Original research Open access

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
Possible causalities between
malnutrition and academic
performances among primary
schoolchildren: a cross-­sectional study
in rural Madagascar
Hirotsugu Aiga ‍ ‍,1 Kanae Abe,2 Emmanuel Randriamampionona,3
Angèle Razafitompo Razafinombana4

To cite: Aiga H, Abe K, ABSTRACT

Randriamampionona E, et al. What this paper adds?
Background The importance of addressing malnutrition
Possible causalities between is increasing in the context of children’s health and
malnutrition and academic ►► This study successfully estimated ‘possible’ causal-
their academic performances. Childhood malnutrition ities between malnutrition and learning capacities
performances among primary
schoolchildren: a cross-­sectional further could reduce a country’s economic productivity. among primary school children, despite employing a
study in rural Madagascar. BMJ No earlier study adequately estimated the causalities cross-­sectional study as its study design.
Nutrition, Prevention & Health between schoolchildren’s malnutrition and their academic ►► A hypothetical causal path indicates that ‘not being
2021;4:e000192. doi:10.1136/ performances. How nutritional status contributes to stunted’ is likely to have caused higher ‘attendance
bmjnph-2020-000192 children’s academic performances has never been rate’ and thereby higher ‘mathematical proficiency’
For numbered affiliations see reported from Madagascar. This study aims to estimate the in a two-­step manner.
end of article. possible causalities between their nutritional status and ►► It is recommended that school feeding program be

copyright.
academic performances in rural Madagascar. strengthened for increasing children’s school atten-
Correspondence to Methods A cross-­sectional household survey was
Professor Hirotsugu Aiga, School
dance rate and thereby their mathematical profi-
conducted in Antananarivo-­Avaradrano district, ciency in Antananarivo-­Avaradrano district.
of Tropical Medicine and Global
Health, Nagasaki University,
Madagascar, from November to December 2017, by
Nagasaki 852-8523, Japan; targeting 404 first and second graders 5–14 years of
​hirotsugu.​aiga@​nagasaki-​u.​ age enrolled in 10 public primary schools. Children’s schoolchildren’s nutritional status and their academic
ac.j​ p anthropometric measurements and structured interviews performances in Madagascar.
with their mothers/caregivers were conducted. Children’s
academic performances data (mathematical and national
Received 12 October 2020
language proficiencies) were collected at each school. INTRODUCTION
Revised 3 December 2020 Ending hunger and improving nutrition and
Accepted 12 December 2020 To estimate associations between their malnutrition
Published Online First and academic performances, bivariate and multivariate health continue to be a key global develop-
18 January 2021 analyses were conducted. To estimate their possible ment agenda under the Sustainable Devel-
causalities between them, three conditions were examined opment Goal 2 (SDG 2). Ensuring education
(crude covariational relationship, covariational relationship and promoting lifelong learning opportuni-
through controlling for a third variable and temporal ties (SDG 4) serve as an equally important
precedence). SDG.1 The World Bank emphasises the
Results Four independent variables produced significantly importance of comprehensively addressing
positive coefficients with mathematical proficiency in both malnutrition and inadequate learning
multivariate analysis. Of the four, ‘not being stunted’ and during childhood to enable each country to
‘attendance rate’ were estimated to be possible causes have healthier and economically productive
of higher mathematical proficiency because they satisfied
populations.2 3 In line with this, Scaling Up
all the three conditions for a causality. On the other
Nutrition, a global multisectoral movement
hand, three independent variables produced significantly
on nutrition, expanded its target monitoring
positive coefficients with national language proficiency in
multivariate analysis. Yet, none of them were estimated to groups beyond children under 5 years of age
be possible causes of higher national language proficiency. to children 13–15 years of age and older.4
© Author(s) (or their Conclusions A hypothetical causal path indicates Yet, globally children aged 5–12 years often
employer(s)) 2021. Re-­use remain excluded from the monitoring frame-
that ‘not being stunted’ is likely to have caused higher
permitted under CC BY-­NC. No
commercial re-­use. See rights ‘attendance rate’ and thereby higher ‘mathematical works as if they were a neglected subpopu-
and permissions. Published by proficiency’ in a two-­step manner. This study is the first lation,5 6 despite the significant impacts of
BMJ. attempt to estimate the possible causalities between nutritional status on their health, academic

18 Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
achievements and future socioeconomic productivity.7 For over 20 years, Madagascar has been challenged by
It is only UNICEF who started employing prevalence of consistently higher prevalence of stunting among children
malnutrition among 5–19 years of age as the standard under 5 years of age for decades. The latest data (49% as
indicators for its annual publication in 2019.8 of 2018) are the fifth highest in the world.8 To address the
The relationship between childhood nutritional status country’s long-­lasting high malnutrition prevalence, the
and learning capacities is an old and new research topic. government of Madagascar increasingly focuses on multi-
It has been analysed and reported since 1970s9 and sectoral coordination for greater coverage and impacts
recently redrawing global attention from the perspective of the national nutrition programme by launching the
of future economic productivity.1 2 10 A number of earlier third National Nutrition Action Plan (PNAN III). PNAN
studies attempted to estimate the associations between III calls for multisectoral interventions also for school-­age
childhood malnutrition and academic performances.11–16 children, for example, community-­ based deworming,
Yet, a majority of them address the associations school feeding, household food security and crop diver-
between school-­ age children’s malnutrition and their sification.18 Yet, it was not until 2017 that the first nutri-
enrolment in, dropping out from and repeating a grade tion study targeting school-­age children was conducted in
in schooling.11–15 There are few studies that go beyond Madagascar. The study reported the malnutrition preva-
it by addressing the associations between malnutrition lence among school-­age children and the risk factors for
and school attendance. The level of school attendance being stunted, underweight and thin.5 This study attempts
serves as the proxy more sensitive to quality and density to explore possible causalities between schoolchildren’s
of their routine learning at schools and home than those nutritional status and their academic performances by
of enrolment, dropping out and repeating a grade. The assessing whether and how nutritional status contributes
only earlier study that addresses the associations between to their school attendance and further to learning capac-
malnutrition and school attendance employed an ambig- ities in rural Madagascar.
uously defined variable of school attendance, that is, a
dichotomous variable of whether a student is ‘currently
METHODS
attending’ in school.16 It is not clear whether the study
A cross-­sectional survey was conducted to estimate the
defined ‘currently attending’ as being present at school
association between malnutrition (ie, stunting, under-
just on the day of the survey or for a specific number of
weight, thinness and overweight) among schoolchildren
days during the predesignated period. To precisely esti-

copyright.
and their academic performances (school attendance
mate the relationship between malnutrition and school
and learning capacities) in Antananarivo-­ Avaradrano
attendance, the variable representing attendance must
district, Analamanga region, Madagascar.
be more clearly and precisely defined, even ideally taking
seasonality into account. More importantly, note that all Study area and target group
the earlier studies were limited to estimating the associ- Antananarivo-­ Avaradrano district is a typical moun-
ations between school-­age children’s malnutrition and tainous area located in central inland Madagascar, where
their academic performances and that none of them temperature is relatively stable all year round (minimum
attempted to estimate their causalities. Though intu- 10°C to maximum 27°C). Eighty-­one per cent of popu-
itively plausible, it is difficult to ascertain or quantify lations of the district live in rural area. All the 10 public
the causalities between children’s nutritional status and primary schools having been considering improving
their academic performances.17 Hence, there is a clear school feeding programme in the district were selected as
need for attempting to estimating causalities between the target recruitment points (ie, Ambohimarina, Ambo-
children’s nutritional status and their academic perfor- hitrarahaba, Ambohitsoa, Ambovona, Ankadinandriana,
mances without focusing only on their associations. Isahafa, Ikianja, Soavinandriamanitra, Tsarahasina and

Figure 1 Causal models between malnutrition and learning capacities.

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192 19

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
Table 1 Characteristics of sampled schoolchildren 5–14 Table 1 Continued
years of age, their malnutrition prevalence, and learning Characteristics in form of categorical
capacities and ordinal variables n (%)
Characteristics in form of categorical
Household Dietary Diversity Score (HDDS) 6.3 1.33
and ordinal variables n (%)
(pt)*
Gender Malnutrition prevalence n Prevalence (%)
 Male 210 53.4  Stunting 137 34.9 (95% CI
 Female 183 46.6 30.2 to 39.8)
Household’s ownership of land  Underweight† 145 38.0 (95% CI
33.1 to 43.0)
 Own land for housing or farming 85 21.6
 Thinness 44 11.2 (95% CI
 Don’t own land for housing or farming 308 78.4 8.3 to 14.7)
Household’s ownership of mobile phone  Overweight 4 1.0 (95% CI 0.3
 Own mobile phone 271 69.0 to 2.6)
 Don’t own mobile phone 122 31.0 Learning capacities
  n (%)
Major income source of child’s household  Mathematical proficiency‡
 Daily labour 140 35.6  Level 1: beginner 237 62.5
 Agriculture and crop sales 118 30.0  Level 2: simple addition 49 12.9
 Selling, trading and commerce 45 11.5  Level 3: simple subtraction 86 22.7
 Skilled wage labour 28 7.1  Level 4: 2-­digit addition/subtraction 6 1.6
 Unskilled wage labour 26 6.6  Level 5: multiplication 1 0.3
 Handicraft and small-­scale industry 25 6.4  National language proficiency§
 Livestock and sales of animal 5 1.3  Level 1: beginner 259 68.7
 Others 6 1.5  Level 2: letter 53 14.1
Education attainment of mother or  Level 3: word 44 11.7
caregiver  Level 4: paragraph 17 4.5
 Never gone to school 15 3.8

copyright.
 Level 5: essay 4 1.1
 Dropped out from primary school 185 47.1
*Score calculated based on 12 food groups consumed at a household on the
 Primary school 149 37.9 previous day (min 0–max 12).24
†n=388. Eleven children older than 120.8 months of age were excluded, as
 Secondary school 37 9.4
their z-­score for weight for age cannot be calculated.19
 High school 6 1.5 ‡n=379. Fourteen children were excluded, as they were absent from the
examinations of mathematics.
 University and other higher education 1 0.3 §n=377. Sixteen children were excluded, as they were absent from the
Primary school a child enrolled in examinations of national language.

 Ambohimarina primary school 25 6.4

 Ambohitrarahaba primary school 106 27.0 Viliahazo primary schools). Students’ academic perfor-
 Ambohitsoa primary school 38 9.7 mance data were readily accessible at those schools. This
 Ambovona primary school 9 2.3 was key to enabling us to identify their associations with
 Ankadinandriana primary school 21 5.3 students’ nutritional status. Of all children enrolled in
 Isahafa primary school 45 11.5
the 10 schools, exclusively first and second graders were
targeted in this study for two reasons. First, targeting first
 Ikianja primary school 31 7.9
and second graders would help minimise the number
 Soavinandriamanitra primary school 35 8.9
of dropouts at the postintervention stage. They will be
 Tsarahasina primary school 50 12.7
more contactable at the time of postintervention survey
 Viliahazo primary school 33 8.4 to be conducted 4–5 years later. They are more likely to
Grade stay enrolled in the same schools without going on to
 First grade 186 47.3 secondary schools or getting a job elsewhere. Second, a
 Second grade 207 52.7 great majority of first and second graders were children
Characteristics in form of interval ratio Mean SD 5–14 years of age, for whom AnthroPlus (a WHO’s soft-
variables ware) can calculate z-­scores for height for age and body
Age (month) 86.1 16.1 mass index (BMI) for age.19
Household size (person/household) 5.4 1.73
Sample size and sampling methods
School attendance rate (%) 90.5 14.4
National prevalence of stunting among children under
Number of meals on the previous day 2.9 0.16
(meal/day)
5 years of age has been around 50% for decades.20 There-
fore, we assumed that its prevalence among first and
Continued second graders could be 50%, under which sample size

20 Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
Figure 2 Comparison of attendance rate between malnutrition and not malnutrition schoolchildren 5–14 years of age.

becomes the greatest (ie, sufficient sample size also for 5–14 years were randomly selected from the respective
estimating prevalence rates of underweight, thinness, strata using student registries of the 10 primary schools.
overweight and obesity). The sample size was calculated
with α (error)=0.05 and d (precision)=0.05. Thus, 384 Data collection

copyright.
schoolchildren were determined as the required sample Data collection was conducted during the period from
size. Assuming 5% of non-­response rate,21 404 children 20 November to 11 December 2017, early rainy season in
were employed as the final sample size. To ensure equal Madagascar. Anthropometric measurements of sampled
representativity of children in each stratum (school, first and second graders and subsequent structured inter-
grade and sex), probability proportional to size approach views with their parents were conducted. Weights of the
was applied. A total of 404 first and second graders aged children were measured to the nearest 0.1 kg using the
electronic scale for mother/child 150 kg × 100 g (UNICEF
reference model: S0141021). Heights were measured to
the nearest 0.1 cm using the UNICEF standard height
scale. Two repeated household visits were made when
target children and mothers (or other caregivers) were
either absent or not available on the initial visits. When
they were totally unavailable despite three visits, a substi-
tute schoolchild was selected by mechanically sampling
the next eligible child in student registries. The data of
birth dates and academic performances of the selected
children were collected from student registries and
academic records, respectively, at the target primary
schools.

Data analysis
The data collected in the household survey were analysed
by using AnthroPlus (WHO, Geneva) for calculation of
z-­scores for height for age, weight for age and BMI for
age,19 and SPSS for Windows V.22 (IBM/SPSS) for all the
other statistical analyses.
The causal model between malnutrition and learning
capacities this study hypothesised was composed of two
Figure 3 Association between attendance rate and learning steps, that is, first step: ‘malnourished students are likely
capacities among schoolchildren 5–14 years of age (n=393). to more poorly attend schools’, and second step: ‘poorly

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192 21

 

22
Table 2 Ordinal regressions on subject proficiencies with background variables
Mathematical proficiency National language proficiency
Background variables B: Coefficient estimate 95% CI P-­value B: Coefficient estimate 95% CI P-­value
Academic variables
 Attendance rate (%) 0.027 (0.005–0.048) 0.014* 0.017 (0.042–0.076) <0.001**
 Grade * (Dropped due to multicollinearity) (Dropped due to multicollinearity)
Nutrition and food security variables
 
Stunting
  
Stunted (Reference) (Reference)
  
Not stunted 0.620 (0.105–1.136) 0.018* 0.925 (-0.499–0.549) 0.925
 
Underweight
  
Underweight (Reference) (Reference)
  
Not underweight (Dropped due to multicollinearity) (Dropped due to multicollinearity)
 
Thinness
  
Thin (Reference) (Reference)
  
Not Thin 0.681 (-0.076–1.432) 0.076 1.007 (0.160–1.853) 0.020*
 
Overweight
  
Overweight (Reference) (Reference)
  
Not overweight −0.917 (-2.988–1.154) 0.385 0.936 (-1.764–3.636) 0.497
 Number of meals on the previous day −0.439 (-2.226–1.349) 0.630 −0.700 (-2.438–1.039) 0.430
 Household diet diversity score (%) 0.160 (-0.026–0.345) 0.091 0.080 (-0.114–0.274) 0.419
Socio-­demographic variables
 
Gender
  
Male −0.152 (-0.597–0.505) 0.505 −0.425 (-0.899–0.050) 0.079
  
Female (Reference) (Reference)
 Age(months) 0.046 (0.030–0.062) <0.001** 0.059 (0.042–0.076) <0.001**
 Number of household members(person) 0.113 (-0.026–0.251) 0.111 0.059 (-0.084–0.203) 0.417
Socio-­economic variables
 
Household’s ownership of land
  Own land for house/farm (Reference) (Reference)
  Don’t own land for house/farm −0.279 (-0.273–0.832) 0.321 0.378 (-0.197–0.953) 0.197
 
Household’s ownership of mobile phone
  Own mobile phone (Reference) (Reference)
  Don’t own mobile phone −0.172 (-0.655–0.321) 0.494 −0.487 (-1.012–0.038) 0.069
 
Major income source of child’s household
  
Daily labour (Reference) (Reference)
BMJ Nutrition, Prevention & Health

Continued

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

copyright.
BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
 Table 2 Continued
Mathematical proficiency National language proficiency
Background variables B: Coefficient estimate 95% CI P-­value B: Coefficient estimate 95% CI P-­value
  Agriculture and crop sales −0.180 (-0.789–0.428) 0.561 −0.288 (-0.933–0.357) 0.381
BMJ Nutrition, Prevention & Health

  Selling, trading and commerce −0.864 (-1.578–−0.149) 0.018 * −0.318 (-1.073–0.437) 0.409
  Skilled wage labour −0.557 (-1.397–0.282) 0.193 −0.238 (-1.135–0.659) 0.603
  Unskilled wage labour −0.138 (-1.094–0.819) 0.778 0.044 (-0.999–1.087) 0.934
  Handicraft and small-­scale industry −0.786 (-1.695–0.124) 0.090 −0.359 (-1.304–0.585) 0.456

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

  Livestock and sales of animal 0.010 (-1.947–1.966) 0.992 −1.170 (-2.932–0.591) 0.193
  
Others † 1.139 (-1.231–3.510) 0.346 15.454 (15.454–15.454) n.a.‡
 
Education attainment of mother/caregiver
  Dropped out from primary school (Reference) (Reference)
  Never gone to school 0.148 (-1.421–1.124) 0.819 −0.150 (-1.502–1.201) 0.827
  
Primary school −0.455 (-0.952–0.043) 0.073 −0.351 (-0.883–0.181) 0.196
  
Secondary school −0.571 (-1.347–0.206) 0.150 0.575 (-1.393–0.242) 0.168
  
High school 15.703 (15.703–15.703) n.a.‡ 15.278 (15.278–15.278) n.a.‡
  University and other higher education −0.727 (-4.440–2.986) 0.701 −1.104 (-4.741–2.533) 0.552
Regression model parameters R2=0.220, P<0.001 R2=0.239, P<0.001

*Indicates either first grade or second grade.

†Includes: (1) remittance from family members; (2) government allowance (eg, pension); and (3) begging and assistance.
‡SPSS was unable to calculate p value.
§ *P <0.05, **P <0.01
n.a., not applicable.

23
copyright.
BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
Table 3 Possible causalities between malnutrition and learning capacities
 Three conditions for a causality
 (Condition 2)
 Multivariate
analysis:
Covariational
relationship
 (Condition 1) through
Bivariate analysis: controlling for  (Condition 3)  Possible
Causality between two variables Crude covariational relationship third variables  Temporal precedence causality
Causal path A: on mathematical proficiency
       
(Step 1)  Yes  Yes  Yes Yes
From not being stunted to  P=0.003*  P=0.041*  While stunting is chronic
attendance rate  Mann-­Whitney U test*  Ordinal undernutrition that
regression¶ occurs generally before
3 years of age,24 school
attendance takes place
after enrolment at 5 or 6
years of age at primary
schools. Thus, clearly,
school attendance follows
becoming stunted.
(Step 2)  Yes  Yes  Yes Yes
From attendance rate to  P=0.013*  P=0.014*  Without attending
mathematical proficiency  Spearman’s correlation†  Ordinal schools, children cannot
regression‡ take mathematics
examinations. Thus,
clearly, mathematical
proficiency follows
attendance.
Causal path B: on national language proficiency
       

copyright.
(Step 1)  No  Yes  Not necessarily No
From not being thin to attendance  P=0.141  P=0.020*  Thinness is acute
rate  Mann-­Whitney U test*  Ordinal undernutrition that can
regression¶ occur after attendance
at school, too. Thus,
school attendance does
not necessarily follow
becoming thin.
(Step 2)  No  No  Yes No
From attendance rate to national  P=0.104  P=0.483  Without attending
language proficiency  Spearman’s correlation (rs)†  Ordinal schools, children cannot
regression§ take national language
examinations. Thus,
national language
proficiency follows
attendance.

*See figure 2.
†See figure 3.
‡Coefficient estimate for not being stunted in multiple regression model with attendance rate (%) and other all background variables as dependent
and independent variables, respectively.
§Coefficient estimate for not being thin in multiple regression model with attendance rate (%) and other all background variables as dependent and
independent variables, respectively.
¶See table 2.
** * P <0.05

attending students are likely to have poorer learning indicated that attendance rate was not normally distrib-
capacities’ (figure 1). uted. In this study, attendance rate was defined as the
First, with the two steps in mind (figure 1), we exam- proportion of the number of schooldays on which a given
ined bivariate associations between children’s nutritional student fully attended to the total number of schooldays
status and their attendance and further between their in academic year 2017 as of the data collection day. At the
attendance and learning capacities. To examine the bivar- time of conducting the survey, only about a month had
iate associations between children’s nutritional status passed since the academic year 2017 started. Thus, the
and their attendance rates, Mann-­ Whitney U test was rate indicates each student’s attendance status not on a
undertaken. This is because the results of Levene’s test year-­round basis but during the initial 1 month only. Yet,

24 Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
Figure 4 Hypothetical causal paths between malnutrition and learning capacities.

it is likely to appropriately represent the children’s and correlation coefficient (Spearman’s rho: rs). When
their parents’ updated behaviours towards the new school |rs|>0.4 with P<0.05 was detected between two indepen-
year. Then, to examine bivariate associations between dent variables, one of the two variables was excluded from
children’s attendance rates and their learning capacities, the independent variables to be entered into an ordinal

copyright.
Spearman’s correlation coefficients (Spearman’s rho) regression model.
were employed. This is because learning capacities of two Third, nutritional status variables producing signif-
subjects (ie, mathematical and national language profi- icant associations with learning capacities in ordinal
ciencies) are ordinal variables whose data were recorded regressions, if any, were further assessed on their possible
in a 5-­point scale, while attendance rate is an interval ratio causalities with mathematical and national language
variable. proficiencies. To detect a causality between two vari-
Second, to identify the child-­specific determinants of ables, generally three conditions need to be satisfied as
learning capacities of each subject, ordinal regression minimum requirement, that is, crude covariational rela-
was performed by employing learning capacities as the tionship, covariational relationship through controlling
dependent variable and its background variables as the for a third variable and temporal precedence.23 In this
independent variables. The background variables were study, we assumed a possible causality exists between
composed of four types of variables in an appropriately two variables, only when all these three conditions were
balanced manner22: (1) academic variables (attendance satisfied.
rate and grade); (2) nutrition and food security vari-
ables (stunting, underweight, thinness, overweight, meal
frequency and Household Dietary Diversity Score); (3) RESULTS
sociodemographic variables (ie, gender, age, household Of 404 sampled first and second graders aged 5–14 years,
size and mother’s/caregiver’s education attainment); and 11 did not participate in the study due to refusal or
(4) socioeconomic variables (household’s ownerships of absence. Thus, 399 children participated in the study. It
land and mobile phone, and major source of household’s was found that another 11 of the 393 children were over
income). For all the categorical or ordinal variables to 120.8 months of age (ie, 10.07 years of age). They were
be employed as the independent variables composed of excluded from analyses related to underweight, since
three or more categories (‘Major income sources’, ‘Educa- AnthroPlus allows z-­scores for weight for age to be calcu-
tion attainment of mother or caregiver’), dummy variables lated exclusively for children up to 120.8 months of age.19
were created as the independent variables for ordinal Table 1 shows the characteristics of children having
regressions. The category with the greatest frequency participated in the study and their malnutrition prev-
was designated as the reference for the dummy variables. alence and academic performances.19 24 The overall
Moreover, prior to applying them to ordinal regressions, prevalence rates of stunting, underweight, thinness and
the multicollinearity between those possible indepen- overweight were 34.9% (95% CI 30.2% to 39.8%), 38.0%
dent variables was examined by assessing Spearman’s (95% CI 33.1% to 43.0%), 11.2% (95% CI 8.3% to 14.7%)

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192 25

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
and 1.0% (95% CI 0.3% to 2.6%), respectively (table 1). DISCUSSION
There was no case of obesity (z-­score for BMI for age >2). The results of the ordinal regression on mathematical
Figure 2 shows the results of bivariate analyses proficiency (table 2) indicate that a child not being
between children’s nutritional status and their atten- stunted is 1.86 times more likely (=exp[0.620]) to achieve
dance rates. School attendance rates among children higher level of mathematical proficiency. It was also
not being stunted and among those not being under- found that 1% increase in attendance rate would lead to
weight were significantly higher than among those being slightly higher proficiency in mathematics, that is, 1.027
stunted (p=0.003) and among those being underweight (=exp[0.027]) times higher.
(p=0.005), respectively. Yet, no significant difference was In addition to the aforementioned associations,
identified in attendance rates between those being thin possible causalities between ‘not being stunted’ and ‘atten-
and not being thin and between those being overweight dance rate’, and between ‘attendance rate’ and ‘mathemat-
and not being overweight. Figure 3 shows the results of ical proficiency’ were detected (table 3). Figure 4 presents
bivariate analyses between children’s attendance rates hypothetical causal paths between malnutrition and
and their learning capacities in the two subjects (ie, math- learning capacities based on these findings. It is assumed
ematics and national language). A significantly positive that ‘not being stunted’ is likely to have led to higher ‘mathe-
association was detected between attendance rate and matical proficiency’ in a two-­step manner through syllogism
mathematical proficiency (rs=0.128; P=0.0013). Similarly, (Causal path A in table 3 and figure 4). Similar possible
a slightly positive association was detected between atten- causalities were not detected between other three types
dance rate and national language proficiency, yet it was of non-­malnutrition (ie, not being underweight, thinness
not statistically significant (rs=0.064; P=0.104). and overweight) and learning capacities. First, ‘not being
The bivariate associations shown in figures 2 and 3 underweight’ was unable to be assessed on its causalities
were assessed without controlling possible confounders. with learning capacities because it was excluded from
Therefore, ordinal regressions were further performed as the ordinal regression model due to multicollinearity
multivariate analyses by employing learning capacities as with ‘not being stunted’. Second, causality was not detected
their dependent variables. Table 2 shows the results of the between ‘not being thin’ and ‘national language profi-
ordinal regressions on mathematical proficiency and on ciency’ (Causal path B in table 3 and figure 4), though
national language proficiency. In the ordinal regression it produced a significantly positive coefficient estimate
on mathematical proficiency, four independent variables in the ordinal regression model (table 2). Third, there

copyright.
produced significantly positive coefficients (P<0.05), that is no room for possible causality between ‘not being over-
is, ‘Attendance rate’, ‘Not being stunted’, ‘Age’ and ‘Selling, weight’ and learning capacities, since it was associated
trading and commerce as major income source’. In the ordinal neither with mathematical proficiency nor with national
regression on national language proficiency, three inde- language proficiency in the ordinal regression (table 2).
pendent variables produced significantly positive coeffi- Earlier studies in Ethiopia, Sri Lanka and Uganda
cients (P<0.05), that is, ‘Attendance rate’, ‘Not being thin’ reported that being stunted was negatively associated
and ‘Age’. with mathematical proficiency of primary schoolchil-
Of the four independent variables having produced dren.26–28 The results of our study support this trend. Yet,
significant coefficient estimates in the ordinal regression our study found that being stunted was significantly asso-
on mathematical proficiency (table 2), ‘not being stunted’ ciated exclusively with poorer mathematical proficiency,
and ‘attendance rate’ are the variables that compose the while two earlier studies detected the similar associations
causal model between malnutrition and learning capac- with both mathematical and national language profi-
ities (figure 1). Thus, they are likely to have logically ciencies27 28 or exclusively with national language profi-
caused higher mathematical proficiency. Therefore, the ciency.29 Thus, the type of subjects whose proficiencies are
causalities between not being stunted and attendance associated with being stunted differs between the school-
rate, and further between attendance rate and mathe- children settings and contexts (eg, countries, grades and
matical proficiency were examined. As shown in ‘Causal age). An earlier study in Chile reported that being over-
path A’ of figure 4, it is assumed that ‘not being stunted’ is weight was not associated with poorer mathematical and
one of the causes of higher ‘attendance rate’, and higher national language proficiencies.30 Our study did not find
‘attendance rate’ is further one of the causes of higher the association between being overweight and academic
‘mathematical proficiency’. This is because all the three performance either, despite its plausibility.
conditions for a causality (crude covariational relation- Evidence suggests that stunting is generally irrevers-
ship, covariational relationship through controlling for ible after the first 1000 days from conception (ie, up to
a third variable and temporal precedence) were satisfied 2 years of age)31 32 despite its slight possibilities reported
(table 3).5 8 25 Similarly, we assessed causalities between by several earlier studies.33 34 Therefore, step 1 in Causal
not being thin and attendance rate, and further between path A (figure 4) would not be able to be effectively and
attendance rate and national language proficiency were efficiently addressed during school age. Yet, there should
examined. Since parts of three conditions were not satis- be other channels which lead to the increase in school
fied, possible causality in ‘Causal path B’ was not detected attendance rate and thereby mathematical proficiency.
(table 3).25 A number of earlier studies reported school feeding

26 Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
programmes are likely to have increased attendance rate future studies. As the data were collected in relatively food
and further mathematical proficiency.35–38 Currently, 10 secure season (ie, November to December),39 there were
primary schools this study targeted implement school some risks of underestimation of malnutrition prevalence
feeding programme.5 Yet, frequencies of school lunch (particularly thinness, proxy for acute undernutrition)
provisions at those schools are limited only to 1 or 2 days and food security situation. To explore more generalis-
a week primarily during food insecure seasons. There- able causalities in view of volatility of local food security
fore, strengthening school feeding programme would be situation, a follow-­up survey should be longitudinally
an effective intervention towards higher attendance rate conducted by targeting the children having participated
and thereby higher mathematical proficiency. in this study during less food secure season. Second, this
‘Age’ was the variable producing a significantly posi- study has the limitations in presenting the length of the
tive coefficient estimate commonly with mathematical lags between being not stunted and increased attendance
and national language proficiencies (figure 4). In the rate and between increased attendance rate and higher
10 primary schools, the levels of both mathematical and mathematical proficiency. Also to appropriately estimate
national language proficiencies are defined based not the lengths of the lags, a longitudinal study needs to be
on grade-­specific learning targets but on overall learning conducted. Third, this study did not employ children’s
targets for the entire 6-­year primary education (table 1). IQ scores as the variable for bivariate and multivariate
Therefore, a greater proportion of older children partic- analyses. Thus, schoolchildren’s congenital intellectual
ipating in the study tend to attend mathematics and abilities to be approximated largely by IQ should be
national language classes for second graders. In fact, considered and addressed in the future studies.
‘age’ and ‘grade’ are positively correlated with signifi-
cance (rs=0.610; p<0.001). Probably, for this reason, ‘age’
was significantly associated with both mathematical and
national language proficiencies in the ordinal regression CONCLUSIONS AND RECOMMENDATIONS
models. This study identified ‘not being stunted’ and ‘attendance
It is difficult to interpret the significant association between rate’ as the determinants or predictors of ‘mathematical
mathematical proficiency and ‘selling, trading and commerce’ as proficiency’ and as possible causes of higher ‘mathematical
the major household income source. Needless to say, those proficiency’. While admitting need and room for improve-
practising petty selling, trading and commerce should be ment of study design, note that this is the first study that

copyright.
competent enough in basic arithmetic for daily cash trans- reports possible nutrition-­related causalities of learning
actions. Most likely part of those from the households whose capacities among primary schoolchildren in Madagascar.
greatest income source is ‘selling, trading and commerce’ some- Malnutrition, in particular stunting, has been drawing
times are either taught mathematics at home or relatively global attention due to its determinativeness in cognitive
strongly advised to study mathematics at school, by their performances during preschool40 and later academic perfor-
parents. This is one of the possible interpretations of this mances during primary education.41 42Recently, the World
result of the ordinal regression on mathematical proficiency. Bank has been highlighting a need for addressing malnutri-
Yet, we are unable to further deepen the discussion on this tion among school-­age children as a foundational investment
result, since we could identify no earlier studies on this topic. each country should make by launching the Human Capital
Some may say such a cross-­sectional study as this cannot Measurement Project.2 3 43 Thus, a global momentum is being
identify causalities. Yet, this study successfully estimated created towards generation of evidence on malnutrition
at least ‘possible’ causalities between malnutrition and among school-­age children. In this context, further studies
learning capacities. There are two major reasons for the should be conducted to more precisely identify not just asso-
successful estimation. First, both crude covariational ciations but causalities between malnutrition and academic
relationship and covariational relationship through performances. That will help health and education stake-
controlling third variables were systematically undertaken holders more effectively collaborate for addressing malnu-
by using bivariate and multivariate analyses. Second, trition among schoolchildren in relation to their academic
temporal precedence between step 1 and step 2 of the performances in all countries suffering from greater malnu-
causal model (figure 1) was self-­evidenced enough. For trition prevalence.
instance, generally children become stunted when they It is recommended that the following actions be taken
are 2 years of age or younger25 31 32 and schoolchildren in view of the implications and significance of the results
physically cannot take examinations without attending of this study.
schools (table 3). ►► Strengthen the existing school feeding programme
There are two types of limitations in this study. First, so as to ensure more frequent and sustainable provi-
this study may have the limitations in reliability and sion of school lunch. All the 10 primary schools this
generalisability of the causalities between malnutrition study targeted are currently attempting to introduce a
and learning capacities due to its study design (ie, cross-­ community-­based school feeding programme. Thus,
sectional study). To increase the reliability of Causal path design and strategic planning of the programme
A in figure 4, data should be longitudinally collected from should be strengthened by proactively promoting the
schoolchildren in a time series manner when conducting use of locally produced, procured and donated foods,

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192 27

  BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
for instance, through readily functioning school includes any translated material, BMJ does not warrant the accuracy and reliability
management committees. of the translations (including but not limited to local regulations, clinical guidelines,
terminology, drug names and drug dosages), and is not responsible for any error
►► Regularly, measure at least height of children at each and/or omissions arising from translation and adaptation or otherwise.
primary school. Though height scale to be used for
Open access This is an open access article distributed in accordance with the
nutrition survey is generally expensive (US$150– Creative Commons Attribution Non Commercial (CC BY-­NC 4.0) license, which
US$300),44 a simple height scale can be handmade permits others to distribute, remix, adapt, build upon this work non-­commercially,
on wall in or out of a classroom by setting gradua- and license their derivative works on different terms, provided the original work is
properly cited, appropriate credit is given, any changes made indicated, and the use
tion scale every 1 or 5 mm. Plotting a dot on growth
is non-­commercial. See: http://​creativecommons.​org/​licenses/​by-​nc/​4.​0/.
chart will serve as an invaluable opportunity to better
equip students with understanding and skills on two-­ ORCID iD
dimensional axes (an essential part of mathematics). Hirotsugu Aiga http://​orcid.​org/​0000-​0001-​5910-​3379
Regular monitoring of height of children will also
provide each school with opportunities to increase
awareness of growth monitoring as an essential part REFERENCES
of school health. 1 United Nations (UN). Sustainable development goals. New York:
UN, 2020. https://​sust​aina​bled​evel​opment.​un.​org/​sdgs. (accessed
November 25, 2020).
Author affiliations 2 World Bank. World development report 2019: changing nature of
1
School of Tropical Medicine and Global Health, Nagasaki University, Nagasaki, work. Washington DC: World Bank, 2019. https://www.​worldbank.​
Japan org/​en/​publication/​wdr2019. (accessed November 25, 2020).
2 3 World Bank. Insights from disaggregating the human capital
IC Net Ltd, Saitama, Japan
3 index. Washington DC: World Bank, 2020. http://​documents1.​
Association des consultants a la recherche au devellopement economique et social worldbank.​org/​curated/​en/​306651578290912072/​pdf/​Insights-​from-​
(ACREDES), Ministry of National Education, Antananarivo, Madagascar Disaggregating-​the-​Human-​Capital-​Index.​pdf
4
JICA Participatory and Decentralized School Management Support Project, 4 Scaling Up Nutrition (SUN). Sun movement monitoring, evaluation,
Antananarivo, Madagascar accountability, and learning. sun: Geneva, 2017. Available: https://​
scalingupnutrition.​org/​progress-​impact/​monitoring-​evaluation-​
accountability-​and-​learning-​meal/ [Accessed November 25, 2020].
Acknowledgements The authors gratefully thank Vonjy Nirina Andrianome for her 5 Aiga H, Abe K, Andrianome VN, et al. Risk factors for malnutrition
overall support to this study. The authors are grateful to Masahiro Hara, Toshiyasu among school-­aged children: a cross-­sectional study in rural
Murai and Nobuhiro Kunieda for their valuable technical support. The authors Madagascar. BMC Public Health 2019;19:773.
further thank Japan International Cooperation Agency (JICA) for its funding support 6 Best C, Neufingerl N, van Geel L, et al. The nutritional status
to the study. This work is sincerely dedicated to all the school-­age children, their of school-­aged children: why should we care? Food Nutr Bull
2010;31:400–17.

copyright.
parents and caregivers, and teaching staff of primary schools in rural areas of
7 Adair LS, Fall CHD, Osmond C, et al. Associations of linear growth
Madagascar. This study was conducted as a part of the JICA technical cooperation and relative weight gain during early life with adult health and human
'Project for Participatory and Decentralized School Management Support'. capital in countries of low and middle income: findings from five birth
Contributors All the authors made substantial intellectual contributions to cohort studies. Lancet 2013;382:525–34.
the study. HA and KA conceptualised and designed the study. ER, HA, KA and 8 United Nations Children’s Fund (UNICEF). State of world’s children
2019. New York: UNICEF, 2019. https://www.​unicef.​org/​sites/​default/​
ARR collected the data. HA analysed and interpreted the data. HA drafted and files/​2019-​12/​SOWC-​2019.​pdf. (accessed November 25, 2020).
finalised the manuscript. KA, ER and ARR critically commented on and revised the 9 Birch HG. Malnutrition. Learning, and intelligence. Am J Public
manuscript. The authors took full and final responsibility for submitting this article Health 1972;62:773–84.
for publication. All the authors reviewed and approved the final version of the 10 Chutikul S. Malnourished Children: An Economic Approach to the
manuscript. Causes and Consequences in Rural Thailand. Technical Report 102.
Honolulu: East-­West Center, 1986.
Funding This work was supported by Japan International Cooperation Agency 11 Jamison DT. Child malnutrition and school performance in China. J
(JICA). Dev Econ 1986;20:299–309.
Map disclaimer JICA agreed on the study design but played no roles in data 12 Moock PR, Leslie J. Childhood malnutrition and schooling in the
Terai region of Nepal. J Dev Econ 1986;20:33–52.
collection, data analysis and interpretation of the results. 13 Pollitt E. Malnutrition and Infection in the Classroom. Paris: UNESCO,
Competing interests None declared. 1990.
14 Harbison R, Hanushek E. Educational Performance of the Poor:
Patient consent for publication Not required. Lessons from Rural North-­east Brazil. Washington DC: Oxford
Ethics approval The study protocol was submitted to the National Committee University Press, 1992.
15 Behrman JR. The impact of health and nutrition on education. World
for Biomedical Research Ethics, the Malagasy Ministry of Health, for its ethical
Bank Res Obs 1996;11:23–37.
clearance. The Committee officially approved the study protocol by exempting it 16 Khanam R, Nghiem HS, Rahman MM. The impact of childhood
from an ethical clearance process due to low-­level invasiveness of the study design malnutrition on schooling: evidence from Bangladesh. J Biosoc Sci
(Exemption letter No 52). An advanced informed consent to participate in both 2011;43:437–51.
anthropometric measurements and structured household interviews was obtained 17 López-­Casasnovas G, Rivera B, Currais L. Health and economic
in written form from the parents of sampled children. In addition, an additional growth: findings and policy implications. Cambridge MA: The MIT
verbal informed consent to participate in the study was further obtained from the Press: 177–84.
parents upon the household visit for a structured interview, in case they might 18 L’Office National de Nutrition (ONN). Plan National d’Action pour la
NutrtionⅢ 2017-2021. Antananarivo: ONN, 2017
change their minds about participation in the study. 19 World Health Organization (WHO). AnthroPlus for personal computers
Provenance and peer review Not commissioned; externally peer reviewed. manual. Geneva: WHO, 2009. http://www.​who.​int/​growthref/​tools/​
who_​anthroplus_​manual.​pdf?​ua=1
Data availability statement Data are available upon reasonable request. 20 International Food Policy Research Institute (IFPRI). Global nutrition
Deidentified participant data are available upon reasonable request. report 2016 – from promise to impact, ending malnutrition by 2030.
Washington DC: IFPRI, 2016. http://​ebrary.​ifpri.​org/​utils/​getfile/​
Supplemental material This content has been supplied by the author(s). It has
collection/​p15738coll2/​id/​130354/​filename/​130565.​pdf. (accessed
not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been November 25, 2020).
peer-­reviewed. Any opinions or recommendations discussed are solely those 21 ACTwatch. Household Survey Madagascar 2012 Survey Report.
of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and Washington DC: ACTwatch, 2013. http://www.​actwatch.​info/​sites/​
responsibility arising from any reliance placed on the content. Where the content default/​files/​content/​publications/​attachments/​ACTwatch%​20HH%​

28 Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192

 BMJ Nutrition, Prevention & Health

BMJNPH: first published as 10.1136/bmjnph-2020-000192 on 18 January 2021. Downloaded from http://nutrition.bmj.com/ on June 30, 2021 at Nagasaki University Library. Protected by
20Report%​20Madagascar%​202012.​pdf. (accessed November 25, 35 Powell CA, Walker SP, Chang SM, et al. Nutrition and education: a
2020). randomized trial of the effects of breakfast in rural primary school
22 Simkins C, Paterson A. Learner performance in South Africa: children. Am J Clin Nutr 1998;68:873–9.
Social and economic determinants of success in language and 36 Grillenberger M, Neumann CG, Murphy SP, et al. Food supplements
mathematics. Cape Town: Human Sciences Research Council, 2005. have a positive impact on weight gain and the addition of animal
23 Oppewal H. Concept of causality and conditions for causality. Wiley source foods increases lean body mass of Kenyan schoolchildren. J
International encyclopedia of marketing. Hoboken: John Wiley & Nutr 2003;133:3957S–64.
Sons, Inc, 2010. 37 Ahmed AU. The impact of feeding children in school: evidence
24 Food and Agriculture Organization of the United Nations (FAO). from Bangladesh. Washington, DC: International Food Policy
Guidelines for measuring household and individual dietary diversity. Research Institute, 2004. http://​e brary.​i fpri.​o rg/​u tils/​g etfile/​
Rome: FAO, 2010. collection/​p 15738coll2/​i d/​1 26329/​f ilename/​1 26540.​p df.
25 Ray S. Relationship between stunting and wasting in children. Med J (accessed November 25, 2020).
Armed Forces India 2013;69:410. 38 Jomaa LH, McDonnell E, Probart C. School feeding programs in
26 Haile D, Nigatu D, Gashaw K, et al. Height for age Z score and developing countries: impacts on children's health and educational
cognitive function are associated with academic performance outcomes. Nutr Rev 2011;69:83–98.
among school children aged 8-11 years old. Arch Public Health 39 Famine Early Systems Network (FEWS Net). Madagascar food
2016;74:17. security outlook June 2017-­January 2018. Antananarivo: FEWS
27 Sarma MSG, Wijesinghe DGNG, Sivananthawerl T. The effects of net, 2017. Available: https://​fews.​net/​sites/​default/​files/​documents/​
nutritional status on educational performance of primary school reports/​Madagascar_​Outlook_​June2017_​Draft_​Final._​0.​pdf
children in the plantation sector in Nuwara Eliya educational zone. [Accessed 25 Nov 2020].
Trop. Agric. Res. 2013;24:203–14. 40 Silva TM, Bueno NB, Azevedo MdeLdaSGde, et al. Cognitive
28 Acham H, Kikafunda JK. Height, weight, body mass index and performance of stunted pre-­s chool children undergoing
learning achievement in Kumi district, East of Uganda. Sci Res Essay nutritional recovery treatment. Rev Paul Pediatr 2018;36:39–44.
2008;3:1–8. 41 Wild CP, Miller JD, Groopman JD. Mycotoxin control in low- and
29 Shariff ZM, Bond JT, Johnson NE. Nutrition and educational middle-­income countries: IARC Working Group Reports, No. 9
achievement of urban primary schoolchildren in Malaysia. Asia Pac J Chapter 2. Child stunting in developing countries. International
Clin Nutr 2000;9:264–73. Agency for Research on Cancer (IARC: Lyon, 2015.
30 Flores OC, Orellana YZ, Leyton BD, et al. Overnutrition and 42 Perkins JM, Kim R, Krishna A, et al. Understanding the association
scholastic achievement: is there a relationship? an 8-­year follow-­up between stunting and child development in low- and middle-­income
study. Obes Facts 2018;11:344–59. countries: next steps for research and intervention. Soc Sci Med
31 Crookston BT, Penny ME, Alder SC, et al. Children who recover from 2017;193:101–9.
early stunting and children who are not stunted demonstrate similar 43 World Bank. Human capital summit 2017- Committing to Action to
levels of cognition. J Nutr 2010;140:1996–2001. Drive Economic Growth. Washington DC: World Bank, 2017. http://
32 Black RE, Victora CG, Walker SP, et al. Maternal and child www.​worldbank.​org/​en/​news/​feature/​2017/​10/​20/​countries-​commit-​
undernutrition and overweight in low-­income and middle-­income to-​strong-​action-​on-​human-​capital-​to-​drive-​economic-​growth.
countries. Lancet 2013;382:427–51. (accessed November 25, 2020).
33 Golden MH. Is complete catch-­up possible for stunted malnourished 44 United Nations Children’s Fund (UNICEF). Unicef target product
children? Eur J Clin Nutr 1994;48 Suppl 1:S58–70. profile Height/length measurement devices. New York: UNICEF, 2017.
34 Prentice AM, Ward KA, Goldberg GR, et al. Critical windows https://www.​unicef.​org/​supply/​media/​1291/​file/​Target%​20product%​

copyright.
for nutritional interventions against stunting. Am J Clin Nutr 20profile%​20(​TPP):%​20Height%​20length%​20measurement%​
2013;97:911–8. 20device.​pdf. (accessed November 25, 2020).

Aiga H, et al. bmjnph 2021;4:e000192. doi:10.1136/bmjnph-2020-000192 29