Journal of the New Zealand Medical Association, 29-June-2012, Vol 125 No 1357
Lavinia Tan, Tony A Blakely
The measurement and monitoring of ethnic inequalities in health have long been of interest in New Zealand, both for research and policy. There are large differences in mortality between ethnicities in New Zealand.1–4
Māori and Pacific have higher mortality rates than non-Māori non-Pacific non-Asian (nonMPA), while mortality rates of Asian people (people from East, South East and South Asia, but excluding those from the Middle East and Central Asia) are less than nonMPA in New Zealand.1 The main contributors to these inequalities are mortality in the older age groups (45+ years), particularly for cancer and cardiovascular disease.1
Mortality rates change over time with changes in society, culture and economy, and perhaps differentially by ethnicity. Māori mortality decreased markedly up to the 1980’s then slowed its decline, such that the gap between Māori and non-Māori increased. Since the late 1990s, mortality rates have declined at a faster rate among Māori, and absolute differences in rates (and life expectancy) are again closing between Māori and non-Māori.2,3
In the past, we have only linked 3 years of mortality data to each census to secure more rapid monitoring data after each census and also because linkage success decayed with time following census. However, we noted in the 2001–04 mortality linkage and in parallel linkage of 5 years of cancer registrations to the census (CancerTrends; www.uow.otago.ac.nz/CancerTrends-info.html) that linkage now deteriorates little with time for events 4 to 5 years after census night.
This improvement is due to numerous sources of residential geocodes, enabling us to better select a geocode for the decedent at about census night even when their death occurred 4 to 5 years after census night. To provide more complete data on social group trends in mortality in New Zealand, we linked 2004–06 mortality records to the already existing 2001–04 census-mortality cohort.
This paper presents trends in mortality, by ethnicity, for the period 1981–2006, with a particular focus on comparisons of the first and second 30 months post-2001 Census (hereafter 2001–03 and 2003–06, respectively) with existing 1996–99 NZCMS data. We investigate the validity of the extended census-cohort, and changes in ethnic-specific all-cause and cause-specific mortality rates and inequalities.
Linkage and weighting—Methods for linkage of each of the 1981, 1986, 1991, 1996 and 2001 Censuses with 3 years of subsequent mortality data have been described elsewhere.4–9 A technical note on the 2004-06 linkage can be found at the NZCMS website.10 Briefly, probabilistic record linkage was used to link census and mortality records anonymously using the variables sex, date of birth, country of birth and ethnic group. The blocking variable used for matching was address or census area unit, using multiple health dataset sources for address geocodes to attempt to bracket the date of the 2001 Census.
Mortality records for 2001–06 eligible for linkage were those where the decedent was alive at the 2001 Census and was living in New Zealand on 2001 Census night according to the duration in New Zealand variable recorded on the mortality file.
The percentage of eligible mortality records linked in 2001–04 was 79.6%, and in 2004-06 was 79.8%. Because not all mortality records were linked to a corresponding census record, it was necessary to correct for any linkage bias and consequent underestimation of mortality rates.
Weights were calculated based on variables that were predictors of linkage in logistic regression analyses: age at census, sex, prioritised ethnicity, rurality, residential mobility of area unit, Territorial Authority, NZ deprivation index, months since census night at death, and cause of death.
To improve efficiency of data processing, the method for weight calculation for the full 2001–06 data differed from previous years. Cells within a stratum that met the numerical criterion of > 5 linked records, were separated and assigned an independent weight, whereas the remaining cells were collapsed.
The order of collapsing of strata variables to ensure sufficient cell sizes was based on the strength of their relationship with linkage (see10). The initial weights were secondarily weighted (in the same manner as previous cohorts) by age, sex and ethnicity so that the sum of all the weighted linked mortality records was forced to equal the total number of linked and unlinked mortality records within age, sex and ethnicity strata. Overall, the revised weighting procedure was found to perform as satisfactorily as previous methods (see10 for a more detailed description of the weighting process).
Due to larger than expected reductions in mortality rates from 2001–03 to 2003–06 (as reported in Results section of this paper), we undertook an additional weighting to force the exact sum of weighted and eligible deaths for the first and second 30-month periods following the 2001 Census to exactly agree (see 10). That is, we were concerned that despite including ‘months since census’ in our primary weighting strategy, our above algorithm of aggregating cells did not adequately correct for any residual deterioration in linkage success – particularly for the younger Māori and Pacific age groups. However, there were no systematic differences between our preferred weights and the additional weights based on duration since census night.
We therefore concluded that residual linkage bias by time since census was not occurring. Consequently, the preferred linkage weights were used for all analyses by the 30 month split (i.e. 2001–03 and 2003–06). We also include previously published 2001–04 rates for comparison.2 11
Cohort analyses—As in previous NZCMS reports and publications (e.g.11) and consistent with ethnicity standards,12 a “Total” definition of ethnicity was used for Māori, Pacific and Asian in analyses, using (all) ethnic groups self-identified by the census question. Thus individuals could be assigned to up to three of Māori, Pacific and Asian ethnicities. A mutually exclusive group of non-Māori non-Pacific non-Asian (hereafter called European/Other) was used as the reference group for ethnicity comparisons.
Age-standardised rates were calculated for all four ethnic groups, using the WHO World population as the standard. We examined changes in mortality rates and relative (standardised rate ratios; SRR) and absolute (standardised rate differences; SRD) inequalities in rates by sex and age group. All data analyses were conducted in SAS.
All-cause mortality rates—Figure 1 shows age standardised all-cause mortality rates for 1–74 year old males and females. Each observation is plotted at the midpoint of its period of follow-up according to the X-axis, meaning that the 2001–03 and 2003–06 series are closer together than the 5-year gaps in the existing 1981–84 to 2001–04 series. The 2001–03 rates commencing the new series agree reasonably closely with the previous 2001–04 rates, with any difference being due to slightly different follow-up periods (30 versus 36 months) and the modified linkage weights.
Mortality rates were consistently highest for Māori, followed by Pacific, then European/Other ethnicities, and lowest for Asians, and inequalities were large at all points in time. The long-run trend of steadily decreasing mortality rates over time for European/Other and Asian continued and recent accelerations in Māori mortality decline seemed to be maintained.
Pacific mortality rates were more variable pre-1996, but also exhibited the same decrease from 1996 onwards. More specifically, observed all-cause mortality rates fell for all ethnic groups from 2001–03 to 2003–06, but more so for Pacific (18.3% and 21.7% for males and females, respectively) and Asian (22.2%, 16.7%), than for Māori (13.2%, 14.2%) and European/Other (13.0%, 10.4%) (Table 1 and Table 2).
Figure 1. All cause age-standardised mortality rate (per 100,000) for males and females by ethnicity
The first line (starting from the left) for each ethnic group is for the previously published 1981–84, 1986–89, 1991–94, 1996–99 and 2001–04 series (i.e. three years of mortality data each linked to a census). The second line is that for the new series, 2001-03 and 2003-06. Error bars are 95% confidence intervals.
Table 1. Person years and numbers of weighted deaths for all ages (1-74 years) combined for 2001–03 and 2003–06
Table 2. Standardised mortality rates and percentage change from preceding cohort for all cause mortality by ethnicity, period and sex. (Full tables of rates for all age groups and all causes of death are available as web annex tables at http://www.uow.otago.ac.nz/nzcms-info.html.)
† 2001–04 compared to 1996–99; 2001–03 compared to 1996–99; and 2003–06 compared to 2001–03,
We suspect that there is a systematic bias in the NZCMS due to the inability to censor respondents who migrate out from New Zealand after census night, and we also suspect this bias might be greater for Pacific and Asian people. We consider this further in the Discussion. Thus, the ‘observed’ rates, SRRs and SRDs for Asian and Pacific people in 2003–06 need to be treated with considerable caution and accordingly, the remainder of the Results section is mostly focussed on rates calculated for Māori and European/Other.
Observed rate ratios for Māori, compared to European/Other, were 2.43 (95%CI 2.31–2.57) for males and 2.72 (2.56–2.89) for females in 2003–06, the same as in 2001–03 for males (2.44) and 7% less for females (2.84) (Table 3 and Figure 2). The observed SRDs decreased by 13% for males (from 434 to 376 per 100,000) and by 16% for females (from 335 to 281 per 100,000).
Table 3. Age-standardised rate ratios (SRRs) and rate differences (SRD) for selected age groups and causes of death, by sex by ethnic group, for 1996-99 onwards. (Full tables are available as web annex tables at http://www.uow.otago.ac.nz/nzcms-info.html)
Mortality rates by age group—Mortality rates were greatest for Māori, followed by Pacific, then European/Other ethnicities, and lowest for Asians for adults aged 15+ yrs (Figure 3; actual rates and percentage changes in Web Annex Table 1 at www.uow.otago.ac.nz/nzcms-info.html). Across all age groups and for both sexes, there is a long-term, steadily decreasing trend in mortality rates.
Rates for Pacific tend to show larger than plausible decreases from 2001–03 to 2003–06 for 45–64 year olds (22.2% and 21.8%, males and females respectively) and for 65–74 year olds (17.1% and 18.6%). By way of comparison, the corresponding percentage declines for European/Other were 12.4% and 7.9% for 45–64 year olds, and 13.8% and 10.4% for 65–74 year olds. Percentage declines from 2001–03 to 2003–06, were also particularly notable among these older age groups for Asian and, to a lesser extent, Māori.
Figure 2. Trends in relative (SRR; left) and absolute (SRD; right) inequalities in all cause mortality for Māori, Pacific and Asian each compared to European/Other, by sex, for all ages (1–74 years)
The first line (starting from the left) for each ethnic group is for the previously published 1981-84, 1986-89, 1991-94, 1996-99 and 2001-04 series (i.e. three years of mortality data each linked to a census). The second line is that for the new series, 2001-03 and 2003-06. Error bars are 95% confidence intervals.
Trends in relative inequalities (SRRs) between Māori and European/Other by age group from 1996–99 to 2001–03 to 2003–06 were unclear across age groups, and therefore arguably best assumed as stable over time (Web Annex Table 3).
Given the background trends for mortality reduction in all ethnic by age groups, there was necessarily a pattern of reducing absolute inequalities between Māori and European/Other over time across age groups. For example, the SRDs among 45–64 year olds reduced by 29% for males and 21% for females from 1996–99 to 2003–06 (Web Annex Table 3), although these may be somewhat overestimated due to the suspected migration bias within NZCMS cohorts (see Discussion).
Mortality rates by specific cause of death— Figure 4 shows cause-specific mortality rates. Declines in cardiovascular disease and injury mortality were the main drivers of all-cause mortality rate reductions for all ethnic groups (Figure 4). Relative inequalities in CVD between Māori and European/Other remain very high (three to four-fold relative risks), but reduced by 8% for both males and females from 2001–03 to 2003–06, which in turn means that absolute inequalities closed by as much as 20%.
Each line plots the previously published 1981-84, 1986-89, 1991-94, and 1996-99 series (but not 2001-04), and then continues with the new 2001-03 and 2003-06 series. Error bars are 95% confidence intervals.
Each line plots the previously published 1981-84, 1986-89, 1991-94, and 1996-99 series (but not 2001-04), and then continues with the new 2001-03 and 2003-06 series. Error bars are 95% confidence intervals.
Māori unintentional injury mortality rates are higher than other ethnicities, for both males and females. Inequalities were unstable over time, and not measured with sufficient precision to make confident conclusions about trends in Māori European/Other inequalities. Injury rates were comparable between Pacific and European/Other ethnicities since 1996–99, and lower among Asian.
Rates for other causes of death can be found at the NZCMS website: www.uow.otago.ac.nz/nzcms-info.html
Observed all-cause mortality rates continued to fall in 2003–06 compared to previous periods for all ethnic groups, but especially so for Pacific and Asian people (and probably affected by bias—see below). The falling mortality rates are largely driven by falls in CVD and unintentional injury.
Relative inequalities in mortality between Māori and European/Other have stabilised since the late 1990s—and possibly decreased for females. Given this stability or slight reduction in relative inequalities, and the overall decreasing trend in mortality for all ethnic groups, the absolute inequalities between Māori and European/Other have necessarily decreased.
Methodologically, we believe we have encountered a bias in the NZCMS when results are reported by time since census. As shown in Figure 1, Pacific and Asian rates seem to fall by an implausibly large amount from 2001–03 to 2003–06. We thoroughly checked our linkage bias weights (see Methods and elsewhere10), and do not believe that inadequate adjustment for (any) decline in linkage success is the problem. Rather, we think that the number of deaths among the people alive on census night is progressively underestimated with time since census night, due to our inability to censor New Zealanders as they emigrate.
Put more simply, we do not identify deaths among those New Zealand residents who completed the 2001 Census and subsequently emigrated, and possibly died overseas. As there is relatively frequent migration to and from the Pacific 13 14, and this likely applies to a more recent immigration population such as Asian people, we suspect this bias is greater among Pacific and Asian people.
A direct test of this hypothesis would require data on emigration by ethnicity to allow censoring, or identification of deaths among those emigrating. We do not have such data. However, we do have data on ‘permanent and long-term departures’ from New Zealand by country/region of destination (personal communication, Robert Didham, Statistics New Zealand, July 2010).
If our hypothesis about differential emigration by ethnicity is correct, we might also expect emigration of older people (at great risk of death) to be more common among Pacific and Asian populations. Those people emigrating to the Pacific or Asia are skewed towards older age groups, both using simple counts and also a ‘crude’ proportion measure using the 2006 New Zealand Pacific, Asian and all population count data as denominators (Web Annex Figure 1 and 2).
If our suspicions are correct and our observed mortality rates are a result of emigrant bias, then mortality rates for Asian and Pacific in 2003–06 are too low, and SRRs in 2003–06 comparing either Pacific or Asian with European/Other are also too low (due to misclassification bias of the mortality outcome that is differential by ethnicity). We do not have enough evidence to suggest any difference in this bias (in relative terms) for Māori and European/Other; consequently, the SRRs comparing Māori with European/Other are probably valid.
A possibly greater underestimation of Pacific and Asian mortality rates is consistent with a study that examined “unhealthy return migrant” (URM) bias mortality rates for Pacific.15 Tobias showed that it is possible to estimate the extent of this bias using lung cancer as a “tracer condition” whereby spuriously high survival could only be explained by emigration, and in turn allowed an estimate of approximately 20% undercount for Pacific deaths in older people. However, its accuracy is unknown and immeasurable. For example, the URM adjustor might account for both return migration and missed NHI links between the National Cancer Registry and mortality files.
There are potential methods to correct this bias in the NZCMS in the future, if we are to present mortality rates by year since census night. First, we currently only attempt to link deaths when the mortality record states they have been in New Zealand long enough to have answered the previous census. Thus, if we were to assume that immigration and emigration roughly cancel each other out (not currently true for Asian people), then we could make these excluded mortality records ‘eligible’, and when they fail to be linked to a census record they will then contribute to a large linkage bias weight, thus also (potentially) correcting for this emigrant bias. Second, we could attempt to link NZCMS data to migration data, and ‘properly’ censor the cohorts. However, this would be a large undertaking, and possibly not justifiable from either a SNZ or a funders perspective.
Third, we could attempt to create a database of New Zealand residents (as of census night) dying overseas in the following 5 years. This would also be a very large undertaking, and probably unreliable. Fourth, we could use external data on emigration rates by sex, age and ethnicity to undertake quantitative bias analyses (or sensitivity analyses) of the results presented here.
Unfortunately, migration data is not collected by ethnic group (the closest approximation is country of birth), and we would have to estimate the mortality rate among the emigrants (which would not be the same as non-emigrants). Finally, and currently our preferred option, once 5 years of mortality data have been linked to the 2006 census, we will have a continuous series of 10 years of linked data, with the fifth and sixth years bracketing the beginning of the new cohort. By using regression modelling, we would be able to estimate the magnitude of unexpected ‘jumps’ in mortality rates across this boundary by sex, age and ethnic group, and thereby estimate sex by age by ethnic group specific adjustment factors.
We suspect that analyses comparing mortality rates over time within one of the closed NZCMS cohorts (e.g. 2001–03 compared to 2003–06) is prone to bias due to our inability to censor people when they migrate out of New Zealand. This limitation to the NZCMS means mortality rates are increasingly underestimated with time since census night, particularly for Pacific and Asian people. However, previously published NZCMS trends remain valid as the duration of follow-up (3 years) is short, and cohorts were not split by time since census.
Nevertheless, it is safe to conclude that mortality rates continued to decline from 2001–03 to 2003–04 for all four ethnic groups. All-cause mortality inequalities for Māori compared to European/Other over this time were probably stable in relative terms and decreasing in absolute terms, but CVD inequalities probably decreased in both absolute and relative terms.
Statistics NZ Security Statement: Access to the data used in this study was provided by Statistics New Zealand under conditions designed to give effect to the security and confidentiality provisions of the Statistics Act 1975. The results presented in this study are the work of the author, not Statistics New Zealand’s.
Author information: Lavinia Tan, Research Fellow; Tony A Blakely, Director; Health Inequalities Research Programme, University of Otago, Wellington
Acknowledgements: We acknowledge comments on drafts of this paper from Martin Tobias, Robert Didham and June Atkinson. The New Zealand Census-Mortality Study was initially funded by the Health Research Council of New Zealand and is now part of the HRC-funded Health Inequalities Research Programme (08/048). The NZCMS is now principally funded by the Ministry of Health.
Correspondence: Tony Blakely, Health Inequalities Research Programme, University of Otago, Wellington, PO Box 7343, Wellington, New Zealand. Fax: +64 (0)4 3895319; email: email@example.com
issue | Search journal |
Archived issues | Classifieds
| Hotline (free ads)
Subscribe | Contribute | Advertise | Contact Us | Copyright | Other Journals