Nutrition in transition: historical cohort analysis summarising trends in under- and over-nutrition among pregnant women in a marginalised population along the Thailand–Myanmar border from 1986 to 2016

Abstract The objective of the present study is to summarise trends in under- and over-nutrition in pregnant women on the Thailand–Myanmar border. Refugees contributed data from 1986 to 2016 and migrants from 1999 to 2016 for weight at first antenatal consultation. BMI and gestational weight gain (GWG) data were available during 2004–2016 when height was routinely measured. Risk factors for low and high BMI were analysed for <18·5 kg/m2 or ≥23 kg/m2, respectively. A total of 48 062 pregnancies over 30 years were available for weight analysis and 14 646 pregnancies over 13 years (2004–2016) had BMI measured in first trimester (<14 weeks’ gestational age). Mean weight at first antenatal consultation in any trimester increased over the 30-year period by 2·0 to 5·2 kg for all women. First trimester BMI has been increasing on average by 0·5 kg/m2 for refugees and 0·6 kg/m2 for migrants, every 5 years. The proportion of women with low BMI in the first trimester decreased from 16·7 to 12·7 % for refugees and 23·1 to 20·2 % for migrants, whereas high BMI increased markedly from 16·9 to 33·2 % for refugees and 12·3 to 28·4 % for migrants. Multivariate analysis demonstrated low BMI as positively associated with being Burman, Muslim, primigravid, having malaria during pregnancy and smoking, and negatively associated with refugee as opposed to migrant status. High BMI was positively associated with being Muslim and literate, and negatively associated with age, primigravida, malaria, anaemia and smoking. Mean GWG was 10·0 (sd 3·4), 9·5 (sd 3·6) and 8·3 (sd 4·3) kg, for low, normal and high WHO BMI categories for Asians, respectively.

These problems are particularly felt in Asian populations that are at increased risk of diet-related non-communicable disease (CVD, stroke, hypertension and diabetes) for similar measures of nutritional status compared with Western populations (7)(8)(9) . Globally, and increasingly within Asia, trends in BMI and gestational weight gain (GWG) are important determinants of disease related to nutrition in pregnancy: overweight and excessive GWG can lead to gestational hypertension, gestational diabetes and macrosomia complicating delivery, while being underweight in pregnancy is associated with pre-eclampsia, intra-uterine growth restriction and low infant birth weight (10)(11)(12)(13)(14)(15)(16) . Furthermore, we continue to learn how the nutritional status of the mother and its relation to the intra-uterine environment leads to intergenerational, long-term effects on non-communicable disease incidence in adulthood, perpetuating a cycle of risk (1,17) .
The nutrition transition and epidemiological shift from infectious diseases such as malaria to a larger burden of diet-related non-communicable diseasea new development over the past 20 yearsis well underway in Asia, including places such as Thailand (4,8,(18)(19)(20) . A 2011 study showed that 40·0 % of reproductive age women are estimated to be overweight and 11·1 % obese in Thailand, with women in neighbouring Myanmar approaching 47·0 % overweight and 11·3 % obeselikely underestimates as the BMI categories used were not adjusted for Asian populations (5) .
Although the nutrition transition is well underway in LMIC settings, there remains a paucity of data documenting trends in nutritional status of marginalised populations in these contexts, particularly for pregnant refugee and migrant women. Most of the published data on these populations come from refugee and migrant populations resettled in high-income countries (21,22) . With a growing global need to address the health of refugees, displaced persons and migrant communities in both LMIC and high-income countries, we examine changes in nutritional status among pregnant women from the Thailand-Myanmar border: in the setting of a protracted refugee situation and among a large and growing migrant worker population from Myanmar. The objective of this manuscript is to describe trends in over-and under-nutrition using weight  and BMI at first antenatal presentation (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016) and GWG (2004GWG ( -2016 in these marginalised populations.

Methods
A brief history of the Shoklo Malaria Research Unit and populations along the Thailand-Myanmar border Refugee populations. Mass movement of persons from Myanmar began in 1984 as 'persons fleeing fighting' amassing in informal shelters along the border region were permitted entry into neighbouring Thailand through formalised 'temporary shelters' (23) . The Border Consortium, tasked with providing basic services to refugees along the border, documented a peak in this population at 165 901 in 2006, housed in ten temporary shelters within Thailand (24) . The Shoklo Malaria Research Unit (SMRU) began work in the Shoklo refugee camp in 1986, which was closed in 1998 and amalgamated to Mae La refugee camp where it is currently the most populous of the remaining nine refugee camps. As of December 2016, the Border Consortium aided 37 518 persons living in Mae La (25) . More than 90 % of the refugee pregnant population in Mae La received antenatal care (ANC) services at the SMRU clinic and more than 75 % of women delivered at the SMRU birth centre (26)(27)(28) .
Karen are the predominant ethnic groups (24,28) , comprised primarily of Sgaw and Pwo Karen, and religious affiliations are predominantly Buddhist or Christian, but with a significant proportion of Muslims, who also generally self-report their ethnicity as 'Muslim'. Literacy rates among the pregnant women in Mae La camp have remained stable around 50 % (29) . Refugees from Myanmar have historically had poor access to maternal health services; however, once in Thailand refugees have access to food assistance and medical services within refugee camps but are barred from work (30)(31)(32) .
Migrant populations. Informal networks of low-skilled, migrant labour developed along the Thailand-Myanmar border in the context of persistent economic hardship within Myanmar coupled with lax labour law enforcement in neighbouring Thailand (23) . Migrants have limited access to health services within Thailand (23,(31)(32)(33) . In the face of a growing migrant population with a significant burden of malaria, SMRU began ANC for migrants in Maw Ker Thai village in 1998, south of SMRU headquarters in Mae Sot, and subsequently in the north in Wang Pha village in 2004 (Fig. 1).
As of December 2009, the Thailand Ministry of Interior counted nearly 1·1 million registered migrant workers from Myanmar, of which 45 % were women. SMRU estimates that its clinics serve a catchment of approximately 200 000 migrants that has remained relatively stable since the late 1990s (34) . The predominant ethnic groups among migrants are Burman and Karen (Sgaw and Pwo). They are predominantly Buddhist. Approximately 45 % of migrant women presenting for ANC are literate (29) .
Shoklo Malaria Research Unit antenatal care and labour and delivery services. Home birth with traditional birth attendants remained widely practised in the refugee camps when SMRU began its ANC and obstetric programme, with SMRU promoting facility-based deliveries with trained midwives beginning in 1994 (35) . Ultrasound was introduced in 2001 using Dynamic Imaging, Fukuda Denshi UF 4100 (since 2002) and Toshiba Powervision 7000 machine (Toshiba, Tokyo, Japan since 2006), all with a 3·75-MHz convex probe, with all women routinely offered two scans. The booking visit scan (between 8 and 14 weeks' gestation) calculates estimated gestational age (EGA) by crown-rump length measurement using the British ultrasound guidelines and with regular quality control in place (36,37) . Labour and delivery wards were established in Wang Pha in 2008 and Maw Ker Thai in 2010. SMRU's entire operation has grown to include locally trained medical assistants, nurses, midwives, sonographers, laboratory technicians, homevisitors and a small number of expatriate doctors, about half of whom are from Myanmar (35) .

Study design
Anonymous digital clinical records were available from a historical cohort of 72 622 pregnant women presenting to SMRU ANC between January 1986 and December 2016. A series of analyses were performed to assess: weight at first ANC, BMI in the first trimester (<14 weeks' gestation) and GWG. We included women with appropriate EGA and delivery of live, term (37-41 completed weeks), congenitally normal, singleton newborns. All analyses excluded women with spontaneous abortions, lost to follow-up before pregnancy outcome was known and inaccurate EGA or weight measurements. BMI analysis included only those women with weight and height measurements at first ANC visit. GWG analysis included women with last weight measured between 34 and 41 completed weeks and Karen (Sgaw and Pwo Karen) and Burman ethnicity.
Variables available for analysis Anthropometric variables. Trained midwives measured weight at first ANC consultation and at each follow-up visit by mechanical scale with 0·5 kg precision on Salter medical grade scales. Maternal weight at delivery was defined as the last weight taken between 34 + 0 and 41 + 6 weeks of gestation. Standing mechanical scales were available at the main clinic, but weight scales purchased locally in 2002 and 2003 were excluded from this analysis as they gave significantly higher weights (2-3 kg) compared with other years included in this analysis.
Height measurement at first ANC was introduced in 2004 using a portable, mechanical stadiometer with 0·5 cm precision, with newer digital, stationary stadiometers allowing for precision up to 1 mm since 2010, using the SECA brand. Both weight and height measurements were taken according to standard procedures (38) .
BMI was calculated when both weight and height measurements were available, beginning in 2004. BMI for this analysis was categorised as defined by the WHO expert consultation on Asian populations: low (BMI <18·5 kg/m 2 ), normal (BMI 18·5-22·9 kg/m 2 ) and high (BMI ≥ 23 kg/m 2 ) (8) . Height and weight measurements from the first ANC visit for the first trimester were used to calculate BMI and categorise each woman. It has been shown that first trimester weight as a proxy for pre-pregnancy weight is accurate in up to 95 % of pregnant women (39) .
GWG was summarised in this cohort and compared with the Institute of Medicine recommendations for pre-pregnancy BMI and GWG (40) . In addition, BMI and GWG in this cohort were also compared with a cohort of healthy pregnant Asian women in Vietnam using Asian WHO criteria, as this population was considered similar to the population analysed here (15) .
Estimation of gestational age. Different techniques were used to calculate EGA over the course of this historical cohort. Fundal height was used until 1992, followed by the Dubowitz method of neonatal assessment. Beginning in 2001, ultrasound became the preferred method of calculating EGA, with Dubowitz used for those women without a reliable dating ultrasound (e.g. late ANC attenders) (28,37,41) . The variation in EGA between these methods in this population is minimal (42) .
Demographic, malaria and haematocrit variables. Available demographic variables collected at first ANC were: length of residency at current address, refugee or migrant status, age, gravidity and parity. In 1999, self-report of smoking status (cigarettes or cheroots) was added as a dichotomous variable. Beginning in 2010, women self-reported literacy as 'able to read' and 'able to write'. Clinical data included malaria infection (any species) and anaemia (haematocrit less than 30 %) at first ANC visit.

Statistical analysis
Descriptive statistics were performed according to data type (continuous or categorical) and distribution (normal or skewed). Univariate analysis determined associations between demographic variables and nutritional status indicators. Proportions were calculated as the number of refugee or migrant women with low, normal or high BMI that year, over the total number of refugee or migrant women with a BMI measurement in the first trimester that year. Patterns over time (unit = 1 year) were assessed using a non-parametric test for trend for ordered groups (an extension of the Wilcoxon rank-sum test) or using a 1-df test for trend. Average changes in weight or BMI per 5-year intervals were estimated using linear regression. Comparisons were made using the normal BMI group as the reference group against the low or high BMI group. To identify significant predictors of low or high BMI, factors significant on univariate analysis (P < 0·05) were included in multivariate logistic regression models: age (<20 years), length of residence (months), ethnicity (Karen, Muslim, Burman or others), refugee or migrant status, parity (gravidity 1 and parity 0 v. all others), malaria infection at any time during pregnancy, anaemia (yes/no), smoking status (yes/no) and literacy (can read or not). Associations were quantified as adjusted OR (AOR) with 95 % CI. Data were analysed using STATA, version 14.2 (StataCorp).

Ethical standards
The present study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects/patients were approved by the Oxford Tropical Research Ethics Committee (OXTREC 28-09, amended 19 April 2012). Written informed consent was obtained from all subjects. This research also received ethical approval from the Tak Province Community Advisory Board, Thailand (reference: TCAB-9/2/2015).

Results
A total of 52 222 pregnant women had known gestational age and delivered a singleton, term, live newborn. There were 4160 (8·0 %) women excluded: 3341 with inaccurate weight measurements from 2002 to 2003 and 819 for whom first trimester weight was not measured. Of the 48 062 women included in the analysis, 32 507 (67·6 %) were refugees and 15 555 (32·4 %) migrants (Fig. 2).

Gestational weight gain
Of the 14 646 women with BMI measured in the first trimester, 7098 were included in the GWG analysis. Mean GWG was 10·0 (SD 3·4, range −1·0 to 24·0), 9·5 (SD 3·6, range −6·0 to 24·1) and 8·3 (SD 4·3, range −4·8 to 25·0) kg for low, normal and high BMI, respectively, with some women losing weight over the course of their pregnancy. GWG was compared with international recommendations from Institute of Medicine for appropriate GWG ranges determined by pre-pregnancy BMI (40) . Women with low or normal BMI in this cohort fell short of Institute of Medicine recommendations for GWG (Table 3). In addition, GWG in this cohort was compared with population means from a Vietnamese cohort of healthy pregnant women, which were analysed based on Asian BMI categories (Table 3). GWG for women in the present study was lower for each Asian BMI category than the Vietnamese cohort (15) .

Discussion
The present study offers a unique perspective: it documents trends in nutritional status of marginalised populations of pregnant women in a protracted refugee situation and among migrants in a socially, politically and economically turbulent region in Southeast Asia. Taken from over 40 000 women and with data spanning 30 years in refugees and 19 years in migrants along the Thailand-Myanmar border, the present study adds to the literature due to its size and scope. It adds to the dearth of literature highlighted by a study of thirty-seven LMIC, by summarising important trends in nutritional status of pregnant women within a marginalised population in an LMIC context (43) , against the backdrop of a nutrition and epidemiological transition. The trends toward overweight in these marginalised communities of refugees and migrants along the Thailand-Myanmar border reflect global trends (1) and present significant challenges. In 2004, 16·6 % of refugee and migrant women had a high BMI (≥23 kg/m 2 ). By 2016, over one-fourth (28·5 %) had BMI ≥23 kg/m 2 and 6 % of the women included in the present study had BMI ≥27 kg/m 2 , putting them at higher risk of pregnancy and delivery complications and poor perinatal outcomes as demonstrated elsewhere in Asia (11,13,15,44,45) . We observed this increase in overweight and obesity among both refugees and migrants, with an earlier and more pronounced trend among refugees. Here, we briefly outline a few considerations for these trends, including underlying biology, access to foods of varying nutritional quality, and the social, economic and political context along the border within which we observe these nutritional trends. The 'Developmental Origins of Adult Disease' may help explain this trend (46,47) . This hypothesis posits that under-nutrition in early life leads to increased risk of diet-related chronic disease later in life. As we cover a generation by study end, we may be witnessing this 'programming' effect among pregnant women with high BMI linked to malnutrition in early life. However, these trends would not be possible without being enhanced by environmental and behavioural factors (48,49) . The 'nutrition transition' documented in Thailand and in Asian populations elsewhere provides evidence of overweight and obesity and their association with increasingly sedentary lifestyles and diets high in animal and saturated fats, sweetened beverages and industrially and locally processed foods (4,5,6,19) . A recent study in this population links sweetened beverage consumption within the prior 24 h with high BMI in pregnancy (50) , with more controlled studies needed to understand the effects of other processed foods on maternal nutrition and pregnancy outcomes in this population. With the nutrition transition well underway in Thailand and its effects more pronounced among the marginalised, the introduction of and easier access to unhealthy, processed foods is compounded by limited access to health information through media; language barriers; low literacy and health literacy among pregnant women; and a lack of awareness among health workers and the communities as to the ill effects of poor nutrition (50,(51)(52)(53) .
Trends and differences in nutritional status in refugees and migrantsnamely in relation to risk factors for low and high BMIcan also be considered in the context of fluctuating political and civil stability along the border. Periods of stability are marked by more consistent access to health care services, education and greater food security, more clearly documented for refugees compared with migrants (29,32,54,55) . As far as the authors are aware, this is the largest cohort demonstrating a significant association with low BMI and malaria among pregnant women (56) and, therefore, the history of malaria control efforts along the border over the present study period bears mentioning. Between 2004 and 2008, the higher incidence of malaria among migrants and its association with the nutritional status of pregnant women occurred over a period marked by a fraught ceasefire process in Eastern Myanmar and continued political  (40) . † WHO BMI categories for Asian populations (8) and consistent with Ota et al. (15) . and civil instability leading to greater disruption of access to health services. Active malaria programmes could be established more easily in refugee camps, leading to greater malaria reduction and its concomitant reduction in risk of low BMI among refugees, while less effective programming allowed malaria to persist in the more mobile migrant communities over the same time period (57) . Occupational and behavioural risk factors and inadequate housing conditions common among seasonal migrants in the border region likely exacerbated these trends (31,32,58) . Although recognising that this is not a complete explanation for trends in nutrition among these marginalised groups, the authors view fluctuating political and civil stability as contributing to the observed trends. Therefore, alongside the relative political and civil stability seen in the refugee camps beginning in 1998, we observe an increase in the proportion of refugee women with high BMI. Earlier studies on nutritional status in this refugee population implicate limited dietary diversity and poor food security as risks for low BMI (59)(60)(61) . However, with greater food security through provided rations, studies have demonstrated trends in improved maternal micronutrient status among refugees and improved birth outcomes (60,62) . Stuetz et al. demonstrated that ownership of livestock and gardens in addition to food rations was significantly associated with improved refugee maternal micronutrient status (62) . These findings reflect greater stability in living and livelihood in the refugee communities. In addition, restrictions on field work outside the camps and an accompanying decrease in physical activity lend risk for sedentary lifestyles associated with over-nutrition in this population. The relative risks for poor nutrition attributable to long-standing armed conflict, occupational and behavioural risk factors, unstable residency, food insecurity and limited health care access are far from clear. However, the suggestion that nutritional trends follow fluctuations in stability in this region over time may shed light on similar dynamics in areas of conflict elsewhere in developing contexts.
However, even as malaria, anaemia and smoking during pregnancy as risk factors for low BMI have decreased over time, the overall rates of low BMI still persist. Therefore, the present study adds to the literature on the double burden of malnutrition in LMIC settings. Although the present study did not find an association between teenage pregnancy and low BMI (likely due to the small sample size), we note that teen pregnancy hovers consistently around 20 % over the study period and has been demonstrated to be associated with low BMI and under-nutrition in a larger cohort in this same population (63) . Therefore, as a risk factor for low BMI, programming around maternal nutrition in communities along the Thailand-Myanmar border may do well to pay specific attention to teen pregnancy for the reduction in the burden of maternal underweight.
The present study is not without limitations. Given that this analysis takes place over a long period of time, the variables analysed have been assessed through different methods over the course of SMRU's work in the region, reflecting the most accurate methods of anthropometric data collection given logistical constraints and changes in appropriate assessments over time. For example, missing data from 2002 to 2003 are due to problems with weight scales that were corrected in 2004. Despite these challenges, the large sample size allows for a robust characterisation of changes in nutritional status among these refugee and migrant populations. The authors note that the present study may have limited generalisability as this is specific to Myanmar refugees and migrants, but hope that the basic indicators of weight and BMI presented here will prove useful for other crisis situations globally. In addition, these findings may help inform the care of pregnant women in high-income countries receiving refugees and migrants from LMIC by providing a plausible context for understanding nutritional status of refugee and migrant women and targeting care to ameliorate the effects of poor nutrition on pregnancy outcomes.
As the double burden of underweight and over-nutrition has been linked to complications in pregnancy and adverse birth outcomes in Asian populations, future analyses will assess optimal BMI and GWG in relation to pregnancy outcomes among refugee and migrant populations along the Thailand-Myanmar border.