Optimising recovery from childhood moderate acute malnutrition: a randomized controlled trial

Optimising recovery from childhood moderate acute malnutrition: a randomized controlled trial | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Optimising recovery from childhood moderate acute malnutrition: a randomized controlled trial Justin O'Sullivan, Theo Portlock, Catriiona Miller, Daniel Ho, and 12 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8585503/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Optimising nutritional strategies to improve growth recovery from moderate acute malnutrition (MAM) remains a global priority. We conducted a community-based, open-label, randomised controlled trial in Bangladesh to evaluate whether an enhanced ready-to-use supplementary food (ERUSF), enriched with prebiotics, Long Chain PolyUnsaturated Fatty Acids (LC-PUFA) and carotenoids, improved recovery outcomes compared with a locally produced RUSF (L-RUSF). Children aged 12 ± 1 months with MAM were treated with RUSF or ERUSF for up to three months or until recovery, followed by nine months of maintenance supplementation with standard or enhanced small-quantity lipid-based nutrient supplements (SQ-LNS or ESQ-LNS) in addition to their daily normal complementary diet. Using WHO criteria (WLZ/WHZ > −2), 69% of participants achieved recovery during the rapid catch-up phase, with no significant difference between intervention arms. When a more stringent, pre-specified biologically informed recovery threshold (WLZ/WHZ > −1) was applied to examine recovery trajectories beyond WHO-defined resolution of wasting, approximately 44% of children in both arms achieved this target. Among those who did recover to this stricter threshold, ERUSF was associated with a significantly shorter time to attainment, indicating more rapid catch-up growth rather than an increased probability of recovery. Relapse rates during the subsequent maintenance phase were low and did not differ between SQ-LNS formulations. To explore biological correlates of recovery heterogeneity, we integrated anthropometric, microbiome, plasma metabolomic, and genetic data using mixed models and exploratory machine-learning approaches. Baseline anthropometric severity, early weight gain, and polygenic risk profiles were associated with recovery status, whereas sociodemographic variables were largely uninformative. Post-hoc counterfactual digital twin modelling suggested that variation in baseline plasma amino acid and vitamin profiles may partially explain non-recovery under locally produced nutritional protocols. These simulations identified model-consistent feature shifts that, if achievable, were predicted to increase recovery to the WLZ/WHZ > −1 threshold; however, these findings are hypothesis-generating and not evidence of causal or clinically actionable effects. Together, these results demonstrate that enhanced RUSF accelerates recovery among children who respond but does not increase overall recovery rates under WHO criteria. Persistent heterogeneity in recovery trajectories appears to reflect underlying biological and genetic factors. These are not fully addressed by current supplementation strategies, highlighting the need for future trials explicitly designed to test biologically stratified nutritional interventions. Health sciences/Medical research/Clinical trial design/Clinical trials Health sciences/Health care/Nutrition Moderate acute malnutrition Digital twins Multi-omics Machine learning Randomized controlled trial Amino acid metabolism fecal microbiome Child growth Nutritional supplementation Bangladesh Figures Figure 1 Figure 2 Figure 3 Figure 4 Main Malnutrition is a quintessential early life adversity and constitutes an ongoing global crisis (UN SDG2: Zero Hunger) 1 . In 2025, of the estimated 47 million children younger than five with acute malnutrition, ~70% were affected by Moderate Acute Malnutrition (MAM), the most common category and most treatable form of undernutrition 2 . Repeated episodes of acute malnutrition, both moderate and severe, in children can lead to stunting over time with irreversible effects that include intergenerational growth failure 3 . MAM in childhood delays the development of multiple systems in the body, including the immune, gastrointestinal, fecal microbial, and central nervous systems, but every system is impaired 4 . Collectively, impairments of these systems increase childhood morbidity and mortality. The treatment goal for MAM is the restoration of normal weight for height (WLZ/WHZ > -2 Z score or mid-upper arm circumference, MUAC > 125 mm) 5 . Among the small proportion of children with MAM who receive supplementary feeds to achieve target weight for height within 3 months of initiation, many frequently fail to achieve anthropometric success and do not re-establish developmental trajectories 6 . Ready to use therapeutic food (RUTF) formulations were designed to treat children with severe acute malnutrition. RUTFs were recently adapted for use in the management of moderate wasting 7,8 and these products are referred to as ready-to-use supplementary foods (RUSFs). The nutritional composition of home-administered RUSF is designed to provide macro-micronutrients in an energy-dense formulation to support rapid catch-up growth. Water activity in RUSF is low compared to any other specialized food used in MAM treatment, so it can prevent bacterial growth and can be stored safely at home without refrigeration 9 . In supplementary feeding programs, children typically receive specially formulated foods in the form of fortified blended flour (FBFs) or lipid-based nutrient supplements (LNSs) dosed at approximately 75 kcal of food per kg of body weight per day or 500 kcal of food per day, in addition to the normal home diet. FBFs need to be cooked with water before consumption. By contrast, LNSs are ready-to-use foods consisting of lipid-based pastes (typically including ground peanuts) with added micronutrients that do not require refrigeration or preparation. Recovery rates from MAM using RUSF can vary widely depending on the study's location, the specific RUSF formula, and other contributing factors (range 20 – 80% 10,11 ). Successful management of MAM rests on many factors where feeding is fundamental; however, other social and economic factors, family dynamics, and underlying medical conditions play variable roles. Further development and testing of optimised feeds is one key research focus that will improve treatment outcomes. Insight into potential feed modifications that can enhance recovery rates is achievable from the results of advanced analytics. For example, digital twin technology is a cost-effective approach for simulating and analysing the behaviour of complex systems under varying conditions 12 . Digital twins have been successfully applied in medicine to improve the prediction, prevention, and treatment of diseases 13–15 . Importantly, digital twins allow for counterfactual analysis, which is not possible with conventional neural networks that are limited to predictive modelling. For example, Sizemore et al. (2024) developed a microbiome digital twin that not only predicted infant head circumference growth but also determined patient-specific clinical interventions 13 . In this study, we examine heterogeneity in recovery from moderate acute malnutrition and the limitations of current supplementation strategies, alongside exploratory modelling to generate hypotheses for biologically informed stratification of recovery. We report a randomized controlled trial in one-year-old Bangladeshi children with MAM comparing a ready-to-use supplementary food (ERUSF) enhanced with bioactive molecules including specific carotenoids and prebiotics, plus fish oil, with a locally produced RUSF, plus corn oil. Children received supplementation for up to three months or until they reached a predetermined weight-for-length/height target (WLZ/WHZ > −1) 16 . This rapid catch-up phase was followed by a nine-month maintenance period with either enhanced or standard small-quantity lipid-based nutrient supplements (SQ-LNS). The primary objective was to determine whether enhanced supplementation accelerated or increased recovery using either the WHO definition of recovery (WLZ/WHZ > −2) or a predefined biologically informed threshold (WLZ/WHZ > −1; hereafter biologically informed recovery threshold) 16 . A secondary objective was to assess whether continued SQ-LNS supplementation reduced recurrence of acute malnutrition through two years of age. Finally, in post-hoc analyses, we used counterfactual digital twin models to explore whether baseline biological profiles could help explain recovery heterogeneity and identify model-consistent states associated with successful recovery. Results Study population characteristics Dhaka, the capital of Bangladesh, is the one of the most densely populated cities in the world and has one of the highest rates of childhood malnutrition globally 16 . The study was conducted in the Mirpur region, where children with MAM (n = 159) and well-nourished controls (n = 75) were recruited at 12 ± 1 months of age. Participants were enrolled as part of a community-based, open-label randomised clinical trial [NCT05629624] to determine the impact of enriching the refeed diet with carotenoids, polyunsaturated fats, and prebiotics during recovery from MAM ( Fig. 1a and Table S1 ) 17 . MAM was defined according to WHO guidelines as a weight-for-length/height z-score (WLZ/WHZ) between –2 and –3 standard deviations 1 . Eligibility criteria included no history of chronic medical conditions, no known congenital anomalies, and no antibiotic use within the past month. A wide range of biological, behavioural, environmental, and developmental datasets were collected, including anthropometry; fecal microbiome composition (species, alpha diversity, and functional pathways); plasma amino acids, lipids, and water-soluble vitamins; polygenic risk scores (PRS); results of sleep surveys; and parental and household characteristics (economics, education, family structure, media consumption, sanitation practices) ( Fig. S1) . Baseline characteristics were comparable between the two malnutrition arms (local RUSF and ERUSF), with no significant differences observed across most demographic, socioeconomic, or household variables (q>0.05; Table 1 ). However, Caesarean delivery was significantly more common among well-nourished children (Fisher’s exact test, p = 0.007), and fathers of malnourished children had fewer years of education (MWU, p = 0.019). Although total household income did not differ significantly, households with malnourished children reported lower monthly expenditures (MWU, p = 0.032) but higher maternal income (MWU, p = 0.016), supporting the hypothesis that children of mothers who go out to work (and leave their young children in the care of others) are at higher risk of morbidity including development of acute malnutrition. Families of malnourished children also reported longer residence in their current homes (MWU, p = 0.011), which might be a marker of family stability, or unresolved or intractable poverty. Other variables, including breastfeeding duration, family structure, and household amenities, did not differ significantly between arms. Table 1. Baseline demographic, socioeconomic, and household characteristics of study participants by study arm (i.e. nutritional group). Values are presented as mean (SD) for continuous variables and n (%) for categorical variables. Comparisons were made using appropriate statistical tests (MWU for numeric-categorical, and Fisher’s Exact Test for categorical-categorical comparisons) between Local RUSF (A) and ERUSF (B) groups, and between malnourished children (A+B) and well-nourished controls (C). Bolded and asterisked p-values indicate statistical significance (p < 0.05). Local RUSF (A) (n=79) ERUSF (B) (n=80) Well-nourished (C) (n=75) A vs B (pval) A+B vs C (pval) Demographics Sex, Female 37 (46.8%) 40 (50.0%) 28 (37.3%) 0.752 0.123 Delivery Mode, Caesarean 26 (32.9%) 29 (36.2%) 40 (53.3%) 0.739 *0.007 Parental Education and Economics Years of fathers’ education 5.34 (3.84) 4.58 (3.48) 6.45 (4.29) 0.252 *0.019 Years of mothers’ education 5.71 (2.89) 5.39 (3.78) 6.56 (4.35) 0.577 0.252 Total monthly income (taka) 19112.66 (7307.36) 19923.75 (9832.73) 25220.00 (20895.05) 0.817 0.266 Total monthly expenditure (taka) 14460.76 (6403.13) 15243.12 (6389.58) 18469.33 (10791.88) 0.548 *0.032 Fathers’ monthly income (taka) 17877.22 (7501.62) 18310.00 (10628.15) 24620.00 (20947.95) 0.813 0.067 Mothers’ monthly income (taka) 1235.44 (2748.94) 1613.75 (3433.11) 600.00 (2046.75) 0.795 *0.016 Family Structure Number of living children 2.01 (1.04) 1.71 (0.94) 1.88 (0.85) *0.036 0.528 Family type - Nuclear 47 (59.5%) 44 (55.0%) 49 (65.3%) 0.632 0.256 Household characteristics Members in household 5.06 (1.81) 4.84 (1.69) 4.76 (1.65) 0.402 0.303 Years lived in current household 5.14 (5.75) 5.64 (6.80) 3.83 (5.22) 0.771 *0.011 Rooms in current household 1.57 (0.83) 1.51 (0.83) 1.60 (0.85) 0.564 0.594 Cooking gas - Yes 59 (74.7%) 65 (81.2%) 61 (81.3%) 0.344 0.609 Working TV - Yes 48 (60.8%) 44 (55.0%) 49 (65.3%) 0.522 0.317 Other Characteristics Exclusive Breastfeeding (months) 5.30 (1.39) 5.30 (1.50) 5.12 (1.71) 0.775 0.336 Place of Birth - Home 30 (38.0%) 32 (40.0%) 22 (29.3%) 0.796 0.152 Recovery from malnutrition We measured anthropometric recovery at two points in the study: 1) following rapid catch-up growth; and 2) following sustained micronutrient and polyunsaturated fatty acids supplementation (Fig. 1a). The WHO treatment goal for MAM is the restoration of normal weight for height (WLZ/WHZ > -2 Z or mid-upper arm circumference, MUAC > 125 mm) 5 . According to WHO criteria, we achieved rapid catch-up recovery of RUSF: 55/79 (69.6%) and ERUSF: 55/80 (68.75%) (Fig. S2). Time to recovery was not significantly different between intervention groups when calculated from weekly anthropometric measurements (Wilcoxon rank-sum test, p = 0.774), despite a trend of shorter mean recovery time in the ERUSF group. Relapse to < -2 SD WLZ/WHZ during the maintenance period did not significantly differ between SQ-LNS (2/79, 2.53%) and ESQ-LNS (1/80, 1.25%) (Fisher’s, p = 1.00). To gain a better understanding of the differences in individual recovery, we predefined a biologically informed recovery threshold, not a clinical standard, of WLZ/WHZ > -1 SD within 3 months 17 . Approximately 40% of treated children achieved the goal weight for height ( Fig. 1b and 1c, Table 2 ). The number of children who achieved the goal weight for height (reaching WLZ/WHZ > -1 SD within 15 weeks) was not significantly different between children receiving the locally produced RUSF and those receiving the ERUSF (44.3% and 43.8%, respectively; Fig. 1b and Table 2 ). However, amongst those children who did reach the biologically informed recovery target, children receiving the ERUSF feed did so faster than those receiving the local RUSF (MWU, p = 0.043; 47.3 ± 21.8 days compared to 57.0 ± 20.4 days, respectively; Table 2 ). This equates to a modest difference of approximately ten days ( Fig. 1d ). However, when considering this outcome, it should be noted that time-to-biologically informed recovery threshold is vulnerable to measurement and expectation bias as the trial was open label. The biologically informed recovery threshold (WLZ/WHZ > −1) was used as a treatment stopping criterion, however, it was also analysed as a biologically informed recovery threshold to examine durability and heterogeneity of response beyond the WHO-defined wasting. Therefore, we used the enhanced biologically informed recovery threshold (WLZ/WHZ > −1) to define three patterns of anthropometric recovery: 1) ‘sustained recovery’, in which children achieved the biologically informed recovery target threshold during the rapid catch-up phase (within 15 weeks) and maintained it through two years of age; 2) ‘delayed recovery’, in which children did not reach the threshold during rapid catch-up growth but achieved it by two years of age; and 3) ‘unsustained recovery’, in which children initially achieved the threshold during rapid catch-up growth but subsequently fell below it by two years of age. There were no significant differences in the proportions of individuals who were present in each of the recovery subcategories following supplementation with the local or enhanced RUSF ( Fig. 1c and Table 2 ). Notably, the proportion of days without feeding failure (i.e. supplement not taken) was not significantly different between groups ( Table 2 ). Table 2: Biologically informed threshold recovery outcomes for children receiving local RUSF (A) and ERUSF (B). Data are presented as counts (percentages) or means (standard deviations). ‘Delayed recovery’ was achieved by two years of age; ‘sustained recovery’ was achieved by rapid catch-up growth and maintained from three months to two years of age; and ‘unsustained recovery’ occurred in individuals who had initial recovery following rapid catch-up growth, but recovery was not maintained at two years of age. Differences in proportions were assessed using Fisher’s exact test. Differences in continuous variables were evaluated using the Mann–Whitney U test. Significant p-values are indicated in bold. Local RUSF (A) (n=79) ERUSF (B) (n=80) p value No recovery 44 (55.7%) 45 (56.2%) 1.000 Recovered (WLZ/WHZ >-1 SD) 35 (44.3%) 35 (43.8%) 1.000 Delayed recovery 1 (1.3%) 2 (2.5%) 0.617 Sustained recovery 9 (11.4%) 8 (10.0%) 1.000 Unsustained recovery 25 (31.6%) 25 (31.2%) 1.000 Days to recovery 57.03 (20.41) 47.29 (21.77) **0.043 Proportion of trial days without feeding failure 0.95 (0.10) 0.93 (0.13) 0.924 The feeds influenced the rate of achievement of target weight for height but did not affect the eventual acquisition of this goal. We also investigated whether sociodemographic factors significantly contributed to reaching the biologically informed recovery threshold ( Table 3 ). There were no significant differences in demographic, socioeconomic, or household characteristics between those children with MAM who reached the threshold and those who did not ( Table 3 ). The proportions of female participants (51.7% vs. 44.3%) and caesarean deliveries (34.8% vs. 34.3%) were similar between the groups who did (recovered) and did not (unrecovered) reach the threshold, respectively. Parental education levels and household economic indicators, including income and expenditure, did not differ significantly between groups. Likewise, family structure and household size showed no clear association with recovery status. However, the duration of residence in the current household was longer amongst children who reached the biologically informed recovery threshold when compared to those who did not (6.9 ± 7.6 vs. 4.2 ± 4.7 years, p = 0.042). Collectively, these results support the conclusion that anthropometric recovery to the biologically informed recovery threshold was not strongly influenced by demographic or socioeconomic parameters, except for one marker of residential stability. Table 3: Demographic, socioeconomic, and household characteristics were not significantly different between children who reached the biologically informed recovery threshold ( recovered ) and those who did not (unrecovered) during the treatment. Values are presented as n (%) for categorical variables and mean (SD) for continuous variables. p-values were calculated using Fisher’s exact test for categorical variables and the Mann–Whitney U test for continuous variables. Significant p-values are indicated in bold with an asterisk. No recovery (n=89) Recovered (n=70) p value Demographics Sex - Female 46 (51.7%) 31 (44.3%) 0.425 Delivery Mode - Caesarean 31 (34.8%) 24 (34.3%) 1.000 Parental Education and Economics Years of fathers' education 5.17 (3.76) 4.69 (3.56) 0.533 Years of mothers' education 5.74 (3.42) 5.30 (3.29) 0.390 Total monthly income (taka) 19797.75 (9062.92) 19168.57 (8152.13) 0.625 Monthly total expenditure (taka) 15298.31 (6593.31) 14290.00 (6117.57) 0.366 Fathers' monthly income (taka) 18646.07 (9425.31) 17394.29 (8880.44) 0.393 Mothers’ monthly income (taka) 1151.69 (2653.60) 1774.29 (3594.23) 0.381 Family Structure Number of living children 1.76 (0.94) 1.99 (1.07) 0.205 Family type - Nuclear 53 (59.6%) 38 (54.3%) 0.523 Household characteristics Number of members in household 4.89 (1.82) 5.03 (1.66) 0.378 Number of years lived in current household 4.20 (4.70) 6.90 (7.63) *0.042 Number of rooms in current household 1.58 (0.86) 1.49 (0.78) 0.566 Cooking gas - Yes 71 (79.8%) 53 (75.7%) 0.568 Working TV - Yes 52 (58.4%) 40 (57.1%) 0.873 Household food availability - Sometimes deficit 45 (50.6%) 26 (37.1%) 0.109 Other Exclusive Breastfeeding (months) 5.20 (1.50) 5.43 (1.36) 0.188 PoB - Home 33 (37.1%) 29 (41.4%) 0.579 As the measured sociodemographic factors (except for number of years lived in current household) were not significantly associated with reaching the biologically informed recovery threshold, we next examined whether biological factors from the gut and plasma might be. We conducted a redundancy analysis (RDA) to investigate how recovery to the biologically informed threshold and other covariates were associated with the variance in each dataset at each time point (Table S9). At year one (i.e. at baseline), none of the variables significantly explained variance in the economic or faecal microbiome pathways profile datasets, although modest trends were observed for supplementation and place of birth in the genetics dataset (p = 0.059 and p = 0.025, respectively). In the analyses, after one year of follow-up on children receiving SQ-LNS, similar patterns emerged, with limited explanatory power across most covariates. However, the association of the supplementation with plasma amino acid concentrations at year two is an expected outcome of dietary supplementation for one year. We also observed that ethnicity was associated with lipid levels (p = 0.035) and sex with anthropometric outcomes (p = 0.001). However, recovery status did not significantly explain variation in any dataset at either timepoint (p > 0.05). Overall, the RDA analysis findings indicate that while demographic and treatment-related variables may influence specific biological or developmental domains, reaching the biologically informed recovery threshold was independent of these covariates within the constraints of the current sample (Fig. S3). To explore whether features from the multi-omic datasets (e.g. microbiome and plasma metabolites) were associated with recovery to the biologically informed threshold, we conducted a mixed modelling analysis ( Fig. 2 ). Across the cohort, anthropometric measurements at baseline were not statistically significantly associated with reaching the biologically informed threshold after correction for multiple testing (FDR). Differences in weight-for-length z-scores (WLZ/WHZ), weight, and length at baseline were small and non-significant (all q > 0.05) between children who did and did not reach the biologically informed threshold. Similarly, neither baseline head circumference (coef = 0.22, q = 1) nor nasion-inion length (coef = 0.44, q = 0.44) was significantly associated with rapid catch-up recovery following multiple test correction (q > 0.05). Female children tended to have smaller head circumference and overall anthropometric measures than males at baseline (coef = -1.07, q = 0.13), consistent with expected sex-related differences. Head sizes that were below 42 cm at one year of age (i.e. recruitment) are consistent with those individuals having suffered significant past, possibly in utero, nutritional stress 18 . Change in weight after one week of rapid catch-up and genetic predispositions were associated with reaching the biologically informed threshold ( Fig. 2, Table S2 ). By contrast, other factors, including differences in complementary food composition and comorbidities such as diarrhoea, fever, cough, or antibiotic use, were not significantly associated with recovery ( Table S2 ). Treatment-specific improvements were observed for sleep quality and microbial composition following enhanced nutritional supplementation. We hypothesized that the baseline factors we measured should change in children who reached the biologically informed threshold, becoming more like those observed in the well-nourished cohort. However, we observed that the WLZ/WHZ score alters with age in the well-nourished cohort ( Fig. 1b ). Therefore, to assess the impact of nine months of sustained supplementation, after the rapid catch-up growth in the first three months, out to 24 months of age on these features, we calculated centroid distances for each measure relative to the well-nourished, age-matched control group at each trial arm and timepoint ( Fig. 3a ). For each dataset, the shrinking or widening of the inter-centroid distance across the two timepoints was used to indicate improvement or deterioration of the feature relative to the well-nourished controls, respectively ( Fig. 3a ). This approach was taken because these features (e.g. microbiome species, microbiome functional pathways, sleep, head circumference) alter with biological age 18–21 . Supplementation with ESQ-LNS, providing micronutrients, minimal energy and inulin and fish oils, after rapid catch-up growth, for approximately nine months (one year follow up) resulted in microbiome and sleep patterns that were more like those observed in the well-nourished cohort ( Fig. 3 ). Notably, the shifts in gut species and sleep patterns following supplementation with ESQ-LNS diverged from what was seen with the traditional SQ-LNS ( Fig. 3b ). The shift in gut species included depletion of S. wadsworthensis and L. fermentum ( Table S3 ). Importantly, the difference in microbial species was not mirrored by a refeed-specific separation of microbial encoded gut pathways ( Fig. 3b, Table S3 ). Continued supplementation with standard or enhanced SQ-LNS for nine months following rapid catch-up growth showed a trend toward convergence upon the well-nourished weight ( Fig. 3b ), suggesting a reduction in the recurrence of malnutrition. To determine whether microbial functions converged toward those of well-nourished children, we compared baseline and post-supplementation pathway abundances, one year later, with those of age-matched well-nourished controls. Most pathways that were initially depleted in malnourished children, including amino acid, coenzyme, and vitamin biosynthesis (e.g., PWY-7242, PWY-5972, PWY-7340), increased following refeeding, indicating functional recovery toward the well-nourished state (Table S3). In contrast, pathways linked to host-derived sugar catabolism (e.g., GLUCUROCAT-PWY) decreased, consistent with reduced dysbiosis. However, the ESQ-LNS group exhibited attenuated longitudinal changes in these pathways relative to non-enhanced SQ-LNS, suggesting slower or less complete convergence toward healthy microbiome function 22 . Collectively these results indicate that enhanced supplementation alters the trajectory of microbiome maturation rather than accelerating convergence. Counterfactual digital twins suggest optimal baseline profiles for recovery Building on our findings of baseline factors that impact recovery, we next developed digital twins to identify the individualized adjustments that could improve anthropometric recovery (WLZ/WHZ) in individuals who did not reach the biologically informed threshold during the randomized controlled trial. A feature matrix composed of multi-omic datasets associated with growth was constructed ( Fig. 4a and Table S4 ). Due to the high dimensionality of microbiome data, principal component analysis (PCA) was carried out first, and the top ten principal components (PCs) were included in the feature matrix ( Table S5 ). A variational auto-encoder (VAE) was trained to learn a latent space representation of the feature space (z), capable of reproducing the original matrix with a reconstruction mean squared error (MSE) < 0.05 and an average Kullback–Leibler (KL) divergence of 4.2 nats (~6 bits) per latent dimension. These results suggest that the latent variables retain substantial information about each individual, making the VAE more appropriate for a digital twin focused on generating realistic counterfactuals for existing individuals, rather than sampling entirely new synthetic individuals from the prior distribution. Using a regression neural network head trained on this latent feature space to predict WLZ/WHZ at one-year post-baseline, a root mean squared error (RMSE) of 0.59 was achieved within the holdout test set ( Fig. 4b ). Given the units are z-scaled (WLZ/WHZ), this is equivalent to 0.59 SD. Gradient-based optimisation was applied, using the trained VAE and regression head to run counterfactual simulations. All non-modifiable features ( Table S4 ) were frozen, and the model prioritised the smallest changes required to bring each individual to -1 WLZ/WHZ (i.e. recovery). The features with the largest predicted changes for simulated recovery included four amino acids, two amino acid related compounds (i.e. ornithine, a non-proteinogenic amino acid; and taurine, which is a sulfonic acid), and four vitamin cofactors ( Fig. 4c ). Whilst the magnitude of changes varied between individuals, the average individual required increased levels of all the top ten features except for ornithine, serine, and alanine ( Fig. 4c ). Applying the average feature change to unrecovered individuals identifies model-consistent intervention targets, not validated nutritional prescriptions, that led to simulated recovery in 12/83 (14.5%) of individuals who did not recover during the trial. Hierarchical clustering separated the individuals into two groups ( Fig. 4d ), which differed by the changes required to their baseline features to reach the biologically informed threshold ( Table S6 ). Of the 12 simulated individuals who recovered from the average suggested feature change, nine were in cluster one and three were in cluster two. Applying the two cluster-specific feature changes (1.5 SD above average required change, Table S6 ) led to simulated recovery in 9/12 (75%) and 48/71 (67.6%) individuals in clusters one and two, respectively. Importantly, for the individuals where these changes were possible (i.e. the average reduction in concentration did not bring them to ≤ 0 µmol/L), the large majority fell within the physiologically acceptable range ( Table S6 ). Positioning of individuals within either of the two clusters was not explained by simple differences in metadata ( Table S7 ). However, analysis of the ratio of plasma concentrations for non-essential and essential/conditionally essential amino acids ( Table S6 ) identified that individuals within cluster 1 had a significantly increased mean ratio (0.46) when compared to individuals within both cluster 2 (0.41) and the well-nourished cohort (0.40) ( Fig. S4 ). We also observed that individuals within the two clusters had differing genetic profiles ( Fig. 4e, Table S8 ). Following false discovery rate (FDR) correction, the difference in average BMI and IBD UC PRSs between the two clusters was significant (both adjusted p = 0.05). Crucially, these clusters were formed based on the modifiable features, meaning the genetics were not directly considered. Discussion Our trial of nutrition supplementation in one-year-old children with MAM compared the rapid catch-up growth promoted by standard WHO-prescribed RUSF with that of an enhanced supplement containing: i. added carotenoids; ii. prebiotics (inulin); and 1 g fish oil. These additions were selected to improve anthropometric and neurological recovery. Additionally, SQ-LNS and an enhanced SQ-LNS were administered to children for an additional 9 months, bringing the total nutrition treatment period to 12 months. When using the WHO criteria, we observed ~69% recovery. This is consistent with observations from similar studies in Tanzania, Cameroon, Ethiopia and Sierra Leone which had 83.7%, 85%, 73%, and 62.1–64.5% (across different food types, including RUSF) recovery, respectively 23,24 . Similarly, 74% recovery was obtained using Microbiota-Directed Complementary Food (MDCF) in Bangladesh 25 . Critically, assessment against a predefined biologically informed recovery threshold (WLZ/WHZ > –1SD) resulted in recovery of 44.3% and 43.8% with either the standard or enhanced refeed, respectively. We contend that our results indicate that beyond initial recovery (> -2SD WHZ) there are different trajectories with some individuals unable to reach > -1SD WHZ even after 12 months of intervention. The trajectory of anthropometric recovery is likely affected by one of many determinants, spanning the feed composition and organoleptic properties, availability of feeds to the infant, childcare practices, concurrent morbidities such as diarrhoea, RTI, and chronic gut pathology. The inclusion of long-term supplementation and long-term enhanced supplementation with carotenoids, LC-PUFA, and prebiotics, together with minimal amounts of energy, was designed to evaluate whether the standard WHO-prescribed and/or the enhanced SQ-LNS prevented the recurrence of wasting once the child had reached the biologically informed recovery threshold. This appeared to be the case, as whilst the mean WLZ/WHZ score for recovered individuals fell below the -1 SD threshold, it mirrored the trajectory that was observed for the well-nourished children and rarely fell below the –2 SD WHZ threshold. Thus, prolonged supplementation with SQ-LNS and ESQ-LNS resulted in individuals following a typical growth pattern, grounded at the starting WLZ/WHZ score at which the supplementation was started. Stunting (< -2 SD below the median of WHO child growth standards) has been associated with reduced exploration of the environment, greater fatigue, and differences in reported sleep (characterised by shorter night sleep duration and increased night waking) in children from Zanzibar and Nepal, when compared to their peers 26 . The sleep impacts of stunting have been associated with vitamin B12 deficiency and iron deficiency anaemia. Lentils and rice, which form the main constituent of the locally produced RUSF 17 , are plants and do not contain vitamin B12 27 . Therefore, the observed improvement in sleep in children whose diets were supplemented with the enhanced refeed, which contains vitamin B12 ( Table S1, Fig. S5 ) 28 , was expected. Despite the strength of the randomised controlled trial design, this study is not without limitations. The study was powered for comparisons between well-nourished (WLZ/WHZ > -1 SD) and wasted children (-3 ≤ WLZ/WHZ < -2 z-score, and/or 11.5 ≤ MUAC < 12.5 cm) at one and three years, such that the proposed medium level difference between the two groups (i.e. two standard deviations) requires a minimum of 51 participants per group for power equal to 80%, using a level of significance equal to 5% (for a one-tailed test) 16 . It is possible that some children experience regression once supplementation ends. Detecting such washout effects would require longer-term follow-up beyond the study period. Although the randomized controlled trial design provides a robust framework for evaluating anthropometric outcomes, the digital twin and machine-learning analyses were exploratory and post-hoc and were not prospectively powered or pre-registered as primary analytic objectives. As with any high-dimensional modelling approach applied to relatively modest sample sizes, there is an inherent risk of overfitting, instability of learned representations, and sensitivity to modelling choices, including feature selection, dimensionality reduction, and imputation strategies. While we mitigated these risks through train–validation–test splits, regularisation, and restriction of counterfactual analyses to biologically plausible ranges, the resulting predictions should be regarded as hypothesis-generating rather than confirmatory. The counterfactual digital twin framework identifies model-consistent feature perturbations that are sufficient to change predicted outcomes within the learned latent space; it does not establish causality or guarantee that such feature modifications are biologically achievable, safe, or effective in real-world interventions. Incorporation of the clinical and biological results into digital twin counterfactual models identified plasma amino acid and vitamin levels as one potential modifiable factor. However, plasma amino acid homeostasis is a networked equilibrium between diet, tissue metabolism (catabolism and anabolism), excretion, and hormonal regulation that reflects dietary intake, tissue demand and physiological stress 28 . The digital twin models identified two clusters of individuals, where specific simulated alterations to baseline plasma amino acid and vitamin levels in unresponsive individuals were predicted to improve recovery to the biologically informed recovery threshold from 44% (achieved with the refeed methods) to 80% (WHL/WHZ > -1 SD). While the digital twin findings should be considered hypothesis-generating and unstable under resampling, it is notable that the simulated alterations were within physiologically relevant levels (e.g. Sick Kids and Mayo Clinic reference ranges (Table S6)) for age- but not context-matched children. The population ranges of plasma amino acids reflect the interaction between environmentally determined features and the individual’s genetic predisposition (e.g. expression of transporters, metabolic enzymes) 28 . Severe stress and starvation, often alongside co-morbid infections and other syndromes, disrupt homeostasis. This leads to the release of gluconeogenic amino acids, increased urea production, and hormonal changes. Therefore, any form of malnutrition treatment that targets the modification of the plasma amino acid baseline concentrations must simultaneously address any comorbidities that might be affecting the homeostatic state. Critically, although our digital twin model was agnostic to the underlying biochemistry of nutrition, the simulated modifications it suggested outline a clear, biologically sound signature for recovery. The modifications show a shift away from the catabolic state of starvation by supplying a) the B-vitamins required for bioenergetic metabolism and b) the amino acids needed to repair tissues. Simulated increases of pantothenic acid (vitamin B5), nicotinamide (vitamin B3), and pyridoxine (vitamin B6; evidenced through the increase in B6 metabolite 4-pyridoxic acid) show a prioritisation of the model to replenish the cofactors essential for bioenergetic metabolism 29–31 . The model simulated increases of isoleucine and proline, amino acids that are substrates for muscle protein and collagen synthesis, respectively, with proline also important for gut homeostasis 32–36 . Concurrently, the model identified that reaching the biologically informed recovery threshold is associated with the reduction of amino acids that mark a catabolic state. High levels of alanine can be a sign of active muscle wasting (through the glucose-alanine cycle), while high levels of ornithine suggest increased stress on the urea cycle from tissue breakdown 37,38 . Importantly, the model is not necessarily suggesting a direct reduction of these amino acids, but rather that their decline is a biomarker of successful recovery. This is consistent with the fact that plasma amino acid and vitamin profiles, in particular, reflect integrated physiological states influenced by diet, infection, inflammation, hormonal regulation, and tissue metabolism, and may act as markers rather than drivers of recovery. Furthermore, although non-modifiable variables such as genetics were held constant during the optimisation step, the deep learning architecture should ensure that complex non-linear interactions between genetic, metabolic, and environmental features may still influence inferred recovery pathways. The digital twin models were trained on data derived from a single trial conducted in a specific sociocultural (Mipur Slum, Dhaka, Bangladesh) and nutritional context (moderate acute malnutrition), which limits generalisability to other populations, age groups, or phenotypes. External validation in independent cohorts, ideally using prospectively collected multi-omic data and predefined modelling protocols, will be essential to assess robustness, reproducibility, and transportability. Finally, while unsupervised and representation-learning approaches do not lend themselves to conventional statistical power calculations, the modest number of non-recovered children constrains the resolution at which recovery subgroups and cluster-specific effects can be reliably identified. Larger studies designed explicitly to integrate machine-learning frameworks with experimental intervention arms will be required to determine whether biologically stratified nutritional strategies can translate into improved clinical outcomes. Despite these limitations, we contend that future nutritional supplementation trials should test the hypothesis that a short treatment with a maintenance energy and protein supplement will shift the metabolism away from the hypercatabolic state, reflected in the ratio of non-essential: essential amino acids, to a more stable starvation state. This shift will contribute to optimising the amino acid ratios and thus enhancing the rapid catch-up growth and metabolic reprogramming that occurs with high energy supplements, moving children away from the danger zone for relapse. Polygenic risk scores, aggregate predictions from multiple inherited disease-risk variants 39 , are emerging as tools for understanding genetic predispositions to complex traits and multifactorial diseases 40 . The PRSs used in this study are for complex traits (i.e. NAFLD, ADHD, BMI, EA, height, IBD‐CD, IBD‐UC, and intelligence) that have known inter‐relationships, comorbidities and putative mechanistic connections to malnutrition. The emergence of the ability to separate digital twin recovery (i.e. clusters one and two) according to their PRS profiles was unexpected. Superficially, our observations are consistent with established relationships between increased polygenic liability for ADHD, BMI and cognition 41 . Similarly, there is evidence for associations from adiposity (BMI) to NAFLD 42 . Despite the existence of these recognized relationships, the exploratory correlation with model-defined recovery pathways is not easily explained. Rather, this is likely to involve complex pleiotropies that underly individual responses to the targeted nutritional intervention. Thus, there is a need for larger nutritional intervention studies that integrate genetic risk scores with functional genomics within a digital twin model to improve our understanding of how genetic pleiotropy contributes to nutritional recovery. In conclusion, this study demonstrates that an enhanced ready-to-use supplementary food accelerates catch-up growth among one-year-old children with moderate acute malnutrition who achieve anthropometric recovery, without increasing the overall proportion of children meeting WHO recovery criteria. Continued supplementation with small-quantity lipid-based nutrient supplements, whether standard or enhanced, was associated with growth trajectories that broadly paralleled those of well-nourished children but did not eliminate residual deficits or fully prevent downward drift in WLZ/WHZ scores. Exploratory, post-hoc digital twin analyses suggest that heterogeneity in recovery may be partially explained by baseline metabolic state, reflected in plasma amino acid and vitamin profiles, and that model-consistent optimisation of these features could, in principle, increase the proportion of children achieving more stringent anthropometric thresholds. These simulations are hypothesis-generating and do not constitute evidence of causal or clinically actionable effects. However, they suggest that recovery ceilings may be modifiable under biologically informed strategies. Future nutritional intervention trials should prospectively evaluate whether stratification by baseline metabolic features, combined with targeted stabilisation strategies, can safely and effectively improve recovery trajectories in children with MAM. Methods Ethics The protocol (registered on ClinicalTrials.gov, study ID number: NCT05629624) was co-developed by researchers from The University of the West Indies, Jamaica; Boston Children’s Hospital, USA; International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b); and University of Auckland, New Zealand. Ethical approvals were obtained from the Research Review Committees (RRC; August 21, 2021) and Ethical Review Committees (ERC) of icddr,b (protocol no.: PR-21084; September 21, 2021), University of Auckland (approval AH23922; for analyses of collected biological samples), and The University of the West Indies (CREC-MN.51, 21/22). Study Design and Participants This study was performed on the baseline (one-year) and two-year data from three cohorts of infants who were enrolled (between February and December 2022) as part of the M4EFaD intervention within the Mirpur poverty-stricken area Dhaka, Bangladesh. Inclusion criteria included a diagnosis of malnutrition, no history of chronic medical conditions, and no antibiotic use within the past month 17 . Nutritional supplement Following enrolment at one year of age, 159 children with moderate acute malnutrition (MAM) were randomized to one of two intervention arms. The local RUSF (L-RUSF) group (n = 79) received two 50 g servings daily (≈ 264 kcal/packet) of a rice and lentil based RUSF previously shown to be acceptable in Bangladesh 44 . This formulation, produced in a hygienic food processing laboratory at the Mirpur clinic, provided ~250 kcal/50 g with 45–50% of calories from fat and 8–10% from protein. The enhanced RUSF (E-RUSF) group (n = 80) received one 92 g packet daily (≈ 500 kcal/packet) of Nutriset-produced RUSF supplemented with additional microbiome- and brain-directed nutrients, carotenoids (zeaxanthin, lutein, cryptoxanthin), and soluble fibres (inulin + FOS) 45–47 . Children receiving E-RUSF also received fish oil in gel capsules, while the L-RUSF group received placebo capsules containing corn oil. Children in both arms received their assigned RUSF daily until they reached the biologically informed recovery threshold (WLZ/WHZ > –1 SD) or for a maximum of 3 months, after which supplementation continued with small-quantity lipid-based nutrient supplements (SQ-LNS; 20 g/day, Nutriset) for the L-RUSF group and enhanced SQ-LNS (ESQ-LNS; 26 g/day) for the E-RUSF group 17 . These supplements were continued until three years of age (two years of follow-up), delivered daily by study staff for the first month and then daily (L-RUSF) or twice weekly (E-RUSF) thereafter. Mothers were provided with feeding guidance during home visits. The enhanced interventions E-RUSF (Enhanced Ready-to-Use Supplementary Food; developed and manufactured by Nutriset) and ESQ-LNS (enhanced Small Quantity Lipid-based Nutrient Supplements; developed and manufactured by Nutriset) were both formulated to provide key nutrients required for rehabilitating wasted brains ( Fig. S5 ). Each supplement contains 24 essential micronutrients (vitamins and minerals) at recommended daily allowance levels, recommended levels of functional lipids including long-chain polyunsaturated fatty acids (DHA and EPA), neural-specific antioxidants (zeaxanthin, lutein, and cryptoxanthin), and a microbiome-modulating soluble dietary fibre mix (inulin and FOS) amounting to 6 g per 26 g daily dose of ESQ-LNS and within the 92 g daily ration of E-RUSF. Recruitment and anthropometric data collection Enrolment was conducted from February 2022 to December 2022 and follow-up at two years and three years was completed in December 2024. Study surveillance workers (SWs) conducted a door-to-door census (approximately 100,000 households) in Mirpur DNCC wards 2, 3 and 5 between February and December 2022. Verbal consent was obtained to participate in the census. The census identified 5,736 children aged between 11 and 13 months and 2,314 children aged between 34 and 38 months. During the census, if the guardian verbally consented to the study procedure, and the babies met the inclusion and exclusion criteria of the study, the surveillance workers proceeded to measure the mid-upper arm circumference (MUAC) of the child. Mothers of babies who were within the MUAC range (MUAC <12.5 and ≥11.5 cm) were invited to visit the icddr,b study clinic for further assessment and enrolment. Final screening for eligibility was based on weight-for-length/weight-for height (WLZ/WHZ) z-scores and study consent occurred at the icddr,b Mirpur study clinic. The consenting process was tailored to each mother’s literacy level and involved reviewing the inclusion and exclusion criteria. Comprehension of the study was assessed using scripted points and open-ended questions. Following consent, the clinical screening team completed a screening form, capturing the date of enrolment, sex, date of birth (DOB), weight (kg), length/height (cm), head circumference (cm), and MUAC (cm) measurements of the child. The weight-for-length/weight-for height (WLZ/WHZ) z-score for each child was calculated using the WHO anthropometric calculator (v3.2.2). The child’s age was validated using the EPI vaccination card. Biological sample collection Stool samples were collected from each infant at their home at the baseline visit. Samples were collected in DNA/RNA Shield Fecal Collection Tubes (Zymo Research, #R1101). Peripheral venous blood samples were collected in EDTA Vacutainers, separated into plasma and RBCs and immediately frozen at -80 °C. Batches of blood and stool samples were air-freighted on dry ice from Bangladesh to the Liggins Institute, New Zealand, for processing and analysis. Microbiome DNA extraction and sequencing Stool samples (1 mL) were mechanically lysed in bead bashing tubes using the MiniG tissue homogenizer prior to extraction of DNA. DNA was extracted from post-bead bashing sample (200 µL) using the ZymoBIOMICS MagBead DNA/RNA extraction kit (Zymo Research, #R2136) following the standard protocol. DNA was collected in DNAse/RNAse Free Water (50 µL). Repeat extractions were performed for samples with a DNA concentration < 14.5 ng/μL. Sequencing libraries were prepared and sequenced (Illumina NovaSeq 150PE reads) to an average sequencing depth of 20M read-pairs/sample by Custom Science. Metagenomic analysis Metagenomics samples were processed as described by Portlock et al., 2025. Raw DNA sequences were processed using BioBakery3 with default parameters tools 48 . Read quality filtering and human decontamination were performed using KneadData (Version 1). Taxonomic profiling was performed using MetaPhlAn4. Functional profiling was achieved using presence/absence and abundance of microbial pathways (MetaCyc) with HUMAnN3 (Version 3.6). A minimum threshold of > 0.1% relative abundance and > 5% prevalence for all detected species was applied. Low-pass whole-genome sequencing Genotype data for the Bangladesh child cohort were generated by Gencove Ltd. through sequencing and imputation. The genotype data for children with moderate acute malnutrition (MAM) were received on 12 October 2023, and for well-nourished controls on 21 March 2024. Quality control and data cleaning were performed prior to downstream PRS analyses. Specifically, SNPs located on chromosomes 24, 25, and 26 were removed, and variants with heterozygous calls in haploid regions were excluded. Samples with > 5% missing genotype calls were discarded, and SNPs with > 5% missingness across individuals were also removed. After quality control, 299 MAM and 145 well-nourished samples remained for subsequent PRS calculations. Calculation of polygenic risk scores Polygenic risk scores (PRSs) were calculated for eight phenotypes relevant to metabolic and neuropsychiatric outcomes, including non-alcoholic fatty liver disease (NAFLD) 49 , attention-deficit/hyperactivity disorder (ADHD) 50 , Crohn’s disease (IBD_CD) 51 , ulcerative colitis (IBD_UC) 51 , body mass index (BMI) 52 , height 53 , intelligence 54 , and educational attainment 55 . For each phenotype, GWAS summary statistics were obtained from the corresponding published studies 56 . From each summary statistics dataset, only SNPs with association p-values below 1 × 10⁻⁶ and present in the Bangladesh child cohort genotype data were retained. SNP identifiers, effect alleles, and effect sizes were extracted to create effect size files. When effect sizes were reported as odds ratios (OR), they were converted to beta coefficients using the natural logarithm (β = log [OR]). PRSs were then computed for the Bangladesh child cohort using PLINK v1.9 57 . The --score function was applied with the curated effect size files to generate individual-level PRSs. Separate analyses were performed for well-nourished children and those with moderate acute malnutrition (MAM). Finally, PRS values derived from the well-nourished and MAM datasets were merged into a single dataset to generate a unified PRS file for downstream statistical analyses. Plasma lipidomics Plasma lipidomics samples were processed as described by Portlock et al., 2025. Briefly, a 10 μL aliquot of the sample was placed in an amber glass autosampler vial and 300 μL of a mixture of Type 1 water, butanol, methanol, chloroform and SPLASH Lipidomix in a ratio of 4:15:15:20:1 was added. The mixture was vortexed and sonicated at room temperature before the protein precipitate was removed by centrifugation and an aliquot of supernatant transferred to an amber glass autosampler for negative ionisation LC-MS/MS. A second aliquot of supernatant was diluted 5 times with 75% IPA for positive ionisation LC-MS/MS. A 5 µL volume of each sample was injected onto a Phenomenex Kinetex F5 column (100 mm × 2.1 mm × 2.6 m) and lipids were separated using a ternary gradient of Type 1 water, methanol and isopropanol containing ammonium acetate. Lipids were quantified and identified with a Q-Exactive mass spectrometer (Thermo Fisher Scientific, Germany) equipped with a heated electrospray ionisation HESI source. Data was processed using MS-DIAL v4.92 58 . Plasma amino acids and B-vitamins The sample preparation and liquid chromatography - tandem mass spectrometry (LCMSMS) analysis of plasma B-vitamers was carried out using a method published previously 59 . The following vitamers were measured: pantothenic acid (B5), 4-pyridoxic acid (B6), nicotinamide (B3), trimethylamine N-oxide (TMAO), riboflavin (B2), and thiamine (B1). Three plasma quality pools controls were used to monitor analyte recovery, inter- and intra-assay reproducibility. All analytes were recovered between 85 and 105%, with average inter- and intra assay coefficients of variation all less than 8%. Additional B-vitamers were included in the assay but the plasma results were all below the limits of detection. B-vitamers and Limits of Detection (LOD) (nM) are nicotinuric acid (12.5), pyridoxal (15), pyridoxine (2), folic acid (10) and pyridoxamine (2). Free amino acids were assayed from 20 µl of plasma with 15 µM L-Nor-Valine as internal standard by Ultra High-Pressure Liquid Chromatography 60 . Average intra-assay %CV for quality controls across all amino acids was less than 10%, inter-assay %CV less than 12%. For sample preparation, all liquid addition steps were carried out on an EpMotion 5075vt workstation (Eppendorf AG, Hamburg, Germany) equipped with single and eight channel pipetting tools of various volumes (50-1000 µL). Samples (up to 82 per 96-well plate), and plasma quality controls (5 per plate) were precipitated prior to derivatization using 160 µl of 0.04 M sulphuric acid containing the internal standard and then 20 µl 10% sodium tungstate. To complete the precipitation process, the plate was sealed, shaken for three minutes (4 °C) and then centrifuged at 1000 g for 10 minutes (Avanti J-15R, Beckman Coulter, Nyon, Switzerland). Twenty microlitres of pre-prepared standards (5 different concentrations across the physiological range), aqueous quality controls (3 per plate) and each of the protein free supernatants were then transferred to a new deep-well plate in a programmed pattern for derivatization. A 140 µl aliquot of 0.2 M borate buffer (pH 8.8) was then added to each well followed by 20 µl of tagging reagent, 6-Aminoquinolyl-N-Hydroxysuccinimidyl Carbamate (AQC) (2.8 mg/ml in acetonitrile). The plate was then sealed and heated at 55 °C for 10 minutes to complete the tagging reaction, then transferred to the UPLC system for analysis. Statistical analyses Python (Version 3.9.2) was used to perform all analyses unless otherwise specified. Principal Co-ordinate Analysis (PCoA) ordinations (plotted using ‘skbio.stats.ordination.pcoa’ module) were used to visualise the clustering of the Bray-Curtis dissimilarities (calculated using ‘skbio.distance.pdist’) between microbiome samples from their species and functional composition. To quantify the variance of the dataset explained by the covariates, PERMANOVA p-values were calculated from those Bray-Curtis Dissimilarities using the ‘permanova’ function from the ‘skbio.stats.distance’ module. Bray-Curtis were also used to capture the temporal dynamics of each dataset from baseline. Numerical associations between species and metadata were measured with Spearman correlation (calculated using ‘spearmanr’ function from ‘scipy.stats’ module), where significance was defined as p-values of < 0.05. Associations between categorical data were measured with Fisher’s Exact test (calculated using ’fisher_exact’ from ’scipy.stats’ module), where significance was defined as p-values of < 0.05. All mixed modelling, differential analysis was performed using Maaslin3 (ver 0.99.16) 61 . Parameters for the modelling were default unless otherwise specified and were specific to each datatype, executed by ‘maaslin3_baseline.sh’ and ‘maaslin3_healthy.sh’ scripts on the projects Github repository. Redundancy analysis (RDA) was used to quantify the proportion of multivariate variation in the response data explained by measured covariates and was implemented in R using the vegan package (v2.7-1). Response and explanatory matrices were matched by sample identifier, and samples with missing values in either the response matrix or the specified explanatory variables were excluded using complete-case filtering. Explanatory variables were specified a priori as fixed effects; variables with fewer than two unique levels were removed and character variables were coerced to factors prior to analysis. Full RDA models were fitted using rda(), and statistical significance of the overall constrained model was assessed using permutation-based analysis of variance (anova.cca) with 999 permutations. Marginal effects of individual explanatory variables, conditional on all other terms in the model, were evaluated using permutation tests with anova.cca(..., by = "margin"). Model fit was summarised using the coefficient of determination (R²) and adjusted R² calculated via RsquareAdj(), with unconstrained variance defined as 1 − R². Digital Twin Model Development and Counterfactual Analysis Deep learning was used to generate a digital twin for simulating personalised interventions. To manage the high dimensionality of the gut microbiome data, principal component analysis (PCA) was applied to the species-level abundance data from both baseline (one-year) and 52-week (two-year) timepoints. The top ten PCs were kept. A feature matrix was constructed by integrating these microbiome PCs with the other datasets used throughout the analyses presented in this paper (Table S4). To handle missing data, continuous variables were imputed using a multivariate iterative imputer (scikit-learn's ‘IterativeImputer’), while binary and categorical variables were imputed using the most frequent value (scikit-learn's ‘SimpleImputer’) 62 . All continuous features were standardised using a standard scaler (i.e. z-scaled; sci-kit-learn's ‘StandardScaler’). A variational auto-encoder (VAE) was then trained on this pre-processed dataset. The VAE, composed of an encoder and decoder with a 64-dimensional latent space and constructed using PyTorch Lightning 63 , was optimised to reconstruct the input data by minimising a composite loss function of mean squared error (MSE) and the Kullback-Leibler (KL) divergence to create a low-dimensional representation of the data. The data was split into training (70%), validation (15%), and testing (15%) sets. A separate neural network regression head was then appended to the frozen, pre-trained VAE encoder. This predictive head was trained to predict the WLZ/WHZ at two years (52 weeks after baseline), with the objective of minimising the MSE on the validation dataset. The final, trained model was used to generate counterfactuals. All individuals with a predicted WLZ/WHZ below the biologically informed recovery threshold of -1 SD were identified. A gradient-based optimisation algorithm was then used to identify the minimal changes to the modifiable baseline features that would raise their predicted WLZ/WHZ to -1 SD. Individual intervention strategies were hierarchically clustered using Ward distance (seaborn’s ‘clustermap’) 64 . To apply average intervention strategies to all individuals, a ‘consensus threshold’ of 0.4 was chosen, meaning that the feature would only be increased or decreased if at least 40% of individuals required an increase or decrease, respectively, of that feature. A consensus threshold of 0.4 was also used when applying cluster-average interventions. To account for high intra-cluster variance in the required changes, any features with a suggested increase were updated by the average plus 1.5 SD. A pairwise comparison of average PRSs between clusters was undertaken. Fisher and Mann-Whitney U tests were used to compare categorical and continuous metadata variables, respectively, between the two clusters. Benjamini-Hochberg (BH) was used for controlling the False Discovery Rate (FDR) for all cluster analyses. Declarations Code availability All analysis code is freely available on the GitHub repository https://github.com/theoportlock/recovery. Documentation for reproduction is provided in the repository’s README file. Ethics approval and consent to participate Ethical approvals were obtained from the Research Review Committee (RRC; August 21, 2021) and Ethical Review Committee (ERC) of icddr,b (protocol no: PR-21084; September 21, 2021), Institutional Review Board of Boston Children’s Hospital, USA (for analyses of neuropsychological assessments), University of Auckland, New Zealand (approval AH23922; for analyses of collected biological samples) and University of West Indies (CREC-MN.51, 21/22). Data availability All data supporting this study are openly available and deposited on Figshare at (https://doi.org/10.17608/k6.auckland.25560768) Competing interests Mamane Zeilani is an employee of Nutriset. The remaining authors declare that they have no competing interests. Funding Work on this clinical trial is supported by Wellcome Leap (9942 Culver Blvd Unit 1277 Culver City, CA 90232-4167, United States; www.wellcomeleap.org) to PDG, JMO, TF and CAN as part of the 1kD Program. We acknowledge our core donors, the Governments of Bangladesh and Canada, for providing unrestricted support and commitment to icddr,b’s research effort. Author Contributions TP, CM, and JOS drafted and co-wrote the manuscript. DH, CP, HHP, ET, NR, TS, SHK, PDG, MZ, TF, RH, CAN commented on the manuscript. JMO, RH, TF, PDG, CAN designed the study. TSand SHK performed assessments and obtained samples in Dhaka. RH oversaw the Dhaka group. TP performed multi-omic analyses, CM performed digital twin analyses, IS, and HHP performed metagenomics, CP and ET performed metabolomics, DH performed genetic analyses. JOS oversaw the Auckland group. CAN oversaw the Boston group. Acknowledgements The authors would like to acknowledge the participants in Mirpur, Dhaka, Bangladesh for their contributions to this study. The authors would also like to thank the study team within the Infectious Diseases Division, International Centre for Diarrheal Disease Research, Bangladesh for their work in participant recruitment, sample collection and assessments. References (‎UNICEF), U. N. C. F. & Bank, W. H. O. & W. The UNICEF-WHO-World Bank Joint Child Malnutrition Estimates (JME) Standard Methodology Tracking Progress on SDG Indicators 2.2.1 on Stunting, 2.2.2 (1) on Overweight and 2.2.2 (2) on Wasting. (UNICEF-WHO, 2024). Organization, W. H., (UNICEF), U. N. C. F. & Bank, W. Levels and Trends in Child Malnutrition: UNICEF / WHO / The World Bank Group Joint Child Malnutrition Estimates: Key Findings of the 2021 Edition. (World Health Organization, Geneva, 2021). WHO Guideline on the Prevention and Management of Wasting and Nutritional Oedema (acute Malnutrition) in Infants and Children under 5 Years. 208 (2024). Kirolos, A. et al. Neurodevelopmental, cognitive, behavioural and mental health impairments following childhood malnutrition: a systematic review. BMJ Glob Health 7, (2022). Lagrone, L., Cole, S., Schondelmeyer, A., Maleta, K. & Manary, M. J. Locally produced ready-to-use supplementary food is an effective treatment of moderate acute malnutrition in an operational setting. Ann Trop Paediatr 30, 103–108 (2010). Teshome, M. S. et al. Current evidence on the effectiveness of Ready-to-Use Supplementary Foods in children with moderate acute malnutrition: a systematic review and meta-analysis. J Nutr Sci 12, (2024). Tapía, M. S., Alzamora, S. M. & Chirife, J. Effects of Water Activity (a w ) on Microbial Stability as a Hurdle in Food Preservation. Water Activity in Foods: Fundamentals and Applications 323–355 (2020) doi:10.1002/9781118765982.CH14. Steenkamp, L., Lategan, R. & Raubenheimer, J. The impact of Ready-to-Use Supplementary Food (RUSF) in targeted supplementation of children with moderate acute malnutrition (MAM) in South Africa. South African Family Practice 57, 322–325 (2015). Heymsfield, G. et al. Effectiveness of acute malnutrition treatment with a simplified, combined protocol in Central African Republic: An observational cohort study. Matern Child Nutr 20, e13691 (2024). Katsoulakis, E. et al. Digital twins for health: a scoping review. NPJ Digit Med 7, 1–11 (2024). Sizemore, N. et al. A digital twin of the infant microbiome to predict neurodevelopmental deficits. Sci Adv 10, 400 (2024). Li, X. et al. A dynamic single cell-based framework for digital twins to prioritize disease genes and drug targets. Genome Med 14, 1–21 (2022). Barbiero, P., Viñas Torné, R. & Lió, P. Graph Representation Forecasting of Patient’s Medical Conditions: Toward a Digital Twin. Front Genet 12, 652907 (2021). Ahmed, T., Michaelsen, K. F., Frem, J. C. & Tumvine, J. Malnutrition: Report of the FISPGHAN working group. J Pediatr Gastroenterol Nutr 55, 626–631 (2012). Shama, T. et al. Multidimensional evaluation of the early emergence of executive function and development in Bangladeshi children using nutritional and psychosocial intervention: A randomized controlled trial protocol. PLoS One 19, (2024). Yeasmin, K. & Yeasmin, T. Assessment of Nutritional Status of Preschool Children: Head Circumference and Other Anthropometric Indices. Bangladesh Med Res Counc Bull 44, 152–159 (2018). Fahur Bottino, G. et al. Early life microbial succession in the gut follows common patterns in humans across the globe. Nature Communications 16, 1–12 (2025). Hamadani, J. D. et al. Integrating an early childhood development programme into Bangladeshi primary health-care services: an open-label, cluster-randomised controlled trial. Articles Lancet Glob Health 7, 366–75 (2019). Goel, P. & Goel, A. Exploring the Evolution of Sleep Patterns From Infancy to Adolescence. Cureus 16, (2024). Makori, N. et al. Efficacy of ready-to-use food supplement for treatment of moderate acute malnutrition among children aged 6 to 59 months. Matern Child Nutr 20, (2024). Medoua, G. N. et al. Recovery rate of children with moderate acute malnutrition treated with ready-to-use supplementary food (RUSF) or improved corn–soya blend (CSB+): a randomized controlled trial. Public Health Nutr 19, 363–370 (2016). Rashid, M. Y., Kebira, J. Y., Oljira, L. & Dheresa, M. Time to Recovery From Moderate Acute Malnutrition and Its Predictors Among Children 6–59 Months of Age Enrolled in Targeted Supplementary Feeding Program in Darolebu District, Eastern Ethiopia: A Retrospective Cohort Study. Front Public Health 10, 914837 (2022). Griswold, S. P. et al. Effectiveness and cost-effectiveness of 4 supplementary foods for treating moderate acute malnutrition: results from a cluster-randomized intervention trial in Sierra Leone. Am J Clin Nutr 114, 973–985 (2021). Chen, R. Y. et al. A Microbiota-Directed Food Intervention for Undernourished Children. New England Journal of Medicine 384, 1517–1528 (2021). Kordas, K. et al. Maternal reports of sleep in 6-18 month-old infants from Nepal and Zanzibar: Association with iron deficiency anemia and stunting. Early Hum Dev 84, 389–398 (2008). Watanabe, F., Yabuta, Y., Bito, T. & Teng, F. Vitamin B12-Containing Plant Food Sources for Vegetarians. Nutrients 2014, Vol. 6, Pages 1861-1873 6, 1861–1873 (2014). Bröer, S. & Gauthier-Coles, G. Amino Acid Homeostasis in Mammalian Cells with a Focus on Amino Acid Transport. Journal of Nutrition 152, 16–28 (2022). Parra, M., Stahl, S. & Hellmann, H. Vitamin B6 and Its Role in Cell Metabolism and Physiology. Cells 2018, Vol. 7, Page 84 7, 84 (2018). Tardy, A. L., Pouteau, E., Marquez, D., Yilmaz, C. & Scholey, A. Vitamins and Minerals for Energy, Fatigue and Cognition: A Narrative Review of the Biochemical and Clinical Evidence. Nutrients 2020, Vol. 12, Page 228 12, 228 (2020). Camillo, L., Zavattaro, E. & Savoia, P. Nicotinamide: A Multifaceted Molecule in Skin Health and Beyond. Medicina 2025, Vol. 61, Page 254 61, 254 (2025). Wunderle, C. et al. Association of leucine and other branched chain amino acids with clinical outcomes in malnourished inpatients: a secondary analysis of the randomized clinical trial EFFORT. European Journal of Clinical Nutrition 2024 79:1 79, 42–49 (2024). Holeček, M. Branched-chain amino acids in health and disease: metabolism, alterations in blood plasma, and as supplements. Nutrition & Metabolism 2018 15:1 15, 1–12 (2018). Wu, G. et al. Proline and hydroxyproline metabolism: implications for animal and human nutrition. Amino Acids 2010 40:4 40, 1053–1063 (2010). Faure, M. et al. Specific Amino Acids Increase Mucin Synthesis and Microbiota in Dextran Sulfate Sodium–Treated Rats. J Nutr 136, 1558–1564 (2006). Wu, D. et al. Proline uptake promotes activation of lymphoid tissue inducer cells to maintain gut homeostasis. Nat Metab 5, 1953–1968 (2023). Matsumoto, S. et al. Urea cycle disorders—update. J Hum Genet 64, 833–847 (2019). Okun, J. G. et al. Liver alanine catabolism promotes skeletal muscle atrophy and hyperglycaemia in type 2 diabetes. Nature Metabolism 2021 3:3 3, 394–409 (2021). Perrott, S. L. & Kar, S. P. Polygenic risk scores for genomics and population screening. The Lancet 404, 935–936 (2024). Slunecka, J. L. et al. Implementation and implications for polygenic risk scores in healthcare. Hum Genomics 15, 1–18 (2021). Du Rietz, E. et al. Association of Polygenic Risk for Attention-Deficit/Hyperactivity Disorder With Co-occurring Traits and Disorders. Biol Psychiatry Cogn Neurosci Neuroimaging 3, 635–643 (2018). Wang, B., Yang, Y., Yin, Z. & Yang, W. The causal impact of body mass index on metabolic biomarkers and nonalcoholic fatty liver disease risk. Sci Rep 15, 1–42 (2025). Gangwal, A. & Lavecchia, A. Artificial intelligence in preclinical research: enhancing digital twins and organ-on-chip to reduce animal testing. Drug Discov Today 30, 104360 (2025). Ahmed, T. et al. Development and acceptability testing of ready-to-use supplementary food made from locally available food ingredients in Bangladesh. BMC Pediatr 14, 1–8 (2014). Obelitz-Ryom, K. et al. Bovine Milk Oligosaccharides with Sialyllactose Improves Cognition in Preterm Pigs. Nutrients 2019, Vol. 11, Page 1335 11, 1335 (2019). Cusick, S. E. & Georgieff, M. K. The Role of Nutrition in Brain Development: The Golden Opportunity of the “First 1000 Days”. J Pediatr 175, 16 (2016). Charbonneau, M. R. et al. Sialylated Milk Oligosaccharides Promote Microbiota-Dependent Growth in Models of Infant Undernutrition. Cell 164, 859–871 (2016). Beghini, F. et al. Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with biobakery 3. Elife 10, (2021). Namjou, B. et al. GWAS and enrichment analyses of non-alcoholic fatty liver disease identify new trait-associated genes and pathways across eMERGE Network. BMC Med 17, 1–19 (2019). Demontis, D. et al. Discovery of the first genome-wide significant risk loci for attention deficit/hyperactivity disorder. Nat Genet 51, 63–75 (2019). Sun, Y. et al. The Contribution of Genetic Risk and Lifestyle Factors in the Development of Adult-Onset Inflammatory Bowel Disease: A Prospective Cohort Study. Am J Gastroenterol 118, 511–522 (2023). Hoffmann, T. J. et al. A Large Multiethnic Genome-Wide Association Study of Adult Body Mass Index Identifies Novel Loci. Genetics 210, 499–515 (2018). Allen, H. L. et al. Hundreds of variants clustered in genomic loci and biological pathways affect human height. Nature 467, 832–838 (2010). Davies, G. et al. Study of 300,486 individuals identifies 148 independent genetic loci influencing general cognitive function. Nat Commun 9, 1–16 (2018). Lee, J. J. et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nat Genet 50, 1112–1121 (2018). Soh, S. E. et al. Cohort profile: Growing Up in Singapore Towards healthy Outcomes (GUSTO) birth cohort study. Int J Epidemiol 43, 1401–1409 (2014). Purcell, S. et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet 81, 559 (2007). Tsugawa, H. et al. MS-DIAL: Data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nat Methods 12, 523–526 (2015). Andraos, S. et al. Plasma B Vitamers: Population Epidemiology and Parent-Child Concordance in Children and Adults. Nutrients 2021, Vol. 13, Page 821 13, 821 (2021). Milan, A. M. et al. The impact of heat treatment of bovine milk on gastric emptying and nutrient appearance in peripheral circulation in healthy females: a randomized controlled trial comparing pasteurized and ultra-high temperature milk. Am J Clin Nutr 119, 1200–1215 (2024). Nickols, W. A. et al. MaAsLin 3: Refining and extending generalized multivariable linear models for meta-omic association discovery. bioRxiv https://doi.org/10.1101/2024.12.13.628459 (2024) doi:10.1101/2024.12.13.628459. Pedregosa FABIANPEDREGOSA, F. et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 12, 2825–2830 (2011). Paszke, A. et al. PyTorch: An Imperative Style, High-Performance Deep Learning Library. Adv Neural Inf Process Syst 32, (2019). Waskom, M. L. seaborn: statistical data visualization. J Open Source Softw 6, 3021 (2021). Additional Declarations Yes there is potential Competing Interest. Mamane Zeilani is an employee of Nutriset. The remaining authors declare that they have no competing interests. Supplementary Files suppmethod.docx Supplementary Methods PortlocksuppTables.xlsx Supplementary Tables 1-9 Supplementarymaterials.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8585503","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":574718391,"identity":"6f437d34-4b2c-43c8-9008-a78cc134f854","order_by":0,"name":"Justin O'Sullivan","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0003-2927-450X","institution":"Liggins Institute, University of Auckland","correspondingAuthor":true,"prefix":"","firstName":"Justin","middleName":"","lastName":"O'Sullivan","suffix":""},{"id":574718392,"identity":"ec3a2502-a3b1-4af3-b1a3-a35bdd791393","order_by":1,"name":"Theo Portlock","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Theo","middleName":"","lastName":"Portlock","suffix":""},{"id":574718393,"identity":"d61a91c9-d2e9-4730-992d-0159a5e978d2","order_by":2,"name":"Catriiona Miller","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Catriiona","middleName":"","lastName":"Miller","suffix":""},{"id":574718394,"identity":"7085db00-833e-4d6e-81ac-9877ebd37cb6","order_by":3,"name":"Daniel Ho","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Ho","suffix":""},{"id":574718395,"identity":"516a7943-7916-4a00-b591-125d1dd01e35","order_by":4,"name":"Chris Pook","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Chris","middleName":"","lastName":"Pook","suffix":""},{"id":574718396,"identity":"c1144b62-16df-4be0-a378-2ab84d5d0537","order_by":5,"name":"Inoli Shennon","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Inoli","middleName":"","lastName":"Shennon","suffix":""},{"id":574718397,"identity":"d4129956-a937-4bcf-a8ff-374b84221a8e","order_by":6,"name":"Talat Shama","email":"","orcid":"","institution":"International Centre for Diarrhoeal Disease Research","correspondingAuthor":false,"prefix":"","firstName":"Talat","middleName":"","lastName":"Shama","suffix":""},{"id":574718398,"identity":"023687db-ace1-4ca4-8829-1497ef2922bc","order_by":7,"name":"Shahria Kakon","email":"","orcid":"","institution":"International Centre for Diarrhoeal Disease Research","correspondingAuthor":false,"prefix":"","firstName":"Shahria","middleName":"","lastName":"Kakon","suffix":""},{"id":574718399,"identity":"8cb85c5b-159d-48fb-a1d9-19a587c61d64","order_by":8,"name":"Hui hui Phua","email":"","orcid":"https://orcid.org/0000-0002-8645-012X","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"hui","lastName":"Phua","suffix":""},{"id":574718400,"identity":"480327fa-1898-4e8e-849b-cd3340865fef","order_by":9,"name":"Eric Thorstensen","email":"","orcid":"","institution":"Liggins Institute, University of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Eric","middleName":"","lastName":"Thorstensen","suffix":""},{"id":574718401,"identity":"2bb21f31-e88c-4dcd-a670-b4314aa450fa","order_by":10,"name":"Navin Rahman","email":"","orcid":"","institution":"Department of Pediatrics, Boston Children’s Hospital and Harvard Medical School; Harvard Graduate School of Education","correspondingAuthor":false,"prefix":"","firstName":"Navin","middleName":"","lastName":"Rahman","suffix":""},{"id":574718402,"identity":"0cb8d497-3223-4c3b-a1f7-feb3b6a9143c","order_by":11,"name":"Peter Gluckman","email":"","orcid":"","institution":"The Univsrity of Auckland","correspondingAuthor":false,"prefix":"","firstName":"Peter","middleName":"","lastName":"Gluckman","suffix":""},{"id":574718403,"identity":"8530c1a9-08f5-4dbf-a907-635b55e4bafa","order_by":12,"name":"Mamane Zeilani","email":"","orcid":"","institution":"Department of External Research and Nutrition Strategy, Nutriset SAS","correspondingAuthor":false,"prefix":"","firstName":"Mamane","middleName":"","lastName":"Zeilani","suffix":""},{"id":574718404,"identity":"a50fea7b-0d01-49f5-80cc-f3b66a6851ab","order_by":13,"name":"Terrence Forrester","email":"","orcid":"","institution":"Faculty of Medical Sciences, UWI Solutions for Developing Countries, The University of the West Indies","correspondingAuthor":false,"prefix":"","firstName":"Terrence","middleName":"","lastName":"Forrester","suffix":""},{"id":574718405,"identity":"9127d143-e431-49ed-acd3-7fb1eaad2ed0","order_by":14,"name":"Rashidul Haque","email":"","orcid":"","institution":"Infectious Diseases Division, International Centre for Diarrhoeal Disease Research","correspondingAuthor":false,"prefix":"","firstName":"Rashidul","middleName":"","lastName":"Haque","suffix":""},{"id":574718406,"identity":"83edb345-e862-4872-8063-b378290b0973","order_by":15,"name":"Charles Nelson","email":"","orcid":"","institution":"Department of Pediatrics, Boston Children’s Hospital and Harvard Medical School; Harvard Graduate School of Education","correspondingAuthor":false,"prefix":"","firstName":"Charles","middleName":"","lastName":"Nelson","suffix":""}],"badges":[],"createdAt":"2026-01-12 22:10:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8585503/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8585503/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101405901,"identity":"e0093898-ac1e-4303-bec2-ba9da16e7fdc","added_by":"auto","created_at":"2026-01-29 10:41:59","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":230338,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEnhanced RUSF reduced time to biologically informed recovery threshold (WLZ/WHZ \u0026gt;-1) during rapid catch-up growth in children with MAM.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, Study design for the evaluation of anthropometric recovery to the biologically informed recovery threshold (WLZ/WHZ \u0026gt;-1 SD) in children with MAM between 12 and 24 months of age, randomized to either an enhanced refeed and supplement arm (ERUSF followed by ESQ-LNS) or a local refeed and supplement arm (RUSF followed by SQ-LNS; Table S1). A well-nourished control group was also recruited. Baseline (T1, 12 months of age) and post-refeeding (T2, 24 months of age) sampling included anthropometry, stool microbiome, and plasma metabolites, with additional one-time measures of household, educational, and genetic covariates. \u003cstrong\u003eb,\u003c/strong\u003e Line graph illustrating individual (fine lines) and mean growth trajectories (solid lines with shaded area representing 95% confidence interval around the mean) for well-nourished (purple), recovered (orange), and non-recovered (red) children over the course of the intervention. \u003cstrong\u003ec,\u003c/strong\u003e most participants who achieved the biologically informed recovery threshold following rapid catch-up growth (age 15 months) had unsustained recovery at age 24 months. Similarly, those who did not the reach the biologically informed recovery thresholdduring the rapid catch-up growth phase remained unrecovered at age 24 months, irrespective of whether they received the local or enhanced refeed supplement. \u003cstrong\u003ed\u003c/strong\u003e, there was a significant difference (p = 0.043) in average time to reach the biologically informed recovery threshold during rapid catch-up growth between individuals provided with the local RUSF and ERUSF supplements. Box plots show the median, the 25th-75th percentiles (IQR), and whiskers extending to 1.5 x IQR.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/4f6bff3dd3f26a5e575ac774.png"},{"id":101405823,"identity":"a3be2c2b-ec3e-4603-9de4-4b126e24c1fc","added_by":"auto","created_at":"2026-01-29 10:41:43","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":85655,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRecovery after one year of refeeding and supplementation (i.e. at two years of age) is mostly driven by the severity of malnutrition at baseline and genetic predisposition.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e,Scatterplot of baseline weight versus change in weight after one week for recovered and non-recovered children. All mixed modelling results are listed in Table S2. \u003cstrong\u003eb\u003c/strong\u003e, Radar plot of mean polygenic risk scores (PRSs) for both recovered and non-recovered children.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/ccd55c9d2fe3cf88c006e80f.png"},{"id":101405853,"identity":"a495e46b-8b46-4053-94cf-4ab9f635d157","added_by":"auto","created_at":"2026-01-29 10:41:47","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":81851,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEnhanced nutritional supplementation caused significant shifts in fecal microbiome species and sleep patterns towards those seen in well-nourished children.\u003c/strong\u003e \u0026nbsp;\u003cstrong\u003ea\u003c/strong\u003e, Illustration of the method for calculating the impact of sustained local or enhanced nutritional intervention to 24 months of age. δ1 is the difference between well-nourished (H1) and malnourished (M1) children at baseline. δ2 and δ3 are the differences between well-nourished (H2) and either recovered malnourished (Mr2) or non-recovered malnourished (Mnr2) children, respectively, after one year of treatment. \u003cstrong\u003eb,\u003c/strong\u003e Bar plot of percentage change in distance from well-nourished controls before and after refeeding (1yr and 2yr). Positive and negative percentages represent improvement and deterioration, respectively. Hashtag indicates no significant difference from the well-nourished arm at the second year timepoint (PERMANOVA, p \u0026gt; 0.05). Fold change (%) is plotted on a symlog\u003csub\u003e10\u003c/sub\u003e scale.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/7977518c4e9e9695fd67b40c.png"},{"id":101405683,"identity":"852b5efb-978c-44bf-9a07-d2ff5ca08b3f","added_by":"auto","created_at":"2026-01-29 10:41:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":465802,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDigital twin counterfactual analysis identifies modifiable baseline factors that enhance recovery to\u003c/strong\u003e \u003cstrong\u003ethe biologically informed threshold.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, A directed acyclic graph (DAG) showing the known contributions to WLZ/WHZ. \u003cstrong\u003eb\u003c/strong\u003e, Methods for developing the digital twin. A variational auto-encoder was trained to create a latent representation of the individuals’ multi-omic datasets (Table S4). The latent vectors were used to train a neural network to predict WLZ/WHZ before counterfactual simulations were run on all 159 malnourished individuals. \u003cstrong\u003ec\u003c/strong\u003e, the top ten modifiable factors with the highest absolute average change required to bring each individual to recovery (-1 WLZ/WHZ). All values are z-scaled. \u003cstrong\u003ed\u003c/strong\u003e, Heatmap showing the change in amount of each factor in c (increase/decrease in z-scaled amount) required to take the individual to recovery. Hierarchical clustering separated the individuals into two clusters. \u003cstrong\u003ee,\u003c/strong\u003e Polar plot showing the average polygenic risk scores (PRSs) for both clusters in D, across the same PRSs as in Fig. 2c. * adjusted p-value = 0.05\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/dd4a4f66a1bfca7c769f62ff.png"},{"id":101753723,"identity":"fdca8c8d-3bb7-4053-b2ca-724612bf83c9","added_by":"auto","created_at":"2026-02-03 10:40:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2590371,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/d0f4127c-0738-44b8-ad58-da4cac2dbc49.pdf"},{"id":101405808,"identity":"cb9d8b56-043a-4a38-892a-0177473fd0f5","added_by":"auto","created_at":"2026-01-29 10:41:35","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":39926,"visible":true,"origin":"","legend":"Supplementary Methods","description":"","filename":"suppmethod.docx","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/27dc4772bc89ddab23983994.docx"},{"id":101405855,"identity":"90c259c0-e2c4-4120-80e0-13c61ed3ece8","added_by":"auto","created_at":"2026-01-29 10:41:49","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1047263,"visible":true,"origin":"","legend":"Supplementary Tables 1-9","description":"","filename":"PortlocksuppTables.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/50e19bcb5ff8b4ed0248d4ae.xlsx"},{"id":101405858,"identity":"b74bfc8d-dd5d-4746-8ebc-87927ab0cf50","added_by":"auto","created_at":"2026-01-29 10:41:49","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":681474,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-8585503/v1/e7bdd161e5dd38ebe8572529.docx"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nMamane Zeilani is an employee of Nutriset. The remaining authors declare that they have no competing interests.","formattedTitle":"Optimising recovery from childhood moderate acute malnutrition: a randomized controlled trial","fulltext":[{"header":"Main","content":"\u003cp\u003eMalnutrition is a quintessential early life adversity and constitutes an ongoing global crisis (UN SDG2: Zero Hunger) \u003csup\u003e1\u003c/sup\u003e. In 2025, of the estimated 47 million children younger than five with acute malnutrition, ~70% were affected by Moderate Acute Malnutrition (MAM), the most common category and most treatable form of undernutrition \u003csup\u003e2\u003c/sup\u003e. Repeated episodes of acute malnutrition, both moderate and severe, in children can lead to stunting over time with irreversible effects that include intergenerational growth failure \u003csup\u003e3\u003c/sup\u003e. MAM in childhood delays the development of multiple systems in the body, including the immune, gastrointestinal, fecal microbial, and central nervous systems, but every system is impaired \u003csup\u003e4\u003c/sup\u003e. Collectively, impairments of these systems increase childhood morbidity and mortality.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe treatment goal for MAM is the restoration of normal weight for height (WLZ/WHZ \u0026gt; -2 Z score or mid-upper arm circumference, MUAC \u0026gt; 125 mm)\u003csup\u003e5\u003c/sup\u003e. Among the small proportion of children with MAM who receive supplementary feeds to achieve target weight for height within 3 months of initiation, many frequently fail to achieve anthropometric success and do not re-establish developmental trajectories \u003csup\u003e6\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eReady to use therapeutic food (RUTF) formulations were designed to treat children with severe acute malnutrition. RUTFs were recently adapted for use in the management of moderate wasting\u0026nbsp;\u003csup\u003e7,8\u003c/sup\u003e and these products are referred to as ready-to-use supplementary foods (RUSFs). The nutritional composition of home-administered RUSF is designed to provide macro-micronutrients in an energy-dense formulation to support rapid catch-up growth. Water activity in RUSF is low compared to any other specialized food used in MAM treatment, so it can prevent bacterial growth and can be stored safely at home without refrigeration\u0026nbsp;\u003csup\u003e9\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn supplementary feeding programs, children typically receive specially formulated foods in the form of fortified blended flour (FBFs) or lipid-based nutrient supplements (LNSs) dosed at approximately 75 kcal of food per kg of body weight per day or 500 kcal of food per day, in addition to the normal home diet. FBFs need to be cooked with water before consumption. By contrast, LNSs are ready-to-use foods consisting of lipid-based pastes (typically including ground peanuts) with added micronutrients that do not require refrigeration or preparation.\u003c/p\u003e\n\u003cp\u003eRecovery rates from MAM using RUSF can vary widely depending on the study's location, the specific RUSF formula, and other contributing factors (range 20 – 80%\u0026nbsp;\u003csup\u003e10,11\u003c/sup\u003e). Successful management of MAM rests on many factors where feeding is fundamental; however, other social and economic factors, family dynamics, and underlying medical conditions play variable roles. Further development and testing of optimised feeds is one key research focus that will improve treatment outcomes.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInsight into potential feed modifications that can enhance recovery rates is achievable from the results of advanced analytics. For example, digital twin technology is a cost-effective approach for simulating and analysing the behaviour of complex systems under varying\u0026nbsp;conditions \u003csup\u003e12\u003c/sup\u003e. Digital twins have been successfully applied in medicine to improve the prediction, prevention, and treatment of diseases \u003csup\u003e13–15\u003c/sup\u003e. Importantly, digital twins allow for counterfactual analysis, which is not possible with conventional neural networks that are limited to predictive modelling. For example, Sizemore et al. (2024) developed a microbiome digital twin that not only predicted infant head circumference growth but also determined patient-specific clinical interventions \u003csup\u003e13\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn this study, we examine heterogeneity in recovery from moderate acute malnutrition and the limitations of current supplementation strategies, alongside exploratory modelling to generate hypotheses for biologically informed stratification of recovery. We report a randomized controlled trial in one-year-old Bangladeshi children with MAM comparing a ready-to-use supplementary food (ERUSF) enhanced with bioactive molecules including specific carotenoids and prebiotics, plus fish oil, with a locally produced RUSF,\u0026nbsp;plus corn oil. Children received supplementation for up to three months or until they reached a predetermined weight-for-length/height target (WLZ/WHZ \u0026gt; −1)\u003csup\u003e16\u003c/sup\u003e. This rapid catch-up phase was followed by a nine-month maintenance period with either enhanced or standard small-quantity lipid-based nutrient supplements (SQ-LNS).\u003c/p\u003e\n\u003cp\u003eThe primary objective was to determine whether enhanced supplementation accelerated or increased recovery using either the WHO definition of recovery (WLZ/WHZ \u0026gt; −2) or a predefined biologically informed threshold (WLZ/WHZ \u0026gt; −1; hereafter biologically informed recovery threshold)\u0026nbsp;\u003csup\u003e16\u003c/sup\u003e. A secondary objective was to assess whether continued SQ-LNS supplementation reduced recurrence of acute malnutrition through two years of age. Finally, in post-hoc analyses, we used counterfactual digital twin models to explore whether baseline biological profiles could help explain recovery heterogeneity and identify model-consistent states associated with successful recovery.\u003c/p\u003e"},{"header":"Results","content":"\u003ch2\u003eStudy population characteristics\u003c/h2\u003e\n\u003cp\u003eDhaka, the capital of Bangladesh, is the one of the most densely populated cities in the world and has one of the highest rates of childhood malnutrition globally \u003csup\u003e16\u003c/sup\u003e. The study was conducted in the Mirpur region, where children with MAM (n = 159) and well-nourished controls (n = 75) were recruited at 12 \u0026plusmn; 1 months of age. Participants were enrolled as part of a community-based, open-label randomised clinical trial [NCT05629624] to determine the impact of enriching the refeed diet with carotenoids, polyunsaturated fats, and prebiotics during recovery from MAM (\u003cstrong\u003eFig. 1a\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Table S1\u003c/strong\u003e) \u003csup\u003e17\u003c/sup\u003e. MAM was defined according to WHO guidelines as a weight-for-length/height z-score (WLZ/WHZ) between \u0026ndash;2 and \u0026ndash;3 standard deviations \u003csup\u003e1\u003c/sup\u003e. Eligibility criteria included no history of chronic medical conditions, no known congenital anomalies, and no antibiotic use within the past month.\u003c/p\u003e\n\u003cp\u003eA wide range of biological, behavioural, environmental, and developmental datasets were collected, including anthropometry; fecal microbiome composition (species, alpha diversity, and functional pathways); plasma amino acids, lipids, and water-soluble vitamins; polygenic risk scores (PRS); results of sleep surveys; and parental and household characteristics (economics, education, family structure, media consumption, sanitation practices) (\u003cstrong\u003eFig. S1)\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBaseline characteristics were comparable between the two malnutrition arms (local RUSF and ERUSF), with no significant differences observed across most demographic, socioeconomic, or household variables (q\u0026gt;0.05;\u0026nbsp;\u003cstrong\u003eTable 1\u003c/strong\u003e). However, Caesarean delivery was significantly more common among well-nourished children (Fisher\u0026rsquo;s exact test, p = 0.007), and fathers of malnourished children had fewer years of education (MWU, p = 0.019). Although total household income did not differ significantly, households with malnourished children reported lower monthly expenditures (MWU, p = 0.032) but higher maternal income (MWU, p = 0.016), \u0026nbsp;supporting the hypothesis that children of mothers who go out to work (and leave their young children in the care of others) are at higher risk of morbidity including development of acute malnutrition. Families of malnourished children also reported longer residence in their current homes (MWU, p = 0.011), which might be a marker of family stability, or unresolved or intractable poverty. Other variables, including breastfeeding duration, family structure, and household amenities, did not differ significantly between arms.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Baseline demographic, socioeconomic, and household characteristics of study participants by study arm (i.e. nutritional group).\u003c/strong\u003e Values are presented as mean (SD) for continuous variables and n (%) for categorical variables. Comparisons were made using appropriate statistical tests (MWU for numeric-categorical, and Fisher\u0026rsquo;s Exact Test for categorical-categorical comparisons) between Local RUSF (A) and ERUSF (B) groups, and between malnourished children (A+B) and well-nourished controls (C). Bolded and asterisked p-values indicate statistical significance (p \u0026lt; 0.05).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"598\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLocal RUSF (A)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=79)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eERUSF (B)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=80)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWell-nourished (C)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=75)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA vs B (pval)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA+B vs C (pval)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Sex, Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e37 (46.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e40 (50.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e28 (37.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.752\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.123\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Delivery Mode, Caesarean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e26 (32.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e29 (36.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e40 (53.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.739\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.007\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParental Education and Economics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Years of fathers\u0026rsquo; education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.34 (3.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.58 (3.48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e6.45 (4.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.252\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.019\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Years of mothers\u0026rsquo; education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.71 (2.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.39 (3.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e6.56 (4.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.577\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.252\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Total monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e19112.66 (7307.36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e19923.75 (9832.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e25220.00 (20895.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.817\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.266\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Total monthly expenditure (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14460.76 (6403.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e15243.12 (6389.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e18469.33 (10791.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.548\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.032\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Fathers\u0026rsquo; monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e17877.22 (7501.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e18310.00 (10628.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e24620.00 (20947.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.067\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Mothers\u0026rsquo; monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1235.44 (2748.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1613.75 (3433.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e600.00 (2046.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.795\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.016\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFamily Structure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Number of living children\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e2.01 (1.04)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.71 (0.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.88 (0.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.036\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.528\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Family type - Nuclear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e47 (59.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e44 (55.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e49 (65.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.632\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.256\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHousehold characteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Members in household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.06 (1.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.84 (1.69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.76 (1.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.402\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.303\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Years lived in current household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.14 (5.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.64 (6.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e3.83 (5.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.771\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.011\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Rooms in current household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.57 (0.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.51 (0.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.60 (0.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.564\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.594\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Cooking gas - Yes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e59 (74.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e65 (81.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e61 (81.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.344\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.609\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Working TV - Yes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e48 (60.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e44 (55.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e49 (65.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.522\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.317\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOther Characteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Exclusive Breastfeeding (months)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.30 (1.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.30 (1.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.12 (1.71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.775\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.336\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 182px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Place of Birth - Home\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e30 (38.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e32 (40.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e22 (29.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.796\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.152\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch2\u003eRecovery from malnutrition\u003c/h2\u003e\n\u003cp\u003eWe measured anthropometric recovery at two points in the study: 1) following rapid catch-up growth; and 2) following sustained micronutrient and polyunsaturated fatty acids supplementation (Fig. 1a). The WHO treatment goal for MAM is the restoration of normal weight for height (WLZ/WHZ \u0026gt; -2 Z or mid-upper arm circumference, MUAC \u0026gt; 125 mm)\u003csup\u003e\u003cspan lang=\"EN-GB\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u0026nbsp;According to WHO criteria, we achieved rapid catch-up recovery of RUSF: 55/79 (69.6%) and ERUSF: 55/80 (68.75%) (Fig. S2).\u0026nbsp;Time to recovery was not significantly different between intervention groups when calculated from weekly anthropometric measurements (Wilcoxon rank-sum test, p = 0.774), despite a trend of shorter mean recovery time in the ERUSF group. Relapse to \u0026lt; -2 SD WLZ/WHZ during the maintenance period did not significantly differ between SQ-LNS (2/79, 2.53%) and ESQ-LNS (1/80, 1.25%) (Fisher\u0026rsquo;s, p = 1.00).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo gain a better understanding of the differences in individual recovery, we predefined a biologically informed recovery threshold, not a clinical standard, of WLZ/WHZ \u0026gt; -1 SD within 3 months \u003csup\u003e17\u003c/sup\u003e. Approximately 40% of treated children achieved the goal weight for height (\u003cstrong\u003eFig. 1b and 1c, Table 2\u003c/strong\u003e). The number of children who achieved the goal weight for height (reaching WLZ/WHZ \u0026gt; -1 SD within 15 weeks) was not significantly different between children receiving the locally produced RUSF and those receiving the ERUSF (44.3% and 43.8%, respectively;\u003cstrong\u003e\u0026nbsp;Fig. 1b\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eTable 2\u003c/strong\u003e). However, amongst those children who did reach the biologically informed recovery target, children receiving the ERUSF feed did so faster than those receiving the local RUSF (MWU, p = 0.043; 47.3 \u0026plusmn; 21.8 days compared to 57.0 \u0026plusmn; 20.4 days, respectively;\u003cstrong\u003e\u0026nbsp;Table 2\u003c/strong\u003e). This equates to a modest difference of approximately ten days (\u003cstrong\u003eFig. 1d\u003c/strong\u003e). However, when considering this outcome, it should be noted that time-to-biologically informed recovery threshold is vulnerable to measurement and expectation bias as the trial was open label.\u003c/p\u003e\n\u003cp\u003eThe biologically informed recovery threshold (WLZ/WHZ \u0026gt; \u0026minus;1) was used as a\u0026nbsp;treatment stopping criterion, however, it was also analysed as a biologically informed recovery threshold to examine durability and heterogeneity of response beyond the WHO-defined wasting. Therefore, we used the enhanced biologically informed recovery threshold (WLZ/WHZ \u0026gt; \u0026minus;1) to define three patterns of\u0026nbsp;anthropometric recovery: 1) \u0026lsquo;sustained recovery\u0026rsquo;, in which children achieved the biologically informed recovery target threshold during the rapid catch-up phase (within 15 weeks) and maintained it through two years of age; 2) \u0026lsquo;delayed recovery\u0026rsquo;, in which children did not reach the threshold during rapid catch-up growth but achieved it by two years of age; and 3) \u0026lsquo;unsustained recovery\u0026rsquo;, in which children initially achieved the threshold during rapid catch-up growth but subsequently fell below it by two years of age. There were no significant differences in the proportions of individuals who were present in each of the recovery subcategories following supplementation with the local or\u0026nbsp;enhanced RUSF (\u003cstrong\u003eFig. 1c\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eTable 2\u003c/strong\u003e). Notably, the proportion of days without feeding failure (i.e. supplement not taken) was not significantly different between groups (\u003cstrong\u003eTable 2\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2:\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eBiologically informed threshold recovery\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;outcomes for children receiving local RUSF (A) and ERUSF (B).\u0026nbsp;\u003c/strong\u003eData are presented as counts (percentages) or means (standard deviations). \u0026lsquo;Delayed recovery\u0026rsquo; was achieved by two years of age; \u0026lsquo;sustained recovery\u0026rsquo; was achieved by rapid catch-up growth and maintained from three months to two years of age; and \u0026lsquo;unsustained recovery\u0026rsquo; occurred in individuals who had initial recovery following rapid catch-up growth, but recovery was not maintained at two years of age. Differences in proportions were assessed using Fisher\u0026rsquo;s exact test. \u0026nbsp;Differences in continuous variables were evaluated using the Mann\u0026ndash;Whitney U test. Significant p-values are indicated in bold.\u003c/p\u003e\n\u003cdiv align=\"Left\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"435\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLocal RUSF (A)\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e(n=79)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eERUSF (B)\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e(n=80)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo recovery\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e44 (55.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e45 (56.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecovered\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;(WLZ/WHZ \u0026gt;-1 SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e35 (44.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e35 (43.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Delayed recovery\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e1 (1.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e2 (2.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.617\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Sustained recovery\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e9 (11.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e8 (10.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Unsustained recovery\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e25 (31.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e25 (31.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDays to recovery\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e57.03 (20.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e47.29 (21.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e**0.043\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 228px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eProportion of trial days without feeding failure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 84px;\"\u003e\n \u003cp\u003e0.95 (0.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e0.93 (0.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.924\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe feeds influenced the rate of achievement of target weight for height but did not affect the eventual acquisition of this goal. We also investigated whether sociodemographic factors significantly contributed to reaching the biologically informed recovery threshold (\u003cstrong\u003eTable 3\u003c/strong\u003e). There were no significant differences in demographic, socioeconomic, or household characteristics between those children with MAM who reached the threshold and those who did not (\u003cstrong\u003eTable 3\u003c/strong\u003e). The proportions of female participants (51.7% vs. 44.3%) and caesarean deliveries (34.8% vs. 34.3%) were similar between the groups who did (recovered) and did not (unrecovered) reach the threshold, respectively. Parental education levels and household economic indicators, including income and expenditure, did not differ significantly between groups. Likewise, family structure and household size showed no clear association with recovery status. However, the duration of residence in the current household was longer amongst children who reached the biologically informed recovery threshold when compared to those who did not (6.9 \u0026plusmn; 7.6 vs. 4.2 \u0026plusmn; 4.7 years, p = 0.042). Collectively, these results support the conclusion that anthropometric recovery to the biologically informed recovery threshold was not strongly influenced by demographic or socioeconomic parameters, except for one marker of residential stability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Demographic, socioeconomic, and household characteristics were not significantly different between children who\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ereached the biologically informed recovery threshold (\u003c/strong\u003e\u003cstrong\u003erecovered\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;and those who did not\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e(unrecovered)\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;during the treatment.\u003c/strong\u003e Values are presented as n (%) for categorical variables and mean (SD) for continuous variables. p-values were calculated using Fisher\u0026rsquo;s exact test for categorical variables and the Mann\u0026ndash;Whitney U test for continuous variables. Significant p-values are indicated in bold with an asterisk.\u0026nbsp;\u003c/p\u003e\n\u003cdiv align=\"Left\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"491\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo recovery (n=89)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecovered (n=70)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Sex - Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e46 (51.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e31 (44.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.425\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Delivery Mode - Caesarean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e31 (34.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e24 (34.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParental Education and Economics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Years of fathers\u0026apos; education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e5.17 (3.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.69 (3.56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.533\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Years of mothers\u0026apos; education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e5.74 (3.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.30 (3.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.390\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Total monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e19797.75 (9062.92)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e19168.57 (8152.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.625\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Monthly total expenditure (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e15298.31 (6593.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14290.00 (6117.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.366\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Fathers\u0026apos; monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e18646.07 (9425.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e17394.29 (8880.44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.393\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Mothers\u0026rsquo; monthly income (taka)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e1151.69 (2653.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1774.29 (3594.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.381\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFamily Structure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Number of living children\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e1.76 (0.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.99 (1.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.205\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Family type - Nuclear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e53 (59.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e38 (54.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.523\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHousehold characteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Number of members in household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e4.89 (1.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.03 (1.66)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.378\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Number of years lived in current household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e4.20 (4.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e6.90 (7.63)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e*0.042\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Number of rooms in current household\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e1.58 (0.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.49 (0.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Cooking gas - Yes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e71 (79.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e53 (75.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.568\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Working TV - Yes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e52 (58.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e40 (57.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.873\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Household food availability - Sometimes deficit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e45 (50.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e26 (37.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.109\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOther\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Exclusive Breastfeeding (months)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e5.20 (1.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5.43 (1.36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.188\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 234px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; PoB - Home\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 106px;\"\u003e\n \u003cp\u003e33 (37.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e29 (41.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e0.579\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAs the measured sociodemographic factors (except for number of years lived in current household) were not significantly associated with reaching the biologically informed recovery threshold, we next examined whether biological factors from the gut and plasma might be. We conducted a redundancy analysis (RDA) to investigate how recovery to the biologically informed threshold and other covariates were associated with the variance in each dataset at each time point (Table S9). At year one (i.e. at baseline), none of the variables significantly explained variance in the economic or faecal microbiome pathways profile datasets, although modest trends were observed for supplementation and place of birth in the genetics dataset (p = 0.059 and p = 0.025, respectively).\u003c/p\u003e\n\u003cp\u003eIn the analyses, after one year of follow-up on children receiving SQ-LNS, similar patterns emerged, with limited explanatory power across most covariates. However, the association of the supplementation with plasma amino acid concentrations at year two is an expected outcome of dietary supplementation for one year. We also observed that ethnicity was associated with lipid levels (p = 0.035) and sex with anthropometric outcomes (p = 0.001). However, recovery status did not significantly explain variation in any dataset at either timepoint (p \u0026gt; 0.05). Overall, the RDA analysis findings indicate that while demographic and treatment-related variables may influence specific biological or developmental domains, reaching the biologically informed recovery threshold was independent of these covariates within the constraints of the current sample (Fig. S3).\u003c/p\u003e\n\u003cp\u003eTo explore whether features from the multi-omic datasets (e.g. microbiome and plasma metabolites) were associated with recovery to the biologically informed threshold, we conducted a mixed modelling analysis (\u003cstrong\u003eFig. 2\u003c/strong\u003e). Across the cohort, anthropometric measurements at baseline were not statistically significantly associated with reaching the biologically informed threshold after correction for multiple testing (FDR). Differences in weight-for-length z-scores (WLZ/WHZ), weight, and length at baseline were small and non-significant (all q \u0026gt; 0.05) between children who did and did not reach the biologically informed threshold. Similarly, neither baseline head circumference (coef = 0.22, q = 1) nor nasion-inion length (coef = 0.44, q = 0.44) was significantly associated with rapid catch-up recovery following multiple test correction (q \u0026gt; 0.05). Female children tended to have smaller head circumference and overall anthropometric measures than males at baseline (coef = -1.07, q = 0.13), consistent with expected sex-related differences. Head sizes that were below 42 cm at one year of age (i.e. recruitment) are consistent with those individuals having suffered significant past, possibly in utero, nutritional stress \u003csup\u003e18\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eChange in weight after one week of rapid catch-up and genetic predispositions were associated with reaching the biologically informed threshold (\u003cstrong\u003eFig.\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;2,\u003c/strong\u003e \u003cstrong\u003eTable S2\u003c/strong\u003e). By contrast, other factors, including differences in complementary food composition and comorbidities such as diarrhoea, fever, cough, or antibiotic use, were not significantly associated with recovery (\u003cstrong\u003eTable S2\u003c/strong\u003e).\u003c/p\u003e\n\u003ch2\u003eTreatment-specific improvements were observed for sleep quality and microbial composition following enhanced nutritional supplementation.\u003c/h2\u003e\n\u003cp\u003eWe hypothesized that the baseline factors we measured should change in children who reached the biologically informed threshold, becoming more like those observed in the well-nourished cohort. However, we observed that the WLZ/WHZ score alters with age in the well-nourished cohort (\u003cstrong\u003eFig. 1b\u003c/strong\u003e). Therefore, to assess the impact of nine months of sustained supplementation, after the rapid catch-up growth in the first three months, out to 24 months of age on these features, we calculated centroid distances for each measure relative to the well-nourished, age-matched control group at each trial arm and timepoint (\u003cstrong\u003eFig. 3a\u003c/strong\u003e). For each dataset, the shrinking or widening of the inter-centroid distance across the two timepoints was used to indicate improvement or deterioration of the feature relative to the well-nourished controls, respectively (\u003cstrong\u003eFig. 3a\u003c/strong\u003e). This approach was taken because these features (e.g. microbiome species, microbiome functional pathways, sleep, head circumference) alter with biological age \u003csup\u003e18\u0026ndash;21\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSupplementation with ESQ-LNS, providing micronutrients, minimal energy and inulin and fish oils, after rapid catch-up growth, for approximately nine months (one year follow up) resulted in microbiome and sleep patterns that were more like those observed in the well-nourished cohort (\u003cstrong\u003eFig. 3\u003c/strong\u003e). Notably, the shifts in gut species and sleep patterns following supplementation with ESQ-LNS diverged from what was seen with the traditional SQ-LNS (\u003cstrong\u003eFig. 3b\u003c/strong\u003e). The shift in gut species included depletion of \u003cem\u003eS. wadsworthensis\u0026nbsp;\u003c/em\u003eand\u003cem\u003e\u0026nbsp;L. fermentum\u003c/em\u003e (\u003cstrong\u003eTable S3\u003c/strong\u003e). Importantly, the difference in microbial species was not mirrored by a refeed-specific separation of microbial encoded gut pathways (\u003cstrong\u003eFig. 3b, Table S3\u003c/strong\u003e). Continued supplementation with standard or enhanced SQ-LNS for nine months following rapid catch-up growth showed a trend toward convergence upon the well-nourished weight (\u003cstrong\u003eFig. 3b\u003c/strong\u003e), suggesting a reduction in the recurrence of malnutrition.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo determine whether microbial functions converged toward those of well-nourished children, we compared baseline and post-supplementation pathway abundances, one year later, with those of age-matched well-nourished controls. Most pathways that were initially depleted in malnourished children, including amino acid, coenzyme, and vitamin biosynthesis (e.g., PWY-7242, PWY-5972, PWY-7340), increased following refeeding, indicating functional recovery toward the well-nourished state \u003cstrong\u003e(Table S3).\u003c/strong\u003e In contrast, pathways linked to host-derived sugar catabolism (e.g., GLUCUROCAT-PWY) decreased, consistent with reduced dysbiosis. However, the ESQ-LNS group exhibited attenuated longitudinal changes in these pathways relative to non-enhanced SQ-LNS, suggesting slower or less complete convergence toward healthy microbiome function \u003csup\u003e22\u003c/sup\u003e. Collectively these results indicate that enhanced supplementation alters the trajectory of microbiome maturation rather than accelerating convergence.\u003c/p\u003e\n\u003ch2\u003eCounterfactual digital twins suggest optimal baseline profiles for recovery\u003c/h2\u003e\n\u003cp\u003eBuilding on our findings of baseline factors that impact recovery, we next developed digital twins to identify the individualized adjustments that could improve anthropometric recovery (WLZ/WHZ) in individuals who did not reach the biologically informed threshold during the randomized controlled trial. A feature matrix composed of multi-omic datasets associated with growth was constructed (\u003cstrong\u003eFig. 4a\u003c/strong\u003e and \u003cstrong\u003eTable S4\u003c/strong\u003e). Due to the high dimensionality of microbiome data, principal component analysis (PCA) was carried out first, and the top ten principal components (PCs) were included in the feature matrix (\u003cstrong\u003eTable S5\u003c/strong\u003e). A variational auto-encoder (VAE) was trained to learn a latent space representation of the feature space (z), capable of reproducing the original matrix with a reconstruction mean squared error (MSE) \u0026lt; 0.05 and an average Kullback\u0026ndash;Leibler (KL) divergence of 4.2 nats (~6 bits) per latent dimension. These results suggest that the latent variables retain substantial information about each individual, making the VAE more appropriate for a digital twin focused on generating realistic counterfactuals for existing individuals, rather than sampling entirely new synthetic individuals from the prior distribution. Using a regression neural network head trained on this latent feature space to predict WLZ/WHZ at one-year post-baseline, a root mean squared error (RMSE) of 0.59 was achieved within the holdout test set (\u003cstrong\u003eFig. 4b\u003c/strong\u003e). Given the units are z-scaled (WLZ/WHZ), this is equivalent to 0.59 SD.\u003c/p\u003e\n\u003cp\u003eGradient-based optimisation was applied, using the trained VAE and regression head to run counterfactual simulations. All non-modifiable features (\u003cstrong\u003eTable S4\u003c/strong\u003e) were frozen, and the model prioritised the smallest changes required to bring each individual to -1 WLZ/WHZ (i.e. recovery). The features with the largest predicted changes for simulated recovery included four amino acids, two amino acid related compounds (i.e. ornithine, a non-proteinogenic amino acid; and taurine, which is a sulfonic acid), and four vitamin cofactors (\u003cstrong\u003eFig. 4c\u003c/strong\u003e). Whilst the magnitude of changes varied between individuals, the average individual required increased levels of all the top ten features except for ornithine, serine, and alanine (\u003cstrong\u003eFig. 4c\u003c/strong\u003e). Applying the average feature change to unrecovered individuals identifies model-consistent intervention targets, not validated nutritional prescriptions, that led to simulated recovery in 12/83 (14.5%) of individuals who did not recover during the trial.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHierarchical clustering separated the individuals into two groups (\u003cstrong\u003eFig. 4d\u003c/strong\u003e), which differed by the changes required to their baseline features to reach the biologically informed threshold (\u003cstrong\u003eTable S6\u003c/strong\u003e). Of the 12 simulated individuals who recovered from the average suggested feature change, nine were in cluster one and three were in cluster two. Applying the two cluster-specific feature changes (1.5 SD above average required change, \u003cstrong\u003eTable S6\u003c/strong\u003e) led to simulated recovery in 9/12 (75%) and 48/71 (67.6%) individuals in clusters one and two, respectively. Importantly, for the individuals where these changes were possible (i.e. the average reduction in concentration did not bring them to \u0026le; 0 \u0026micro;mol/L), the large majority fell within the physiologically acceptable range (\u003cstrong\u003eTable S6\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003ePositioning of individuals within either of the two clusters was not explained by simple differences in metadata (\u003cstrong\u003eTable S7\u003c/strong\u003e). However, analysis of the ratio of plasma concentrations for non-essential and essential/conditionally essential amino acids (\u003cstrong\u003eTable S6\u003c/strong\u003e) identified that individuals within cluster 1 had a significantly increased mean ratio (0.46) when compared to individuals within both cluster 2 (0.41) and the well-nourished cohort (0.40) (\u003cstrong\u003eFig. S4\u003c/strong\u003e). We also observed that individuals within the two clusters had differing genetic profiles (\u003cstrong\u003eFig. 4e, Table S8\u003c/strong\u003e). Following false discovery rate (FDR) correction, the difference in average BMI and IBD UC PRSs between the two clusters was significant (both adjusted p = 0.05). Crucially, these clusters were formed based on the modifiable features, meaning the genetics were not directly considered.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur trial of nutrition supplementation in one-year-old children with MAM compared the rapid catch-up growth promoted by standard WHO-prescribed RUSF with that of an enhanced supplement containing: i. added carotenoids; ii. prebiotics (inulin); and 1 g fish oil. These additions were selected to improve anthropometric and neurological recovery. Additionally, SQ-LNS and an enhanced SQ-LNS were administered to children for an additional 9 months, bringing the total nutrition treatment period to 12 months. When using the WHO criteria, we observed ~69% recovery. This is consistent with observations from similar studies in Tanzania, Cameroon, Ethiopia and Sierra Leone which had 83.7%, 85%, 73%, and 62.1–64.5% (across different food types, including RUSF) recovery, respectively \u003csup\u003e23,24\u003c/sup\u003e. Similarly, 74% recovery was obtained using Microbiota-Directed Complementary Food (MDCF) in Bangladesh\u0026nbsp;\u003csup\u003e25\u003c/sup\u003e. \u0026nbsp;Critically, assessment against a predefined biologically informed recovery threshold (WLZ/WHZ \u0026gt; –1SD) resulted in recovery of 44.3% and 43.8% with either the standard or enhanced refeed, respectively. We contend that our results indicate that beyond initial recovery (\u0026gt; -2SD WHZ) there are different trajectories with some individuals unable to reach \u0026gt; -1SD WHZ even after 12 months of intervention. The trajectory of anthropometric recovery is likely affected by one of many determinants, spanning the feed composition and organoleptic properties, availability of feeds to the infant, childcare practices, concurrent morbidities such as diarrhoea, RTI, and chronic gut pathology.\u003c/p\u003e\n\u003cp\u003eThe inclusion of long-term supplementation and long-term enhanced supplementation with carotenoids, LC-PUFA, and prebiotics, together with minimal amounts of energy, was designed to evaluate whether the standard WHO-prescribed and/or the enhanced SQ-LNS prevented the recurrence of wasting once the child had reached the biologically informed recovery threshold. This appeared to be the case, as whilst the mean WLZ/WHZ score for recovered individuals fell below the -1 SD threshold, it mirrored the trajectory that was observed for the well-nourished children and rarely fell below the –2 SD WHZ threshold. Thus, prolonged supplementation with SQ-LNS and ESQ-LNS resulted in individuals following a typical growth pattern, grounded at the starting WLZ/WHZ score at which the supplementation was started.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStunting (\u0026lt; -2 SD below the median of WHO child growth standards) has been associated with reduced exploration of the environment, greater fatigue, and differences in reported sleep (characterised by shorter night sleep duration and increased night waking) in children from Zanzibar and Nepal, when compared to their peers \u003csup\u003e26\u003c/sup\u003e. The sleep impacts of stunting have been associated with vitamin B12 deficiency and iron deficiency anaemia. Lentils and rice, which form the main constituent of the locally produced RUSF \u003csup\u003e17\u003c/sup\u003e, are plants and do not contain vitamin B12 \u003csup\u003e27\u003c/sup\u003e. Therefore, the observed improvement in sleep in children whose diets were supplemented with the enhanced refeed, which contains vitamin B12 (\u003cstrong\u003eTable S1, Fig. S5\u003c/strong\u003e) \u003csup\u003e28\u003c/sup\u003e, was expected.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDespite the strength of the randomised controlled trial design, this study is not without limitations. The study was powered for comparisons between well-nourished (WLZ/WHZ \u0026gt; -1 SD) and wasted children (-3 ≤ \u0026nbsp;WLZ/WHZ \u0026lt; -2 z-score, and/or 11.5 ≤ \u0026nbsp;MUAC \u0026nbsp;\u0026lt; 12.5 cm) at one and three years, such that the proposed medium level difference between the two groups (i.e. two standard deviations) requires a minimum of 51 participants per group for power equal to 80%, using a level of significance equal to 5% (for a one-tailed test)\u0026nbsp;\u003csup\u003e16\u003c/sup\u003e. It is possible that some children experience regression once supplementation ends. Detecting such washout effects would require longer-term follow-up beyond the study period. Although the randomized controlled trial design provides a robust framework for evaluating anthropometric outcomes, the digital twin and machine-learning analyses were exploratory and post-hoc and were not prospectively powered or pre-registered as primary analytic objectives. As with any high-dimensional modelling approach applied to relatively modest sample sizes, there is an inherent risk of overfitting, instability of learned representations, and sensitivity to modelling choices, including feature selection, dimensionality reduction, and imputation strategies. While we mitigated these risks through train–validation–test splits, regularisation, and restriction of counterfactual analyses to biologically plausible ranges, the resulting predictions should be regarded as hypothesis-generating rather than confirmatory.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe counterfactual digital twin framework identifies model-consistent feature perturbations that are sufficient to change predicted outcomes within the learned latent space; it does not establish causality or guarantee that such feature modifications are biologically achievable, safe, or effective in real-world interventions. Incorporation of the clinical and biological results into digital twin counterfactual models identified plasma amino acid and vitamin levels as one potential modifiable factor. However, plasma amino acid homeostasis is a networked equilibrium between diet, tissue metabolism (catabolism and anabolism), excretion, and hormonal regulation that reflects dietary intake, tissue demand and physiological stress \u003csup\u003e28\u003c/sup\u003e. The digital twin models identified two clusters of individuals, where specific simulated alterations to baseline plasma amino acid and vitamin levels in unresponsive individuals were predicted to improve recovery to the biologically informed recovery threshold from 44% (achieved with the refeed methods) to 80% (WHL/WHZ \u0026gt; -1 SD). While the digital twin findings should be considered hypothesis-generating and unstable under resampling, it is notable that the simulated alterations were within physiologically relevant levels (e.g. Sick Kids and Mayo Clinic reference ranges (Table S6)) for age- but not context-matched children. The population ranges of plasma amino acids reflect the interaction between environmentally determined features and the individual’s genetic predisposition (e.g. expression of transporters, metabolic enzymes) \u003csup\u003e28\u003c/sup\u003e. Severe stress and starvation, often alongside co-morbid infections and other syndromes, disrupt homeostasis. This leads to the release of gluconeogenic amino acids, increased urea production, and hormonal changes. Therefore, any form of malnutrition treatment that targets the modification of the plasma amino acid baseline concentrations must simultaneously address any comorbidities that might be affecting the homeostatic state. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCritically, although our digital twin model was agnostic to the underlying biochemistry of nutrition, the simulated modifications it suggested outline a clear, biologically sound signature for recovery. The modifications show a shift away from the catabolic state of starvation by supplying a) the B-vitamins required for bioenergetic metabolism and b) the amino acids needed to repair tissues. Simulated increases of pantothenic acid (vitamin B5), nicotinamide (vitamin B3), and pyridoxine (vitamin B6; evidenced through the increase in B6 metabolite 4-pyridoxic acid) show a prioritisation of the model to replenish the cofactors essential for bioenergetic\u0026nbsp;metabolism \u003csup\u003e29–31\u003c/sup\u003e. The model simulated increases of isoleucine and proline, amino acids that are substrates for muscle protein and collagen synthesis, respectively, with proline also important for gut homeostasis\u0026nbsp;\u003csup\u003e32–36\u003c/sup\u003e. Concurrently, the model identified that reaching the biologically informed recovery threshold is associated with the reduction of amino acids that mark a catabolic state. High levels of alanine can be a sign of active muscle wasting (through the glucose-alanine cycle), while high levels of ornithine suggest increased stress on the urea cycle from tissue breakdown\u0026nbsp;\u003csup\u003e37,38\u003c/sup\u003e. Importantly, the model is not necessarily suggesting a direct reduction of these amino acids, but rather that their decline is a biomarker of successful recovery. This is consistent with the fact that plasma amino acid and vitamin profiles, in particular, reflect integrated physiological states influenced by diet, infection, inflammation, hormonal regulation, and tissue metabolism, and may act as markers rather than drivers of recovery. Furthermore, although non-modifiable variables such as genetics were held constant during the optimisation step, the deep learning architecture should ensure that complex non-linear interactions between genetic, metabolic, and environmental features may still influence inferred recovery pathways.\u003c/p\u003e\n\u003cp\u003eThe digital twin models were trained on data derived from a single trial conducted in a specific sociocultural (Mipur Slum, Dhaka, Bangladesh) and nutritional context (moderate acute malnutrition), which limits generalisability to other populations, age groups, or phenotypes. External validation in independent cohorts, ideally using prospectively collected multi-omic data and predefined modelling protocols, will be essential to assess robustness, reproducibility, and transportability. Finally, while unsupervised and representation-learning approaches do not lend themselves to conventional statistical power calculations, the modest number of non-recovered children constrains the resolution at which recovery subgroups and cluster-specific effects can be reliably identified. Larger studies designed explicitly to integrate machine-learning frameworks with experimental intervention arms will be required to determine whether biologically stratified nutritional strategies can translate into improved clinical outcomes. Despite these limitations, we contend that future nutritional supplementation trials should test the hypothesis that a short treatment with a maintenance energy and protein supplement will shift the metabolism away from the hypercatabolic state, reflected in the ratio of non-essential: essential amino acids, to a more stable starvation state. This shift will contribute to optimising the amino acid ratios and thus enhancing the rapid catch-up growth and metabolic reprogramming that occurs with high energy supplements, moving children away from the danger zone for relapse.\u003c/p\u003e\n\u003cp\u003ePolygenic risk scores, aggregate predictions from multiple inherited disease-risk variants \u003csup\u003e39\u003c/sup\u003e, are emerging as tools for understanding genetic predispositions to complex traits and multifactorial diseases \u003csup\u003e40\u003c/sup\u003e. The PRSs used in this study are for complex traits (i.e. NAFLD, ADHD, BMI, EA, height, IBD‐CD, IBD‐UC, and intelligence) that have known inter‐relationships, comorbidities and putative mechanistic connections to malnutrition. The emergence of the ability to separate digital twin recovery (i.e. clusters one and two) according to their PRS profiles was unexpected. Superficially, our observations are consistent with established relationships between increased polygenic liability for ADHD, BMI and cognition \u003csup\u003e41\u003c/sup\u003e. Similarly, there is evidence for associations from adiposity (BMI) to NAFLD \u003csup\u003e42\u003c/sup\u003e. Despite the existence of these recognized relationships, the exploratory correlation with model-defined recovery pathways is not easily explained. Rather, this is likely to involve complex pleiotropies that underly individual responses to the targeted nutritional intervention. Thus, there is a need for larger nutritional intervention studies that integrate genetic risk scores with functional genomics within a digital twin model to improve our understanding of how genetic pleiotropy contributes to nutritional recovery.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn conclusion, this study demonstrates that an enhanced ready-to-use supplementary food accelerates catch-up growth among one-year-old children with moderate acute malnutrition who achieve anthropometric recovery, without increasing the overall proportion of children meeting WHO recovery criteria. Continued supplementation with small-quantity lipid-based nutrient supplements, whether standard or enhanced, was associated with growth trajectories that broadly paralleled those of well-nourished children but did not eliminate residual deficits or fully prevent downward drift in WLZ/WHZ scores. Exploratory, post-hoc digital twin analyses suggest that heterogeneity in recovery may be partially explained by baseline metabolic state, reflected in plasma amino acid and vitamin profiles, and that model-consistent optimisation of these features could, in principle, increase the proportion of children achieving more stringent anthropometric thresholds. These simulations are hypothesis-generating and do not constitute evidence of causal or clinically actionable effects. However, they suggest that recovery ceilings may be modifiable under biologically informed strategies. Future nutritional intervention trials should prospectively evaluate whether stratification by baseline metabolic features, combined with targeted stabilisation strategies, can safely and effectively improve recovery trajectories in children with MAM.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eEthics\u003c/h2\u003e\n\u003cp\u003eThe protocol (registered on ClinicalTrials.gov, study ID number: NCT05629624) was co-developed by researchers from The University of the West Indies, Jamaica; Boston Children’s Hospital, USA; International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b); and University of Auckland, New Zealand. Ethical approvals were obtained from the Research Review Committees (RRC; August 21, 2021) and Ethical Review Committees (ERC) of icddr,b (protocol no.: PR-21084; September 21, 2021), University of Auckland (approval AH23922; for analyses of collected biological samples), and The University of the West Indies (CREC-MN.51, 21/22).\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eStudy Design and Participants\u003c/h2\u003e\n\u003cp\u003eThis study was performed on the baseline (one-year) and two-year data from three cohorts of infants who were enrolled (between February and December 2022) as part of the M4EFaD intervention within the Mirpur poverty-stricken area Dhaka, Bangladesh. Inclusion criteria included a diagnosis of malnutrition, no history of chronic medical conditions, and no antibiotic use within the past month \u003csup\u003e17\u003c/sup\u003e.\u003c/p\u003e\n\u003ch2\u003eNutritional supplement\u003c/h2\u003e\n\u003cp\u003eFollowing enrolment at one year of age, 159 children with moderate acute malnutrition (MAM) were randomized to one of two intervention arms. The local RUSF (L-RUSF) group (n = 79) received two 50 g servings\u0026nbsp;daily (≈ 264 kcal/packet) of a rice and lentil based RUSF previously shown to be acceptable in Bangladesh\u0026nbsp;\u003csup\u003e44\u003c/sup\u003e. This formulation, produced in a hygienic food processing laboratory at the Mirpur clinic, provided ~250 kcal/50 g with 45–50% of calories from fat and 8–10% from protein. The enhanced RUSF (E-RUSF) group (n = 80) received one 92 g packet daily (≈ 500 kcal/packet) of Nutriset-produced RUSF supplemented with additional microbiome- and brain-directed nutrients, carotenoids (zeaxanthin, lutein,\u0026nbsp;cryptoxanthin), and soluble fibres (inulin + FOS) \u003csup\u003e45–47\u003c/sup\u003e. Children receiving E-RUSF also received fish oil in gel capsules, while the L-RUSF group received placebo capsules containing corn oil.\u003c/p\u003e\n\u003cp\u003eChildren in both arms received their assigned RUSF daily until they reached the biologically informed recovery threshold (WLZ/WHZ \u0026gt; –1 SD) or for a maximum of 3 months, after which supplementation continued with small-quantity lipid-based nutrient supplements (SQ-LNS; 20 g/day, Nutriset) for the L-RUSF group and enhanced SQ-LNS (ESQ-LNS; 26 g/day) for the E-RUSF group \u003csup\u003e17\u003c/sup\u003e. These supplements were continued until three years of age (two years of follow-up), delivered daily by study staff for the first month and then daily (L-RUSF) or twice weekly (E-RUSF) thereafter. Mothers were provided with feeding guidance during home visits.\u003c/p\u003e\n\u003cp\u003eThe enhanced interventions E-RUSF (Enhanced Ready-to-Use Supplementary Food; developed and manufactured by Nutriset) and ESQ-LNS (enhanced Small Quantity Lipid-based Nutrient Supplements; developed and manufactured by Nutriset) were both formulated to provide key nutrients required for rehabilitating wasted brains (\u003cstrong\u003eFig. S5\u003c/strong\u003e). Each supplement contains 24 essential micronutrients (vitamins and minerals) at recommended daily allowance levels, recommended levels of functional lipids including long-chain polyunsaturated fatty acids (DHA and EPA), neural-specific antioxidants (zeaxanthin, lutein, and cryptoxanthin), and a microbiome-modulating soluble dietary fibre mix (inulin and FOS) amounting to 6 g per 26 g daily dose of ESQ-LNS and within the 92 g daily ration of E-RUSF.\u003c/p\u003e\n\u003ch2\u003eRecruitment and anthropometric data collection\u003c/h2\u003e\n\u003cp\u003eEnrolment was conducted from February 2022 to December 2022 and follow-up at two years and three years was completed in December 2024. Study surveillance workers (SWs) conducted a door-to-door census (approximately 100,000 households) in Mirpur DNCC wards 2, 3 and 5 between February and December 2022. Verbal consent was obtained to participate in the census. The census identified 5,736 children aged between 11 and 13 months and 2,314 children aged between 34 and 38 months. During the census, if the guardian verbally consented to the study procedure, and the babies met the inclusion and exclusion criteria of the study, the surveillance workers proceeded to measure the mid-upper arm circumference (MUAC) of the child. Mothers of babies who were within the MUAC range (MUAC \u0026lt;12.5 and ≥11.5 cm) were invited to visit the icddr,b study clinic for further assessment and enrolment. Final screening for eligibility was based on weight-for-length/weight-for height (WLZ/WHZ) z-scores and study consent occurred at the icddr,b Mirpur study clinic. The consenting process was tailored to each mother’s literacy level and involved reviewing the inclusion and exclusion criteria. Comprehension of the study was assessed using scripted points and open-ended questions. Following consent, the clinical screening team completed a screening form, capturing the date of enrolment, sex, date of birth (DOB), weight (kg), length/height (cm), head circumference (cm), and MUAC (cm) measurements of the child. The weight-for-length/weight-for height (WLZ/WHZ) z-score for each child was calculated using the WHO\u0026nbsp;anthropometric calculator (v3.2.2). The child’s age was validated using the EPI vaccination card.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eBiological sample collection\u003c/h2\u003e\n\u003cp\u003eStool samples were collected from each infant at their home at the baseline visit. Samples were collected in DNA/RNA Shield Fecal Collection Tubes (Zymo Research, #R1101). Peripheral venous blood samples were collected in EDTA Vacutainers, separated into plasma and RBCs and immediately frozen at -80 °C. Batches of blood and stool samples were air-freighted on dry ice from Bangladesh to the Liggins Institute, New Zealand, for processing and analysis.\u003c/p\u003e\n\u003ch2\u003eMicrobiome DNA extraction and sequencing\u003c/h2\u003e\n\u003cp\u003eStool samples (1 mL) were mechanically lysed in bead bashing tubes using the MiniG tissue homogenizer prior to extraction of DNA. DNA was extracted from post-bead bashing sample (200 µL) using the ZymoBIOMICS MagBead DNA/RNA extraction kit (Zymo Research, #R2136) following the standard protocol. DNA was collected in DNAse/RNAse Free Water (50 µL). Repeat extractions were performed for samples with a DNA concentration \u0026lt; 14.5 ng/μL. Sequencing libraries were prepared and sequenced (Illumina NovaSeq 150PE reads) to an average sequencing depth of 20M read-pairs/sample by Custom Science.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eMetagenomic analysis\u003c/h2\u003e\n\u003cp\u003eMetagenomics samples were processed as described by Portlock et al., 2025. Raw DNA sequences were processed using BioBakery3 with default parameters tools \u003csup\u003e48\u003c/sup\u003e. Read quality filtering and human decontamination were performed using KneadData (Version 1). Taxonomic profiling was performed using MetaPhlAn4. Functional profiling was achieved using presence/absence and abundance of microbial pathways (MetaCyc) with HUMAnN3 (Version 3.6). A minimum threshold of \u0026gt; 0.1% relative abundance and \u0026gt; 5% prevalence for all detected species was applied.\u003c/p\u003e\n\u003ch2\u003eLow-pass whole-genome sequencing\u003c/h2\u003e\n\u003cp\u003eGenotype data for the Bangladesh child cohort were generated by Gencove Ltd. through sequencing and imputation. The genotype data for children with moderate acute malnutrition (MAM) were received on 12 October 2023, and for well-nourished controls on 21 March 2024. Quality control and data cleaning were performed prior to downstream PRS analyses. Specifically, SNPs located on chromosomes 24, 25, and 26 were removed, and variants with heterozygous calls in haploid regions were excluded. Samples with \u0026gt; 5% missing genotype calls were discarded, and SNPs with \u0026gt; 5% missingness across individuals were also removed. After quality control, 299 MAM and 145 well-nourished samples remained for subsequent PRS calculations.\u003c/p\u003e\n\u003ch2\u003eCalculation of polygenic risk scores\u003c/h2\u003e\n\u003cp\u003ePolygenic risk scores (PRSs) were calculated for eight phenotypes relevant to metabolic and neuropsychiatric outcomes, including non-alcoholic fatty liver disease (NAFLD) \u003csup\u003e49\u003c/sup\u003e, attention-deficit/hyperactivity disorder (ADHD) \u003csup\u003e50\u003c/sup\u003e, Crohn’s disease (IBD_CD) \u003csup\u003e51\u003c/sup\u003e, ulcerative colitis (IBD_UC) \u003csup\u003e51\u003c/sup\u003e, body mass index (BMI) \u003csup\u003e52\u003c/sup\u003e, height \u003csup\u003e53\u003c/sup\u003e, intelligence \u003csup\u003e54\u003c/sup\u003e, and educational attainment \u003csup\u003e55\u003c/sup\u003e. For each phenotype, GWAS summary statistics were obtained from the corresponding published studies \u003csup\u003e56\u003c/sup\u003e. From each summary statistics dataset, only SNPs with association p-values below 1 × 10⁻⁶ and present in the Bangladesh child cohort genotype data were retained. SNP identifiers, effect alleles, and effect sizes were extracted to create effect size files. When effect sizes were reported as odds ratios (OR), they were converted to beta coefficients using the natural logarithm (β = log [OR]).\u003c/p\u003e\n\u003cp\u003ePRSs were then computed for the Bangladesh child cohort using PLINK v1.9 \u003csup\u003e57\u003c/sup\u003e. The --score function was applied with the curated effect size files to generate individual-level PRSs. Separate analyses were performed for well-nourished children and those with moderate acute malnutrition (MAM). Finally, PRS values derived from the well-nourished and MAM datasets were merged into a single dataset to generate a unified PRS file for downstream statistical analyses.\u003c/p\u003e\n\u003ch2\u003ePlasma lipidomics\u003c/h2\u003e\n\u003cp\u003ePlasma lipidomics samples were processed as described by Portlock \u003cem\u003eet al.,\u0026nbsp;\u003c/em\u003e2025. Briefly, a 10 μL aliquot of the sample was placed in an amber glass autosampler vial and 300 μL of a mixture of Type 1 water, butanol, methanol, chloroform and SPLASH Lipidomix in a ratio of 4:15:15:20:1 was added. The mixture was vortexed and sonicated at room temperature before the protein precipitate was removed by centrifugation and an aliquot of supernatant transferred to an amber glass autosampler for negative ionisation LC-MS/MS. A second aliquot of supernatant was diluted 5 times with 75% IPA for positive ionisation LC-MS/MS. A 5 µL volume of each sample was injected onto a Phenomenex Kinetex F5 column (100 mm × 2.1 mm × 2.6 m) and lipids were separated using a ternary gradient of Type 1 water, methanol and isopropanol containing ammonium acetate. Lipids were quantified and identified with a Q-Exactive mass spectrometer (Thermo Fisher Scientific, Germany) equipped with a heated electrospray ionisation HESI source. Data was processed using MS-DIAL v4.92 \u003csup\u003e58\u003c/sup\u003e.\u003c/p\u003e\n\u003ch2\u003ePlasma amino acids and B-vitamins\u003c/h2\u003e\n\u003cp\u003eThe sample preparation and liquid chromatography - tandem mass spectrometry (LCMSMS) analysis of plasma B-vitamers was carried out using a method published previously \u003csup\u003e59\u003c/sup\u003e. The following vitamers were measured: pantothenic acid (B5), 4-pyridoxic acid (B6), nicotinamide (B3), trimethylamine N-oxide (TMAO), riboflavin (B2), and thiamine (B1). Three plasma quality pools controls were used to monitor analyte recovery, inter- and intra-assay reproducibility. All analytes were recovered between 85 and 105%, with average inter- and intra assay coefficients of variation all less than 8%. \u0026nbsp;Additional B-vitamers were included in the assay but the plasma results were all below the limits of detection. B-vitamers and Limits of Detection (LOD) (nM) are nicotinuric acid (12.5), pyridoxal (15), pyridoxine (2), folic acid (10) and pyridoxamine (2).\u003c/p\u003e\n\u003cp\u003eFree amino acids were assayed from 20 µl of plasma with 15 µM L-Nor-Valine as internal standard by Ultra High-Pressure Liquid Chromatography \u003csup\u003e60\u003c/sup\u003e. Average intra-assay %CV for quality controls across all amino acids was less than 10%, inter-assay %CV less than 12%.\u003c/p\u003e\n\u003cp\u003eFor sample preparation, all liquid addition steps were carried out on an EpMotion 5075vt workstation (Eppendorf AG, Hamburg, Germany) equipped with single and eight channel pipetting tools of various volumes (50-1000 µL). Samples (up to 82 per 96-well plate), and plasma quality controls (5 per plate) were precipitated prior to derivatization using 160 µl of 0.04 M sulphuric acid containing the internal standard and then 20 µl 10% sodium tungstate. To complete the precipitation process, the plate was sealed, shaken for three minutes (4 °C) and then centrifuged at 1000 g for 10 minutes (Avanti J-15R, Beckman Coulter, Nyon, Switzerland). Twenty microlitres of pre-prepared standards (5 different concentrations across the physiological range), aqueous quality controls (3 per plate) and each of the protein free supernatants were then transferred to a new deep-well plate in a programmed pattern for derivatization. A 140 µl aliquot of 0.2 M borate buffer (pH 8.8) was then added to each well followed by 20 µl of tagging reagent, 6-Aminoquinolyl-N-Hydroxysuccinimidyl Carbamate (AQC) (2.8 mg/ml in acetonitrile). The plate was then sealed and heated at 55 °C for 10 minutes to complete the tagging reaction, then transferred to the UPLC system for analysis.\u003c/p\u003e\n\u003ch2\u003eStatistical analyses\u003c/h2\u003e\n\u003cp\u003ePython (Version 3.9.2) was used to perform all analyses unless otherwise specified. Principal Co-ordinate Analysis (PCoA) ordinations (plotted using ‘skbio.stats.ordination.pcoa’ module) were used to visualise the clustering of the Bray-Curtis dissimilarities (calculated using ‘skbio.distance.pdist’) between microbiome samples from their species and functional composition. To quantify the variance of the dataset explained by the covariates, PERMANOVA p-values were calculated from those Bray-Curtis Dissimilarities using the ‘permanova’ function from the ‘skbio.stats.distance’ module. Bray-Curtis were also used to capture the temporal dynamics of each dataset from baseline. Numerical associations between species and metadata were measured with Spearman correlation (calculated using ‘spearmanr’ function from ‘scipy.stats’ module), where significance was defined as p-values of \u0026lt; 0.05. Associations between categorical data were measured with Fisher’s Exact test (calculated using ’fisher_exact’ from ’scipy.stats’ module), where significance was defined as p-values of \u0026lt; 0.05.\u003c/p\u003e\n\u003cp\u003eAll mixed modelling, differential analysis was performed using Maaslin3 (ver 0.99.16)\u0026nbsp;\u003csup\u003e61\u003c/sup\u003e. Parameters for the modelling were default unless otherwise specified and were specific to each datatype, executed by ‘maaslin3_baseline.sh’ and ‘maaslin3_healthy.sh’ scripts on the projects Github repository.\u003cbr\u003e\u0026nbsp;Redundancy analysis (RDA) was used to quantify the proportion of multivariate variation in the response data explained by measured covariates and was implemented in R using the vegan package (v2.7-1). Response and explanatory matrices were matched by sample identifier, and samples with missing values in either the response matrix or the specified explanatory variables were excluded using complete-case filtering. Explanatory variables were specified a priori as fixed effects; variables with fewer than two unique levels were removed and character variables were coerced to factors prior to analysis. Full RDA models were fitted using rda(), and statistical significance of the overall constrained model was assessed using permutation-based analysis of variance (anova.cca) with 999 permutations. Marginal effects of individual explanatory variables, conditional on all other terms in the model, were evaluated using permutation tests with anova.cca(..., by = \"margin\"). Model fit was summarised using the coefficient of determination (R²) and adjusted R² calculated via RsquareAdj(), with unconstrained variance defined as 1 − R².\u003c/p\u003e\n\u003ch2\u003eDigital Twin Model Development and Counterfactual Analysis\u003c/h2\u003e\n\u003cp\u003eDeep learning was used to generate a digital twin for simulating personalised interventions. To manage the high dimensionality of the gut microbiome data, principal component analysis (PCA) was applied to the species-level abundance data from both baseline (one-year) and 52-week (two-year) timepoints. The top ten PCs were kept.\u003c/p\u003e\n\u003cp\u003eA feature matrix was constructed by integrating these microbiome PCs with the other datasets used throughout the analyses presented in this paper (Table S4). To handle missing data, continuous variables were imputed using a multivariate iterative imputer (scikit-learn's ‘IterativeImputer’), while binary and categorical variables were imputed using the most frequent value (scikit-learn's ‘SimpleImputer’) \u003csup\u003e62\u003c/sup\u003e. All continuous features were standardised using a standard scaler (i.e. z-scaled; sci-kit-learn's ‘StandardScaler’).\u003c/p\u003e\n\u003cp\u003eA variational auto-encoder (VAE) was then trained on this pre-processed dataset. The VAE, composed of an encoder and decoder with a 64-dimensional latent space and constructed using PyTorch Lightning \u003csup\u003e63\u003c/sup\u003e, was optimised to reconstruct the input data by minimising a composite loss function of mean squared error (MSE) and the Kullback-Leibler (KL) divergence to create a low-dimensional representation of the data.\u003c/p\u003e\n\u003cp\u003eThe data was split into training (70%), validation (15%), and testing (15%) sets. A separate neural network regression head was then appended to the frozen, pre-trained VAE encoder. This predictive head was trained to predict the WLZ/WHZ at two years (52 weeks after baseline), with the objective of minimising the MSE on the validation dataset.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe final, trained model was used to generate counterfactuals. All individuals with a predicted WLZ/WHZ below the biologically informed recovery threshold of -1 SD were identified. A gradient-based optimisation algorithm was then used to identify the minimal changes to the modifiable baseline features that would raise their predicted WLZ/WHZ to -1 SD.\u003c/p\u003e\n\u003cp\u003eIndividual intervention strategies were hierarchically clustered using Ward distance (seaborn’s ‘clustermap’) \u003csup\u003e64\u003c/sup\u003e. To apply average intervention strategies to all individuals, a ‘consensus threshold’ of 0.4 was chosen, meaning that the feature would only be increased or decreased if at least 40% of individuals required an increase or decrease, respectively, of that feature. A consensus threshold of 0.4 was also used when applying cluster-average interventions. To account for high intra-cluster variance in the required changes, any features with a suggested increase were updated by the average plus 1.5 SD.\u003c/p\u003e\n\u003cp\u003eA pairwise comparison of average PRSs between clusters was undertaken. Fisher and Mann-Whitney U tests were used to compare categorical and continuous metadata variables, respectively, between the two clusters. Benjamini-Hochberg (BH) was used for controlling the False Discovery Rate (FDR) for all cluster analyses.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eCode availability\u003c/p\u003e\n\u003cp\u003eAll analysis code is freely available on the GitHub repository https://github.com/theoportlock/recovery. Documentation for reproduction is provided in the repository\u0026rsquo;s README file.\u003c/p\u003e\n\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eEthical approvals were obtained from the Research Review Committee (RRC; August 21, 2021) and Ethical Review Committee (ERC) of icddr,b (protocol no: PR-21084; September 21, 2021), Institutional Review Board of Boston Children\u0026rsquo;s Hospital, USA (for analyses of neuropsychological assessments), University of Auckland, New Zealand (approval AH23922; for analyses of collected biological samples) and University of West Indies (CREC-MN.51, 21/22).\u003c/p\u003e\n\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eAll data supporting this study are openly available and deposited on Figshare at (https://doi.org/10.17608/k6.auckland.25560768)\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eMamane Zeilani is an employee of Nutriset. The remaining authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eWork on this clinical trial is supported by Wellcome Leap (9942 Culver Blvd Unit 1277 Culver City, CA 90232-4167, United States; www.wellcomeleap.org) to PDG, JMO, TF and CAN as part of the 1kD Program. We acknowledge our core donors, the Governments of Bangladesh and Canada, for providing unrestricted support and commitment to icddr,b\u0026rsquo;s research effort.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eTP, CM, and JOS drafted and co-wrote the manuscript. DH, CP, HHP, ET, NR, TS, SHK, PDG, MZ, TF, RH, CAN commented on the manuscript. JMO, RH, TF, PDG, CAN designed the study. TSand SHK performed assessments and obtained samples in Dhaka. RH oversaw the Dhaka group. TP performed multi-omic analyses, CM performed digital twin analyses, IS, and HHP performed metagenomics, CP and ET performed metabolomics, DH performed genetic analyses. JOS oversaw the Auckland group. CAN oversaw the Boston group.\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eThe authors would like to acknowledge the participants in Mirpur, Dhaka, Bangladesh for their contributions to this study. The authors would also like to thank the study team within the Infectious Diseases Division, International Centre for Diarrheal Disease Research, Bangladesh for their work in participant recruitment, sample collection and assessments.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003e(\u0026lrm;UNICEF), U. N. C. F. \u0026amp; Bank, W. H. O. \u0026amp; W. The UNICEF-WHO-World Bank Joint Child Malnutrition Estimates (JME) Standard Methodology Tracking Progress on SDG Indicators 2.2.1 on Stunting, 2.2.2 (1) on Overweight and 2.2.2 (2) on Wasting. (UNICEF-WHO, 2024).\u003c/li\u003e\n\u003cli\u003eOrganization, W. H., (UNICEF), U. N. C. F. \u0026amp; Bank, W. Levels and Trends in Child Malnutrition: UNICEF / WHO / The World Bank Group Joint Child Malnutrition Estimates: Key Findings of the 2021 Edition. (World Health Organization, Geneva, 2021).\u003c/li\u003e\n\u003cli\u003eWHO Guideline on the Prevention and Management of Wasting and Nutritional Oedema (acute Malnutrition) in Infants and Children under 5 Years. 208 (2024).\u003c/li\u003e\n\u003cli\u003eKirolos, A. et al. Neurodevelopmental, cognitive, behavioural and mental health impairments following childhood malnutrition: a systematic review. BMJ Glob Health 7, (2022).\u003c/li\u003e\n\u003cli\u003eLagrone, L., Cole, S., Schondelmeyer, A., Maleta, K. \u0026amp; Manary, M. J. Locally produced ready-to-use supplementary food is an effective treatment of moderate acute malnutrition in an operational setting. Ann Trop Paediatr 30, 103\u0026ndash;108 (2010).\u003c/li\u003e\n\u003cli\u003eTeshome, M. S. et al. Current evidence on the effectiveness of Ready-to-Use Supplementary Foods in children with moderate acute malnutrition: a systematic review and meta-analysis. J Nutr Sci 12, (2024).\u003c/li\u003e\n\u003cli\u003eTap\u0026iacute;a, M. S., Alzamora, S. M. \u0026amp; Chirife, J. Effects of Water Activity (a w ) on Microbial Stability as a Hurdle in Food Preservation. Water Activity in Foods: Fundamentals and Applications 323\u0026ndash;355 (2020) doi:10.1002/9781118765982.CH14.\u003c/li\u003e\n\u003cli\u003eSteenkamp, L., Lategan, R. \u0026amp; Raubenheimer, J. The impact of Ready-to-Use Supplementary Food (RUSF) in targeted supplementation of children with moderate acute malnutrition (MAM) in South Africa. South African Family Practice 57, 322\u0026ndash;325 (2015).\u003c/li\u003e\n\u003cli\u003eHeymsfield, G. et al. Effectiveness of acute malnutrition treatment with a simplified, combined protocol in Central African Republic: An observational cohort study. Matern Child Nutr 20, e13691 (2024).\u003c/li\u003e\n\u003cli\u003eKatsoulakis, E. et al. Digital twins for health: a scoping review. NPJ Digit Med 7, 1\u0026ndash;11 (2024).\u003c/li\u003e\n\u003cli\u003eSizemore, N. et al. A digital twin of the infant microbiome to predict neurodevelopmental deficits. Sci Adv 10, 400 (2024).\u003c/li\u003e\n\u003cli\u003eLi, X. et al. A dynamic single cell-based framework for digital twins to prioritize disease genes and drug targets. Genome Med 14, 1\u0026ndash;21 (2022).\u003c/li\u003e\n\u003cli\u003eBarbiero, P., Vi\u0026ntilde;as Torn\u0026eacute;, R. \u0026amp; Li\u0026oacute;, P. Graph Representation Forecasting of Patient\u0026rsquo;s Medical Conditions: Toward a Digital Twin. Front Genet 12, 652907 (2021).\u003c/li\u003e\n\u003cli\u003eAhmed, T., Michaelsen, K. F., Frem, J. C. \u0026amp; Tumvine, J. Malnutrition: Report of the FISPGHAN working group. J Pediatr Gastroenterol Nutr 55, 626\u0026ndash;631 (2012).\u003c/li\u003e\n\u003cli\u003eShama, T. et al. Multidimensional evaluation of the early emergence of executive function and development in Bangladeshi children using nutritional and psychosocial intervention: A randomized controlled trial protocol. PLoS One 19, (2024).\u003c/li\u003e\n\u003cli\u003eYeasmin, K. \u0026amp; Yeasmin, T. Assessment of Nutritional Status of Preschool Children: Head Circumference and Other Anthropometric Indices. Bangladesh Med Res Counc Bull 44, 152\u0026ndash;159 (2018).\u003c/li\u003e\n\u003cli\u003eFahur Bottino, G. et al. Early life microbial succession in the gut follows common patterns in humans across the globe. Nature Communications 16, 1\u0026ndash;12 (2025).\u003c/li\u003e\n\u003cli\u003eHamadani, J. D. et al. Integrating an early childhood development programme into Bangladeshi primary health-care services: an open-label, cluster-randomised controlled trial. Articles Lancet Glob Health 7, 366\u0026ndash;75 (2019).\u003c/li\u003e\n\u003cli\u003eGoel, P. \u0026amp; Goel, A. Exploring the Evolution of Sleep Patterns From Infancy to Adolescence. Cureus 16, (2024).\u003c/li\u003e\n\u003cli\u003eMakori, N. et al. Efficacy of ready-to-use food supplement for treatment of moderate acute malnutrition among children aged 6 to 59 months. Matern Child Nutr 20, (2024).\u003c/li\u003e\n\u003cli\u003eMedoua, G. N. et al. Recovery rate of children with moderate acute malnutrition treated with ready-to-use supplementary food (RUSF) or improved corn\u0026ndash;soya blend (CSB+): a randomized controlled trial. Public Health Nutr 19, 363\u0026ndash;370 (2016).\u003c/li\u003e\n\u003cli\u003eRashid, M. Y., Kebira, J. Y., Oljira, L. \u0026amp; Dheresa, M. Time to Recovery From Moderate Acute Malnutrition and Its Predictors Among Children 6\u0026ndash;59 Months of Age Enrolled in Targeted Supplementary Feeding Program in Darolebu District, Eastern Ethiopia: A Retrospective Cohort Study. Front Public Health 10, 914837 (2022).\u003c/li\u003e\n\u003cli\u003eGriswold, S. P. et al. Effectiveness and cost-effectiveness of 4 supplementary foods for treating moderate acute malnutrition: results from a cluster-randomized intervention trial in Sierra Leone. Am J Clin Nutr 114, 973\u0026ndash;985 (2021).\u003c/li\u003e\n\u003cli\u003eChen, R. Y. et al. A Microbiota-Directed Food Intervention for Undernourished Children. New England Journal of Medicine 384, 1517\u0026ndash;1528 (2021).\u003c/li\u003e\n\u003cli\u003eKordas, K. et al. Maternal reports of sleep in 6-18 month-old infants from Nepal and Zanzibar: Association with iron deficiency anemia and stunting. Early Hum Dev 84, 389\u0026ndash;398 (2008).\u003c/li\u003e\n\u003cli\u003eWatanabe, F., Yabuta, Y., Bito, T. \u0026amp; Teng, F. Vitamin B12-Containing Plant Food Sources for Vegetarians. Nutrients 2014, Vol. 6, Pages 1861-1873 6, 1861\u0026ndash;1873 (2014).\u003c/li\u003e\n\u003cli\u003eBr\u0026ouml;er, S. \u0026amp; Gauthier-Coles, G. Amino Acid Homeostasis in Mammalian Cells with a Focus on Amino Acid Transport. Journal of Nutrition 152, 16\u0026ndash;28 (2022).\u003c/li\u003e\n\u003cli\u003eParra, M., Stahl, S. \u0026amp; Hellmann, H. Vitamin B6 and Its Role in Cell Metabolism and Physiology. Cells 2018, Vol. 7, Page 84 7, 84 (2018).\u003c/li\u003e\n\u003cli\u003eTardy, A. L., Pouteau, E., Marquez, D., Yilmaz, C. \u0026amp; Scholey, A. Vitamins and Minerals for Energy, Fatigue and Cognition: A Narrative Review of the Biochemical and Clinical Evidence. Nutrients 2020, Vol. 12, Page 228 12, 228 (2020).\u003c/li\u003e\n\u003cli\u003eCamillo, L., Zavattaro, E. \u0026amp; Savoia, P. Nicotinamide: A Multifaceted Molecule in Skin Health and Beyond. Medicina 2025, Vol. 61, Page 254 61, 254 (2025).\u003c/li\u003e\n\u003cli\u003eWunderle, C. et al. Association of leucine and other branched chain amino acids with clinical outcomes in malnourished inpatients: a secondary analysis of the randomized clinical trial EFFORT. European Journal of Clinical Nutrition 2024 79:1 79, 42\u0026ndash;49 (2024).\u003c/li\u003e\n\u003cli\u003eHoleček, M. Branched-chain amino acids in health and disease: metabolism, alterations in blood plasma, and as supplements. Nutrition \u0026amp; Metabolism 2018 15:1 15, 1\u0026ndash;12 (2018).\u003c/li\u003e\n\u003cli\u003eWu, G. et al. Proline and hydroxyproline metabolism: implications for animal and human nutrition. Amino Acids 2010 40:4 40, 1053\u0026ndash;1063 (2010).\u003c/li\u003e\n\u003cli\u003eFaure, M. et al. Specific Amino Acids Increase Mucin Synthesis and Microbiota in Dextran Sulfate Sodium\u0026ndash;Treated Rats. J Nutr 136, 1558\u0026ndash;1564 (2006).\u003c/li\u003e\n\u003cli\u003eWu, D. et al. Proline uptake promotes activation of lymphoid tissue inducer cells to maintain gut homeostasis. Nat Metab 5, 1953\u0026ndash;1968 (2023).\u003c/li\u003e\n\u003cli\u003eMatsumoto, S. et al. Urea cycle disorders\u0026mdash;update. J Hum Genet 64, 833\u0026ndash;847 (2019).\u003c/li\u003e\n\u003cli\u003eOkun, J. G. et al. Liver alanine catabolism promotes skeletal muscle atrophy and hyperglycaemia in type 2 diabetes. Nature Metabolism 2021 3:3 3, 394\u0026ndash;409 (2021).\u003c/li\u003e\n\u003cli\u003ePerrott, S. L. \u0026amp; Kar, S. P. Polygenic risk scores for genomics and population screening. The Lancet 404, 935\u0026ndash;936 (2024).\u003c/li\u003e\n\u003cli\u003eSlunecka, J. L. et al. Implementation and implications for polygenic risk scores in healthcare. Hum Genomics 15, 1\u0026ndash;18 (2021).\u003c/li\u003e\n\u003cli\u003eDu Rietz, E. et al. Association of Polygenic Risk for Attention-Deficit/Hyperactivity Disorder With Co-occurring Traits and Disorders. Biol Psychiatry Cogn Neurosci Neuroimaging 3, 635\u0026ndash;643 (2018).\u003c/li\u003e\n\u003cli\u003eWang, B., Yang, Y., Yin, Z. \u0026amp; Yang, W. The causal impact of body mass index on metabolic biomarkers and nonalcoholic fatty liver disease risk. Sci Rep 15, 1\u0026ndash;42 (2025).\u003c/li\u003e\n\u003cli\u003eGangwal, A. \u0026amp; Lavecchia, A. Artificial intelligence in preclinical research: enhancing digital twins and organ-on-chip to reduce animal testing. Drug Discov Today 30, 104360 (2025).\u003c/li\u003e\n\u003cli\u003eAhmed, T. et al. Development and acceptability testing of ready-to-use supplementary food made from locally available food ingredients in Bangladesh. BMC Pediatr 14, 1\u0026ndash;8 (2014).\u003c/li\u003e\n\u003cli\u003eObelitz-Ryom, K. et al. Bovine Milk Oligosaccharides with Sialyllactose Improves Cognition in Preterm Pigs. Nutrients 2019, Vol. 11, Page 1335 11, 1335 (2019).\u003c/li\u003e\n\u003cli\u003eCusick, S. E. \u0026amp; Georgieff, M. K. The Role of Nutrition in Brain Development: The Golden Opportunity of the \u0026ldquo;First 1000 Days\u0026rdquo;. J Pediatr 175, 16 (2016).\u003c/li\u003e\n\u003cli\u003eCharbonneau, M. R. et al. Sialylated Milk Oligosaccharides Promote Microbiota-Dependent Growth in Models of Infant Undernutrition. Cell 164, 859\u0026ndash;871 (2016).\u003c/li\u003e\n\u003cli\u003eBeghini, F. et al. Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with biobakery 3. Elife 10, (2021).\u003c/li\u003e\n\u003cli\u003eNamjou, B. et al. GWAS and enrichment analyses of non-alcoholic fatty liver disease identify new trait-associated genes and pathways across eMERGE Network. BMC Med 17, 1\u0026ndash;19 (2019).\u003c/li\u003e\n\u003cli\u003eDemontis, D. et al. Discovery of the first genome-wide significant risk loci for attention deficit/hyperactivity disorder. Nat Genet 51, 63\u0026ndash;75 (2019).\u003c/li\u003e\n\u003cli\u003eSun, Y. et al. The Contribution of Genetic Risk and Lifestyle Factors in the Development of Adult-Onset Inflammatory Bowel Disease: A Prospective Cohort Study. Am J Gastroenterol 118, 511\u0026ndash;522 (2023).\u003c/li\u003e\n\u003cli\u003eHoffmann, T. J. et al. A Large Multiethnic Genome-Wide Association Study of Adult Body Mass Index Identifies Novel Loci. Genetics 210, 499\u0026ndash;515 (2018).\u003c/li\u003e\n\u003cli\u003eAllen, H. L. et al. Hundreds of variants clustered in genomic loci and biological pathways affect human height. Nature 467, 832\u0026ndash;838 (2010).\u003c/li\u003e\n\u003cli\u003eDavies, G. et al. Study of 300,486 individuals identifies 148 independent genetic loci influencing general cognitive function. Nat Commun 9, 1\u0026ndash;16 (2018).\u003c/li\u003e\n\u003cli\u003eLee, J. J. et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nat Genet 50, 1112\u0026ndash;1121 (2018).\u003c/li\u003e\n\u003cli\u003eSoh, S. E. et al. Cohort profile: Growing Up in Singapore Towards healthy Outcomes (GUSTO) birth cohort study. Int J Epidemiol 43, 1401\u0026ndash;1409 (2014).\u003c/li\u003e\n\u003cli\u003ePurcell, S. et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet 81, 559 (2007).\u003c/li\u003e\n\u003cli\u003eTsugawa, H. et al. MS-DIAL: Data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nat Methods 12, 523\u0026ndash;526 (2015).\u003c/li\u003e\n\u003cli\u003eAndraos, S. et al. Plasma B Vitamers: Population Epidemiology and Parent-Child Concordance in Children and Adults. Nutrients 2021, Vol. 13, Page 821 13, 821 (2021).\u003c/li\u003e\n\u003cli\u003eMilan, A. M. et al. The impact of heat treatment of bovine milk on gastric emptying and nutrient appearance in peripheral circulation in healthy females: a randomized controlled trial comparing pasteurized and ultra-high temperature milk. Am J Clin Nutr 119, 1200\u0026ndash;1215 (2024).\u003c/li\u003e\n\u003cli\u003eNickols, W. A. et al. MaAsLin 3: Refining and extending generalized multivariable linear models for meta-omic association discovery. bioRxiv https://doi.org/10.1101/2024.12.13.628459 (2024) doi:10.1101/2024.12.13.628459.\u003c/li\u003e\n\u003cli\u003ePedregosa FABIANPEDREGOSA, F. et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 12, 2825\u0026ndash;2830 (2011).\u003c/li\u003e\n\u003cli\u003ePaszke, A. et al. PyTorch: An Imperative Style, High-Performance Deep Learning Library. Adv Neural Inf Process Syst 32, (2019).\u003c/li\u003e\n\u003cli\u003eWaskom, M. L. seaborn: statistical data visualization. J Open Source Softw 6, 3021 (2021).\u003cbr\u003e \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Moderate acute malnutrition, Digital twins, Multi-omics, Machine learning, Randomized controlled trial, Amino acid metabolism, fecal microbiome, Child growth, Nutritional supplementation, Bangladesh","lastPublishedDoi":"10.21203/rs.3.rs-8585503/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8585503/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eOptimising nutritional strategies to improve growth recovery from moderate acute malnutrition (MAM) remains a global priority. We conducted a community-based, open-label, randomised controlled trial in Bangladesh to evaluate whether an enhanced ready-to-use supplementary food (ERUSF), enriched with prebiotics, Long Chain PolyUnsaturated Fatty Acids (LC-PUFA) and carotenoids, improved recovery outcomes compared with a locally produced RUSF (L-RUSF). Children aged 12 ± 1 months with MAM were treated with RUSF or ERUSF for up to three months or until recovery, followed by nine months of maintenance supplementation with standard or enhanced small-quantity lipid-based nutrient supplements (SQ-LNS or ESQ-LNS) in addition to their daily normal complementary diet.\u003c/p\u003e\n\u003cp\u003eUsing WHO criteria (WLZ/WHZ \u0026gt; −2), 69% of participants achieved recovery during the rapid catch-up phase, with no significant difference between intervention arms. When a more stringent, pre-specified biologically informed recovery threshold (WLZ/WHZ \u0026gt; −1) was applied to examine recovery trajectories beyond WHO-defined resolution of wasting, approximately 44% of children in both arms achieved this target. Among those who did recover to this stricter threshold, ERUSF was associated with a significantly shorter time to attainment, indicating more rapid catch-up growth rather than an increased probability of recovery. Relapse rates during the subsequent maintenance phase were low and did not differ between SQ-LNS formulations.\u003c/p\u003e\n\u003cp\u003eTo explore biological correlates of recovery heterogeneity, we integrated anthropometric, microbiome, plasma metabolomic, and genetic data using mixed models and exploratory machine-learning approaches. Baseline anthropometric severity, early weight gain, and polygenic risk profiles were associated with recovery status, whereas sociodemographic variables were largely uninformative. Post-hoc counterfactual digital twin modelling suggested that variation in baseline plasma amino acid and vitamin profiles may partially explain non-recovery under locally produced nutritional protocols. These simulations identified model-consistent feature shifts that, if achievable, were predicted to increase recovery to the WLZ/WHZ \u0026gt; −1 threshold; however, these findings are hypothesis-generating and not evidence of causal or clinically actionable effects.\u003c/p\u003e\n\u003cp\u003eTogether, these results demonstrate that enhanced RUSF accelerates recovery among children who respond but does not increase overall recovery rates under WHO criteria. Persistent heterogeneity in recovery trajectories appears to reflect underlying biological and genetic factors. These are not fully addressed by current supplementation strategies, highlighting the need for future trials explicitly designed to test biologically stratified nutritional interventions.\u003c/p\u003e","manuscriptTitle":"Optimising recovery from childhood moderate acute malnutrition: a randomized controlled trial","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-29 10:38:42","doi":"10.21203/rs.3.rs-8585503/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"35c10ac0-1f70-482c-b115-5e58f9b8b4b7","owner":[],"postedDate":"January 29th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":61157515,"name":"Health sciences/Medical research/Clinical trial design/Clinical trials"},{"id":61157516,"name":"Health sciences/Health care/Nutrition"}],"tags":[],"updatedAt":"2026-02-02T23:55:08+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-29 10:38:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8585503","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8585503","identity":"rs-8585503","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}