Introduction
In Aotearoa New Zealand (ANZ), the indigenous Māori population and Pacific peoples (the fourth largest ethnic group when considered as a single population group) have significantly poorer health status than people from other ethnic groups [
1‐
3]. Improving equity in health status between population groups has been a strategic priority in successive health system reforms. In 2020, a comprehensive review of the health and disability system concluded that there were still major – and sometimes increasing - disparities in health status between ethnic groups, and that inequities differed between health district (District Health Board (DHB)) areas [
4] This resulted in another major health reform, through the Pae Ora (Healthy Futures) Act 2022 [
5], which took effect on 1 July 2022. The local DHBs were disestablished and replaced with the national organisations ‘Te Whatu Ora’ (Health New Zealand (HNZ) [
6], ‘Te Aka Whai Ora’(the Māori Health Authority to drive improvement in Māori health) [
7]; a new Public Health Agency (within the Ministry of Health) [
8], and a centralised National Public Health Service (within Te Whatu Ora) instead of the previous 12 Public Health Units [
9]. Instead of the 20 DHBs, 60–80 smaller ‘Localities’ are to be established, [
10] which are geographic areas that are home to communities with their own specific health needs. A population health approach will be embedded in these localities for better implementation of (public) health measures. A new Pacific Health Strategy has also been developed to support Pacific peoples’ health [
11].
Whilst there has been attention to emerging evidence about disparities in health status for Pacific populations over the past 20 years, and several Pacific Health Action Plans have been developed [
11,
12], inconsistent reporting and lack of visibility of Pacific ethnic groups in health data and statistics prevent a clear understanding of the health issues Pacific people face, complicating policy making. Underrepresentation, specifically of Pacific people in population statistics, is well documented [
13], but attention to this issue has usually only occurred when health statistics exceed estimates of 100%, exposing inaccurate denominators. Health system reports often include a caveat that data should be ‘interpreted with care’, without addressing the possible impacts of these inaccuracies [
14,
15].
Having accurate population data is a prerequisite for making informed decisions and promoting equitable health in populations and communities [
16]. High-quality census data collected as part of the official five-yearly count of people and dwellings is required for accurate population estimates [
17]. During the rollout of COVID-19 vaccinations, media interest and public reporting of vaccination coverage led to the Ministry of Health and Statistics New Zealand (StatsNZ) announcing - without much public explanation - that Health Service User data (HSU2020) would provide more reliable estimates of COVID-19 vaccination coverage than the generally used population projections or “best available population” (BAP) measure based on census data [
18]. The impact of the introduction of new biases when using HSU [
19], while continuing use of BAP denominators in other health estimates, have been overlooked.
In this article, we aimed to understand how the choice of denominator population affects estimates of population size and health system performance for Pacific peoples at a local level by describing and analysing the quality and appropriateness of the two denominators currently used in health statistics (HSU and BAP) using two indicators as examples. Furthermore, we discuss how the use of population estimates in other health statistics may also contribute to persistent institutional racism experienced by Pacific people.
Discussion
Our study found substantial discrepancies when using BAP projections and HSU data for the denominator in health care statistics. These discrepancies were not equally distributed across geographic areas nor ethnic groups and were most pronounced in the Pacific population in South Auckland, where Pacific people do better than other ethnic groups with one denominator, and worse when using the other. It is important to understand that both denominators have specific and significant shortcomings.
The gold standard for official population statistics is the 5-yearly census and its derived population statistics. The infrequency of census data collection, the failure to require internal migrants to register changes in residential address [
27], and the decentralisation of the health system into 20 health districts as part of the 2000 health reforms have created the need for more granular subnational population data. Ethnicity is frequently used in funding algorithms as it is an important determinant of health and health service use [
33]. BAP has been developed for (health care) planning at subnational district level. Declining census participation (particularly for ethnic minority groups) inequitably affects not only the accuracy of census population size estimates, but also the quality of collected data. This affects the assumptions upon which BAP projections are modelled. Selective underestimates of the most vulnerable population groups lead to specific underfunding of these groups, and this contributes to increasing inequity in health.
Although population projections were never intended to be used as a denominator for health status statistics, BAP has increasingly been used for this purpose. Projected population statistics are based on population-level rather than individual data so cannot be linked to individual health statistics (as population-based registers can be). Health outcome statistics can only be ecologically compared with population projections at an aggregate level, resulting in ‘numerator-denominator’ discrepancies and biased results [
31]. StatsNZ’s recommendation is to always validate the consistency of the numerator with the denominator in health statistics. This is not possible using BAP denominators, demonstrating that BAP data are not suitable for this purpose. The problems with the use of BAP are exposed when percentages over 100 occur.
In the absence of a population register, HSU data were used as a proxy to estimate COVID-19 vaccination coverages. This solved two problems: the numerator and denominator could be linked and percentages over 100 no longer occurred, and HSU data included addresses that are presumably updated more frequently than census address data and therefore are more accurate. This enabled estimating vaccination coverage by ethnicity, deprivation index and age group in any defined geographic area [
40]. However, new, non-quantifiable biases were introduced. First, health user data are an inappropriate denominator for the uptake of health care, as underserved populations with no or limited health care access are invisible in the dataset. Furthermore, there are quality issues with ‘ethnicity’ and ‘address’ data in the HSU that disproportionately affect Māori and Pacific ethnic groups. Linking Census 2018 to HSU2020 data showed that 15–20% of people registered as Māori or Pacific in census data were not recorded as such in HSU [
36]. Another study comparing HSU2018 to Census 2018 found 16% fewer Māori in HSU, and that Māori were more often misclassified than non-Māori ethnic groups [
41]. A further problem for accurate Pacific health data is that HSU prioritises ethnicity data by Māori instead of reporting by total ethnicity, as recommended by StatsNZ. This under-represents Pacific people by another 15% [
31].
Population undercounts can arise for several reasons, such as non-participation in censuses, under-use of civil services, under-use of health care, misclassification of data used to link datasets, immigrants who are not registered in birth and education registers, gaps in emigration data or linking problems with emigration data, etc. All these reasons can work in concert, and different reasons can apply to different ethnic groups.
Deprivation is related to geographic mobility and the number of changes in residential address increases with increasing deprivation [
42]. Pacific (and Māori) people use health care less frequently and move more often, and therefore have less accurate address updates. Health user data are often used when address data are needed, for example for recalls for national vaccinations or cancer screening programmes. Inaccurate addresses and absence in health data contribute to disparities in health because people will not receive invitations for screening or recalls. Both disproportionately affect Māori and Pacific peoples [
15].
BAP and HSU are not official statistics which means that the datasets are not regularly released or revised. Interim, unplanned adjustments can be made. BAP projections are preliminary and can be adjusted up to three years after publication [
18]. The interim adjustments in both BAP and HSU after publication of the post-hoc review by StatsNZ [
31] are likely to be the reason we could not reproduce the published reviewed data in our calculations. Unanticipated changes in methodology were also applied to Census 2018 (Table
1), which adds to the difficulty in reliably studying trends compared to data based on successive census statistics.
Systematic and consistent collection of data is essential to support the public health surveillance functions of monitoring trends over time, studying determinants for change, detecting (new) risk groups, and assessing the impact of interventions [
43]. To be able to interpret trends reliably, stable denominators that accurately represent population changes are important. Most ANZ health statistics [
35,
39,
44], including the national surveillance of notifiable diseases under the Health Act 1956 [
45,
46] use BAP denominators. Most of these publications acknowledge that estimates are based on projections rather than actual statistics but do not clarify that the degree of inaccuracy differs between ethnic groups and between regions. Furthermore, projection methods can be changed at any time, including retrospectively up to 3 years after publication. In our trend evaluation of PHO enrolment and cervical cancer screening, we found that changes in trends mostly coincided with changes in population projections that had differing effects by ethnic group, rather than in changes in enrolment or screening numbers, making meaningful interpretation impossible.
Both indicator examples, ‘access to primary care’ and ‘uptake of cervical screening’, show that replacing the BAP denominator with HSU has different impacts on estimates in different geographic areas and on different ethnic groups, which cannot be corrected for. The denominator change did however consistently have the largest impact on ‘access to primary care’ and ‘uptake of cervical screening’ for the Pacific populations living in the neighbourhoods with the highest deprivation levels in the Counties Manukau district in South Auckland.
The indicators that are published on government websites and dashboards report that Pacific people have the highest coverage compared with all ethnic groups, and the coverage for Pacific people in Counties Manukau is higher than for Pacific people in other regions. It is more likely that these indicators for Pacific peoples are lowest of all ethnic groups, which is indeed what we found based on estimates using HSU denominators. Although the accuracy of HSU is unknown, it is safe to assume that HSU denominators are underestimates of the true population size, disproportionately so for smaller ethnic groups, making the real uptake for Pacific people even lower than the results we report based on the HSU denominator.
The need for equitable denominators for health statistics: a real-time population register
Experiments are underway to investigate if collated national governmental administrative data in the IDI can contribute to or replace the census in the future [
47,
48]. Combining administrative data has its own quality problems, largely based on non-uniform, inconsistent, and non-transparent methods for data collection [
49]. In the current IDI, in theory, people can be simultaneously registered with as many different addresses and ethnicities as there are administrative databases. IDI deals with the problem of multiple datasets with different addresses and ethnicity by having a “personal details table” which ranks sources of data based on data quality assessments and includes variables such as “source ranked ethnicity”. Discrepancies between IDI and census cohorts were, again, found to be largest for people in the highest deprived neighbourhoods where most Māori and Pacific people live [
47]. Complicated data cleaning procedures are in place which delays the availability of data and with suboptimal results [
47]. It is unlikely that IDI data quality, which is collected by different agencies and not in controllable standardised ways will ever reach the standardised ethnicity data quality as collected on census night.
Under the current health reforms, it is intended to embed a population health approach in smaller locality areas than the previous health districts for better implementation of (public) health measures. Analysing data on a more granular level will magnify these data quality problems and further increase the inaccuracies.
A major improvement in the quality and completeness of population data could be achieved by developing an overarching, central governmental population register that collects ‘real-time’ basic demographic data. Instead of the current suite of ‘unique’ identifiers for health, tax, education etc., one national unique identifier would be introduced for each resident, replacing all other identifiers. When a person interacts with any governmental administrative service, the respective systems should retrieve the most up to date demographics from this central register, ask the person to check their details, correct where necessary, send the corrected data back to the register, update, and log the changes. Although the absence of a population register has been described as one of the main barriers in the transition to a census based on linked administrative registers, this major change towards a linked administrative census is now considered without addressing this barrier [
50]. However, the creation of a comprehensive individual register would be an additional major change to the programme of official statistics, and would require the prior robust assessment of both social acceptability and the practicality of such a database of comprehensive routinely linked information.
Population-based registers that record all vaccinations administered to individuals targeted for vaccination
residing within a certain administrative area is the gold standard to calculate vaccine coverage [
51]. A recent review (2022) found that access to childhood vaccinations depends on enrolment in primary care, which is disproportionately delayed or hampered for Māori (and likely Pacific) children [
52]. This contributes to the rapidly declining vaccination coverage in both ethnic groups. We recommended urgent action to develop a nationwide, centrally governed system that provides antenatal immunisations information, with a focus on vaccination coverage for groups with the lowest coverage. A 2020 review of the national cervical screening programme [
53] also recommended developing a population-based register for ongoing audit and review of cervical cancer cases that utilise a consistent methodology and allow the selection of control groups for case control studies.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.