Early childhood linear growth failure in low- and middle-income countries

Globally 149 million children under five are estimated to be stunted (length more than 2 standard deviations below international growth standards). Stunting, a form of linear growth failure, increases risk of illness, impaired cognitive development, and mortality. Global stunting estimates rely on cross-sectional surveys, which cannot provide direct information about the timing of onset or persistence of growth failure- a key consideration for defining critical windows to deliver preventive interventions. We performed the largest pooled analysis of longitudinal studies in low- and middle-income countries to date (n=31 cohorts, 62,993 children, ages 0-24 months), allowing us to identify the typical age of linear growth failure onset and to investigate recurrent faltering in early life. The highest incidence of stunting onset occurred from birth to age 3 months. From 0 to 15 months, less than 5% of children per month reversed their stunting status, and among those who did, stunting relapse was common. Early timing and low reversal rates emphasize the importance of preventive intervention delivery within the prenatal and early postnatal phases coupled with continued delivery of postnatal interventions through the first 1000 days of life.

There have been marked reductions in child mortality in the past decade, 1 in part due to the 160 emphasis on child survival in the Millennium Development Goals. Yet children born in low-and middle 161 income countries (LMICs) still face enormous challenges to their ability to thrive, including poverty, poor 162 nutrition, poor water and sanitation, poor access to health care and limited stimulation in the home 163 environment. 2 A commonly used, readily measurable indicator of children's overall health and 164 development is their linear growth as measured by a child's length (height or stature). 3 Extensive 165 evidence supports its use as a surrogate for more difficult-to-measure physiological and cognitive 166 deficits that accrue during early life. 4,5 Children in optimal environments have the same growth 167 potential, regardless of their geographic location 6 , and thus the World Health Organization (WHO) has 168 developed international standards for child growth. 7 A child is "stunted" if his or her length is more than 169 2 standard deviations below the median of the growth standard for age and sex. In 2018, 149 million 170 children under 5 years (22% globally) were stunted, with the largest burden in South Asia and Africa. 8,9 171 Early life stunting is associated with increased risk of mortality, 10 diarrhea, pneumonia, and measles in 172 childhood 11,12 and impaired cognition and productivity in adulthood. 3,13,14 Global income would increase 173 by an estimated $176.8 billion per year if linear growth failure could be eliminated. 15 The WHO 2025 174 Global Nutrition Targets 16 and Sustainable Development Goal 2 propose to reduce stunting prevalence 175 from 2012 levels by 40% by 2025, 17 reflecting renewed attention to child growth as a key determinant of 176 overall health and human capital. 18 177 178 The age of stunting onset has direct implications for the timing and nature of preventive public 179 health interventions. In low-resource settings, most linear growth failure occurs in the first 2 years of 180 life, and 70% of absolute length deficits by age 5 years occur before age 2 years. 3 Beyond this age, catch 181 up growth is rare in the absence of radical improvements to a child's nutrition, health, and 182 environment. [19][20][21][22][23] Intrauterine growth restriction and preterm birth are strongly associated with the risk 183 of stunting at 24 months of age, 24 and the first 1000 days of life is considered the critical window in 184 which to intervene to prevent stunting. 6,25 185 186 Granular information about age of linear growth failure onset and its persistence in early life will 187 best inform when and how to intervene with preventive measures. Yet, most studies of the global 188 epidemiology and burden of stunting have pooled nationally representative, cross-sectional surveys -189 predominantly national Demographic Health Surveys (DHS) -to estimate age-specific stunting 190 prevalence. 23,[26][27][28] Cross-sectional studies cannot identify the timing of linear growth failure onset 191 because they measure prevalence, which cannot distinguish between persistent and new cases of 192 growth failure. Few studies have reported stunting incidence, 29,30 and even fewer have estimated age-193 specific incidence within the first two years of life to inform the timing of intervention delivery. [31][32][33] conducted an in-depth study of linear growth failure incidence in a pooled analysis of 31 longitudinal 195 cohorts with multiple, frequent measurements in LMICs. The analysis provides new insights into the 196 timing of onset and duration of linear growth failure, with important implications for interventions. We 197 found that linear growth failure occurs very early in the prenatal and postnatal phase -before the age 198 when most postnatal linear growth interventions begin (age 6 months). Our findings confirm the 199 importance of the first 1,000 days as a critical window to intervene to prevent linear growth failure, but 200 motivate a renewed focus on interventions delivered in the prenatal and early postnatal periods. 201 202 203 204 205 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020 Pooled longitudinal analyses 206 207 Here, we report a pooled analysis of 31 longitudinal cohorts from 15 LMICs in South Asia,  Saharan Africa, and Latin America followed between 1969 and 2014. Our objective was to estimate age-209 specific incidence and prevalence of stunting as well as linear growth velocity from 0 to 24 months. A 210 companion article in this issue reports results for child wasting (weight-for-length Z-score < 2 standard 211 deviations below the reference median). 34 These data were amassed as part of the Knowledge 212 Integration (ki) initiative of the Bill & Melinda Gates Foundation, which aggregated observations on 213 millions of participants from a global collection of studies on child birth, growth and development. 35 We 214 selected longitudinal cohorts from the database that met five inclusion criteria: 1) conducted in LMICs; 215 2) enrolled children between birth and age 24 months and measured their length and weight repeatedly 216 over time; 3) did not restrict enrollment to acutely ill children; 4) enrolled at least 200 children; and 5) 217 collected anthropometry measurements at least every 3 months (Extended Data Fig 1). These inclusion 218 criteria ensured we could rigorously evaluate the timing and onset of stunting among children who were 219 broadly representative of general populations in LMICs. For included randomized trials, if trials found 220 effects on growth within the intervention arms, the analysis was limited to the control arm. Thirty-one 221 cohorts met inclusion criteria, including 62,993 children and 464,345 total measurements (Fig 1). Most 222 cohorts were enrolled within the past 10 years (Extended Data  Fig 2). 228 229 We calculated length-for-age Z-scores (LAZ) using WHO 2006 growth standards. 7 We dropped 969 230 out of 465,314 measurements (0.2%) because LAZ was unrealistic (> 6 or < -6 Z), and we defined 231 stunting as LAZ < -2 and severe stunting as LAZ < -3. 7 Unless otherwise indicated, estimates that pool 232 across cohorts used random effects models fit with restricted maximum likelihood estimation. 36,37 233 Within each cohort the monthly mean LAZ ranged from -3.18 to +1.31, and the monthly proportion 234 stunted ranged from 0.3% to 86.1% (Fig 1). We compared LAZ from included cohorts with those of 235 contemporary population-based, cross-sectional DHS data for children 0-24 months of age in the same 236 countries and regions to assess their representativeness. Cohorts had similar Z-score distributions as 237 their target populations (Fig 2a). Mean LAZ by age was approximately 0.25-0.75 standard deviations 238 lower in included ki cohorts than in DHS surveys in the same countries in South Asia and Africa, and 239 mean LAZ by age was slightly higher in Latin American cohorts (Fig 2b). Analyses therefore include a 240 large set of cohorts from populations with high burdens of growth failure but are not representative of 241 entire world regions. The width of the LAZ distribution was unimodal and similar to that of the WHO 242 child growth standard's LAZ distribution, consistent with the assertion that linear growth failure is a 243 "whole population" condition. 27 244 245 Age-specific patterns of mean LAZ and stunting prevalence were consistent with prior analyses of 246 cross-sectional survey data. 26,27 Most children were born with linear growth deficits: mean LAZ at birth 247 was -0.75 (Fig 2a). LAZ declined steadily from birth, and by 24 months mean LAZ was -1.82. South Asia 248 had lower LAZ than other regions: mean LAZ at birth was -1.04 and the mean at 24 months was -2.01. 249 Yet, the decline in mean LAZ was the largest in Africa, where the mean LAZ at birth was -0.56 and the 250 mean at 24 months was -1.87. Linear growth failure was evident across percentiles of the LAZ 251 distribution; in South Asia, the 95 th percentile of children's LAZ scores dropped below 0 by age 14 252 months (Extended Data Fig 3). At birth, the overall prevalence of stunting (LAZ <-2) was 12%; prevalence 253 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020 increased steadily to 43% at 18 months and remained stable thereafter (Fig 3a). In African cohorts, 254 stunting prevalence increased steadily each month from 10% at birth to 44% at 24 months. In South 255 Asia, 18% of children were born stunted, and stunting prevalence increased steadily to 53% at 18 256 months and then decreased to 48% at 24 months. At most ages, stunting prevalence and incidence were 257 lower in Latin America; wider confidence intervals for this region reflect greater variation between 258 cohorts because stunting was more common in Guatemala than in other countries (https://child-259 growth.github.io/stunting/cohort.html). 260 261 Onset of stunting in early life 262 263 To measure the timing of stunting onset, we classified a child as a new incident case in three-month 264 age periods if their LAZ dropped below -2 for the first time in that age period. Across cohorts, 12% of 265 children were stunted at birth, and 17% experienced incident stunting onset between birth and 3 266 months (Fig 3b); children stunted between birth and 3 months accounted for 40% of all children who 267 experienced stunting by age 24 months (65% of children). Trends were similar for severe stunting 268 (https://child-growth.github.io/stunting/severe-stunting.html). 269 270 In an exploratory analysis, we stratified age-specific mean LAZ by age of stunting onset among 271 children in monthly measured cohorts and observed three distinct subgroups whose mean LAZ followed 272 statistically different trajectories (Fig 3c). 26% of all children were born with linear growth deficits (mean 273 LAZ < -1), and their mean LAZ stabilized around -2 from age 1 month onward, with differences at birth 274 in this group converging through likely regression to the mean. 11% of children were born with small 275 linear growth deficits on average (mean 0 < LAZ < -1 at birth) and reached a mean of -2 at subsequent 276 ages. The remaining 62% of children maintained mean LAZ > -2 at all ages, but mean LAZ declined 277 steadily from birth to 15 months. Overall trends in mean LAZ were consistent with our age-specific 278 incidence and prevalence estimates (Fig 3a-b). Children born with linear growth deficits may benefit 279 more from prenatal interventions, while those whose mean LAZ declined later may benefit from both 280 prenatal and postnatal interventions. A companion article reports characteristics that increase risk of 281 earlier versus later growth failure. 35 282 283 Stunting reversal and relapse 284 285 Stunting prevalence and incidence estimates classify linear growth measures into binary categories, 286 which could mask the dynamics of linear growth failure onset and reversal across the continuous range 287 of LAZ. 5,38,39 We hypothesized that lower than average linear growth (LAZ <0) would persist among 288 children who experienced stunting reversal (i.e., LAZ increased from below -2 to above -2). In addition, 289 we hypothesized that children who experienced stunting reversal would experience stunting relapse at 290 later ages. To test these hypotheses, we classified a child's change in stunting status from birth to 15 291 months among monthly-measured cohorts (measurement frequency beyond 15 months was less 292 consistent) (Fig 4a). Initial onset of stunting occurred before age 6 months for a majority of children. The 293 proportion of children who experienced stunting reversal was less than 5% per month from ages 2 to 7 294 months and was lower at subsequent ages. 295 296 Among children who experienced stunting reversal, we summarized the LAZ distribution at older 297 ages. We then estimated the mean difference in LAZ measured at older ages compared to when stunting 298 was reversed. At the time of stunting reversal, the mode of the distribution of LAZ scores was close to 299 the -2 cutoff (Fig 4b). As children aged, LAZ distributions gradually shifted downwards, illustrating that 300 linear growth deficits continued to accumulate. The mean LAZ decreased by up to 0.71 SDs from 3 to 12 301 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint months after stunting reversal, and the older a child was at the time of reversal, the larger the decline in 302 subsequent LAZ (Fig 4c). Overall, improvements in LAZ among children whose stunting status reversed 303 were neither sustained nor large enough to erase linear growth deficits, indicating that stunting reversal 304 did not resemble a biological recovery process for many children. 305 306 Growth velocity by age and sex 307 308 Age-specific growth velocity provides a complementary view of child growth failure only possible in 309 longitudinal cohorts. We defined linear growth velocity as the change in length between two time points 310 divided by the number of months between the time points (cm/month). We also estimated within-child 311 rates of change in LAZ per month. The change in LAZ per month measures the extent to which a child's 312 length relative to WHO standards for his or her age and sex changed over time. From 0-3 months, length 313 velocity was in between the 15 th and 25 th percentile of the WHO standard for girls and was close to the 314 15 th percentile for boys; velocity for both sexes then improved relative to WHO standards as children 315 aged (Fig 5a). Larger deficits at the youngest ages were consistent with highest incidence of stunting 316 from birth to age 3 months (Fig 3b). LAZ velocity was similar for male and female children (Fig 5b). 317 Patterns were similar across geographic regions (Extended Data Fig 4a, 4b). 318 319 The accumulation of growth velocity deficits was reflected in lower mean LAZ for boys than girls 320 across all ages (Fig 5c). Overall patterns were similar across geographic regions (Extended Data Fig 4c). 321 Worse growth among boys is consistent with higher infant mortality rates among boys found around the 322 world, suggesting that males may be more vulnerable than females in early life, [40][41][42][43][44][45][46] or that evolutionary 323 selection favors females in resource-poor environments. 47 324 325 Discussion 326 327 Estimates of the global burden of stunting and timing of linear growth failure have relied largely on 328 cross-sectional surveys that have indicated the need for interventions targeted to children younger than 329 24 months. 6,25 This large-scale, longitudinal analysis of 31 prospective cohorts from LMICs allowed us to 330 provide new insights into the timing, persistence, and recurrence of linear growth failure within the first 331 2 years of life. Prevalence estimates matched general patterns reported in cross-sectional studies, 332 gradually increasing with age. 23,26-28 Yet, incidence estimates from longitudinal analyses revealed a 333 starkly different age-specific pattern and showed that stunting incidence was highest from birth to 3 334 months and declined thereafter (Fig 3a-b). The vast majority of children who experienced stunting 335 reversal (i.e., crossed above the <-2 LAZ cut-off) continued to experience linear growth deficits, and 336 over 20% were stunted again at later measurements (Fig 4). 337 338 Pre-pregnancy and prenatal risk factors for postnatal linear growth failure include young maternal 339 age, 48 maternal infection, 49 and maternal undernutrition, including low pre-pregnancy body mass index, 340 short stature, and inadequate gestational weight gain. 50 Rigorous evaluations of interventions to address 341 these predictors have been limited to date. Interventions to improve maternal nutrition 51 and reduce 342 adolescent pregnancies in LMICs by delaying the age of marriage and first pregnancy 52 may reduce the 343 risk of linear growth failure. In addition, interventions to reduce prenatal infections associated with 344 growth failure, such as intermittent preventive treatment for malaria, may also increase fetal linear 345 growth in regions where such infections co-occur with linear growth failure. 53 346 347 In the early postnatal phase (age 0-6 months), when growth velocity and stunting incidence were 348 highest, the World Health Organization recommends exclusive breastfeeding to prevent infant infection 349 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint and to support child growth. 54 Exclusive breastfeeding significantly reduces the risk of mortality and 350 morbidity, but exclusive breastfeeding promotion has not been found to reduce the risk of infant 351 stunting. 11,55-58 The extent to which exclusive breastfeeding impacts linear growth may depend on 352 maternal breast milk composition and microbiota, both of which may be impacted by maternal 353 undernutrition. 59-61 In the complementary feeding phase (age 6-24 months), interventions to reduce 354 growth failure include energy, protein, and micronutrient supplementation of children; deworming; 355 water and sanitation improvements; vaccination; and maternal education. However, meta-analyses 356 evaluating the effectiveness of interventions during the complementary feeding phase on stunting 357 prevalence and mean LAZ have reported modest impacts of lipid-based nutrient supplements, 62 modest 358 or no impact of micronutrient supplementation, 63 and no impact of water and sanitation improvements, 359 deworming, or maternal education. 63 The dearth of postnatal interventions that can effectively improve 360 child linear growth motivates renewed efforts to identify alternative, possibly multisectoral 361 interventions, and in parallel improve intervention targeting and implementation. 64,65 362 363 There were several limitations to the analyses. First, length estimates may be subject to 364 measurement error; stunting reversal and relapse analyses that rely on thresholds are more sensitive to 365 such errors. However, detailed assessments of measurement quality indicated that measurement 366 quality was high across cohorts (https://child-growth.github.io/stunting/QA.html). Second, estimates of 367 LAZ at birth using the WHO Child Growth Standards overestimate stunting among children who are born 368 preterm. 66 Accurate estimates of gestational age were not available in included cohorts; seven cohorts 369 measured gestational age by recall of last menstrual period or newborn examination, and one cohort 370 measured gestational age by ultrasound. In a sensitivity analysis adjusting for gestational age pooling 371 across cohorts that measured it, stunting prevalence at birth was 1% lower (Extended Data Fig 5). Third, 372 included cohorts were not inclusive of all countries in the regions presented here, and linear growth 373 failure was more common in included African and South Asian cohorts than in corresponding 374 contemporary representative surveys. However, the consistency between linear growth patterns in this 375 and nationally representative surveys (Fig 2) suggests that overall our results have good external 376 validity. Finally, the included cohorts measured child length every 1-3 months, and the exact ages of 377 measurement varied, so the number of children and cohorts contributing to estimates differs between 378 analysis strata. However, when we repeated analyses in cohorts with monthly measurements from birth 379 to 24 months (n=18 cohorts in 10 countries, 10,830 children), results were similar (https://child-380 growth.github.io/stunting/monthly.html). 381 382 Conclusion 383 The WHO Global Nutrition Targets and Sustainable Development Goals set ambitious targets to 384 improve child linear growth by 2025. Our findings of early linear growth failure onset and infrequent 385 reversal for a majority of children who faltered support interventions delivered from conception to the 386 early postnatal phase along with continued delivery of postnatal interventions to prevent linear growth 387 failure among children in LMICs. 388 389 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. Biol. 202, 65-76 (2000). 484 43. Ingemarsson, I. Gender aspects of preterm birth. BJOG Int. J. Obstet. Gynaecol. 110, 34-38 (2003). 485 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. Mean length-for-age Z-scores by age in months for each cohort. Cohorts are sorted by geographic region 543 and mean length-for-age Z-score. (c) Number of observations contributed by each cohort. (d) Overall 544 stunting prevalence in each cohort, defined as proportion of measurements with length-for-age z-score 545 < -2. 546 547 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint incident stunting from birth to age 15 months (N=14 cohorts that measured children at least monthly 567 between birth and age 15 months, N=7,198 children). "Never" includes children who did not become 568 stunted by age 15 months. Shaded ribbons indicate 95% confidence intervals. Pooled results were 569 derived from random effects models with restricted maximum likelihood estimation. 570 571 572 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. of LAZ at subsequent measurements after stunting reversal. (c) Mean difference in LAZ following 581 stunting reversal at each subsequent age of measurement compared to the age of reversal using 582 random effects models fit with restricted maximum likelihood estimation. (b) and (c) include data from 583 18 cohorts in 10 countries with at least monthly measurement (N = 10,677 children). All panels contain 584 data up to age 15 months because in most cohorts; measurements were less frequent above 15 months. 585 a b c . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. percentile (solid black line). Differences were statistically significant between sexes at ages 0-3, 3-6, 15-592 18, and 18-21 months. (b) Within-child difference in length-for-age Z-score per month by age and sex. 593 Differences were statistically significant between sexes at ages 0- 3, , 9-12, 12-15, 18-21, and 21-24 594 months. (c) Mean length-for-age Z-score by age and sex; all differences were statistically significant. All 595 panels include 29 ki cohorts in 13 countries that measured children at least quarterly (n = 50,022 596 children) pooled using random effects models fit with restricted maximum likelihood estimation. 597 Vertical bars indicate 95% confidence intervals. 598 599 600 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020

602
Study designs and inclusion criteria 603 We included all longitudinal observational studies and randomized trials available through the ki project 604 on April 2018 that met 5 inclusion criteria: 1) conducted in low-or middle-income countries; 2) enrolled 605 children between birth and age 24 months and measured their length and weight repeatedly over time; 606 3) did not restrict enrollment to acutely ill children; 4) enrolled at least 200 children; 5) collected 607 anthropometry measurements at least every 3 months (Extended Data Figure 1). Thirty-one longitudinal 608 cohorts in 15 countries followed between 1969 and 2014 met inclusion criteria. There was no evidence 609 of secular trends in LAZ (https://child-growth.github.io/stunting/secular-trends.html). We calculated 610 cohort measurement frequency as the median days between measurements. If randomized trials found 611 effects on growth within the intervention arms, the analyses were limited to the control arm. We 612 included all measurements under 24 months of age, assuming months were 30.4167 days. We excluded 613 extreme measurements of LAZ > 6 or < -6 following WHO growth standard recommendations. 1 In many 614 studies, investigators measured length shortly after birth because deliveries were at home, but the 615 majority of measurements were within the first 7 days of life (https://child-616 growth.github.io/stunting/age-meas.html); for this reason, we grouped measurements in the first 7 days 617 as birth measurements. Gestational age was measured in only eight cohorts (7 cohorts measured it by 618 recall of last menstrual period or newborn examination; 1 measured it by ultrasound); thus, we did not 619 attempt to exclude preterm infants from the analyses. 620 621 Quality assurance 622 The ki data team assessed the quality of individual cohort datasets by checking the range of each 623 variable for outliers and values that are not consistent with expectation. Z-scores were calculated using 624 the median of replicate measurements and the 2006 WHO Child Growth Standards. 1 In a small number 625 of cases a child had two anthropometry records at the same age, in which case we used the mean of the 626 records. Analysts reviewed bivariate scatter plots to check for expected correlations (e.g., length by 627 height, length/height/weight by age, length/height/weight by corresponding Z-score). Once the 628 individual cohort data was mapped to a single harmonized dataset, analysts conducted an internal peer 629 review of published articles for completeness and accuracy. Analysts contacted contributing 630 investigators to seek clarification about potentially erroneous values in the data and revised the data as 631 needed. 632 633 Outcome definitions 634 We used the following summary measures in the analysis: 635 636 Prevalent stunting was defined as the proportion of measurements within a specific stratum (e.g., age) 637 below the 2006 WHO standard -2 LAZ, and analogously below -3 LAZ for severe stunting. For each age, 638 we included children with LAZ measurements within one month before and after that age in the point 639 prevalence estimate to account for variation in the exact age of child measurement. For example, point 640 prevalence at 6 months included children aged 5-7 months. 641 642 Incident stunting episodes were defined as a change in LAZ from above -2 Z in the prior measurement to 643 below -2 Z in the current measurement. Similarly, we defined severe stunting episodes using the -3 Z 644 cutoff. Children were considered at risk of stunting at birth, so children born stunted were considered 645 to have an incident episode of stunting at birth. Children were also assumed to be at risk of stunting at 646 the first measurement in non-birth cohorts and trials. Children whose first measurement occurred after 647 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint birth were assumed to have experienced stunting onset at the age halfway between birth and the first 648 measurement. The vast majority of children were less than 5 days of age at their first measurement life 649 (https://child-growth.github.io/stunting/age-meas.html). 650 651 Incidence proportion We calculated the incidence proportion of stunting during a defined age range 652 (e.g. 3-6 months) as the proportion of children at risk of becoming stunted who became stunted during 653 the age range (the onset of new episodes). 654 655 Changes in stunting status were classified using the following categories: "Never stunted": children with 656 LAZ -2 at previous ages and the current age. "No longer stunted": children who previously reversed 657 their stunting status with LAZ -2 at the current age. "Stunting reversal": children with LAZ <-2 at the 658 previous age and LAZ  -2 at the current age. "Newly stunted": children whose LAZ was previously 659 always -2 and with LAZ <-2 at the current age. "Stunting relapse": children who were previously 660 stunted with LAZ -2 at the previous age and LAZ <-2 at the current age. "Still stunted": children whose 661 LAZ was <-2 at the previous and current age. 662 663 Growth velocity was calculated as the change in length in centimeters between two time points divided 664 by the number of months between the time points. We compared the change in length in centimeters 665 per month measures to the WHO Child Growth Standards for linear growth velocity. 2 We also estimated 666 within-child rates of change in LAZ per month. 667 668 Measurement frequency 669 Analyses of prevalence, incidence, and growth velocity (Figs 2, 3, 5) included cohorts with at least 670 quarterly measurements in order to include as many cohorts as possible. Analyses of fluctuations in 671 stunting status (Fig 4) were restricted to cohorts with at least monthly measurements to allow 672 evaluation of changes in stunting status with higher resolution. 673 674 Subgroups of interest 675 We stratified the above outcomes within the following subgroups: child age, grouped into one-or three-676 month intervals, (depending on the analysis); the region of the world (Asia, sub-Saharan Africa, Latin 677 America); child sex, and the combinations of those categories. 678 679 Statistical analysis 680 All analyses were conducted in R version 3.4.2. 3 681 682 Estimation of mean LAZ by age in Demographic and Health Surveys and ki cohorts 683 684 We downloaded standard DHS individual recode files for each country from the DHS program 685 website (https://dhsprogram.com/). We used the most recent standard DHS datasets for the individual 686 women's, household, and height and weight datasets from each country. We obtained variables for 687 country code, sample weight, cluster number, primary sampling unit, and design stratification from the 688 women's individual survey recode files. From the height and weight dataset, we used standard recode 689 variables corresponding to the 2006 WHO growth standards for height-for-age. 690 691 After excluding missing observations, restricting to measurements of children 0-24 months of age, 692 and restricting to z-scores within WHO-defined plausible values, surveys collected from 2006 to 2018 in 693 51 countries were included (Extended Data Table 2). Surveys from Afghanistan, Bolivia, Brazil, Central 694 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint African Republic, Colombia, Indonesia, Nicaragua, Philippines, Paraguay, and Vietnam did not meet 695 these inclusion criteria and were excluded from the analysis. 696 697 We classified countries into regions (South Asia, Latin America, and Africa) using the World Health 698 Organization regional designations with the exception of the classification for Pakistan, which we 699 included in South Asia to be consistent with prior linear growth studies using DHS. 4 One included cohort 700 was from Belarus, and we chose to exclude it from region-stratified analyses as it was the only European 701 study. 702 703 We estimated age-stratified mean from ages 0 to 24 months within each DHS survey, accounting for 704 the complex survey design and sampling weights. We then pooled estimates of mean LAZ for each age in 705 months across countries using a fixed effects estimator (details below). We computed two sets of 706 pooled results: 1) DHS measuring children 0-24 months in countries that overlapped with ki study 707 countries and 2) all DHS countries measuring children 0-24 months in each geographic region (as in Fig  708 2) and (https://child-growth.github.io/stunting/DHS.html). We compared DHS estimates with mean LAZ 709 by age in the ki study cohorts, which we estimated using penalized cubic-splines with bandwidth chosen 710 using generalized cross-validation. 5 We used splines to estimate age-dependent mean LAZ in the ki study 711 cohorts to smooth any age-dependent variation in the mean caused by less frequently measured 712 cohorts. 713 714 Fixed and random effects models 715 716 Several analyses pooled results across study cohorts. The primary method of pooling was using random 717 effects models. This modeling approach assumes that studies are randomly drawn from a hypothetical 718 population of longitudinal studies that could have been conducted on children's linear growth in the 719 past or future. We also fit fixed effects models as a sensitivity analysis (https://child-720 growth.github.io/stunting/fixed-effects.html); inferences about estimates from fixed effects models are 721 restricted to only the included studies. 6 722 723 Random effects models assume that the true population outcomes  are normally distributed with 724 mean  and variance  2 (i.e., that  ~ N(,  2 )). To estimate outcomes in this study, the random effects 725 model is defined as follows for each study in the set of i = 1, …, k studies: 726 727 where yi is the observed outcome in study i, ui is the random effect for study i, and ei is the estimated 730 outcome for study i, and ei is the sampling error within study i. The model assumes that ui ~ N(0,  2 ) and 731 ei ~ N(0, vi), where vi is the study-specific sampling variance. We fit random effects models using the 732 restricted maximum likelihood estimator. 7,8 If a model failed to converge, models were fit using a 733 maximum likelihood estimator instead. The estimate of  is the estimated mean outcome in the 734 hypothetical population of studies (i.e., the estimated outcome pooling across study cohorts). 735 736 We also fit inverse variance weighted fixed effects models defined as follows: 737 738 (2) 739 740 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020 where is the weighted mean outcome in the set of k included studies, and wi is a study-specific 741 weight, defined as the inverse of the study-specific sampling variance vi. is the estimated mean 742 outcome in the specific studies included in this analysis. 743 744 For both types of outcomes, we pooled binary outcomes on the logit scale and then back-745 transformed estimates after pooling to constrain confidence intervals between 0 and 1. For cohort-746 stratified analyses, which did not pool across studies, we estimated 95% confidence intervals using the 747 normal approximation (https://child-growth.github.io/stunting/cohort.html). 748 749 For the estimates of changes in stunting status, to ensure that percentages summed to 100%, results 750 in Fig 4a were not pooled using random effects. Results were similar when pooling with random effects 751 (https://child-growth.github.io/stunting/fixed-effects.html#changes-in-stunting-status-by-age). 752 753 Estimation of prevalence and incidence 754 We estimated prevalence and incidence as defined above in 3-month age intervals within specific 755 cohorts and pooled within region and across all studies (Fig 3). Pooled analyses used random effects 756 models for the primary analysis and fixed effects models for sensitivity analyses as described above. 757 758 Estimation of changes in stunting status 759 To assess fluctuations in stunting status over time, we conducted an analysis among cohorts with at 760 least monthly measurements from birth through age 15 months to provide sufficient granularity to 761 capture changes in stunting status. Because only 12 cohorts in 10 countries met these criteria, we 762 performed only pooled analyses (i.e., we did not stratify by region or study cohort) to ensure the sample 763 size was sufficient. We estimated the proportion of children in each stunting category defined above 764 under "Changes in stunting status" at each month from birth to 15 months. To ensure that percentages 765 summed to 100%, we present results that were not pooled using random effects. Analyses using random 766 effects produced similar results (https://child-growth.github.io/stunting/fixed-effects.html#changes-in-767 stunting-status-by-age). 768 769 To examine the distribution of LAZ among children with stunting reversal, we created subgroups of 770 children who experienced stunting reversal at ages 3, 6, 9, and 12 months and then summarized the 771 distribution of the children's LAZ at ages 6, 9, 12, and 15 months. Within each age interval, we estimated 772 the mean difference in LAZ at older ages compared to the age of stunting reversal and estimated 95% 773 confidence intervals for the mean difference. Pooled analyses used random effects models for the 774 primary analysis and fixed effects models for sensitivity analyses as described above. 775 776 Linear growth velocity 777 We estimated linear growth velocity within 3-month age intervals stratified by sex, pooling across 778 study cohorts (Fig 5) as well as stratified by geographic region (Extended Data Fig 4) and study cohort 779 (https://child-growth.github.io/stunting/cohort.html). Analyses included cohorts that measured children 780 at least quarterly. We included measurements within a two-week window around each age in months to 781 account for variation in the age of each length measurement. Pooled analyses used random effects 782 models for the primary analysis and fixed effects models for sensitivity analyses as described above 783 (https://child-growth.github.io/stunting/fixed-effects.html). 784 785 Sensitivity Analyses 786 We conducted three sensitivity analyses; results are available at (https://child-787 growth.github.io/stunting). First, to assess whether inclusion of PROBIT, the single European cohort, 788 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. Bhan, and all other members of the study staffs and field teams. We would also like to thank all study 829 participants and their families for their important contributions.  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint Extended Data Figure 1| ki cohort selection. Analyses focused on longitudinal cohorts to enable the 856 estimation of prospective incidence rates and growth velocity. In April 2018, there were 86 longitudinal 857 studies on GHAP. From this set, we applied five inclusion criteria to select cohorts for analysis. Our 858 rationale for each criterion follows. (1) Studies were conducted in lower income or middle-income 859 countries. Our target of inference for analyses was children in LMICs, which remains a key target 860 population for preventive interventions.
(2) Studies measured length and weight between birth and age 861 24 months. We were principally interested in growth faltering during the first 1,000 days (including 862 gestation), thought to be the key window for linear growth faltering. 7 (3) Studies did not restrict 863 enrollment to acutely ill children. Our focus on descriptive analyses led us to target, to the extent 864 possible, the general population. We thus excluded some studies that exclusively enrolled acutely ill 865 children, such as children who presented to hospital with acute diarrhea or who were severely 866 malnourished. (4) Studies enrolled at least 200 children. Age-stratified incident episodes of stunting and 867 wasting were sufficiently rare that we wanted to ensure each cohort would have enough information to 868 estimate rates before contributing to pooled estimates. (5) Studies collected anthropometry 869 measurements at least every 3 months. We limited studies to those with higher temporal resolution to 870 ensure that we adequately captured incident episodes and recovery. 871 872 873 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint

886
Extended Data Figure 3 | Mean, 5th and 95 th percentile of length-for-age Z-score by age in ki 887 longitudinal cohorts estimated with cubic splines in cohorts with at least monthly measurement (N=14 888 cohorts that measured children at least monthly, N=7,456 children). 889 890 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. 95% confidence intervals for ki cohorts. (b) Within-child difference in length-for-age Z-score per month 896 by age, sex, and region. (c) Mean length-for-age Z-score by age, sex, and region. Results shown in all 897 panels were derived from 29 ki cohorts in 13 countries that measured children at least quarterly (n = 898 50,022 children). 899 900 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint 901 Extended Data Figure 5 | Comparison of stunting prevalence at birth with and without gestational age 902 correction 903 This figure includes the results from correcting at-birth Z-scores in the ki cohorts that measured 904 gestational age (GA) for 37,218 measurements in 8 cohorts. The number in the parentheses following 905 each cohort name indicates the prevalence of pre-term birth in each cohort. The corrections are using 906 the Intergrowth standards and are implemented using the R growthstandards package (https://ki-907 tools.github.io/growthstandards/). Overall, the stunting prevalence at birth decreased slightly after 908 correcting for gestational age, but the cohort-specific results are inconsistent. Observations with GA 909 outside of the Intergrowth standards range (<168 or > 300 days) were dropped for both the corrected 910 and uncorrected data. Prevalence increased after GA correction in some cohorts due to high rates of 911 late-term births based on reported GA. Gestational age was estimated based on mother's recall of the 912 last menstrual period in the Jivita-3, Cebu, New Delhi, INCAP, IRC, and CMC-V-BCS-2002 cohorts, was 913 based on the Dubowitz method (newborn exam) in the Keneba cohort, and was based on ultrasound 914 measurements in the PROBIT trial. 915 916 917 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 11, 2020. . https://doi.org/10.1101/2020.06.09.20127001 doi: medRxiv preprint