Stunting, age at school entry and academic performance in developing countries: A systematic review and meta-analysis

Although many studies have examined the associations between growth problems in infancy and age at school entry, grade repetition, school dropout and schooling level in developing country, no synthesis of the evidence has been conducted. We aim to review evidence of the effects of stunting, or height-for-age, on schooling level and schooling trajectories, dened as the combination of school entry age, grade repetition, and school dropouts. We conducted a systematic review of studies (last update March 20, 2021) estimating that estimate the association between stunting, or height-for-age, and at least one component of the school trajectory, or schooling level, using ve databases (PubMed, Embase, Education Resources Information Center (ERIC), Web of Science and PsycINFO). Study selection and data extraction were performed by two independent reviewers. Pooled effects were calculated using the generic inverse variance weighting random effect model. The studies’ risk of bias was assessed using the ROBINS-I tool for non-randomized studies. We screened 3944 by and and inclusion Meta-analysis showed that an increase in leads to an increase in early [OR: 1.34 (95% CI: 1.07; 1.67)], a reduction in late enrollment [OR: 0.63 (95% CI: 0.51; 0.78)], an increase in schooling level [MD: 0.24 (95% CI: 0.14; 0.34)], and a reduction of school overage [OR: 0.79 (95% CI: 0.70; 0.90)]. The odds of grade repetition increased by 59% (OR = 1.59; 95% CI: 1.18; 2.14) for stunted children compared to those with no stunting. health issues in educational policies. Systematic review registration: PROSPERO CRD42020198346 that from this study do not allow conclusions to be drawn regarding the relationship between stunting or height-for-age and dropping out of school.


Introduction
In many countries, only a minority of children grow up healthy [1]. The 2018 World Nutrition Report indicates that stunting affects 150.8 million children under ve years of age, which represents 22.2% of the world's children [2,3]. The vast majority of stunted children come from developing countries (148.0 of 150.8 million) [3]. These countries also have more of out-of-school children or people with low academic achievement than the global average. The UNESCO Institute for Statistics reports that, in 2018, 17.7% of children of primary school age were out of school in the least developed countries, compared to only 8.2% globally [4]. In the same year, only 54.0% reached the last grade of primary education in developing countries compared to 81.7% globally [4].
In this context, many studies have been carried out in developing countries on the effects of early childhood development on future academic achievement. These studies have shown that stunting in the rst ve years of life leads to cognitive impairment in children [5][6][7][8], poor school performance, fewer years of schooling, and low productivity in adulthood [7,8]. Children who have been stunted in childhood are therefore more likely to delay school enrollment, perform poorly in school, repeat a grade, and drop out of school than those who have not been stunted [9,10]. However, some studies observed no signi cant association between childhood stunting and academic performance [11,12], grade repetition [10,13], and school dropout [14].
Systematic reviews have been undertaken in this eld in developing countries. However, most of them [15][16][17][18][19][20] are qualitative reviews. To our knowledge, only one review [21] carried out a meta-analysis on the effect of linear growth or stunting on child development, but it does not include outcomes on age at school entry, grade repetition, and school dropouts. We aim to review evidence of the effect of stunting or heightfor-age on schooling level and schooling trajectories, de ned as the combination of school entry age, grade repetition, and school dropouts.

Methodology
The protocol of the review was designed according to the "Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 statement" [22] and registered on the International Prospective Register of Systematic Reviews (PROSPERO) on September 20, 2020 (#CRD42020198346). This review was conducted according to the "Cochrane collaborative guidelines for systematic reviews" [23] and is reported according to "The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate healthcare interventions" [24] (see PRISMA check list in Additional le 1).

Eligibility Criteria
The PICOS (Population, Intervention/exposure, Comparator, Outcome and Study design) approach was used to de ne inclusion criteria. Our population of interest was primary school aged children in developing countries. Generally, the legal age of admission to primary school is between ve and seven years old [25]. Primary school usually lasts six years, although it can range from four to seven years [25], and it usually ends between ages 10 and 12 years [25]. We considered studies that include children aged between 5 and 12 years from developing countries. We also included studies on poor child development that resulted in stunting when children were less than ve years old. Studies that use standardized height-for-age ratio (height-for-age z-scores) as a measure of child growth during infancy were included. We also considered studies that used height as a marker of stunting. We considered four outcomes: age at school entry, grade repetition, school dropout, and schooling level. Schooling level was de ned as the highest level of education attained by an individual at the time of study. Eligible studies were observational studies (prospective, retrospective, case-control, case series, and cross-sectional studies). Studies on diverse populations were included if data on the subgroup of primary school age children related to stunting and outcomes could be extracted or if primary school aged children constituted more than 80% of the study population.

Research Strategy
We conducted a comprehensive systematic literature search via Pubmed, Embase, ERIC (via Ovid), Web of Science and PsycINFO (via Ovid) (last updated March 20, 2021). We developed a rigorous search strategy using relevant keywords related to stunting, schooling trajectory, and geographic area. The search strategy was designed using both free and controlled vocabularies in PubMed, and then translated into other databases. No restriction was applied on language or date of publication. We consulted an information specialist to validate our search strategy, and the sensitivity of the strategy was evaluated by verifying the inclusion of ve relevant studies.

Studies management and selection Data management
The bibliographic reference management software package EndNote was used for citation management. We imported references from databases into EndNote and then removed duplicates using both automatic and manual screening based on study titles. Then, citations were transferred to Covidence for selection.

Study Selection process
Two reviewers (JG and LPB) evaluated all studies independently by screening titles, abstracts, and full texts to identify studies that met the inclusion criteria. We rst evaluated inter-reviewer agreement (Kappa) on eligibility using the rst 300 citations to ensure that reviewers had a good understanding of inclusion criteria. Inter-reviewer agreements were assessed after each step of selection. Disagreements between JG and LPB were resolved by consensus or by consulting a third reviewer. At the full text stage, reasons for exclusions were recorded.

Data extraction
An Excel data extraction form and a detailed instruction manual was developed and piloted with a sample of three studies. The same two reviewers (JG and LPB) extracted data independently from the selected studies. Data extracted include study characteristics ( rst author, year of publication, country), population characteristics (sample size, proportion of girls, age), study design, follow-up duration, exposure (type of exposure, age at exposition measurement), outcomes (age at school entry, level of education attained, repetition, dropouts), effect measures, and con dence intervals (adjusted measure of effect, con dence interval, p-value, standard errors, confounding variables). Study authors were contacted with up to three email attempts in case of missing information or unclear data. All extracted data from the two reviewers were crosschecked and disagreements were discussed to reach a consensus or by involving a third reviewer.

Risk of bias
The risk of bias of the included studies was assessed using the ROBINS-I tool for non-randomized studies [26]. The tool covers confounding bias, selection bias, classi cation bias, bias due to missing data, and bias due to measurement of the outcome. Studies were classi ed as low, moderate, or high risk. To assess confounding bias, we considered a model well adjusted if it included demographic characteristics (e.g. child sex, child age), and household characteristics (e.g. socioeconomic status, household size, place of residence), and characteristics of the mother and father (e.g. education, size, ethnic group). The same two reviewers (JG and LPB) extracted data on risk of bias evaluation independently. Disagreements were resolved by discussion or by involvement of a third reviewer.

Statistical analysis and data synthesis
Eligible studies were described in detail according to PICOS parameters. We conducted a meta-analysis to estimate the pooled effect of stunting on the different outcomes and their 95% con dence intervals. Then, all studies with su cient information to estimate the pooled effect were included in the meta-analysis. Pooled effects were calculated according to the domain of the outcome, study design, and effect measured. Thus, for a given outcome domain, more than one pooled effect was estimated if the outcomes were measured in different ways or if different types of effect measures were extracted.

Sensitivity and subgroup analysis
To understand the source of heterogeneity, a sensitivity analysis was performed by removing one study at a time from the pooled effect size estimation. This allowed us to measure the effect of each selected study on the pooled effect heterogeneity. We were not able to conduct sensitivity analysis on studies at low risk of bias or subgroup analysis based on age of child stunting assessment (≤ 2 years of age and > 2 years of age) due to the insu cient number of studies.

Study selection and characteristics
We identi ed 4981 studies, of which 3944 were screened by title and abstract after removing duplicates (Fig. 1). Eighty-seven (87) studies were assessed by full text and 16 were considered eligible for the review. From these studies, six were included in the meta-analysis. The interreviewer agreements (Kappa statistics) were 97.5% and 83.5% respectively, for title and abstract screening and full-text selection. All studies were in English.  (Table 1 and   Table 2). Only two studies were published after 2015 [10,43]. Most of the studies (56.2 %) used a prospective observational design [6, 7, 10, 12, 14, 36, 37, 41, 42], while 43.8 % were cross-sectional studies [33-35, 38-40, 43]. The most commonly studied exposure was standardized height-for-age (56.2 %) [6, 12, 34, 37, 38, 40, 41] followed by stunting (37.4%) [10,14,35,36,39]. Two studies [42,43] used both stunting and height-for-age z-score. Most of the studies analyzed age at school entry [7,10,[36][37][38][39][40][41][42], and schooling level [6, 10,12,34,35,39,43]. Relatively few studies analyzed grade repetition [10,12,14,37,42] and dropout [14,33,38]. One study included four countries [6] and another involved two countries [35]. One study presented results only by sex [36]. In Sunny, DeStavola (10), the exposition was measured at three time points (between 0 and 4 months, between 11 and 16 months, and between 4 and 8 years). Sample sizes of included studies ranged from 325 to 2711. Participants were between 6.2 and 18.0 years old at the time of study (Table 2). *Some studies were counted more than once because they use more than one type of exposition or outcome leading to a number of studies greater than 16 and a total of percentages greater than 100%.  (13) or high (2) risk of confounding bias ( Table 3 in Additional le 2). More than one-fth (4, 25%) of studies have a high risk of bias due to missing data. Missing information was not reported in ve (31%) studies ( Table 3 in Additional le 2). Risks of bias in other domains of bias were low for all studies. Assessment of publication bias by funnel plot was not possible because of low number of studies by outcome of interest and variation in measure of effects estimated. Stunting and age at school entry Nine studies presented associations between height-for-age or stunting and an outcome related to age at school entry [7,10,[36][37][38][39][40][41][42]. These studies could not be combined because exposures, outcomes or effect measures differed. Among these studies, one estimated the association between height-for-age at two years and early or late enrollment by sex [36]. The meta-analysis from this study (Fig. 2) suggests that one unit increase in height-for-age is associated with a 34% increase in the odds of early enrollment [OR: 1.33 (95% CI: 1.07; 1.67), I²=0%] and a reduction of 37% in the odds of late enrollment [OR: 0.63 (95% CI: 0.51; 0.78), I²=0%]. All studies reported an association between height-for-age or stunting and the age at school entry [7,10,[36][37][38][39][40][41][42] (Table 2 and Table 4 in Additional le 3). Stunting and schooling level Two of the four cross-sectional studies [34,35] assessing the association between height-for-age and school overage by mean difference observed that an increase of one unit of height-for-age in a child was associated with an increase in schooling level for their age. The pooled effect of the studies led to similar results [MD: 0.24 (95% CI: 0.14; 0.34), I²=92%] (Fig. 3), but was characterized by high heterogeneity. Two other cross-sectional studies were not used in this pooled effect estimation [39,43]. One [39] found that stunted children were more likely to be overage, and the other reported that stunted children were more likely to be in a low grade; a one unit increase in height-for-age was associated with an increase in grade attainment [43]. In meta-analysis of longitudinal studies, an increase in height-for-age was associated with a reduction in the odds of school overage [OR: 0.79 (95% CI: 0.70; 0.90), I²=76%] with high heterogeneity. Gandhi, Ashorn (12) and Sunny, DeStavola (10) were not included in the pooled effect estimation because of the analysis methods they used and their outcome measurements. Gandhi, Ashorn (12) reported a non-signi cant association between height-for-age and schooling level, and Sunny, DeStavola (10) found a signi cant association ( Table 2 and Table 4 in Additional le 3). Stunting and grade repetition Figure 4 shows that grade repetition is associated with stunting or height-for-age. All included studies used a longitudinal design. The pooled estimates suggests that the odds of grade repetition increase by 59 % [OR: 1.59 (95% CI: 1.18; 2.14), I²=51%) for stunted children compared to non-stunted children with moderate heterogeneity. Two studies, which are not include in the meta-analysis, report an association between stunting and grade repetition [12,37], and one other study did not nd an association [42] (see Table 2 and Table 4 in Additional le 3). Stunting and school dropout

Discussion
This systematic review suggests that stunting determines age at school entry, schooling level, and grade repetition. An increase one unit in standard deviation in height-for-age is associated with an increase in the odds of early enrollment and delayed enrollment. Children with greater height-for-age were less likely to be overage for their grade. We also found that stunted children were more likely to repeat a grade than nonstunted children. Results from this study do not allow conclusions to be drawn regarding the relationship between stunting or height-forage and dropping out of school.
Childhood stunting can be associated with di culties learning the school curriculum. Children with high height-for-age z-scores, or those who were non-stunted started school earlier than those with low height-for-age z-score or those who were stunted. The latter are considered unready to start school at the minimum enrollment age [7,37,44]. Delayed enrollment could also re ect a lter imposed by schools if administrators use height as a sign of school readiness [7]. The high probability of grade repetition for stunted children is due to low school performance. Grade repetition occurs when children's academic performance is deemed unsatisfactory. Schooling levels can be seen as the re ection of age at school enrollment and grade repetition, and are thus dependent on academic performance. Several studies have highlighted that stunted growth and height-for-age are associated, respectively, negatively and positively with test results in mathematics, reading, communication and motor development [6,21,[45][46][47][48][49]. Results showed that impaired growth and development in infancy negatively affects later academic performance and therefore academic trajectory, which leads, overall to low school levels. This may explain why stunted children are more likely to be unemployed, less productive, and to have low social status than non-stunted children [50][51][52][53].
Stunting could lead to a delay in the development of cognitive functions and permanent cognitive impairments, which improve little with age [54]. This relationship between stunting and cognitive abilities is particularly important in the rst years of life when vital human development occurs in all domains, including the brain formation [16,55]. When stunting occurs in this early stage of life, it severely affects attention development, executive functions such as cognitive exibility, working memory, and visuospatial functions like visual construction [54]. Experimental research on animals has also shown that nutrition de ciencies negatively affect brain development and measure of performance [55][56][57][58], but it is di cult to extrapolate this to human cognition [57]. Thus, to establish causality between nutritional status and performances, intervention studies has been undertaken, and they have shown that early intervention on health and nutrition increase child probability to be enrolled on time in primary school, and improve cognitive development [15,59].

Strengths and limitations
This review is one step towards better understanding the effects of growth in early childhood on subsequent school trajectories. It is the rst review to highlight the components of the academic trajectory that are in uenced by stunting. This review does have some weaknesses, however. First, almost all studies were identi ed as having moderate risk of confounding bias, even though some of them used advanced methods to control for confusion. This is due to the tool of bias assessment. Second, outcomes and measures of effect varied widely across studies, which limited our ability to estimate pooled effects. However, this diversity allowed us to explore multiple facets of academic performance. Third, we did not obtain su cient data to estimate pooled effects of stunting or height-for-age on dropouts, which suggests that this outcome has not been su ciently studied in the literature. Fourth, due to the low number of studies, we were not able to perform subgroup analysis which may have shown an effect of the timing of stunting (e.g., stunting before 2 years vs stunting after 2 years).

Conclusion
The results show that stunting in childhood might lead to a delay school enrollment, grade repetition, school dropout and low schooling levels. This study is a step towards understanding the overall effect of stunting or height-for-age on academic trajectory. Results showed that impaired growth and development in infancy is associated with a delay of school age entry, an increased risk of grade repetition, and increased school dropout, which, in turn, lead to children's low levels of education. Although this review provides an overall picture of the educational trajectory of children from developing countries who experienced stunting in childhood, further research is needed on the effect of stunting on educational trajectories among this population. Since stunting affects more children from poor communities than from wealthy communities, Pooled effects were not estimated for school dropout because no two studies used the same effect measures. Nevertheless, results from these studies were mitigated. Mendez and Adair (14) reported that stunted children were more likely to drop out of school than non-stunted children. But Glewwe and Jacoby (38) found that taller children tended to leave school earlier, while Bogin and MacVean (33) reported that school continuation or dropout was not in uenced by health or nutritional environment (Table 2 and Table 4 in Additional le 3). Sensitivity analysis The number of studies was not su cient to conduct sensitivity analyses according to risk of bias or subgroup analyses by age. We performed a sensitivity analysis on schooling level by removing one study or estimated effect at a time from the pooled effect size. When we removed the effect size of Tanzania from The Partnership for Child Development (35) study, the Higgin's I² decreased from 90-0% and the magnitude of the pooled effect increased from 0.24 (Fig. 3) to 0.29 (Fig. 5 in Additional le 4).
future research should also explore the effect modi cation of socioeconomic status on the relationship between stunting and school trajectories to inform the development of effective interventions. The current results imply the need for leaders of developing countries to work more for the prevention of stunting through programs and projects focused on nutrition and health problems in childhood. Similarly, health issues should be integrated into education policies to allow for speci c care of stunted children in order to improve their school performance. Meta-analysis of the association between height-for-age and early or late enrollment Meta-analysis of the association between height-for-age and schooling level Figure 4 Meta-analysis of the association between height-for-age and grade repetition

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.