (regardless of the underlying cause); dx is the standard
number of deaths at age x.
Note that the leading causes of death contribute more to the value of the
denominator of this indicator than other causes of death. As a consequence,
associations involving causes that frequently contribute to deaths due to a cancer or
a disease of the circulatory system are less likely to emerge as strong associations.
d
One possible alternative consists in replacing the dc;xx terms in the denominator by the
d
average value, for the various UCDs, of the uu dc;xx terms. In fact, both formulas lead to
very similar results.9
Finally, a choice needs to be made about how to count causes reported on a given
death certificate that belong to the same group under the classification used for the
8
As the indicator is based on the ratio of averages, age distributions of its components (udc,x/udx and dc,x/
dx) may affect the value of the CDAI. Nevertheless, we find that for cancers as UCD, the correlation
between CDAIs and the age structure of the components is close to zero.
9
More precisely, the two series of indicators computed for a given contributing cause and various
underlying causes are proportional (the denominator is the same for each UCD), so the conclusions are
unchanged. The correlation between the two series of indicators computed for a given UCD and various
contributing causes is generally very close to one.
123
Analysing Multiple Causes of Death
analysis. In this study, for the computation of the SRMUs, these causes are counted
only once. For CDAIs, whatever the underlying cause of death,10 only one mention
of a given group as contributing cause is considered in the computation of the
indicator. Distinct ICD-10 codes that are considered ‘redundant’ with an aggregated
classification, could of course be considered as different if a more detailed
categorization was chosen. This implies that values of the CDAIs computed at
different levels of the classification cannot be compared.
3 Results
Table 2 displays the basic descriptive indicators computed for all deaths involving a
neoplasm. When only the underlying cause of death is accounted for, the
standardized11 mortality rate for neoplasms is only slightly higher in France
(1,957 per million) than in Italy (1,918 per million). As the cancer-specific SRMU is
1.1 in both countries, the situation is unchanged when all mentions of a neoplasm
are accounted for (2,125 per million in France vs. 2,110 per million in Italy). Most
site-specific values of the SRMU are very close to one. If we exclude the secondary
malignant neoplasms that, not surprisingly, are mainly mentioned as a contributing
cause, the highest values are for the ‘malignant neoplasms of ill-defined/unspecified
and independent primary multiple sites’ (9.0 in Italy and 3.2 in France), for ‘benign
neoplasms/neoplasms of uncertain or unknown behaviour’ (1.7 in Italy and 1.6 in
France), and for ‘malignant neoplasms of the skin and the subcutaneous tissue’ (1.5
in both countries) and ‘malignant neoplasms of the prostate’ (1.4 in both countries).
The ranking of the mortality rates by anatomic site produces results that are
almost unchanged when all causes are taken into account. In both countries,
‘malignant neoplasms of larynx, trachea, bronchus and lung’ come first, followed by
‘malignant neoplasms of small intestine, colon, rectum, anus and other/ill-defined
digestive organs’ and ‘malignant neoplasm of lymphoid, haematopoietic and related
tissue’. In Italy, the ranking of the three following sites (‘liver, intrahepatic bile
ducts, gallbladder and other unspecified parts of biliary tract’, ‘breast’ and
‘stomach’) is unchanged. In France, ‘malignant neoplasms of breast’ and ‘malignant
neoplasms of prostate’ exchange their rankings. In both the single and the MCOD
approach, ‘malignant neoplasms of liver, the intrahepatic bile ducts, gallbladder and
other unspecified parts of biliary tract’ and ‘malignant neoplasms of stomach’ rank
lower in France than in Italy. In contrast, rates for malignant neoplasms of prostate,
of oesophagus, and for ‘malignant neoplasms of lip, oral cavity, pharynx’ are higher
in France.
In all, the impact of the MCOD approach in terms of cancer-related mortality
levels is modest. The most notable exception is for neoplasms of the prostate. In a
10
So, we do not exclude associations between an underlying cause and a contributing cause belonging to
the same group.
11
The standard population is the WHO 2003 European population by sex and five-year age groups.
123
A. F. Désesquelles et al.
Table 2 Standardized mortality rates (per million) for each cause reported as underlying cause or
multiple cause and standardized ratio of multiple to underlying cause (SRMU)—Neoplasms, France and
Italy, 2003
Underlying Multiple causea SRMU
cause
In France, there are very few deaths with a secondary malignant neoplasm as underlying cause, and the
corresponding SRMU has not been computed. Data: France: Inserm CépiDc mortality database/Italy: Istat
mortality database
a
Mechanisms of death and ill-defined contributing causes are excluded
b
Excluding deaths due to external causes and infant deaths
recent article based on US mortality data, Fink et al. (2011) also find that neoplasms
of the prostate are frequently reported as contributing cause of death. They suggest
that this might indicate better survival chances for this cancer. But it may also
123
Analysing Multiple Causes of Death
reflect the fact that diagnosis often occurs at advanced ages when people may have
already developed other diseases.
In both countries, the average number of causes reported on death certificates with a
neoplasm as UCD is higher than for all causes of death taken together (3.1 vs. 2.9 in
Italy, 2.5 vs. 2.4 in France), and the share of those certificates with no contributing
cause is lower (10 vs. 16 % in Italy, 23 vs. 33 % in France). When looking at the
standardized prevalence rates of the various combinations of causes (see Tables 3
and 4 in the appendix), we find that for all anatomic sites, the most frequent
associations are with neoplasms, diseases of the circulatory system, diseases of the
respiratory system, as well as—but to a lesser extent—diseases of the digestive
system and endocrine, nutritional and metabolic diseases. As these groups of causes
are the main killers in the two countries (Désesquelles et al. 2010), we suspect that
some of these associations are not specific to cancer-related mortality. And indeed,
the examination of the CDAIs provides a quite different picture. Figure 3a
(respectively Fig. 3b) display the contributing (respectively, underlying) causes of
death for which a neoplasm is mentioned as underlying (respectively contributing)
cause of death more frequently than expected. The next graphs (Fig. 4 for France
and Fig. 5 for Italy) display the CDAIs computed at a more detailed level of
classification12 for all deaths with a malignant neoplasm as UCD.13 The cells of the
table that result from the cross-matching of every site-specific malignant neoplasm
as UCD and every contributing cause of death are coloured according to the value of
the corresponding CDAI. After computing the standard deviation (r) of the
CDAIs,14 we created five classes: [0;100[, [100;100 ? r/4[, [100 ? r/4;100 ? r/
2[, [100 ? r/2;100 ? 3r/4[and [100 ? 3r/4;?[.The first class is white while all
other classes are progressively darker shades of grey.
In Fig. 3a, b, the similarity of the results for the two countries is very striking.
The frequent association we noted earlier between neoplasms and diseases of the
circulatory/respiratory system when commenting the standardized prevalence rates
is no longer visible (CDAIs are under 100). In contrast, this analysis confirms that
the simultaneous reporting of a neoplasm as underlying and contributing cause of
death is a feature of cancer-related mortality. When a neoplasm is mentioned as
contributing cause (Fig. 3b), no other associations have a CDAI over 100. Figures 4
and 5 show that for almost all anatomic sites as UCD, a secondary malignant
neoplasm or a ‘malignant neoplasm of ill-defined/unspecified and independent
primary multiple sites’ is very frequently reported as contributing cause of death.
For some of the emerging associations, the UCD and the contributing cause belong
to the same group (e.g. Intest/Intes, Haem/Haem), or involve proximate anatomic
12
See the list of the groups in the Table 5 in appendix.
13
Only malignant neoplasms with well-defined anatomic sites are presented.
14
The standard deviation was calculated taking into account all the values of the CDAI in the two
countries.
123
A. F. Désesquelles et al.
Ext OIlldef
Other Mecha
OGen Senil
Hyppro Other
OKidn OGen
Renal Hyppro
OMusc OKidn
Arthr Renal
Skin OMusc
ODiges Arthr
Chroliv Skin
Ulcer ODiges
Chroliv
OResp
Ulcer
Lung
OResp
Chrolow
Lung
Asthm Chrolow
Aculow Asthm
Pneu Aculow
Influ Pneu
OCirc Influ
Hypten OCirc
Cereb Hypten
OHeart Cereb
Ischae OHeart
ONerv Ischae
Parki ONerv
Alzh Parki
CONTRIBUTING CAUSE
Alzh
UNDERLYING CAUSE
Epi
OMent Epi
Demen OMent
Drug Demen
Alco Drug
OEndo Alco
OEndo
Thyro
Thyro
Obes
Obes
Malnu Malnu
Diab Diab
Blood Blood
Neoplasm-France
Neoplasm-France
Beni Beni
Neoplasm-Italy
Neoplasm-Italy
ONeo ONeo
UnsNeo UnsNeo
SecNeo SecNeo
Brain Brain
Haem Haem
Blad Blad
Kidn Kidn
Prost Prost
Ova Ova
Uter Uter
Brea Brea
Skin Skin
Mela Mela
Lung Lung
Panc Panc
Liv
Liv
Intes
Intes
Stom
Stom Oeso
Oeso UpAero
UpAero OInfec
OInfec Infint
Infint Sept
Sept Hepa
Hepa AIDS
(b)
(a)
AIDS Tub
Tub
Fig. 3 CDAIs—France and Italy, 2003. a Neoplasms (all anatomic sites) as UCD. b Neoplasms (all
anatomic sites) as contributing cause of death. The area of a circle is the value of the CDAI. Values of the
indicator under 100 and/or corresponding to fewer than 50 cases are not shown. See abbreviations in the
Table 5 in appendix. Data France: Inserm CépiDc mortality database/Italy: Istat mortality database
sites, such as sites of the digestive system (e.g. Intes/Stom, Intes/Intes15) or the
upper aerodigestive tract (UpAero/UpAero), reflecting the fact that cancers of any
tissue are likely to spread to neighbouring tissue. Concomitant use of alcohol and
15
See abbreviations in the Table 5 in appendix.
123
Analysing Multiple Causes of Death
Fig. 4 CDAIs—associations (vertical axis) with neoplasms as UCD (horizontal axis)—France, 2003.
Values of the indicator under 100 and/or corresponding to fewer than 50 cases are not shown. See
abbreviations in the Table 5 in appendix. Data Inserm CépiDc mortality database
123
A. F. Désesquelles et al.
Fig. 5 CDAIs—associations (vertical axis) with neoplasms as UCD (horizontal axis)—Italy, 2003.
Values of the indicator under 100 and/or corresponding to fewer than 50 cases are not shown. See
abbreviations in the Table 5 in appendix. Data Istat mortality database
123
Analysing Multiple Causes of Death
123
A. F. Désesquelles et al.
123
Analysing Multiple Causes of Death
associations do not really challenge medical knowledge, and as they do not emerge
in both countries, we suspect that they reflect differences in French and Italian
certifying practices. The coincidence of thyroid disorders (mainly: hypothyroidism)
and breast cancer (Brea/Thyro), also specific to France, is more challenging, and a
topic of debate in the scientific literature. Hypothyroidism may be a consequence of
radioactive therapies (Daoud et al. 2005) but it is also suggested that thyroid
abnormalities could play a causal role in the development and progression of breast
cancers (Smyth 1997).
5. Among the various consequences/complications of cancers, symptoms may also
be distinguished—though the distinction is probably tenuous. As an example, in
both countries, brain cancer is frequently combined with epilepsy and other
diseases of the nervous system (mainly: benign intracranial hypertension,
compression of brain, cerebral oedema).
This categorization is necessarily a bit oversimplified, and in many cases it is
difficult to say whether an association belongs to one or another category. Frequent
associations involving ‘other diseases of the digestive system’ (Oeso/ODiges, Stom/
ODiges, Intes/ODiges, Liv/ODiges, Panc/ODiges), ‘other diseases of kidney and
ureter’ (Intes/OKidn, Uter/OKidn, Pros/OKidn, Blad/OKidn, Haem/OKidn), ‘other
diseases of the genitourinary system’ (Intes/OGen, Brea/OGen, Uter/OGen, Pros/
OGen, Kidn/OGen, Blad/OGen) or respiratory diseases (Chrolow, Lungdis, Oresp)
as contributing cause of death may well belong to any of these categories.
In the end, very few strong associations remain difficult to interpret. For instance,
it is unclear why, specifically to France, ‘other diseases of the circulatory system’
(mainly: pulmonary embolism, phlebitis and thrombophlebitis) emerge as a
contributing cause for neoplasm of uterus (Uter/OCircu). Similarly, we could not
find any explanation for the observed contribution of Alzheimer’s disease to
neoplasm of bladder (Blad/Alzh), and of ‘other diseases of the nervous system’
(mainly: benign intracranial hypertension, compression of brain, cerebral oedema),
to melanoma (Mela/ONerv) and neoplasm of kidney (Kidn/ONerv). Further research
on these associations is certainly indicated.
4 Conclusions
Our study leads to several firm conclusions. Concerning the quality of the MCOD
data, it is sometimes argued that a higher average number of causes per certificate
signals better diagnosis and more complete recording and, as a consequence, better
quality (White et al. 1989). The fact that the number of reported causes is higher
when death occurs at hospital supports this argument. As the average number of
mentions is higher in Italy, this tends to indicate better quality of the MCOD data in
this country. In any event, the similarity of the results of our study of cancer-related
mortality in France and Italy is very striking. It is tempting to interpret this
similarity as a positive signal for data quality, but caution is required. First, though
we do not consider it very likely, we cannot rule out that similar biases lead to
similar results. Second, we obviously cannot presume that this similarity holds true
123
A. F. Désesquelles et al.
for other causes of death. Note that while it is sometimes difficult to determine
primary and secondary anatomic sites, the reporting of the morbid process involving
a cancer may be less problematic than for other causes of death. People with cancer
usually receive regular medical follow-up, and the circumstances surrounding their
death are likely to be well described.
The fact that most of our results are in line with medical knowledge or with
ongoing debate in the medical literature speaks in favour of good data quality.
Hence, the MCOD approach can help to identify associations which, though not
currently validated by medical knowledge, should be taken seriously and
investigated further. Regarding the causes that frequently contribute to cancer
mortality, we were able to classify our results into five patterns of associations. It
will be interesting to examine how far these patterns apply to diseases other than
neoplasms.
As neoplasms are very often mentioned as the underlying cause of death, the
impact of the MCOD approach in terms of mortality levels is modest. Yet, the case
of cancer of the prostate, for which the impact is greater than for other sites,
suggests that with improved survival and increased incidence of cancers at old ages,
neoplasms may be mentioned more frequently as contributing causes of death. So,
we recommend that the MCOD approach can be used to monitor cancer-related
mortality.
Appendix
123
Analysing Multiple Causes of Death
123
Table 3 Standardized prevalence at death (per 1,000)—Neoplasms as UCD—France, 2003
Infectious Neoplasms Diseases of the Endocrine, Mental and Diseases of Diseases of Diseases of Diseases of Diseases of the Diseases of the Diseases of the Other External
and blood (-forming nutritional and behavioural the nervous the the the digestive skin and musculoskeletal genitourinary diseases causes
parasitic organs), metabolic disorders system circulatory respiratory system subcutaneous system / system
diseases immunol.disorders diseases system system tissue connective tissue
123
Malignant neoplasm of lip, oral cavity &
40 303 81 75 146 140
pharynx
Malignant neoplasm of oesophagus 43 350 75 46 136 160 114 44
Malignant neoplasm of stomach 504 82 167 67 148
Malignant neoplasm of small intestine,
51 617 70 176 65 187 42
colon, rectum & anus
Malignant neoplasm of liver, intrahepatic
bile ducts, gallbladder&other parts of 108 258 89 43 156 49 346
biliary tract
Malignant neoplasm of pancreas 436 80 137 155
Malignant neoplasm of larynx, trachea,
417 57 65 188 160
bronchus & lung
Malignant melanoma of skin 834 51 81 122 69 51
Malignant neoplasm of skin 523 103 209 102
Malignant neoplasm of breast 735 77 45 188 89 62
Malignant neoplasm of cervix uteri & other
50 600 79 179 49 99 98
parts of uterus
Malignant neoplasm of ovary 702 48 131 60 166 42
Malignant neoplasm of prostate 45 633 43 72 40 211 106 53 110
Malignant neoplasm of kidney 696 97 213 89 53 120
Malignant neoplasm of bladder 54 512 76 190 83 72 162
Malignant neoplasm of lymphoid
141 153 155 82 224 158 52 87
/haematopoietic tissue
Malignant neoplasm of eye, brain & other
181 69 192 170 100
parts of central nervous system
Secondary malignant neoplasm
Malignant neoplasm of illdef./unspec./
44 677 72 42 180 97 85 44
independent primary multiple sites
Other Malignant neoplasms 469 71 173 103 54 60
Benign neoplasms and in situ neoplasms or
83 70 81 90 53 108 274 138 114 78 50
uncertain or unknown behaviour
Values of the indicator under 40 per 1,000 and/or corresponding to fewer than 50 cases are not shown
Data Inserm CépiDc mortality database
A. F. Désesquelles et al.
Table 4 Standardized prevalence at death (per 1,000)—Neoplasms as UCD—Italy, 2003
Infectious Neoplasms Diseases of the Endocrine, Mental and Diseases Diseases of Diseases of Diseases of Diseases of the Diseases of the Diseases of the Other External
and blood (-forming nutritional behavioural of the the the the skin and musculoskeletal genitourinary diseases causes
parasitic organs), and disorders nervous circulatory respiratory digestive subcutaneous system / system
diseases immunol.disorders metabolic system system system system tissue connective tissue
diseases
Malignant neoplasm of lip, oral cavity &
677 65 263 151 74
pharynx
Malignant neoplasm of oesophagus 684 57 78 249 182 156
Malignant neoplasm of stomach 700 86 61 245 101 199 61
Malignant neoplasm of small intestine,
725 54 68 265 103 219 87
colon, rectum & anus
Malignant neoplasm of liver,
intrahepatic bile ducts, 103 463 95 248 84 504 79
gallbladder&other parts of biliary tract
Malignant neoplasm of pancreas 682 126 230 78 260 67
Malignant neoplasm of larynx, trachea,
580 64 307 252 44 50
Analysing Multiple Causes of Death
Values of the indicator under 40 per 1,000 and/or corresponding to fewer than 50 cases are not shown
Data Istat mortality database
123
A. F. Désesquelles et al.
123
Analysing Multiple Causes of Death
Table 5 continued
123
A. F. Désesquelles et al.
Table 5 continued
123
Analysing Multiple Causes of Death
References
Aubenque, M. (1975). Un aperçu historique des statistiques sanitaires de la France. Paris: INSEE.
Australian Bureau of Statistics. (2003). Multiple cause of death analysis, 1997–2001. http://www.abs.
gov.au/ausstats/abs@.nsf/mf/3319.0.55.001.
Chamblee, R. F., & Evans, M. C. (1982). New dimensions in cause of death statistics. American Journal
of Public Health, 72(11), 1265–1270.
Coste, J., & Jougla, E. (1994). Mortality from rheumatoid arthritis in France: 1970–1990. International
Journal of Epidemiology, 23(3), 545–552.
Crews, D. E. (1988). Multiple causes of death and the epidemiological transition in American Samoa.
Biodemography and Social Biology, 35(3–4), 198–213.
Crews, D. E. (1990). Multiple causes of death, chronic diseases and aging. Collegium antropologicum,
14(2), 197–204.
Daoud, J., Siala, W., Guermazi, F., Besbes, M., Frikha, M., Ghorbel, M., et al. (2005). Hypothyroidism
following cervical irradiation in the management of carcinoma of the nasopharynx and of the breast,
a prospective study on eighty-four cases. Cancer Radiotherapy, 9, 140–147.
David, H. A. (1974). Parametric Approaches to the Theory of Competing Risks. In F. Proschan &
R. J. Serfling (Eds.), Reliability and biometry (pp. 275–290). Philadelphia: Society for Industrial and
Applied Mathematics.
Désesquelles, A., & Meslé, F. (2004). Intérêt de l’analyse des causes multiples dans l’étude de la
mortalité aux grands âges, l’exemple français. Cahiers que´be´cois de de´mographie, 33(1), 83–116.
Désesquelles, A., Salvatore, M. A., Frova, L., Pace, M., Pappagallo, M., Meslé, M., et al. (2010). Revisiting
the mortality of France and Italy with the multiple-cause-of-death approach. Demographic Research,
23(28), 71–806.
Dorn, H. F., & Moriyama, I. M. (1964). Uses and significance of multiple cause tabulations for mortality
statistics. American Journal of Public Health Nations Health, 54, 400–406.
Eakin, E. G., Youlden, D. R., Baade, P. D., Lawler, S. P., Reeves, M. M., Heyworth, J. S., et al. (2006).
Health status of long-term cancer survivors: Results from an Australian population-based sample.
Cancer Epidemiology Biomarkers Prevention, 15, 1969–1976.
Fink, A. K., German, R. R., Heron, M., Stewart, S. L., Johnson, C. J., Finch, J. L., et al. (2011). Impact of
using multiple causes of death codes to compute site-specific, death certificate-based cancer
mortality statistics in the United States. Cancer Epidemiology, 35(2), 126–131.
Frenzen, P. (2003). Mortality due to gastroenteritis of unknown etiology in the United States. The Journal
of Infectious Diseases, 18, 441–452.
Frova, L., Salvatore, M. A., Pappagallo, & Egidi, V. (2009). The multiple cause of death approach to
analyse mortality patterns. Genus, 65(1), 1–21.
Fuhrman, C., Jougla, E., Nicolau, J., Eilstein, D., & Delmas, M. C. (2006). Deaths from chronic
obstructive pulmonary disease in France, 1979–2002: A multiple cause analysis. Thorax, 61,
930–934.
Fuhrman, C., Jougla, E., Uhry, Z., & Delmas, M. (2009). Deaths with asthma in France, 2000–2005: A
multiple-cause analysis. Journal of Asthma, 46(4), 402–406.
Goldacre, M. J., Duncan, M. E., Cook-Mozaffari, P., & Griffith, M. (2003). Trends in mortality rates
comparing underlying-cause and multiple-cause coding in an English population 1979–1998.
Journal of Public Health Medicine, 25, 249–253.
Goldacre, M. J., Duncan, M. E., Cook-Mozaffari, P., & Griffith, M. (2004a). Trends in mortality for
cancers, comparing multiple- and underlying-cause rates, in an English population 1979–1999.
British Journal of Cancer, 90, 1019–1021.
Goldacre, M. J., Duncan, M., Cook-Mozaffari, P., & Griffith, M. (2004b). Mortality rates for common
respiratory diseases in an English population 1979–1998: Artefact and substantive trends. Journal of
Public Health (Oxford), 26, 8–12.
Goldacre, M. J., Duncan, M. E., Cook-Mozaffari, P., & Neil, H. A. W. (2004c). Trends in mortality rates
for death-certificate-coded diabetes mellitus in an English population 1979–99. Diabetic Medicine,
21, 936–939.
Goldacre, M. J., Duncan, M., Griffith, M., & Cook-Mozaffari, P. (2004d). Alcohol as a certified cause of
death in a ‘middle England’ population 1979–1999: Database study. Journal of Public Health
(Oxford), 26, 343–346.
123
A. F. Désesquelles et al.
Goldacre, M. J., Duncan, M., Griffith, M., & Rothwell, P. M. (2008). Mortality rates for stroke in England
from 1979 to 2004: Trends, diagnostic precision, and artifacts. Stroke, 39, 2197–2203.
Goldacre, M. J., Mant, D., Duncan, M., & Griffith, M. (2005). Mortality from heart failure in an English
population, 1979–2003: Study of death certification. Journal of Epidemiology and Community
Health, 59, 782–784.
Goodman, R. A., Manton, K. G., Nolan, T. F., Bregmen, D. J., & Hinman, A. R. (1982). Mortality data
analysis using a multiple-cause approach. Journal of the American Medical Association, 247(6),
793–796.
Guralnick, L. (1966). Some problems in the use of multiple causes of death. Journal of Chronic Diseases,
19(9), 979–990.
Hedskog, L., Zhang, S., & Ankarcrona, M. (2011). Strategic role for mitochondria in Alzheimer’s disease
and cancer. Antioxidants & Redox Signaling, 15(10), 2150–2563.
Israel, R. A., Rosenberg, H. M., & Curtin, L. R. (1986). Analytical potential for multiple cause-of-death
data. American Journal of Epidemiology, 124, 161–179.
ISTAT. (1958). Indagine statistica sulle concause di morte—1951–1954. Supplemento all’annuario di
statistiche sanitarie del 1955.
ISTAT (2004). Applying ACS to Causes of death Statistics in Italy. Some clues on implementation, bridge
coding and further steps. In Frova L, Marchetti S, Pace M (Eds.), Applying acs to causes of death
statistics in Italy. Rome: ISTAT.
Janssen, T. A. (1940). Importance of tabulating multiple causes of death. American Journal of Public
Health Nations Health, 30(8), 871–879.
Jougla, E., Papoz, L., Balkau, B., Maguin, P., & Hatton, F. (1992). Death certificate coding practices
related to diabetes in European countries—The ‘EURODIAB subarea C’ study. International
Journal of Epidemiology, 21(2), 343–351.
Kaplan, M. H., Armstrong, D., & Rosen, P. (1974). Tuberculosis complicating neoplastic disease.
A review of 201 cases. Cancer, 33, 850–858.
Kohn, R. R. (1982). Cause of death in very old people. Journal of the American Medical Association,
247(20), 2793–2797.
Lee, L., & Thompson, W. A. (1974). Results on failure time and patterns for the series system. In Frank.
Proschan & R. J. Serfling (Eds.), Reliability and biometry (pp. 291–302). Philadelphia: Society for
Industrial and Applied Mathematics.
Mackenbach, J. P., Kunst, A. E., Lautenbach, H., Bijlsma, F., & Oei, Y. B. (1995). Competing causes of
death: an analysis using multiple-cause-of-death data from The Netherlands. American Journal of
Epidemiology, 141(5), 466–475.
Mannino, D. M., Ford, E., Giovino, G. A., & Thun, M. (1998). Lung cancer deaths in the United States
from 1979 to 1992: An analysis using multiple-cause mortality data. International Journal of
Epidemiology, 27(2), 159–166.
Manton, K. G. (1986). Cause specific mortality patterns among the oldest old: Multiple cause of death
trends 1968 to 1980. Journal of Gerontology, 41(2), 282–289.
Manton, K. G., & Myers, G. C. (1987). Recent trends in multiple-caused mortality 1968 to 1992: Age and
cohort components. Population Research and Policy Review, 6, 161–176.
Manton, K. G., & Poss, S. S. (1979). Effects of dependency among causes of death for cause elimination
life table strategies. Demography, 16, 313–327.
Manton, K. G., & Stallard, E. (1982). Temporal trends in U. S. multiple cause of death mortality data
1968 to 1977. Demography, 19(4), 527–547.
Manton, K. G., Tolley, D. H., & Poss, S. S. (1976). Life table techniques for multiple-cause mortality.
Demography, 13, 541–564.
Markush, R. E., & Seigel, D. G. (1968). Prevalence at death. I. A new method for deriving death rates for
specific diseases. American Journal of Public Health Nations Health, 58(3), 544–557.
Matzkin, H., & Soloway, M. S. (1993). Cigarette smoking: A review of possible associations with benign
prostatic hyperplasia and prostate cancer. Prostate, 22, 277–290.
McCoy, L., Redelings, M., Sorvillo, F., & Simon, P. (2005). A multiple cause-of-death analysis of asthma
mortality in the United States, 1990–2001. Journal of Asthma, 42(9), 757–763.
Mele, A., Pulsoni, A., Bianco, E., Musto, P., Szklo, A., Sanpaolo, M. G., et al. (2003). Hepatitis C virus
and B-cell non-Hodgkin lymphomas: An Italian multicenter case–control study. Blood, 102,
996–999.
Moriyama, I. M. (1956). Development of the present concept of cause of death. American Journal of
Public Health, 46, 436–451.
123
Analysing Multiple Causes of Death
123
A. F. Désesquelles et al.
Wong, O. (1977). A competing-risk model based on the life table procedure in epidemiological studies.
International Journal of Epidemiology, 6, 153–159.
Wrigley, J. M., & Nam, C. B. (1987). Underlying versus multiple causes of death: Effects on interpreting
cancer mortality differential by age, sex and race. Population Research and Policy Review, 16(2),
149–160.
Ziadé, N., Jougla, E., & Coste, J. (2010). Population-level impact of osteoporotic fractures on mortality
and trends over time: a nationwide analysis of vital statistics for France, 1968–2004. American
Journal of Epidemiology, 172, 942–951.
123
View publication stats