Introduction

Linear growth faltering in children remains a major global health challenge, with a staggering 148.1 million stunted children under the age of 5 years in low and middle-income countries (LMICs) [1]. Evidence suggests that early-life stunting is associated with an increased risk of childhood mortality [2], infectious diseases such as diarrhea, pneumonia, and measles [3,4], impaired growth and cognitive development [5–8], as well as poor adult economic outcomes [9]. Furthermore, recent studies have demonstrated that impaired linear growth in early life is associated with subsequent deficits in later childhood, with infrequent reversal of stunting [10,11].

While several nutritional interventions have shown potential in addressing childhood stunting, achieving widespread success remains elusive [12]. Moreover, growth during the first thousand days is influenced by several factors, including environmental exposures, infectious diseases, and malnutrition, and is not homogeneous in pattern [13,14]. As children follow heterogeneous growth patterns [15], characterizing childhood growth trajectories and tailoring interventions to distinct groups has become essential. Novel, trajectory-informed strategies may strengthen the effectiveness of interventions aimed at preventing childhood stunting.

Group-based trajectory models can classify individuals with similar features, characteristics, and attributes into discrete, unambiguous subgroups [16–18] and have become increasingly useful in understanding patterns of child growth and development. While prior research has established a link between early growth patterns during the first 24 months of life and subsequent growth, these trajectories have not been adequately explored using the latent class growth modeling (LCGM) technique. LCGM is a unique approach because it models population-level developmental heterogeneity as distinct, interpretable trajectory groups rather than as individual-level random variation [19,20]. Unlike traditional growth models, LCGM identifies groups that cannot be directly observed [20]. Although some studies have explored trends in linear growth in low-income settings [21,22], comprehensive data from LMICs regarding the presence of distinct group-based trajectories of linear growth among young children remain limited. Few studies have applied group-based or latent trajectory approaches to identify heterogeneous patterns, such as early growth faltering, catch-up growth, or persistently adequate growth, within the same population. This knowledge gap limits our understanding of growth heterogeneity in LMICs and constrains the ability to identify critical windows for intervention and to design targeted strategies for children most at risk of sustained linear growth faltering. It also remains unclear to what extent early-life linear growth trajectories influence and set the course for linear growth attained later in childhood.

This study aimed to identify groups of children with distinct linear growth trajectories during the first two years of life in a birth cohort from seven LMICs. It also assessed the association between these early-life linear growth trajectories and attained linear growth and stunting at 60 months of age.

Methods

Results

The MAL-ED study enrolled 1,868 children from seven countries (Bangladesh, India, Nepal, Brazil, Peru, South Africa, and Tanzania). Of these, 1469 children had length data at birth and 24 months, and 1047 children had complete data on all other variables (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flow diagram of participant inclusion and retention in the analysis from the MAL-ED birth cohort study.

Sites: BGD: Bangladesh – Dhaka; INV: India – Vellore; NEB: Nepal – Bhaktapur; BRF: Brazil – Fortaleza; PEL: Loreto, Peru; SAV: South Africa – Venda; TZH: Tanzania – Haydom.

https://doi.org/10.1371/journal.pone.0329596.g001

Of the children, 49% were girls. The average length-for-age z score at birth ranged from −1.02 in INV and TZH to −0.71 in SAV. Overall, median protein consumption was 37.7 g/d with the lowest median protein intake in NEB (30.4 g/d) and the highest median protein intake in BRF (54.0 g/d) at 60 months of age. The seven locations had different median fat intakes, ranging from 21.3 g/d in TZH to 45.4 g/d in BRF. The carbohydrate intakes in the seven areas varied, ranging from 171 g/d in NEB to 318 g/d in TZH (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Study population characteristics.

https://doi.org/10.1371/journal.pone.0329596.t001

Results of LCGM

The population distribution of LAZ progression was best described by a five-class model, with cubic terms for the first three trajectories and quartic terms for the remaining two. LAZ trajectory class 1, termed “severely attenuated linear growth,” represented 9% of the children in the sample. The group was born with a mean LAZ of −2.15 and experienced a marked decline through 24 months (mean LAZ = −2.73, −3.25, −3.60, and −3.69 at 6, 12, 18, and 24 months, respectively). Trajectory class 2, labeled “moderately attenuated linear growth,” accounted for 25% of participants. These children had a mean LAZ of −1.37 at birth, which progressively declined to −1.76, −2.19, −2.50, and −2.56 at 6, 12, 18, and 24 months, respectively. Trajectory class 3, termed “mildly attenuated linear growth,” comprised 34% of participants. This group had a mean LAZ of −0.84 at birth and showed a gradual decline through 24 months. Trajectory class 4, termed “stable linear growth,” included 25% of participants. Their mean LAZ increased slightly during the first five months of life and then gradually declined through 24 months. Trajectory class 5, labeled “improved linear growth,” represented 7% of children. This group was born with a mean LAZ of −0.05, demonstrated substantial improvement up to 12 months of age, and then showed a modest decline through 24 months (mean LAZ = 0.82, 0.91, 0.87, and 0.57 at 6, 12, 18, and 24 months, respectively) (Fig 2) (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Estimated mean length-for-age z-score (LAZ) at different ages by trajectory group, derived using LCGM, with 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0329596.t002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. LCGM-derived latent trajectories of length-for-age z-score (LAZ) from 0-24 months with 95% confidence intervals.

Trajectory group: Class 1: Severely attenuated; Class 2: Moderately attenuated; Class 3: Mildly attenuated; Class 4: Stable; Class 5: Improved.

https://doi.org/10.1371/journal.pone.0329596.g002

Across all trajectories, the average posterior probability of group membership exceeded 0.90, the odds of correct classification were greater than 50, and modeled group probabilities were in good agreement with the observed proportions of assigned group membership (Fig 3) (Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Fit statistics for the final LCGM model of length-for-age z-score (LAZ).

https://doi.org/10.1371/journal.pone.0329596.t003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Mean height-for-age z-score at 60 months by trajectory group across the seven study sites.

Trajectory group: Class 1: Severely attenuated; Class 2: Moderately attenuated; Class 3: Mildly attenuated; Class 4: Stable; Class 5: Improved.

https://doi.org/10.1371/journal.pone.0329596.g003

Effect of LAZ trajectories on HAZ at 5 years

LAZ trajectories during the first two years of life were associated with HAZ at five years of age (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Association of LAZ trajectories during the first two years of life with height-for-age z-score (HAZ) at age 60 months (n = 1047).

https://doi.org/10.1371/journal.pone.0329596.t004

After adjusting for potential confounders, the improved linear growth class had a predicted 0.86 higher HAZ at five years of age (95% CI: 0.67, 1.04), whereas the severely, moderately, and mildly attenuated linear growth classes had significantly lower HAZ at five years compared with the stable linear growth class (severely: β = −2.10; 95% CI: −2.26, −1.95; moderately: β = −1.34; 95% CI: −1.45, −1.22; mildly: β = −0.70; 95% CI: −0.81, −0.59).

Effect of LAZ trajectories on stunting at 5 years

LAZ trajectories during the first two years of life were significantly associated with stunting at five years of age (Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Association of LAZ trajectories during the first two years of life with stunting at age 60 months (n = 1047).

https://doi.org/10.1371/journal.pone.0329596.t005

After adjusting for potential confounders, the severely attenuated linear growth class (OR 132.78; 95% CI: 61.58, 286.52) and the moderately attenuated linear growth class (OR 16.00; 95% CI: 10.28, 24.91) had higher odds of stunting at the age of five years compared with the other three trajectory classes.

Discussion

To our knowledge, this is the first study to identify discrete linear growth trajectories during the first two years of life and to examine their association with attained linear growth and stunting in later childhood in a multi-country birth cohort from low-income settings. We identified five distinct linear growth trajectories from birth to 24 months using latent class growth modeling. Namirembe et al. identified four distinct linear growth trajectories from birth to 12 months among children in rural Uganda [35]. Similarly, a longitudinal birth cohort study of HIV-unexposed children in Zimbabwe identified four distinct linear growth trajectories from birth through two years of age by clustering infants with similar growth patterns [36].

Of the five trajectory classes identified in our study, the “severely attenuated linear growth” group had the lowest mean LAZ at birth (−2.15), already within the stunted range, and declined further to −3.69 by 24 months, reaching the severely stunted range. In contrast, the “improved linear growth” group started as non-stunted (mean LAZ −0.05 at birth), improved through 12 months, and showed only slight tapering by 24 months (mean LAZ 0.57). Both groups represented relatively small proportions of participants (9% and 7%, respectively). The largest proportion of children (34%) belonged to the “mildly attenuated linear growth” trajectory. Although this group started in the non-stunted range (mean LAZ −0.84 at birth), their mean LAZ gradually declined to −1.69 at 24 months, placing them in the at-risk-of-stunting range. The remaining two trajectories each comprised 25% of the cohort. The “moderately attenuated linear growth” group started in the at-risk-of-stunting range (mean LAZ −1.38 at birth), declined steadily, and reached the stunted range by 24 months (mean LAZ −2.56). In contrast, children in the “stable linear growth” trajectory maintained a relatively stable growth and remained within the non-stunted range throughout the first 24 months.

Among the five trajectory classes, the two groups with the lowest mean LAZ at birth, namely the “severely attenuated linear growth” and “moderately attenuated linear growth”, demonstrated markedly elevated odds of stunting at five years of age compared with the other three groups. Complementing this finding, Krebs et al. reported that deficits in birth length were strong predictors of attained LAZ and stunting at the age of two years across sites in the Democratic Republic of Congo, Guatemala, India, and Pakistan [37]. Mild and moderate stunting at birth were associated with a 0.5–1.0 lower mean LAZ and a 40–60% increased risk of stunting at two years [37].

Our findings further suggest a graded association between early-life linear growth trajectories and attained HAZ at five years of age. Compared with the stable linear growth class, the mildly, moderately, and severely attenuated linear growth trajectories were progressively associated with lower HAZ at five years, demonstrating a dose-response pattern. Conversely, the improved linear growth class was associated with higher HAZ at five years. Although the biological mechanism underlying these graded effects remains unclear, accumulating evidence indicates that early life growth faltering has long-term and potentially lifelong consequences for health outcomes [38,39]. Nutrition during the first 1,000 days of life plays a critical role in shaping growth and development across the life course [40,41].

Roth et al., analyzing 179 Demographic and Health Surveys from 64 LMICs using Monte Carlo simulations, presented an important population-level perspective [42]. They demonstrated that as mean HAZ declined from birth to three years, both tails of the distribution (5th and 95th percentiles) shifted downward, resulting in a nearly symmetrical narrowing of the overall distribution [42]. This age-related distributional shift suggests that children across the entire HAZ spectrum experience postnatal linear growth faltering, supporting the view that postnatal growth faltering in LMICs should be considered a whole-population condition [42]. While Roth et al. demonstrated a population-level downward shift in the HAZ distribution in LMICs, our findings extend this perspective by identifying distinct longitudinal growth trajectories within that broader pattern of faltering.

WHO nutritional status categories (normal, stunted, severely stunted) provide cross-sectional classifications based on length/height-for-age at a given time point. In contrast, the trajectory classes identified through LCGM reflect longitudinal growth patterns over time. The practical implication of identifying five trajectory classes is that children with similar HAZ at 24 months may have followed different growth pathways (e.g., persistently faltering versus improving), which may carry different implications for growth at 60 months. Although some trajectory classes, particularly the severely and moderately attenuated groups, may overlap with children classified as stunted under WHO standards at specific ages, the two approaches are conceptually distinct. While WHO standards classify growth status relative to an external reference population at a single time point, our trajectory-based classification captured internally derived, dynamic growth processes across early life and therefore provided additional insight beyond conventional categories.

The distinct patterns identified in this study highlight the importance of tracking early-life growth trajectories and implementing group-specific strategies to prevent growth faltering. Interventions based solely on population averages may overlook children at the highest risk. Consideration of the timing of growth faltering, recovery patterns, and contextual determinants is essential when developing targeted strategies to improve child growth and health. We recommend the development and implementation of trajectory-informed strategies and policies that account for children’s specific growth patterns alongside nutritional status, WASH conditions, environment, and socioeconomic context, consistent with broader policy recommendations [43,44].

This study has several limitations. Although we characterized early-life linear growth trajectories and demonstrated their association with attained linear growth in later childhood, we did not investigate determinants of trajectory membership. Important early-life factors such as parental height, maternal nutritional status, gestational weight gain, placental stress or insufficiency, length of gestation, and size at birth were not fully accounted for due to data limitations. Catch-up growth among infants born small for gestational age and catch-down growth among those born large for gestational age were also not examined because gestational age–specific birth size data were limited. In addition, environmental exposures, including infections, morbidity, and nutritional intake between 25 and 59 months, were not comprehensively captured. However, we adjusted for the WAMI index, a composite indicator reflecting socioeconomic and environmental conditions (including SES and WASH components), as well as dietary intake at 60 months. Future research should examine the determinants and causal pathways underlying distinct growth trajectories to better inform preventive strategies.

Conclusion

This study identified five distinct linear growth trajectories during the first two years of life among children from seven LMICs, ranging from severely attenuated to improved growth patterns. These early-life trajectories were strongly associated with attained linear growth and stunting at five years of age, demonstrating a graded relationship across trajectory classes. While current public health strategies rely primarily on WHO-defined thresholds of stunting, trajectory-based classification may offer complementary insights by capturing longitudinal growth dynamics and identifying children at risk of persistent growth faltering before they meet conventional criteria. Integrating longitudinal growth patterns with existing frameworks may strengthen early identification and inform timely, context-specific interventions aimed at preventing long-term growth deficits.

Supporting information