Malnutrition trends in displaced Rohingya children from 2017 to 2024: a comprehensive 7

Discussion This retrospective cross-sectional analysis of deidentified EHR is the first to comprehensively characterise the nutritional status of 13 219 Rohingya refugee children aged 6–60 months from the 2017 influx to 2024. Although multiple stratified analyses are presented, these were prespecified to characterise population-level patterns across key demographic and temporal dimensions. Multiple linear regressions were conducted to reduce type 1 error (online supplemental table S40). Secondary analyses were considered exploratory and hypothesis-generating. Accordingly, we emphasise overall, internally consistent patterns across age, sex, season and year and the prespecified multivariable regressions (online supplemental table S40), treating subgroup-specific p values as descriptive rather than confirmatory. Importantly, because each calendar year represents a distinct cross-sectional sample of children attending the clinics, observed differences reflect population-level shifts rather than within-child longitudinal change. Accordingly, formal multiplicity correction was not applied, as the stratified analyses and the multiple regression models were prespecified and hypothesis-driven, focusing on coherent population-level trends. Demographic patterns and age distribution In Kutupalong, the mean age was slightly higher than the median, indicating a right-skewed distribution. A similar pattern was seen in Balukhali. Despite high birth rates linked to limited family planning and contraceptive hesitancy,23 age group proportions remained stable from 2017 to 2024, supporting valid year-to-year comparisons. Malnutrition prevalence The incidence of GAM (27.05% in 2024) and SAM (6.38%) exceeded WHO emergency thresholds of 15% and 2%–3%, respectively,24 and surpassed national estimates from Bangladesh and Myanmar. From 2017 to 2024, 23.80% of Rohingya children had GAM (figure 1)—well above recent national estimates of 8.44% wasting in Bangladesh and 6.80% in Myanmar.25 26 However, these national figure estimates predate the COVID-19 pandemic and may have worsened due to pandemic-related food supply disruptions.12 Consistent with prior findings,13 our study demonstrated that males had higher malnutrition prevalence and lower mean and median WFZ (overall GAM 28.11% vs 19.15%; mean WFZ −0.96 vs −0.89), a pattern that remained stable across all 7 years. Biological factors, including faster early growth, higher caloric needs and potentially higher susceptibility to infections, may partly explain this disparity. Behavioural and cultural practices, such as differential feeding or care-seeking behaviours, may further contribute to these differences. Programmatic factors could also play a role: during crises or routine food distributions, interventions may unintentionally prioritise younger infants or children perceived as more vulnerable, leaving male children less prioritised. Together, these biological, behavioural and programmatic factors likely interact to produce the observed sex differences in undernutrition indicators such as WFZ and MUAC, highlighting the need for targeted approaches in nutrition programmes.27 These proposed mechanisms are speculative, as our dataset did not include direct measures of feeding practices, morbidity or gender norms. Importantly, infectious disease data showed no sex-based differences in GI or RTI prevalence, suggesting that disparities in nutrition are unlikely explained by infectious disease burden. Age-stratified nutritional patterns Age stratification showed that children aged 6–24 months consistently had higher WFZ and MUAC z-scores across all years, contrasting with earlier reports in this population.7 13 For example, in 2022, GAM was about 23.3% among children aged 6–24 months compared with 25.1% among those aged 25–36 months, illustrating the elevated burden in older age groups. Possible explanations include extended breastfeeding and delayed weaning under nutritional stress.28 Wet nursing and prolonged breastfeeding observed in the camps, along with national data showing a median breastfeeding duration of 19 months, support this hypothesis.29 In contrast, children aged 24–60 months may face cumulative nutritional deficits due to inadequate diets and higher energy demands, which can impact growth metrics.30 These findings can also reflect gaps in existing nutrition programmes, which predominantly focus on children under 24 months through interventions such as complementary feeding support, growth monitoring and targeted supplementation. As a result, older children may receive less attention and fewer resources, leaving them at elevated risk for malnutrition.29 Notably, among children aged 25–36 months, the population-level difference in mean MUAC z-score between samples assessed in May and samples assessed in August was larger than in any other age group (figure 5, online supplemental table S22), suggesting heightened seasonal vulnerability in this stratum. This group may be especially at risk due to weaning transitions, greater infection exposure from increased mobility and limited inclusion in targeted nutrition programmes despite ongoing physiological needs.11 This vulnerability is underscored by infection data: while 25–36-month-olds had the lowest GI infection rates, they had the highest RTI prevalence among age groups (online supplemental table S36), a mixed picture suggesting seasonal nutritional decline is not solely driven by infectious disease burden. Temporal trends in nutritional status Across calendar years from 2017 to 2024, the population-level prevalence of GAM was higher in later years than in 2017, and mean WFZ was lower during the COVID-19 period (2020–2021) compared with earlier years (figure 2). The spike in GAM observed in 2021 coincided with pandemic-related service disruptions, reduced care-seeking and adjustments to treatment protocols, including reduced follow-up and modified feeding regimens. Although population-level average WFZ improved slightly after 2021, the prevalence of GAM among children assessed in 2024 remained 73% higher than among those assessed in 2017 (figure 2). During the COVID-19 period, malnutrition screening protocols in some settings shifted towards greater reliance on MUAC alone, which may have influenced case detection and comparability across time. Moreover, to this point, annual sample sizes varied substantially, ranging from 614 children in 2021 to 5148 in 2018. This variation likely reflects factors such as clinic service availability, COVID-19 disruptions and population movements within the camps. This variation in sample size also affects annual estimate precision: years with lower sample sizes (notably 2021) yielded the widest Wilson 95% CIs for annual GAM (online supplemental table S41) and therefore contribute proportionally greater uncertainty to year-to-year trend comparisons, although all annual interval widths nonetheless remained less than 10%, with 15% being the standard cut-off used to determine appropriate precision. Additionally, the number of first visits in earlier years of the study likely reflect the 2017 influx,31 with fewer new children to be screened from 2020 onward. Consistent with this pattern, the proportion of first-time visits declined sharply during the COVID-19 period and remained substantially lower than in the early years of the study. This sharper reduction in new screening numbers, compared with the more modest decline in overall clinic visits, likely reflects the limited arrival of new camp residents following the initial displacement period and a gradual shift toward births and internal movement within the camps as the primary sources of first-time screened children, with temporary reductions in clinic access and changes in care-seeking behaviour during the COVID-19 period likely contributing as well. Recognising these differences helps contextualise year-to-year trends in nutritional outcomes.11 Specifically, since the initial influx in August 2017, only 2.18% of the December 2024 catchment population arrived between 2018 and 2024, new children presenting in later years most plausibly represent camp-born infants ageing into the 6–60-month window rather than externally arriving migrants, while older children age out of eligibility. Consistent with this, the age-group composition of successive cross-sectional samples remained stable (online supplemental table S2). Year-to-year differences in nutritional indicators are therefore unlikely to be driven by compositional shifts from external migration, though residual effects of unmeasured intracamp movement within the wider Kutupalong-Balukhali extension settlement for which no consistent public data exist cannot be excluded (see online supplemental appendix for further detail). Importantly, there were no changes to anthropometric protocols or measurement personnel at either HAEFA clinic during this period. In fact, the proportion of children assessed using all three indicators of weight, height and MUAC increased markedly after 2020, reaching over 96% by 2024, supporting the consistency and reliability of the measurements despite pandemic-related challenges. Differential missingness across years was concentrated in MUAC rather than weight or height (online supplemental table S42), and MICE estimates under a missing-at-random framework were highly consistent with complete-case analyses (online supplemental table S39), arguing against meaningful bias in WFZ-based temporal trends (online supplemental tables S41, S42 and S43). The consistent deterioration observed across multiple anthropometric indicators, however, suggests that these trends reflect true nutritional vulnerability rather than measurement changes alone. Age distribution and refugee inflows stayed relatively stable, though 2024 saw a slight rise due to renewed conflict in Myanmar’s Rakhine State.32 High birth rates also have increased population pressure.1 Nutritional decline is likely multifactorial, driven by crisis protraction, donor fatigue and global competing priorities constraining food, healthcare and Water, Sanitation and Hygiene (WASH) services. Additionally, families may sell nutrition supplementation for increased economic flexibility within the household unit. Documented cuts in both caloric and nutrient-rich rations have reduced dietary adequacy. Overcrowding, poor sanitation and limited healthcare sustain high morbidity, reinforcing undernutrition. These conditions, compounded by psychosocial stress from displacement, may have further weakened child feeding practices.11 12 The mismatch between MUAC and WFZ nadirs (2020 and 2021, respectively) and their divergent long-term patterns reflects their differing sensitivities. MUAC responds more rapidly to acute nutritional changes in fat and muscle stores, while WFZ is shaped by both weight and height. Therefore, WFZ is slower to shift and signals prolonged undernutrition. The lower mean WFZ (figure 2) observed among children assessed in 2021 may reflect sustained population-level nutritional stress.33 Purchasing power, market volatility and food security constraints The widening gap between nominal e-voucher values and their inflation-adjusted equivalents paralleled declining WFZ scores (figure 3). As inflation eroded real voucher value, families’ food purchasing power dropped. From 2018 to 2024, the gap widened from −BDT53 to −BDT165, alongside a decline in mean WFZ, suggesting nominal increases in aid failed to preserve food access.14 Prices of staples like rice, lentils and oil rose due to inflation, crop losses and currency depreciation. Because camps rely heavily on local markets, households are highly exposed to these shocks. Even modest price hikes can significantly reduce purchasing capacity under fixed-value e-voucher systems, despite WFP efforts to buffer designated markets from external volatility.34 From 2018 to 2024, the growing gap between nominal and real voucher value highlighted the diminishing adequacy of food assistance (figure 3). Moreover, national data from Bangladesh show strong correlations between food inflation and rates of stunting and wasting.35 Global increases in commodity and fuel prices have further strained household budgets, while camp restrictions on employment and movement limit Rohingya refugees’ ability to supplement food access through labour or subsistence.36 Periods of declining inflation-adjusted e-voucher purchasing power coincided with lower WFZ, suggesting that broader economic constraints may be associated with child nutritional outcomes. Although these findings are ecological and cannot establish causality, they underscore the potential importance of supporting household purchasing power within nutritional programmes. Seasonality of nutritional status Our study demonstrated clear seasonal variation in population-level nutritional status, with mean WFZ lowest among children assessed during the monsoon (May-August) and children assessed during the post-monsoon periods (eg, mean WFZ −0.63 in January samples vs −1.22 in August samples; figure 4). This cross-sectional pattern likely reflects overlapping factors operating during these months: flooding and mobility restrictions disrupt food delivery and healthcare. These seasonal effects, which are well documented in WFP and UNHCR reports, reduce intake and impair nutrient absorption.2 The proportion of children classified as well-nourished was lower in samples assessed between January and August, while the proportions classified as MAM and SAM were highest among children assessed in July and August. Additionally, although clinic attendance was approximately 31% lower at the end of the monsoon season than at the start of the post-Aman season (online supplemental table S18), the differences in mean WFZ and MUAC across seasonal samples remained statistically significant, indicating that these cross-sectional patterns are not solely explained by variation in attendance, though unmeasured seasonal patterns in visit reason or illness-related care-seeking may still contribute. Surprisingly, rates of GI infection and RTI did not increase during the monsoon (online supplemental table S38). Urgent care visits also remained stable. This may reflect persistently high baseline transmission sustained by overcrowding, inadequate WASH infrastructure and poor ventilation—blunting typical seasonal disease patterns.37 Limited mobility may reduce transmission, while WASH interventions and seasonal preparedness may further moderate infection risk. Under-reporting due to poor weather and delayed care-seeking may also play a role. These factors likely explain the absence of monsoon-related infection spikes, despite deteriorating nutrition.38 Mean MUAC z-scores showed limited seasonal variation across cross-sectional samples (figure 5, online supplemental table S6), despite MUAC’s greater sensitivity to acute nutritional shifts.33 This may reflect chronically high baseline undernutrition in the population, obscuring short-term cross-sectional differences. Coping strategies such as rationing or prioritising food for younger children may also dampen month-to-month differences in mean MUAC at the population level.39 Among the 25–36-month group, however, mean MUAC was notably lower in samples assessed in August than in those assessed in May, paralleling WFZ patterns and highlighting the seasonal vulnerability of this age group (figure 5). Overall agreement between MUAC and WFZ was limited. Bland-Altman analysis showed increasing divergence at lower z-scores, suggesting MUAC identifies children with more pronounced deficits. Yet, WFZ captured more GAM cases overall, indicating greater sensitivity to early-stage malnutrition. Overall, 69.78% of children classified as GAM by WFZ were not identified by MUAC, and 67.44% classified by MUAC were not identified by WFZ. These levels of discordance are consistent with prior literature demonstrating substantial disagreement between MUAC and weight-based indicators of acute malnutrition. Multicountry analyses including Bangladesh have reported that only about one-quarter to one-third of children identified as acutely malnourished by either indicator are classified by both, corresponding to discordance rates of approximately 60%–70%.40 This poor agreement has important programmatic implications, as reliance on a single indicator may lead to systematic under-identification or misclassification of malnutrition in specific subgroups. These findings support the use of complementary anthropometric measures in screening and admission criteria to improve case detection and ensure equitable access to nutrition services.33 Lastly, recent population-level evidence from Gaza illustrates how acute malnutrition among children aged 6–59 months can fluctuate dramatically in response to humanitarian aid access. Horino et al conducted near-population-wide MUAC screening of over 200 000 uniquely identified children across multiple health centres and shelters, documenting rapid increases in wasting during periods of aid restriction and declines when aid resumed.41 While the Gaza context differs from Rohingya camps, this work highlights that community-level or population-level surveillance can reveal temporal dynamics and prevalence patterns that may not be fully captured by clinic-based assessments, particularly when care-seeking is selective or access is constrained. These findings reinforce the importance of interpreting our clinic-based results in the context of potential selection effects, programme coverage and seasonal or logistical disruptions. Limitations This study has several limitations, though steps were taken to mitigate them. Its retrospective design limits causal inference, but the large sample size and consistent EHR-based data support robust trend analyses. Selection bias may arise because the study is mainly clinic-based, as children attending clinics may differ systematically from those who do not. Annual sampling fractions of the combined Camp 1W and Camp 09 catchment population also varied across the study period (about 36% in 2018, 4%–6% during the pandemic 2020–2022 and 10%–13% in 2023–2024; online supplemental table S41), and lower-coverage years are more susceptible to selection effects, though Wilson 95% CIs with finite population correction remained less than 10% throughout (online supplemental table S41). These coverage estimates additionally rely on cross-sectional UNHCR population snapshots rather than person-time denominators and may modestly underestimate or overestimate true annual sampling fractions if camp populations shifted within a given year. Conversely, in addition to the fact that more than 80% of the first-time visit data were obtained from the non-illness-related visits as well as from the catchment population, several features of the dataset support adequate representativeness of the population screened at HAEFA health centres: the age-group composition of the sample remained stable across all study years (online supplemental table S2), which mirrors UNHCR data31; Wilson 95% CIs with finite population correction were narrow even in lower-coverage years (range 1.81%–8.17%; online supplemental table S41); and multiple imputation under a missing-at-random framework yielded estimates highly consistent with complete-case analyses (online supplemental table S39), arguing against substantial bias from informative missingness. Information bias could occur due to measurement error in anthropometry or reporting of age and demographic variables. Ecological inference limitations also apply to analyses linking population-level exposures, such as e-voucher purchasing power, with population-level nutritional outcomes, meaning individual-level causal conclusions cannot be drawn. Notably, based on programme monitoring rather than formal coded analysis, over 80% (10 924) of first-visit data were from routine, non-illness-related services such as immunisations, growth monitoring, community-based screening, or caregiver attendance for family planning. Clinic programme records and staff report that this proportion remained broadly stable across years and seasons, with approximately 80% of first visits consistently attributable to routine preventive or benefit-linked services rather than acute illness. This indicates that many children without acute non-nutrition-related medical issues were included. Clinic-based surveillance may still over- or underestimate malnutrition: prevalence could be inflated if attendees are disproportionately vulnerable or underestimated if children enrolled in therapeutic feeding programmes do not return for follow-up. Restricting analyses to first-visit data provides a snapshot of overall nutritional status but does not allow modelling of within-child trajectories. Additionally, the dataset does not provide information on the reasons for clinic visits, which limits interpretation of how care-seeking patterns may have influenced observed seasonal nutrition variation. Multiple imputation under a MAR assumption addressed missing data, though unmeasured factors such as illness severity, breastfeeding or household characteristics may still influence outcomes. Economic data were limited by complex aid dynamics; in addition, because inflation-adjusted purchasing power and calendar year were not modelled jointly, temporal trends and economic effects on nutritional indicators cannot be empirically disentangled. Multiple stratified analyses across age, sex, season and year increase potential type I error; however, primary outcomes were prespecified, and conclusions are based on consistent trends across multiple approaches. Findings may not fully generalise beyond Kutupalong and Balukhali, as camps differ, though not significantly, in health service coverage and infrastructure facilities (online supplemental figures S11 and S12). Clinic records indicate that no children were enrolled in external malnutrition programmes during the study period, and children diagnosed with MAM or SAM at first-visit were referred to WFP facilities and did not return for follow-up. Future trend analysis studies stratifying routine versus illness-related visits will be of interest and may provide insight into potential case-mix effects, which is beyond the scope of the current study. Finally, strict WFZ and MUAC z-score cut-offs (>−4.0) may have slightly underestimated severe malnutrition.