Public Health 123 (2009) 156–162

Contents lists available at ScienceDirect

Public Health
journal homepage: www.elsevierhealth.com/journals/pubh

Original Research

Life expectancy and age–period–cohort effects: analysis and projections of

mortality in Spain between 1977 and 2016
R. Cleries a, b, *, J.M. Martı́nez c, J. Valls a, d, L. Pareja a, L. Esteban a, R. Gispert e, V. Moreno b, f,
J. Ribes a, J.M. Borràs a, b
a
Catalan Cancer Registry, Generalitat de Catalunya/Catalan Institute of Oncology, IDIBELL, Hospitalet de Llobregat, Spain
b
Department of Clinical Sciences, Universitat de Barcelona, Barcelona, Spain
c
Unitat de Recerca en Salut Laboral, Universitat Pompeu Fabra, Barcelona, Spain
d
Departament de Matematiques, Universitat Autònoma de Barcelona, Spain
e
Registre de Mortalitat de Catalunya, Generalitat de Catalunya, Barcelona, Spain
f
Bioinformatics and Biostatistics Unit, Catalan Institute of Oncology, IDIBELL, Hospitalet de Llobregat, Spain

a r t i c l e i n f o s u m m a r y

Article history: Objectives: This study aimed to: (1) assess Spanish mortality trends between 1977 and 2001 and their
Received 28 February 2008
impact on life expectancy; and (2) assess the differences in life expectancy between men and women for
Received in revised form
the period 2002–2016.
19 September 2008
Accepted 28 October 2008 Study design: Time trends study using age–period–cohort (APC) analysis.
Available online 20 January 2009 Methods: A Bayesian APC model was fitted to describe Spanish mortality rates for the period 1977–2001
and to project Spanish mortality rates for 2002–2016. Life expectancy was predicted through Chiang’s
Keywords: method using projected mortality rates.
Mortality Results: There was a significant cohort effect for Spanish mortality, showing a slight increase in mortality
Life expectancy
among men aged 20–39 years between 1986 and 1997 (birth cohorts 1940–1970). Life expectancy is
Age–period–cohort models
Projections
expected to increase by approximately 0.5% in men and women between 1977 and 2016 (1 year per 5-
year period). Life expectancy for males born between 2012 and 2016 will be 77.15 years, compared with
84.95 years for females born during the same period.
Conclusions: The rising trend in mortality among the 1940–1970 cohorts may be due to the increased risk
of avoidable causes of death related to acquired immunodeficiency syndrome, traffic accidents, and drug
and alcohol abuse during the mid 1980s. The decline in mortality rates in recent years could lead to
a mean increase in life expectancy of 1 year per 5-year period in both genders between 2002 and 2016.
An increase in life expectancy for women and a levelling off for men is expected for age groups older than
79 years.
Ó 2008 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.

Introduction
Life expectancy (LE) trends over time provide major public
health information that reflects social and political changes in
populations. Different changes have been observed in European
Union countries since the beginning of the 21st Century. Estimates
of LE during recent decades showed that Sweden, Spain and France
have the highest LE in Europe.1 In particular, Spain compared
favourably with other low-mortality countries between 1970 and
2000.2 In 2004, LE at birth reached 83.8 years for Spanish women,
* Corresponding author. Catalan Cancer Registry, Generalitat de Catalunya/
Catalan Institute of Oncology, IDIBELL, Av Gran Via Km 2,7 Hospitalet de Llobregat
08907, Spain. Tel.: þ34 932 607 812.
E-mail address: r.cleries@iconcologia.net (R. Cleries).
almost 7 years higher than that for Spanish men,1 and close to that
in Japan (85.5 years); the country which ranks the highest
worldwide.3
In Spain, a previous study on the impact of avoidable causes of
death, such as acquired immunodeficiency syndrome (AIDS), traffic
accidents and alcohol and drug abuse, between 1987 and 2001
showed its effect on LE over the last two decades.4 Projections of
future LE based on recent trends are relevant for major health
policy decisions related to retirement funding and planning of
health services.
This study aimed to: (1) assess recent Spanish mortality trends
and LE by age group for each gender, and (2) project LE by gender
taking into account the effects of age, period and birth cohort on the
predicted mortality trends using a Bayesian approach, which allows
the effect of the precision of a country’s future population to be
taken into account.5

0033-3506/$ – see front matter Ó 2008 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.puhe.2008.10.026
 R. Cleries et al. / Public Health 123 (2009) 156–162 157

Methods
A time trend study was carried out using a Bayesian age–
period–cohort (APC) analysis to estimate the effects of age, period
of death and birth cohort on mortality.5–13 Age effects represent
different risks associated with different age groups. Period effects
reflect changes in treatment or exposure to a certain risk factor that
would increase mortality in all age groups at the same time.12,13
Cohort effects are associated with long-term habits or long-term
exposures, whereby different generations have been exposed to
different risks. A change in exposure to risk factors or a change in
population habits in time is detected through these effects when
they show curvature (non-linear increase or decrease in mortality
risk)7,12,13 at a certain point in time. As such, a key point in APC
analysis is to detect these curvatures.7,10,12,13
Population distribution
The National Institute of Statistics of the Spanish Government
provided mortality data and population age distribution data for
the period 1975–2004, as well as projected future population data
for the period 2002–2016. The first 2 years (1975–1976) and the last
3 years (2002–2004) were used for internal validation of projec-
tions. Data were grouped into five 5-year periods (1977–1981,
1982–1986, 1987–1991, 1992–1996 and 1997–2001) and 18 5-year
age groups (0–4 to 85–89 years, data for persons older than 89
years were excluded from the analysis due to non-precision of the
population at risk estimates). Table 1 shows the population distri-
bution used in this study and the percentage increase for each age
group by consecutive 5-year periods. These age groups and
calendar periods involved 22 overlapping 10-year cohorts due to
the relationship cohort ¼ period-age.6,7 The cohort groups were
defined by their mid-years (starting with 1889 and finishing with
1999).
Statistical methods
A Bayesian APC analysis was undertaken to describe mortality
time trends and projections. Although different methods have been
proposed to obtain LE forecasts based on age-specific mortality
rates, these do not take into account the birth cohort effect on
mortality rates, when this exists. In this case, LE projections can be
improved by including the birth cohort effect. Under this scenario,
the Bayesian framework is a flexible alternative to obtain LE fore-
casts including age, period and birth cohort.5
The analysis started by estimating recent mortality trends
between 1977 and 2001 through modelling the ratio of number of
deceased cases (numerator) and population at risk (denominator)
using a Poisson distribution.12,13 Once the trends were estimated,
predictions of mortality rates were obtained based on these trends
and using the future population at risk12,13 between 2002 and 2016.
Finally, future mortality rates by age group were used to predict LE
based on Chiang’s method.14,15
Age, cohort and period effects were assessed using generalized
linear models, assuming that the number of deaths follows a Pois-
son distribution.6,7 Under the Bayesian framework, trends corre-
sponding to age, period and birth cohort were smoothed using
second-degree autoregressive smoothing (non-parametric
smoothing with autoregressive error component). This resulted in
linear extrapolations for age, period and cohort.8–12 In APC analysis,
the relationship cohort ¼ period-age is associated with the problem
of non-identifiability in the parameter estimates.6,7,13 Estimation of
parameters extracted from the full APC model was undertaken
assuming zero slopes for cohort effect, using 1977–1981 as the
reference period and 1949 as the reference cohort. Graphical
representation of the exponential of each of these parameters and
their 95% credibility intervals (CRI, the Bayesian confidence inter-
vals) has been reported. The number of predicted deaths for 2002–
2016 was obtained using the APC model. Once the expected
number of deaths was calculated, Chiang’s method for constructing
life tables was used to calculate LE in each projected 5-year
period.14,15 Standard errors for predicted LE were obtained by
means of the posterior distribution of the mortality projections
obtained through the Bayesian APC model. Estimation of LE based
on mortality predicted from an APC model requires the analyst to
make strong parametric assumptions about the degree of data
smoothing needed in order to obtain posterior standard errors of
data functions such as LE. The Bayesian method extracts the
necessary information from the data to describe the observed
trend, projecting it into the future in the smoothest possible
way,5,12 and achieving sensible predictions in situations when other
methods would fail.12,13 This Bayesian APC model is particularly
interesting because it allows the uncertainty associated with
functions of the parameters to be readily explored.12

Table 1
Population distribution in Spain by gender and for age groups 0–89 years during 1997–2016, and comparison of evolution (%) between consecutive 5-year periods from 1997–
2001 to 2012–16

Age Men Women

group
Period Period Period Period Evolution (%) Evolution (%) Period Period Period Period Evolution (%) Evolution (%)
(years)
1997– 2002– 2007– 2012– 2002–2006 vs 2012–2016 vs 1997– 2002– 2007– 2012– 2002–2006 vs 2012–2016 Vs
2001 2006 2011 2016 1997–2001 2007–2011 2001 2006 2011 2016 1997–2001 2007–2011
0–4 943,345 1,098,496 1,162,856 1,231,607 16.45 5.91 890,346 1,037,432 1,096,540 1,162,294 16.52 6.00
5–9 1,011,847 1,007,487 1,057,678 1,135,608 0.43 7.37 956,029 956,164 1,002,720 1,074,511 0.01 7.16
10–14 1,127,778 1,076,978 1,065,078 1,073,696 4.50 0.81 1,069,422 1,019,189 1,008,574 1,017,947 4.70 0.93
15–19 1,397,046 1,202,193 1,176,093 1,172,201 13.95 0.33 1,329,175 1,140,663 1,112,894 1,107,878 14.18 0.45
20–24 1,669,800 1,532,922 1,459,361 1,409,414 8.20 3.42 1,598,718 1,465,320 1,391,446 1,348,488 8.34 3.09
25–29 1,685,106 1,883,533 1,899,661 1,857,511 11.78 2.22 1,627,137 1,781,430 1,782,697 1,749,795 9.48 1.85
30–34 1,643,018 1,874,666 2,004,187 2,109,119 14.10 5.24 1,611,053 1,777,266 1,870,654 1,955,786 10.32 4.55
35–39 1,548,887 1,782,869 1,881,315 1,995,145 15.11 6.05 1,542,538 1,729,348 1,792,785 1,875,043 12.11 4.59
40–44 1,387,934 1,645,601 1,750,917 1,850,575 18.56 5.69 1,395,217 1,632,005 1,717,860 1,796,407 16.97 4.57
45–49 1,237,487 1,446,421 1,558,405 1,664,731 16.88 6.82 1,247,161 1,457,066 1,562,805 1,661,313 16.83 6.30
50–54 1,169,236 1,260,594 1,325,153 1,423,174 7.81 7.40 1,198,366 1,287,278 1,353,579 1,453,825 7.42 7.41
55–59 990,073 1,168,186 1,210,072 1,243,241 17.99 2.74 1,033,626 1,223,472 1,266,519 1,299,750 18.37 2.62
60–64 947,031 973,073 1,061,310 1,130,025 2.75 6.47 1,038,836 1,045,832 1,140,710 1,217,992 0.67 6.77
65–69 963,090 900,825 869,044 906,512 6.47 4.31 1,103,561 1,029,260 983,803 1,014,867 6.73 3.16
70–74 783,301 864,556 869,815 839,435 10.37 3.49 977,083 1,058,215 1,059,927 1,020,231 8.30 3.75
75–79 544,481 642,452 680,368 713,216 17.99 4.83 779,257 887,246 927,438 960,726 13.86 3.59
80–84 293,007 389,419 428,671 457,679 32.90 6.77 521,452 635,903 684,621 723,695 21.95 5.71
85–89 204,926 237,399 265,248 299,465 15.85 12.90 460,854 540,992 587,440 640,483 17.39 9.03

Population for 2007–2016 is based on projection method (see Spanish National Institute of Statistics40).
158 R. Cleries et al. / Public Health 123 (2009) 156–162

WinBUGS 1.416 software was used for the analyses, which allowed and 187,300 for men. Crude mortality rates decreased between
Bayesian inference to be performed using Gibbs sampling. R2Win- 1977 and 2001 (Estimated annual percent change (EAPC) men
BUGS17 library was used as an interface to run WinBUGS within the -1.20%, 95% CRI -1.35 to -1.05%; EAPC women -1.95%, 95% CRI -2.05
statistical software R.18 For each model, three chains were run with to -1.85%). However, these rates showed a levelling off trend for
100,000 iterations, discarding the first 10,000 burning samples.19,20 men and women aged 85–89 years (see Fig. 1), and an increase in
After convergence of all parameters, the posterior median value for mortality during the period 1987–1996 for those cohorts of men
the expected number of total deaths was estimated. and women aged between 20 and 39 years (birth cohorts 1940–
1970), although this was less pronounced in women.
Results The APC effects are depicted graphically in Fig. 2. The age effect
showed a linear increase in mortality rate, whereas the cohort
The annual average numbers of deaths between 1977 and 1981 effect showed a curvature (a change point) after the 1949 birth
were 151,800 for men and 136,851 for women, whereas in cohort and for each gender (more pronounced in men). The latter
1997–2001, the average annual numbers were 166,392 for women effect is related to the increase in mortality in those cohorts born

a Men

85-89 85-90
10000 80-84 10000 80-85
75-79 75-80

Mortality rate per 100,000 men-years

70-75
Crude rate per 100,000 men-years

70-74 65-70
65-69

60-64 60-65
1000 55-59 1000 55-60
50-54 50-55
45-49 45-50
40-44 40-45
30-35
35-39 30-34 35-40 0-5
100 25-29 0-4 100 25-30 20-25
15-19 20-24 15-20

5-9 5-10
10-15
10-14
10 10

1 1

1880 1900 1920 1940 1960 1980 2000 1975 1980 1985 1990 1995 2000 2005
Central year of birth Period of death

b Women

85-89 85-90
10000 10000
80-84 80-85
Mortality rate per 100,000 women-years

75-79 75-80
Crude rate per 100,000 women-years

65-69 70-74 70-75

65-70
1000 1000
55-59 60-64 55-60 60-65

45-49 50-54 45-50 50-55

0-5
40-45
100 40-44 0-4 100
35-39 35-40
30-35
25-29 30-34 25-30
15-19 15-20 20-25
5-9 20-24 5-10
10-15
10-14
10 10

1 1

1880 1900 1920 1940 1960 1980 2000 1975 1980 1985 1990 1995 2000 2005
Central year of birth Period of death

Figure 1. Age-specific 5-year mortality rates per 100,000 person-years in Spain by birth cohort and period of death according to age groups 0–89 years and during the period 1977–
2001. (a) Men, (b) women.
 R. Cleries et al. / Public Health 123 (2009) 156–162 159

between 1940 and 1970, with the maximum risk for males in the
1959 cohort. These cohorts did not show a continuous decreasing
risk of death with time. Regarding the period effect, there was
a slight curvature in mortality rate for women, which was almost
non-existent for men. The lesser curvature associated with the
period effect could be due to a major role of cohort effects in terms
of interpreting mortality rates in Spain.
Once mortality projections have been carried through the
Bayesian APC model, Table 2 shows LE for 1997–2001 and projected
LE for 2002–2016. Taking into account the effect of age, LE will
increase by approximately 0.5% in men and women between 1997
and 2016 (close to 1 year per 5-year period), although women show
a slightly higher increase than men. For the period 2002–2006,
these projections present large variability, ranging from 1.1 (age
group >85 years) to 4.9 (age groups 40–44, 50–54 and 55–59 years).
In Table 2, the columns entitled ‘2002–2006 minus 1997–2001’,
‘2012–2016 minus 2007–2011’ and ‘2012–2016 minus 1997–2001’
compare differences in LE between these periods. The greatest
differences among men are found in those aged 50–69 years,
whereas among women, the greatest differences are found in those
over 70 years of age. Projected LE for men born between 2012 and
2016 is 77.15 years, compared with 84.95 years for women born in
the same period. LE is expected to increase by 2.55 years in men
(from 74.61 to 77.15 years) and by 2.85 years in women (from 82.10
to 84.95 years) between 1997–2001 and 2012–2016. These increases
range from 3.24 to 5.25 years for men and from 2.41 to 3.63 for
women in age groups 5–64 years, and the gains in LE can be more
than 3 years from 1997 to 2016 for age groups 65–79 years. Based on
these projections, gender differences are found for the period 1997–
2016 among age groups over 79 years, where LE is expected to
remain stable for men (column 2012–16 minus 1997–2001: 0.31
and 0.15 years) and increase for women (column 2012–16 minus
1997–2001: 3.74 and 0.09 years). However, these differences cannot
be tested statistically because they were obtained through projec-
tions (non-observed data) and a function of an APC model.12
Discussion
This study shows that mortality declined by 1.2% for men and
1.95% for women between 1977 and 2001, which leads to a mean

a Men
10000 3
95% Credibility Interval
Point Estimate
Mortality rate per 100,000 men-years

Rate Ratio
1000 1

50 0.3

5 0.01
10
20

30
40

50
60
70

80

1890
90
00
10

20
30

40
50

60
70
80

90
00
19
19

19
19

19
19

19
19
19

19
20

Age Birth Cohort Period

b Women

10000 3
95% Credibility Interval
Mortality rate per 100,000 women-years

Point Estimate
Rate Ratio

1000 1

50 0.3

5 0.01
10
20

30
40

50
60
70

80

1890
90
00
10

20
30

40
50

60
70
80

90
00
19
19

19
19

19
19

19
19
19

19
20

Age Birth Cohort Period

Figure 2. Age, period and cohort effects in Spanish men (a) and women (b) during the period 1977–2001. Thick lines, point estimates; narrow lines, 95% credibility intervals.
160 R. Cleries et al. / Public Health 123 (2009) 156–162

Table 2
Life expectancy for those periods where mortality is projected, by age group and for men and women. Life expectancy for 1997–2001 was estimated with observed mortality
rates, and that for 2002–2016 was based on mortality projections extracted from the age–period–cohort model

Age Men Women

group
Period Perioda Perioda Period a
2002–2006 2012–2016 2012–2016 Period Perioda Perioda Perioda 2002–2006 2012–2016 2012–2016
(years)
1997– 2002– 2007– 2012– minusb minusb minusb 1997– 2002– 2007– 2012– minusb minusb minusb
2001 2006 2011 2016 1997–2001 2007–2011 1997–2001 2001 2006 2011 2016 1997–2001 2007–2011 1997–2001
0–4 74.61 75.40 76.25 77.15 0.80 0.90 2.55 82.1 83.31 84.25 84.95 1.20 0.70 2.85
5–9 67.34 69.40 71.15 72.55 2.10 1.40 5.25 75.4 76.71 77.51 77.8 1.30 0.30 2.41
10–14 63.12 64.10 65.21 66.3 1.10 1.10 3.3 70.7 71.82 72.93 74.04 1.12 1.11 3.34
15–19 58.71 59.85 60.98 62.10 1.15 1.10 3.38 66 67.14 68.26 69.36 1.14 1.10 3.36
20–24 54.41 55.52 56.61 57.65 1.12 1.05 3.26 61.4 62.61 63.76 64.85 1.21 1.09 3.45
25–29 50.11 51.21 52.29 53.34 1.11 1.05 3.24 56.8 58.05 59.25 60.41 1.25 1.15 3.61
30–34 45.80 46.98 48.07 49.07 1.18 1.00 3.27 52.2 53.51 54.72 55.83 1.31 1.11 3.63
35–39 41.50 42.91 44.12 45.13 1.41 1.01 3.63 47.6 48.82 49.99 51.11 1.22 1.12 3.51
40–44 37.30 38.65 39.88 40.98 1.35 1.10 3.68 43.1 44.35 45.52 46.62 1.25 1.10 3.52
45–49 33.20 34.51 35.72 36.81 1.31 1.10 3.62 38.6 39.87 41.05 42.15 1.27 1.10 3.55
50–54 29.10 30.81 32.42 33.91 1.71 1.50 4.82 34.2 35.41 36.54 37.62 1.20 1.08 3.42
55–59 25.00 26.81 28.42 29.81 1.81 1.40 4.82 29.8 31.25 32.14 33.22 1.20 1.08 3.42
60–64 21.10 22.70 24.15 25.45 1.60 1.30 4.35 25.5 26.72 27.88 28.99 1.22 1.11 3.49
65–69 17.20 18.40 19.55 20.65 1.20 1.10 3.45 21.3 22.53 23.70 24.82 1.23 1.12 3.52
70–74 13.50 14.30 15.05 15.75 0.80 0.70 2.25 17.3 18.61 19.85 21.05 1.30 1.20 3.75
75–79 10.00 10.30 10.55 10.75 0.30 0.20 0.75 13.6 14.95 16.23 17.44 1.35 1.21 3.84
80–84 3.30 3.40 3.50 3.61 0.10 0.10 0.31 2.5 3.83 5.075 6.235 1.33 1.16 3.73
85–89 2.20 2.25 2.30 2.35 0.05 0.05 0.15 0.40 0.45 0.47 0.49 0.05 0.02 0.09
a
Projected median value of the life expectancy for periods 2002–2016.
b
Differential of median life expectancy between periods. Note that projected life expectancy depends on projected mortality and its posterior standard deviation. These
values range between 1.1 and 3.7 for period 2002–2006, 2.0 and 4.1 for 2007–2011, and 2.5 and 4.8 for 2012–2016 for men and women (data not shown). As such, these
differentials could present certain variability.

increase in LE of 1 year per 5-year period for both genders. Birth
cohort effects explain mortality trends better than period effects.
Cohort effect on Spanish mortality
A notable decreasing trend in mortality was observed in Spain
between 1977 and 2001 with the exception of the age group 85–89
years, for which mortality did not decrease at the same pace. The
mortality rate increased between 1987 and 1996 for those cohorts
of men and women aged between 20 and 39 years (birth cohorts
1940–1970). This phenomenon was a retrogression of health at
young ages, which means a loss rather than a gain in young ages,
and was particularly seen in men. Explanations for this trend could
include the increased risk of avoidable mortality during the mid
1980s.2,21–24 This includes deaths associated with AIDS, drug and
alcohol consumption (including cirrhosis and liver cancer25) and
accidental mortality rates, as has been described previously.2,21,26,27
Deaths from these causes have had a major impact in reducing the
active/working and reproductive age populations, reflected in the
negative LE gains in the mid 1980s in these age groups.2 Inspection
of the cohort effect showed the increased risk in these cohorts, with
a significant curvature in those cohorts of men born between 1950
and 1960. These cohorts were aged 20–39 years during the 1980s,
and experienced political and structural crises, such as unem-
ployment, at the end of the 1970s and during the mid 1980s.2
During the 1980s, fatal traffic injuries increased dramatically in
age groups 18–35 years. In 1994, a decrease was observed after the
introduction of laws to restrict speed limits.28,29 On the other hand,
AIDS was a major cause of death among young Spanish citizens
during the study period,2 and its standardized death rate increased
until 1995 (23 cases per 100,000 men-years and five cases per
100,000 women-years).30 Although this started to decline after
1995, mainly due to effective treatment,25,31 by 2001, Spain had the
highest death rate from AIDS among European countries.32 AIDS
and drugs that are mainly administered intravenously share similar
pathways. The highest mortality rates from these causes of death
were observed in age groups 30–40 years between 1987 and
1996.30 An APC analysis of liver disease in Catalonia between 1980
and 1997 showed that cirrhosis mortality (which is associated with
hepatitis B and C infections, and alcohol consumption) decreased in
all age groups with the exception of men aged 25–35 years.25 The
fact that the increase in deaths due to traffic accidents and AIDS
appeared at the same time as the cohort effect in mortality among
young Spaniards reveals that the increased risk of death in men
aged 20–39 years during 1987–1996 could be partially related to
these causes. This produced an important loss in expected years of
life in the affected age groups, and had a remarkable impact on the
subsequent age groups that limited their expected life gains.26
On the other hand, the APC analysis in women showed the non-
decreasing trend in mortality for these young cohorts. Traffic
accidents, AIDS, and drug- and alcohol-related mortality rates are
lower for women than men in these cohorts.25 These gender
differences translated into a LE that was 6–7 years higher for
women compared with men during the 1990s, and a leveling off of
cancer and cardiovascular mortality rates among middle-aged
Spanish populations.26 Evidence suggests that at least half of the
contributions to the LE time differential between 1977 and 2001
come from age groups over 50 years, and in women, the contri-
butions are primarily concentrated in age groups above 70 years.2,5
These differences may continue for the next 15 years at the same
pace as detected in the mortality projections (Table 2), but these are
subject to variability due to modifications in the structure of the
future Spanish population.
Projected life expectancy
A clear increasing trend for LE in both genders was seen when
comparing 1997–2001 with 2012–2016. In men, the projections
show an increase in LE of more than 2.5 years for age group 0–4
years, a mean increase of 3.8 years for age groups 5–64 years,
a mean increase of 2.1 years for age groups 65–79 years, and
a levelling off for those aged 80 years or more. These differences are
higher in women than in men (0–4 years, 2.9 years; 5–64 years, 3.6
years; 65–79 years, 3.71 years; 80 years, 1.8 years). These results
are in concordance with those of a previous study of trends in LE in
Spain, which concluded that the impact on LE projections when
restricted to non-external causes was larger for men than for
women.22 In the present study we have observed that gains in LE
 R. Cleries et al. / Public Health 123 (2009) 156–162
still benefit older women, continuing the recover of their biological
advantage in front of death when comparing to LE of older
men.33,34 Future increments in LE for the entire study period will
depend on the process against cancers, the continuing decline of
diseases of the circulatory system35 and the advances in preventive
care.36,37 In women, it seems that only the incorporation of lifestyle
characteristics of men could emerge in the future into a regressive
profile.
This study has certain limitations, including how future pop-
ulation numbers can affect LE. These models project the future
number of deaths depending on the trend of previous mortality
rates in the future Spanish population.12–15 Accuracy regarding the
precision of future population age groups can be affected by vari-
ations in migration rates and birth rates.5,12 Predictions for periods
longer than 10 years should be interpreted as predictions with
some uncertainty due to possible variations in the future pop-
ulation.5,12 Birth rate is expected to increase due to migrations and
a high LE,38 and this could lead to gains in LE as has been observed
in previous studies.2 However, changes in migration rates may
affect mortality projections during the time frame considered in
this study, as migrants to Spain are generally young or middle-
aged.39 To depict this issue, Table 1 shows future population
distributions in Spain until 2016. These estimates are based on data
from the National Institute of Statistics of the Spanish Government
using specific methods for modelling birth rates, mortality and
migration.40 Based on this analysis, immigration rates in Spain will
increase up to 2010 and level off afterwards. This phenomenon
could affect cohorts of individuals aged 20–49 years between 1997
and 2016. These cohorts are affected by immigration37–40 and they
show a higher increase when comparing 2002–2006 with 1997–
2001 (increases between 11.48% and 18.56% in men) with respect to
2007–2011 and 2012–2016 (maximum increase is 6.82% in men
aged 45–49 years, see Table 1). This effect will be similar in women.
In addition, the percentage of males in the population aged 70–89
years between 2002 and 2006 increased between 10% and 32%
when compared with the increase in 2012–2016 with respect to
that of 2007–2011. The expected levelling off of immigration after
2010 and improvements in health care could explain stabilization
of the increase in population around 2015.39,40 In addition, these
immigration rates show geographic variability in Spain, being non-
homogeneous among Spanish regions.40
Another limitation to this study is that geographic variability
among regions and how this could affect the analysis were not
taken into consideration. Since 1980, Spain has been an example in
Europe of the ceding of central competences of the Spanish
Government, such as health care, to 17 autonomous communities.41
One of the aims of these processes is to converge health equalities
among regions. Evidence suggests that some regions have clearly
improved their situation between 1980 and 2001. Some regions
with poor initial indicators have overtaken other regions that had
a better initial situation, making it difficult to confirm convergence
at the beginning of 2000,42 and leading to health inequity between
regions which affects mortality and LE.42 For that reason, a more
detailed analysis should include a regional effect in the prediction
model to take regional heterogeneity into account. However, age,
period, cohort and regional effects cannot be analysed with the
same aim as in the present study due to identifiability problems
between parameters.43 It is only possible to analyse the age–
period–region or age–cohort–region data, although some of these
models are still under development.43,44 A future study could
investigate the framework of spatiotemporal studies of mortality.
The decrease in mortality rates since 1997 and the expected
increases in LE until 2016 may be a result of improvements in
treatments for AIDS,5,25,27,30,31 reduced consumption of lethal
drugs,5,25 and reduced traffic accidents.5,23,28 Mortality projections
show that the difference in LE between men and women is
161
expected to stabilize. Based on mortality projections, projected LE
for men born between 2012 and 2016 will be 77.1 years, compared
with 85.4 years for women born during the same period. It is
predicted that LE will increase for the period 2002–2016 with
a mean increase of 1 year per 5-year period for both genders. The
increasing LE anticipated in the coming years should inform health
planning needs45 and the Spanish public health system in terms of
retirement funding and health policy.
Acknowledgements
The authors wish to thank Meritxell Nomen for secretarial
assistance; her efforts are greatly appreciated.
Ethical approval
None sought.
Funding
Partial support was received from the Instituto de Salud Carlos
III of the Spanish Government (C03/09, C03/10) and from the
Agencia de Gestio d’Ajuts Universitaris i de Recerca of the Catalan
Government (2003XT 00023).
Competing interests
None declared.
References
1. Eurostat. Europe: life expectancy and mortality 2004. European Comission.
Luxembourg. Available from: http://epp.eurostat.ec.europa.eu/portal/page?_
pageid¼2053,62710155&_dad¼portal&_schema¼PORTAL; 2007 [last accessed
03.04.07].
2. Gómez-Redondo R, Boe C. Decomposition analysis of Spanish life expectancy at
birth: evolution and changes in the components by sex and age. Demograph Res
2005;13:521–46.
3. Human Mortality Database. University of California, Berkeley (USA), and Max
Planck Institute for Demographic Research (Germany). Available from: http://
www.mortality.org or http://www.humanmortality.de [last accessed 01.02.07].
4. Gispert R, Serra I, Bares MA, Puig X, Puigdefabregas A, Freitas A. The impact of
avoidable mortality on life expectancy at birth in Spain: changes between three
periods, from 1987 to 2001. J Epidemiol Community Health 2008;62:783–9.
5. Stoto M. The accuracy of population projections. J Am Stat Assoc 1983;78:13–20.
6. Clayton D, Schifflers E. Models for temporal variation in cancer rates. I: age-
period and age-cohort models. Stat Med 1987;6:449–67.
7. Clayton D, Schifflers E. Models for temporal variation in cancer rates. II: age-
period-cohort models. Stat Med 1987;6:469–81.
8. Breslow NE, Clayton DG. Approximate inference in generalized linear mixed
models. J Am Stat Assoc 1993;88:9–25.
9. Berzuini C, Clayton D. Bayesian analysis of survival on multiple time scales. Stat
Med 1994;13:823–38.
10. Bray I, Brennan P, Boffetta P. Recent trends and future projections of lymphoid
neoplasms: a Bayesian age-period-cohort analysis. Cancer Causes Control
2001;12:813–20.
11. Bray I. Application of Markov chain Monte Carlo methods to projecting cancer
incidence and mortality. Appl Statist 2002;51:151–64.
12. Bashir SA, Estève J. Projecting cancer incidence and mortality using Bayesian
age-period-cohort models. J Epidemiol Bioestat 2001;6:287–96.
13. Holford TR. The estimation of age, period and cohort effects for vital rates.
Biometrics 1983;39:311–24.
14. Chiang CL. The life table and its construction. Introduction to stochastic processes
in biostatistics. New York: John Wiley & Sons; 1968. p. 189–214.
15. Chiang CL. Competing risks in mortality analysis. Annu Rev Publ Health
1991;12:281–307.
16. Spiegelhalter DJ, Thomas A, Best N, Lunn D. WinBUGS user manual, version 1.4.
Cambridge: MRC Biostatistics Unit; 2002.
17. Sturtz S, Ligges U, Gelman A. R2WinBUGS: a package for running WinBUGS
from R. J Stat Soft 2005;12:1–16.
18. R Development Core Team. R: a language and environment for statistical
computing. Vienna: R Foundation for Statistical Computing. Available from:
http://www.R-project.org; 2006 [last accessed 15.12.07].
19. Gilks WR, Richardson S, Spiegelhalter DJ. Markov chain Monte Carlo in practice.
London: Chapman & Hall; 1996.
162 R. Cleries et al. / Public Health 123 (2009) 156–162
20. Clèries R, Ribes J, Esteban L, Martinez JM, Borras JM. Time trends of breast
cancer mortality in Spain during the period 1977–2001 and Bayesian approach
for projections during 2002–2016. Ann Oncol 2006;17:1783–91.
21. Valkonen T, Vallin J, Meslé F, Valkonen T. Tendances en matière de mortalité
differentialle. In: Council of Europe EdClS, editor. Tendances en matière de
mortalité et mortalité differentialle. Études demographiques n 36; 2001.
22. Guillen M, Anguera A. Forecasting Spanish natural life expectancy. Risk Anal
2005;25:1161–70.
23. Saiz-Sanchez C, Bautista-Rentero D, Corella-Piquer D, Cortina-Birlanga S,
Gonzalez-Arraez JI. Age-period-cohort analysis of mortality caused by traffic
accidents in Spain. Salud Publica Mex 1999;41:170–6.
24. Aran BM, Gispert R, Puig X, Puigdefabregas A, Tresserras R. Geographical
distribution and time trends in avoidable mortality in Catalonia, Spain (1986–
2001). Gac Sanit 2005;19:307–15.
25. Ribes J, Cleries R, Borras J, Galceran J, Bosch FX. Time trends in incidence and
mortality for chronic liver disease and liver cancer in the interval 1980–1997 in
Catalonia, Spain. Eur J Gastroenterol Hepatol 2004;16:865–72.
26. Serra I, Gispert R, Puig X, Torné MM, Puigdefabregas A. Impacte de l’edat i les
causes de mort en els canvis de l’esperança de vida. Catalunya 1987–2002.
Available from: http://teleregions.gencat.es/salut/depsalut/pdf/imcauses2006.
pdf; 2006 [last accessed 15.12.07].
27. Puig X, López-Abente G, Gispert R, Freitas A, Puigdefàbregas A. Tendències de la
mortalitat a Catalunya, 1978–2002. Models edat/perı́ode/cohort. Available from:
Barcelona: Departament de Salut Generalitat de Catalunya, http://www.gencat.
net/salut/depsan/units/sanitat/pdf/t1978-2002.pdf; 2005 [last accessed 15.12.07].
28. Gine JM. Mortality by traffic accidents in Catalonia and in other autonomous
communities. Gac Sanit 1992;6:164–9.
29. Puig X, Gispert R, Puigdefabregas A. Mortalitat per accidents de trànsit. Cata-
lunya, 1983–1998. Butlletı́ Epidemiológic de Catalunya 2002;6:73–6.
30. Mortalidad en España y Comunidades Autònomas. Instituto de Salud Carlos III .
Centro Nacional de Epidemiologı́a. Madrid 2007. Available from: http://www.isciii.
es/htdocs/centros/epidemiologia/anexos/ww9201_cau.htm [accessed 01.02.07].
31. Garcia dO Cayla JA, Brugal MT, Galdos H, Jansa JM, Clos R. The evolution of AIDS
mortality and survival in Barcelona (1981–1997). Med Clin (Barc) 1999;113:
169–70.
32. World Health Organization. Highlights on health in Spain. WHO Regional Office
for Europe. Copenhagen. Available from: http://www.euro.who.int/eprise/
main/WHO/Progs/CHHSPA/burden/20050131_4; 2005 [last accessed 01.02.07].
33. Waldron I. Sex differences in illness incidence, prognosis and mortality: issues
and evidence. Soc Sci Med 1983;17:1107–23.
34. Waldron I. Sex differences in human mortality: the role of genetic factors. Soc
Sci Med 1983;17:321–33.
35. Lawes CM, Bennett DA, Feigin VL, Rodgers A. Blood pressure and stroke: an
overview of published reviews. Stroke 2004;35:1024.
36. Brotons C, Iglesias M, Martin-Zurro A, Martin-Rabadan M, Gene J. Evaluation of
preventive and health promotion activities in 166 primary care practices in
Spain. The Coordinating Group for Prevention and Health Promotion in Primary
Care in Spain. Fam Pract 1996;13:144–51.
37. Nolte E, McKee M. Measuring the health of nations: analysis of mortality
amenable to health care. BMJ 2003;327:1129.
38. Arroyo Pérez A. Tendencias demográficas durante el siglo XX en España. Madrid:
Universidad de Sevilla/Instituto nacional de Estadı́stica; 2003.
39. Jansa JM. Immigration and aging: new challenges in Public Health. Gac Sanit
2006;20(Suppl. 1):10–4.
40. Instituto Nacional de Estadı́stica. Proyecciones a corto plazo. Metodologı́a.
Available from: http://www.ine.es/metodologia/t20/t20269_m.pdf; 2007 [last
accessed 10.08.08].
41. Montero-Granados R, Jimenez JD, Martin J. Decentralisation and convergence in
health among the provinces of Spain (1980–2001). Soc Sci Med 2007;64:
1253–64.
42. Regidor E, Calle ME, Navarro P, Dominguez V. Trends in the association
between average income, poverty and income inequality and life expectancy in
Spain. Soc Sci Med 2003;56:961–71.
43. Lagazio C, Biggeri A, Dreassi E. Age-period-cohort models and disease mapping.
Environmetrics 2003;14:475–90.
44. Schmid V, Held L. Bayesian age-period-cohort modeling and prediction –
BAMP. J Stat Soft 2007;21:1–18.
45. Montero GR, Jimenez Aguilera JD, Martin Martin JJ. Estimation of an index of
regional health needs in Spain using count regression models with filter. Health
Policy 2007;81:4–16.