  |
|
|
||||||||||||
|
||||||||||||||
| Journal of the New Zealand Medical Association,
21-April-2006, Vol 119 No 1232
|
||||||||||||||
|
How low can it go? Projecting ischaemic heart disease
mortality in New Zealand to 2015
Martin Tobias, Kerry Sexton, Stewart Mann, Norman
Sharpe
Ischaemic heart disease (IHD,
coronary heart disease) mortality has been falling steadily in New Zealand for
at least the past three decades1 yet this disease still accounts for almost
one-quarter of all deaths.1 Also, large ethnic inequalities
continue to exist,
with Māori IHD mortality rates at least twice those of
non-Māori.2–4
Projections of future trends
in IHD mortality (overall and by ethnicity) are therefore of continuing interest
to national health policy advisors and funders, District Health Board planners,
and health service providers.
The aim of this study is to
identify how IHD mortality rates have varied in the past between successive
birth cohorts and time periods, and to use this analysis of cohort and period
effects to project these rates (and counts) into the future.
MethodsData
sources—Unit record mortality data from 1956 to 2000 was made
available by the New Zealand Health Information Service. To minimise diagnostic
and coding bias over the study period, a broad definition of IHD was used,
corresponding to ICD-9 codes 410–414. These codes may still omit IHD
deaths attributed to codes for ‘sudden death’, ‘heart
failure’, ‘cardiovascular disease not otherwise specified’ and
‘(type 2) diabetes mellitus’. All deaths so coded over the study
period were assessed using multiple cause of death coding, and (where possible)
were also linked to hospital discharges (for mention of IHD). This process
yielded relatively few additional IHD deaths, however (approximately an
additional 10%).
Māori
IHD deaths were adjusted for undercounting of Māori ethnicity on death
certificates using the New Zealand Census – Mortality Study (NZCMS)
adjustors.3 These adjustors were derived by linking census to mortality
records for the 3 years following each
census from
1981onwards. So for the ethnic (Māori – non-Māori) analysis, the
study was restricted to 1981–2000 (rather than
1956–2000).
Midyear population estimates for 1956–2000 and
projections (series 4) for 2001–2015 were obtained from Statistics New
Zealand. Māori
populations were interpolated intercensally, with the 1996 censal population
being re-estimated from the 1991 and 2001 censal populations (as the ethnicity
item varied in 1996 from earlier and later censuses).
Rates were age standardised for summarisation by the
direct method, with the World Health Organization (WHO) World population as the
standard.5
Age/period/cohort
(APC) modelling—For readers unfamiliar with APC modelling, the
following brief explanation is provided. Mortality rates can be thought of as
realisations of three dimensions of time: age (at death), period (calendar year
of death) and cohort (year of birth). Age, period and cohort are proxies for the
real drivers of IHD mortality (e.g. ‘age’ captures the cumulative
process of atherosclerosis over the life course, ‘period’ captures
developments in treatment and prevention, and ‘cohort’ captures risk
exposures related to birth cohorts such as tobacco use or diet). Given a
sufficiently long time series of IHD mortality data, regression models can be
constructed to project rates based on the historical trends in age, period and
cohort effects.
APC models were fit to the available data, using 5-year
age groups and 5-year calendar periods, so defining 10-year overlapping birth
cohorts, using the statistical package S Plus. However, the entire dataset could
not be modelled because of poor fit. Instead, the data had to be truncated to
the 35–74 age range to obtain good fits, omitting the substantial
proportion of IHD deaths that occur in very old age.
Both classical (frequentist) and Bayesian models were
constructed. For the frequentist models, the assumption was made that the
underlying risk of IHD mortality increases exponentially with age, allowing
period-cohort models (rather than full APC models) with identifiable (i.e.
unique) period and cohort effects to be fit.6 Projections were then obtained by
linear regression of future period and cohort effects on the most recent three
observed effects.
For the Bayesian models, the ‘random walk
2’ full APC models were fit.7 These models have unidentifiable effects but
identifiable projections, and were used:
Ex-post tests were carried out by
fitting both sets of models to a reduced dataset that omitted the most recent
observed period (1996–2000). Mortality for this period was then projected
and the projections compared to the observed values.
ResultsDescriptivePeriod—For the
total population, rates for almost all age by sex groups increased from
1956–61 to peak in 1966–70 and then declined steadily to
1996–2000 (Figure 1).
Rates are
now well below those seen at the beginning of the study period, having fallen on
average about 60% from the peak in 1966–70 (slightly more for females and
younger age groups). This corresponds to an average annual percentage change of
approximately –3.5% over the observation period.
Figure 1. Ischaemic heart disease (IHD) mortality rates
by period (1956–2000) and age (35–54), total population
 
For
Māori, age-specific rates have declined more slowly over the
1981–2000 period (both sexes). Only in the older age groups have
substantial falls been seen, with rates in younger age groups declining
relatively little over the observation period (data not shown).
Cohort—For the
total population, each cohort experienced lower IHD mortality rates at
corresponding ages than preceding cohorts (Figure 2).
For
Māori, the pattern was essentially similar (for the included cohorts),
although with less change in the age specific rates of successive cohorts (data
not shown).
Figure 2. IHD mortality rates by cohort
(1891–1961) and age (35–74), total population
 
ModellingPeriod
effects—Period effects derived from the frequentist model are
presented in Figure 3. The effects are expressed as relative risks ie an effect
<1 is protective (lowers the IHD mortality rate) while an effect >1 is
adverse (increases the IHD mortality rate).
For the total population,
period effects become progressively less adverse from the 1971–75 period
onwards and by the most recently observed period are protective and large This
steady trend in the period effect over three or more decades provides support
for our assumption of linear projection over the 2001–2015 period.
For
Māori, there has also been an improving (ie downward) trend in period
effects from 1981–85 onwards, although the trend is shallower than that
for non-Māori, especially among males. The period effect reaches 1 in the
most recent observed period (1996–2000) or shortly before this, and is
thereafter projected to become protective, more so for females than
males.
Figure 3. Period effects, 1961–2015 (projected),
ages 35–74
Males: Females:
 Figure 4. Cohort effects, 1891 to 1976 (projected),
ages 35–74
Males:
 Females:
 Cohort
effects—Cohort effects derived from the frequentist model are
presented in Figure 4. The effects are again expressed as relative risks ie an
effect <1 is protective while an effect >1 is adverse.
For the total population
(both sexes), the striking finding is the absence of any strong cohort effects
(at least from the 1891 cohort onwards). Nevertheless, there are interesting
albeit minor variations in cohort effects to be seen. From the 1891 cohort
(females) or 1896 cohort (males), the cohort effects become increasingly adverse
(greater than 1), although still small, up to the 1916 cohort, whereafter the
trend reverses and the cohort effect becomes increasingly protective (although
still small) up to the 1946 cohort.
The most recently observed
cohorts (i.e. the 1951, 1956 and 1961 cohorts) reverse the trend once more, and
become increasingly adverse—although only the 1956 and 1961 effects
(males) or 1961 effect (females) actually exceed 1.
Although the recent trend is
far less clear than that for the period effects (see Figure 3), linear
projection over recent cohorts produces a substantive adverse cohort trend, at
least for the next few cohorts (both sexes). Thus our projection indicates the
possible emergence, for the first time ever, of a substantive (and adverse)
cohort effect (as shown by the dotted line in Figure 4).
For
Māori, the pattern is essentially similar, with no strong cohort effects
being detected. Nevertheless, the projection is for an upward trend in
cohort effects over the next fifteen years, although this is of much smaller
magnitude for males than females.
Projections—Projections
were done using both the frequentist model (which required the assumption that
recent trends in period and cohort effects would continue linearly into the
future) and the Bayesian model (which required no such assumption). In fact,
both models gave almost identical projections across all age by sex by ethnicity
groups, supporting the linearity assumption (data available from the authors).
Further validation of both models was provided by the ex post test: using the
reduced dataset, both models produced near identical projections for the
1996–2000 period, which agreed closely with the observed rates (data
available from the authors). Only the Bayesian projections are reported
here.
For
the total population (both sexes), and for both ethnic groups (Māori and
non-Māori), age-specific and age-standardised IHD mortality rates are
projected to continue to decrease until 2015. However, the rate of decrease will
progressively slow (Figure 5). This reflects
the interaction of increasingly protective period effects with increasingly
adverse cohort effects.
For both total and ethnic
populations, age standardised within the 35–74 age range, IHD mortality
rates are projected to continue their
long-term downward
trend for both sexes. It should be noted that the 90% credible interval is wide
for Māori (reflecting small numbers), but does not encompass an actual
increase in rates.
IHD
burden—Projections of the IHD burden were done by applying the
projected age-specific IHD mortality rates to the projected population (within
the 35–74 age range); see Figure 6.
For the total population (both sexes), the count of IHD
deaths (within the age range 35–74) is projected to continue to decrease,
albeit slightly more slowly than in previous decades. The projected average
annualised count in 2001–05 is 1447 (males) and 507 (females), decreasing
to 1103 and 345 in 2011–15 respectively (reductions of 24% and 32%
respectively).
Figure 6A. Average annualised IHD mortality count, ages
35–74, by sex, total population, 1956–2015 (projected)
 Figure 6B. Average annualised IHD mortality count, ages
35 – 74, by sex, Maori population, 1981 – 2015 (projected)
 This lesser reduction in burden than risk of IHD mortality
results because the declining risk is partially offset by increasing population
size together with a small contribution from population ageing (within the
35–74 age range) as the large ‘baby boom’ cohorts reach late
middle and early old age.
For
Māori, the decline in IHD mortality risk is (relatively) smaller and the
growth (and ageing) of the population is (relatively) greater. As a result, the
number of Māori deaths (both sexes) is projected to actually increase over
the next ten to fifteen
years. By
2011–15, the average annualised count of IHD deaths among Māori is
projected to reach approximately 560, a 15% increase from the 480 (average
annualised) deaths estimated for 2000–05.
DiscussionAlthough the analysis had to
be limited to the 35–74 age group, this study nevertheless provides new
and valuable information. It confirms that the peak of the IHD epidemic occurred
in New Zealand in the late 1960s, as was already known from earlier research.8
Since then, rates have fallen substantially
(by approximately 60%)
at all ages, although much less steeply for Māori. However, our models
project that the rate at which IHD mortality declines in the next decade will
progressively slow among both sexes and both major ethnic groups. To avoid this
outcome, improvements in the coverage, quality, and effectiveness of
prevention and treatment interventions will be required over and above those
anticipated by projecting the historical trend.
This projected slower decline
in IHD mortality risks, coupled with a growing and ageing population, leads us
to forecast that the burden of IHD mortality (i.e. counts as opposed to rates)
will decrease by only 25–30% (approximately) over the next decade. Indeed,
the burden (and corresponding need for preventive and therapeutic coronary care
services) is projected to
increase
for Māori. This finding has major policy implications, not least the need
to urgently improve access for Māori to and through coronary care, if
worsening of inequality in heart health between Māori and non-Māori is
to be avoided.
Our projections relate only
to the burden of IHD mortality. Trajectories for non-fatal burdens (including
need for acute coronary care and management of people with heart failure) may be
very different. Furthermore, trends in the 35–74 age group may differ from
those in the 75+ age group.
Our study reveals an
interesting pattern with regard to cohort effects. The absence of any strong
cohort effects from the 1891 to the 1951 cohorts contradicts the hypothesis
advanced by Barker9 (which states that the risk of IHD is largely predetermined
in utero), at least at the population
level. Under the fetal origins hypothesis, dramatic increases followed by
decreases in cohort effects should have been detected—yet no strong cohort
effects were found at all. This finding does not mean that the hypothesised
relationship does not exist at the individual level, merely that it is unlikely
to have had a substantive impact on the IHD epidemic at the population
level.
Our model does, however, suggest the possible emergence of a
rising cohort effect among those born since the early 1950s. If confirmed, this
would provide an explanation for the projected slowing in the secular trend of
IHD mortality over the next 10 to 15 years. That is, recent cohorts are
projected to experience higher underlying risks of IHD mortality than their
preceding cohorts—so partially offsetting the benefits that would
otherwise accrue to them from the projected continuing (protective) trend in
period effects.
What might explain these projected trends in period and
cohort effects? Continuing improvement in period effects is likely to reflect
better coverage and quality of preventive and therapeutic interventions for IHD.
Our model is unable to disaggregate the period effect into incidence reduction
and case fatality reduction components. However, analysis of Auckland MONICA
data suggests that approximately half may be attributable to downshifts in
population risk factor distributions and half to more effective and accessible
treatments (including secondary prevention and thrombolysis in
particular).10
The emergence of a substantive adverse cohort effect is
harder to explain. Firstly, it may simply be an artefact of our frequentist
model, specifically the linear regression of future cohort effects. However, the
Bayesian model, which requires no such assumption, gave almost identical
projections. Secondly, it could reflect changing proportions of different ethnic
groups in the population. This explanation is unlikely as the same pattern is
seen in the ethnic specific analyses (although
less convincingly so
for Māori males). Thirdly, it could reflect the emergence of a
‘core’ of people who are less responsive to health promotion
messages such as not smoking cigarettes—although this would be expected to
produce stabilisation of cohort effects rather than actual reversal of
the prior trend.
A more likely explanation
relates to the emergence, since the 1970s, of the epidemic of obesity (and
consequential type 2 diabetes) in New Zealand and indeed throughout the
developed world.11 In fact, similar slowing in IHD mortality declines has been
observed recently in some other developed countries, and a rising cohort effect
has been detected in both Australia12 and Sweden.13 If this explanation is
confirmed, our study will have provided the first signal of an impact of the
obesity epidemic on IHD mortality rates and burdens in New Zealand.
Regardless of the
explanation, our projections for the next decade have clear implications for
policy. At the very least, these projections imply that there is no room for
complacency in regard to the prevention and treatment of IHD—especially if
we are concerned about reducing inequalities in health between
Māori and
non-Māori.
Note:
This paper is published with the permission of the Deputy Director General of
Health (Public Health). However, opinions are the authors’ own and do not
necessarily reflect Ministry of Health policy advice.
Author
information: Martin Tobias, Public Health Physician, Ministry of Health,
Wellington; Kerry Sexton, Public Health Registrar, Ministry of Health,
Wellington; Stewart Mann, Associate Professor of Cardiovascular Medicine,
Wellington School of Medicine and Health Sciences, University of Otago,
Wellington; Norman Sharpe, Medical Director, The National Heart Foundation of
New Zealand, Auckland
Acknowledgements:
The authors gratefully acknowledge statistical assistance from Sue Paul and
Craig Wright (Ministry of Health).
Correspondence: Dr
Martin Tobias, Ministry of Health, PO Box 5013, Wellington. Fax: (04) 495 4401;
email: martin_tobias@moh.govt.nz
References:
|
|
||||||||||||
| | Current
issue | Search journal |
Archived issues | Classifieds
| Hotline (free ads) Subscribe | Contribute | Advertise | Contact Us | Copyright | Other Journals |
|