Grouping women of South Asian ethnicity for pregnancy research in New Zealand

Background The New Zealand (NZ) Ministry of Health ethnicity data protocols recommend that people of South Asian (SAsian) ethnicity, other than Indian, are combined with people of Japanese and Korean ethnicity at the most commonly used level of aggregation in health research (level two). This may not work well for perinatal studies, as it has long been observed that women of Indian ethnicity have higher rates of adverse pregnancy outcomes, such as perinatal death. It is possible that women of other SAsian ethnicities share this risk. Aims This study was performed to identify appropriate groupings of women of SAsian ethnicity for perinatal research. Materials and Methods National maternity and neonatal data, and singleton birth records between 2008 and 2017 were linked using the Statistics NZ Integrated Data Infrastructure. Socio‐demographic risk profiles and pregnancy outcomes were compared between 15 ethnic groups. Recommendations were made based on statistical analyses and cultural evaluation with members of the SAsian research community. Results Similarities were observed between women of Indian, Fijian Indian, South African Indian, Sri Lankan, Bangladeshi and Pakistani ethnicities. A lower‐risk profile was seen among Japanese and Korean mothers. Risk profiles of women of combined Indian‐Māori, Indian‐Pacific and Indian‐New Zealand European ethnicity more closely represented their corresponding non‐Indian ethnicities. Conclusions Based on these findings, we suggest a review of current NZ Ministry of Health ethnicity data protocols. We recommend that researchers understand the risk profiles of participants prior to aggregation of groups in research, to mitigate risks associated with masking differences.


INTRODUCTION
Women of South Asian (SAsian) ethnicity in Western societies are at higher risk for adverse pregnancy outcomes compared with other ethnicities, including stillbirth, preterm delivery, and gestational diabetes mellitus (GDM). [1][2][3][4] Geographically South Asia includes Afghanistan, Bangladesh, Bhutan, India, the Maldives, Nepal, Pakistan, and Sri Lanka. 5 Current international research combines women of SAsian ethnicities for analysis based on these definitions, 1,6,7 even though 'South Asian' encompasses people of varied backgrounds. Historical heterogeneity of SAsian peoples and more recent migration patterns have resulted in sociocultural diversity both within South Asia, and in the rest of the world. The Indian diaspora is additionally the largest worldwide. 8 Migrants from the Indian subcontinent made up for 3.4% of the total New Zealand (NZ) population at the time of the 2013 Census, increasing to 4.8% in 2018. 9 Consecutively, while the total sum of births in NZ has declined over the last two decades, a growing number of births to women of Indian ethnicity has been observed each year. 10 Considering the current birth-trend, it has become increasingly important to understand and mitigate this risk of adverse pregnancy outcomes among women of SAsian ethnicity.
Some key characteristics of ethnicity recorded in NZ include that it is self-defined, and that an individual may identify with more than one ethnic group. The use of ethnicity data in health research is addressed by the Ethnicity Data Protocols for the Health and Disability Sector by the Ministry of Health (MOH). 11 According to this protocol, ethnicity data can be categorised at four different 'levels'. In aggregation a reasonable level of detail is maintained for some ethnicities (such as Māori, Pacific Peoples, Chinese or Indian), while other minority groups are merged together despite large heterogeneity (such as other Asian ethnicities, African or Latin American). All SAsian groups, except Indian, are merged with Japanese and Korean ethnicity at level two aggregation, which is most commonly used in maternity research. If an individual identifies with multiple ethnicities, three forms of output are recommended: total response (ie a person is counted in each reported ethnic group), prioritised (ie a single ethnic group is allocated based on prioritisation tables outlined by the MOH), and sole/combination (ie categories of women with one or a combination of ethnic groups). 11 Although 'prioritised' output of ethnicity data is often used in health research with the intention to fairly represent Māori, Pacific Peoples and ethnic minorities, while ensuring large enough groups for analyses, the appropriateness of this method has been questioned. 12,13 This study was performed to identify groupings of women of SAsian ethnicity for pregnancy research, with a comparison to current NZ ethnicity categorisation, as understanding the facilitators and barriers to health outcomes for ethnic minorities require populations with small numbers to be appropriately aggregated.
Based on historical diaspora, we hypothesised that women of Indian, Sri Lankan, Bangladeshi and Pakistani ethnicity would be suitable for grouping. We identified a mother as Fijian Indian, when ethnicity was coded as 'Fijian' and 'Indian', 'Fijian Indian', or 'Fijian Indian' and 'Indian'.

MATERIALS AND METHODS
Similarly mothers were classified as South African Indian, when ethnicity was coded as 'South African' and 'Indian'. No births to women of Bhutanese or Maldivian ethnicities were found. Other combined groups were excluded from this study as these were either too small or too heterogenous.

Statistical analyses
Analyses were performed using SAS 8.3 Enterprise Guide.
Demographic characteristics were compared in univariate analyses using Kruskal-Wallis and χ 2 or Fisher's exact tests. Post hoc analyses were adjusted for multiple comparisons by applying the Dwass-Steel-Critchlow-Fligner procedure or the Stepdown Bonferroni method. Investigated pregnancy outcomes included perinatal related mortality (deaths from 20 weeks gestation up to the 28th day after birth per 1000 total births), preterm birth (PTB; births <37 weeks gestation per 100 live births), caesarean section (CS; both elective and emergency procedures), assisted deliveries (AD; forceps and vacuum), induction of labour (IOL), hypertensive disorders of pregnancy (pregnancy-induced hypertension, pre-eclampsia and eclampsia) 15 and gestational diabetes mellitus (GDM). 16 If hypertensive disorders or GDM were not flagged, these were assumed 'absent'. Births to women registered with a district health board (DHB; hospital providers of maternity care) were excluded from all analyses including body mass index (BMI), smoking, parity or trimester of booking. Analyses on CS, AD and IOL were based on nulliparous women only. Pregnancy outcome rates were compared by Mantel-Haenszel test and simple logistic regression methods. Odds ratio (OR) or adjusted odds ratio (aOR) estimates, and profile-likelihood confidence intervals were computed. Women of Indian ethnicity, as the largest SAsian category, were used as the referent group in all analyses.
No between-group analyses were performed.

RESULTS
There were 606 435 singleton births in NZ between 2008 and 2017.
An increasing number of births to women of SAsian ethnicity was observed over time, representing 3.7% of total births in 2008 and 7.4% in 2017 (Fig. 1). There were 31 074 births among mothers identifying as solely SAsian (Table 1). Ethnicity data was sourced from BDM for the majority of women (98.3%). Demographic characteristics of all individual groups are shown in Table 1. Pregnancy outcome rates are shown in Table 2. Some results were suppressed due to low counts, or secondarily suppressed to prohibit re-calculation, following Statistics NZ guidelines. 17

Maternal characteristics
Significant differences in maternal characteristics such as mater-  As seen in Figure 2, women of Māori, Pacific and NZE ethnicities have lower GDM rates across all BMI categories, compared to Indian mothers.

DISCUSSION
Many similarities in demographic characteristics and pregnancy complications were observed between women of Indian, Fijian Indian, South African Indian, Sri Lankan, Bangladeshi and  Hypertensive disorders of pregnancy adjusted for maternal age, maternal body mass index, and parity; gestational diabetes adjusted for maternal body mass index. † Denominator: all vaginal births (excluding caesarean section); S: suppressed value due to low count, or to prohibit re-calculation.
Pakistani ethnicity. Therefore, we consider it appropriate to group these women for perinatal studies in the NZ setting. We acknowledge that some women of Fijian Indian ethnicity may identify more as Pacific than Indian due to historic events over the past 140 years. 18 Although we observed a marginal shift in pregnancy risk factors between women of Indian and Fijian Indian ethnicities toward a Pacific risk profile, this was not associated with a significant alteration in pregnancy outcomes. In order to better represent people from these SAsian groups in NZ, we therefore propose a revision of the current collection and aggregation methods of the MOH ethnicity data protocols, reflecting concerns raised during public consultation by Statistics NZ in 2019. 13 An alternative grouping method, based on similar analyses as those performed in this study, could include aggregation by Central, South, South-East, and East Asian ethnicities. Furthermore, although this study aims to group women with similar risk profiles, we acknowledge that a high level of diversity still exists between people of SAsian ethnicity.
Current ethnicity data collection methods that record 'Indian' or 'other Asian' do not identify this internal diversity and therefore prohibit more nuanced discussion. An example of internal diversity includes religion, although a genetic study of SAsian people suggests that geographic location explains genetic variation better than religion, highlighting the importance of ancestry. 22 It has been suggested to reconstruct classification based on people of north Indian and south Indian descent, emphasising this ancestral link. 23 If the quality of ethnicity data collection in NZ were improved and included more detail, the accuracy of grouping could be increased further.  Table 1. previously published. By performing this research, we were able to challenge commonly accepted protocols, with the aim to improve maternity research in NZ. This may ultimately lead to better understanding of risk and development of intervention strategies tailored to specific at-risk groups.
There were some data limitations to this study. Exploratory analyses by level four ethnicity codes suggest low-quality data collection for women of SAsian ethnicity. We suspect that Fijian Indian mothers were often incorrectly coded as 'Indian' and 'Pacific' separately, as in preliminary exploration of the dataset per year, the number of women coded as either Fijian Indian or two separate ethnicities were largely inversely correlated. This issue has been addressed by the ethnicity data protocols. 11 In addition, even though NZ is known to have a growing population of South African Indians over the last decade, 24 none were identified as such in our dataset. We expect similar inconsistencies may have occurred among other ethnic groups. Such incorrect coding may happen with inappropriate data collection. For example, women might identify as 'Punjabi', but are reported solely as 'Indian', and some healthcare professionals may be unaware that 'Fijian Indian' is acknowledged as a unique ethnicity. Furthermore, births to women registered with a DHB were excluded from all analyses including BMI, smoking, parity or booking trimester, due to a large amount of missing data. Anecdotally the variable 'booking trimester' may not accurately represent the timing of registration with a healthcare provider and it has not been validated. However, the MOH does use this variable in annual reports on clinical indicators for quality control. 25 Additionally, ethnicities with poorer socio-economic status generally booked later in pregnancy, as expected. 26 Analyses on CS, AD and IOL were based on nulliparous women only, since we were not able to adjust for previous obstetric outcomes as an important risk factor. 27 Further limitations were specific to the IDI.
Missing BDM data on perinatal deaths, and a conservative linking method by Statistics NZ, restricted identification of clinical data in many perinatal death cases. 28 In addition, data quality from the various sources is variable, with general inconsistencies in metadata, and no administrative data are available for those who did not access government services or do not reside in NZ. 29 Confounding between ethnicity and ancestry (often corresponding to country of birth) is especially important when considering metabolic disorders such as GDM, where genetics or epigenetics may play a role. This study has shown that some risk factors can be culturally determined, as seen among the combined ethnic groups. In contrast, the high rate of GDM among Fijian Indian mothers (similar to Indian women) may indicate ancestral importance. Even though these intergenerational differences may explain some variance in pregnancy outcome, 2 no analyses were performed between women of first-and secondgeneration migrants in this study due to the relatively low number of SAsian women born in NZ. With increasing migration, ethnic groups will change over time, thus analyses should be repeated in future studies to continue correct grouping of individuals. As this study is specific to the NZ setting, we recommend other countries perform similar analyses within their unique population. While NZ-based health research is generally conducted by ethnicity, surrogate variables are often used internationally, such as country of birth, 6 nationality, 30 race, 7 or a combination. 31 We argue that ethnicity together with ancestry captures the influence on health outcomes best, as beside a genetic component, major modifiable risk factors may be culturally determined. 32 In conclusion, within the definition of 'South Asian', most subgroups can be combined for pregnancy research in NZ. However, as we identified some groups with differing socio-demographic background and risk profiles, our data emphasises the need to justify aggregation by ethnicity. The importance of accurate and detailed ethnicity data collection is highlighted.

DISCLAIMER
These results are not official statistics. They have been created for research purposes from the Integrated Data Infrastructure (IDI), which is carefully managed by Stats NZ. For more information about the IDI please visit https://www.stats.govt.nz/integ rated-data/.