Preventing NCDs in Adolescence Crucial

Introduction

The prevalence of non-communicable diseases (NCDs) is increasing rapidly in lowand middleincome countries (LMICs) as the epidemiological transition from infectious diseases to NCDs occurs(1).The consequences of NCDs threaten the social and economic development and public health(2–5). In 2016, almost 75% of all NCDs deaths were from LMICs(6).Also, NCDs are one of the major causes of death before the age of 70, which is known as ‘premature death’(7). Approximately 82% of NCDs caused premature deathoccurred in LMICs(8).The major risk factors have been identified concerning the various incidences of risk of NCDs which included utilisation of tobacco, harmful use of alcohol, physical inactivity, unhealthy diet and overweight/obesity(9). Those risk factors are related to modifiable and preventable behaviours, which are also known as ‘modifiable behavioural risk factors’(9).An important strategy to prevent and control NCDs is to target on reducing those behavioural risk factors(10–14). Increasing evidences showed that addressing thebehaviouralrisk factors during adolescence can offeran important opportunity as a primary prevention to reduce the risk of NCDs in theirlateradulthood(15,16).Adolescence is a crucial phase ofthe transition from childhood to adulthoodwith rapid changes in one’s physical, cognitive and emotional conditions(17). This is also an optimal period of time to establish good lifestyle behaviours, and those behaviours will have a better chance to be kept for the whole lifespan(16). World Health Organization (WHO) defines adolescentsas the people in the ages between 10 to 19 years old(15).Globally, 11.7% adolescents are drinking alcohol, 81% adolescents have insufficient physical activity, 150 million youths use tobacco, and 124millionadolescents are overweight and obese(6,9,18).Compared with high income countries, adolescents from LMICs have a higher chance of being exposed to the behavioural risk factors, as the level of socioeconomic status (SES)are negatively associated with thoserisk factors(19). Also, the rapid adoption of those behaviours among adolescents are closely related with rapid urbanization and globalization in LMICs(16,20).On the other hand, more attention is needed for this population group in LMICs as theycurrently contribute to a large proportion of thetotal population(21). There are an estimated 1.2 billion adolescents globally and 86% of them are living in LMICs, see Figure 1(22).

With the urgent need to tackle behavioural risk elements and factors which are integral to NCDs among adolescents in LMICs, this is critical to investigate their determinants.The development of adolescent behavioural risk factors are mainly determined by the wider social and familyfactors such as household food insecurity(17,23).Household food insecurity is defined as ‘limited or uncertain availability of nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways’(24).There was an estimate of 222 million people from least-developed countries suffer from severe food insecurity(25).However, most of the evidences on investigating the correlation between household food insecurity and the development of adolescent behavioural risk factors are from high income countries, and their association in LMICs setting is still under investigated(26).There is a critical gap inliterature and policy-making in the contexts where food insecurity is highest incident globally and the largest share of the world’s adolescents live(26). Thus, this study aims to provide new evidence on the association between household food insecurity and adolescent NCDs behavioural risk factorsin LMICs. The following part will review the previous researches related to household food insecurity and adolescent NCDs behavioural risk factors.Then the limitation of previous literatures, the theoretical pathways of their linkage in LMICs and the role of thisproposed study to fill the research gap will also be discussed.The hypothesis, aim and objectives of this study will be listed at the end of this section.

Household food insecurity and adolescent NCDs behavioural risk factors

Adolescents have been recognised as a ‘forgotten group’ in the current food insecurity studies, as most of the literatures conducted based on early child or adult population(27).Previous literatures investigated the association between household food insecurity and NCDs behavioural risk factorsamong adolescents werevery limited(26).A cross-sectional study done by Romo in 2016 determined the association between hunger and increased risk of adolescent NCDs risk behaviours aged 16 in Bolivia(28). By using the data from a 2012 global school-based student health self-administered survey, this study finds that hunger was associated with a significant increased risk of tobacco use (Adjusted odds ratio(AOR): 2.12, 95%CI: 1.50-2.99), insufficient physical activity (AOR: 1.21, 95%CI: 1.08-1.47) and no daily fresh vegetable and fruit consumption (AOR: 1.21, 95%CI: 1.11-1.31)(28). Even though this study provides some significant results, the scope of this study is to focus on hunger to represent the food security status. Food insecurity not only refers to being hungry due to insufficient amount of food to eat, but also it includes the quality of food intake, frequency of worry about lack of food and any coping actions to deal with food insecure situation(29–31).Moreover, the results of this study lack the generalisation since the sample population only from adolescents at school, and missed the results from adolescents who are absenteeism.In addition, the results only provide a picture in a very specific country context, and the relation can be different in different contexts. Furthermore,a cross-sectional study done by Swahn et al. in 2012 examined the correlation between hunger and risk behaviours in adolescents aged between 13-16 in four African countries (Bostwana, Kenya, Uganda and Zambia) by using Global School-Based Student Health Survey. The significant association between hunger and current alcohol use only found in Zambia(AOR: 1.60, 95%CI: 1.08-2.36). Regarding the utilisation of the models of multivariable logistical regression, which could be adjusted regarding the incidence of innate factors such as regional affiliation, ethnicity, age extents and gender identities as well as multiplicity of different risk factors related with the behavioural aspects, the analysis of the scenario, where the involved research propositions had been the presence of hunger and absence of hunger, the apparent complications could be understood to be the extensive probability of the variations of consumption of basket of calorific consumptions. Such consumptions are mostly reflective of the categories of both vegetable as well as non-vegetable substance. In this regard, the statistical observations could divulge the factor that the Adjusted Odds Ratio [AOR] is equal to 1:21 with the Confidence Interval [CI] being 97%, (1.11–1.31; P < .001). Furthermore, the AOR of the Deficiency of Physical Activity has been 1.23 with the Confidence Interval or CI being that of 95% (1.08–1.35; P = .001) as well as the current ratio of tobacco utilisation being reflective of the statistical dimensions of (AOR = 1.32; 95% CI, 1.18–1.47). Apart from this, the extent of consumption of soda which could be sweetened by sugar, could be demonstrated as (AOR = 0.95; 95% CI, 0.86–1.04; P = .23) and the current rate of utilisation of alcohol could be understood to be (AOR = 0.95; 95% CI, 0.84–1.07; P = .38) (28). Likewise, the factor of obesity is also measurable as (AOR = 0.97; 95% CI, 0.77–1.21; P = .76). The evaluation of the frequency of hunger could be effective from two perspectives. The first one is associated with the increasing effect of exposure based responses which could be directly reflective of the consumption of vegetable as well as other substances. The second one is related to the dearth of physical movement and activities. This second one could be considerately influenced with the probability of the effect of exposure as well as responses for the measurements of utilisation of tobacco [AOR = 1.21; 95% CI, 0.93–1.56; P = .15]. This measurement is related to the absence or meagre presence of hunger. The rarity of hunger is related to [AOR 1.83; 95% CI: 1.34–2.50; P = <.001]. The medium frequency of occurrence of hunger is related to the measure of [AOR = 2.12; 95% CI, 1.50–2.99; P < .001]. The precision of the results could be delineated through the multiplicity of imputation and these also could confirm the magnitude of the outcomes. It could be observed to be similar to the aspects stated in Romo’s study, that, hunger is not a comprehensive indicator to assess the food security situation.The generalisation of the result in this study is also too narrow in adolescents at school.Another cross-sectional study conducted by Kim et al. in 2016 investigated the association between household food insecurity and the prevalence of smoking based on the young adult population aged 18 to 30 with low SES in the United Statesby using the data from 2011-2012 California Health Interview Survey(32). The household food security status was measured by 6-item short form adopted from US Household Food Security Survey. This study showed that there was a significant association between food insecure status and current smoking status (AOR: 1.87, 95%CI: 1.25-2.28)(32). However, the result from this study is hard to generalise due to two reasons. First, this study only provides evidence as to a particular population, which is the low SES group; however, this status tends to have a higher risk of smoking than the general population(19). Second, the prevalence of smoking is usually positively related to age, and this may cause young adults to have a higher chance of smoking than adolescents in general(33).

There are more studies seeking to determine the association between household food insecurity and overweight/obesity among adolescents. Overweight/obesity also often used as a proxy measure for inadequate physical activity and unhealthy diet(16). However, the results of their association are inconsistent, and most of them are done based on the sample from United States (34,35). The positive association between household food insecurity and increased risk of being overweight or obesity has been found in some studies(36–40). For example, a recently published cross-sectional survey done by Holben et al. in 2015 which is based on the data from a large-scale nationally representative survey‘the National Health and Nutrition Survey Examination (NHANES)’ tested the association of household food insecurity and being overweight or obese on adolescents aged 12 to 18 in the United States(40). The study identified that the risk of being overweight(AOR: 1.44, 95%CI: 1.12-1.87) and obese (AOR: 1.32, 95%CI: 1.01-1.74) among the adolescents from marginal food insecure households is significantlyhigher than from the food secure households(40). On the other hand, there are severalcross-sectional studies indicating that there is no association between household food security and overweight or obesity among adolescents(41–44). The inconsistent results may be partially due to the various ways of measuring food insecurity status and the different category standards for defining overweight and obesity, which can be either based on weight-for-age or BMI-for-age z-score to compare with CDCgrowth chart or International Obesity Task Forceweight categorisation.

Research gap on household food insecurity and adolescents NCDs risk factors

Previousresearches investigated the association between household food insecurity and adolescent NCDs behavioural risk factors have several limitations. First of all, as mentioned above, the current evidences on the association between household food security and NCDs behavioural risk factors among adolescents mostly comes from the United States and other high income countries.There is a lack of clear evidence in LMICs, andthis may also partly due to lack of available data from LMICs(45). Moreover, for the studies conducted based on LMICs, the study population usually based on one country. The results may have lack of generalisation to apply into other LMICs settings. On the other hand, the way of measuring household food insecurity is inconsistent across studies, which includes either short- or full-length of USDA Household Food Security Model (HFSM), Household Food Insecurity Access Scale (HFIAS) or simpler measures of hunger. The tool is used to measure food security status that may have a direct impact on the results. Also, previous literatures usually investigated the association between household food insecurity with a single behavioural risk factor. There is a lack of complete evidence of the correlation between household food insecurity on the risk of developing NCDs behavioural risk factors in general.

Theoretical pathways in LMICs

The potentialconnection between household food insecurity tothe lack of well-being among children/adolescents is indicated in a conceptual framework obtained from Fram et al. (2015) in Figure 2(46). There are three potential hypothesised mechanisms which explain that household food insecurity is related to the adolescent NCDs behavioural risk factors in LMICs. First, food shortages can cause anxiety, distress and shameless among adolescents(47). Long-term mental stress caused by the worry about the availability of food may accelerate adolescents’ adoption onconsumption and abuses of substances such asalcohol and tobacco to cope up with stress and relief with the painful feeling(26,46,48).A second channel is due to the hunger caused by household food insecurity. Hunger has an adverse effect on people’ physical and cognition, such as fatigue(45). The feeling of fatigue may directly lead to a decreased level of physical activity(45). A further channel links to lower quality of dietsas the price of fast food or food containing high sugar and saturated fat is usually lower than the nutritional food(26). Household suffers from food insecurity tend to purchase the cheaper food and compromise on food quality and nutritional value(26). Thus, this may encourage unhealthy food among adolescents and they tend to chooselessnutritionaland less diverse with the reduced consumption of animal products, dairy products, and fruits and vegetables(46). As a result, both physical inactivity and inadequate nutritional food intake will eventually lead to the development of overweight and obesity among adolescents.

On the other hand, adolescents can be considered a vulnerable population group in food insecure households in LMICs(49). Evidence shows a high prevalence of food insecurity among adolescents from LMICs, which may due to the fact that they are excluded from the household’s food insecurity buffering strategy(49). When food is available in food insecure households, their parents tend to give the food to the youngest children first, as a bufferingstrategy since they consider younger children more susceptible to the lack of food consumption, and adolescents usually are considered to be affected less by the food insecurity(49).Eventually, this may further deepen the severity of adolescents’ lack of food intake in food insecure families. As a result, adolescents from food insecure households also tend to develop some coping strategies to manage their situation with insufficient food consumption by themselves(45,50).Some possible coping strategies included trying to reduce their daily physical activityand maintain sedentary to save their energy in order to diminish the amount of food need. Some adolescents start smoking as the nicotine contained in the tobacco can help to suppress the feeling of hunger(50).

Features of the proposed research

This study obtained the data from Young Lives (YL), a longitudinal study of child poverty conducted in Ethiopia, India (Andhra Pradesh and Telangana), Peru and Vietnam. The survey followed two cohorts of children and their families for 15 years from 2002 to 2017. The study decided to use the cross-sectional data from round 3 older cohort, which collected in 2009-2010 and children in this cohort turn into the age of 15. The data for adolescent behavioural risk factors included current tobacco and alcohol use and overweight/obesity were available. This study mainly focused on assessing the association between household food insecurity and overall NCDs behavioural risk factors by combing all the three behavioural risk factors together as a composite outcome indicator among the pooled sample from four countries. By doing this, the evidence can provide a direct idea of the correlation between household food insecurity and the occurrences of NCDs behavioural risk factors in LMICs in general. This also provides a direct evidence to support the potential policy implementation in LMICs. Household food security was measured by a commonly-used scale survey based on the Household Food Insecurity Access Scale (HFIAS), which his a validated measure of food insecurity in LMICs setting, and assessed household situation from multiple domains(29).In addition, previous study showed the different experiences of the household food insecurity with adolescent boys and girls, and with adolescents from rural and urban area in LMICs setting(51–53). There is a potential modification effect between household insecurity and a differentadolescent’sgender or different area of residence onthe association with NCDs behavioural risk factors. Therefore, this study also examines the potential differences in the associationbetween household food insecurity and adolescents NCDs risk behavioural among boys and girls and adolescents living in the rural and urban area. The four countries includes in this study, Ethiopia, India, Peru and Vietnam provide a diverse picture by region and socioeconomics of LMICs settings(54). The correlation may also be different in the different country setting; therefore, this study also examines the association between household food insecurity and adolescent NCDs risk factors in those four countries separately. On the other hand, all of the four countries has a high proportion of the adolescent population and also are facing the critical issue of NCDs as indicated in Table1. The evidence from this study can be used to assist the local policy NCDs implication in order to solve the issues as well(55,56).

The total adolescent population,the proportion of the adolescent population,the total number of NCD deaths,percentage of deaths from NCDs,and risk of premature death from NCDsin Ethiopia, India, Peru and Vietnam in 2017(55,56).

The subsequent research propositions could be ascertained to be of particular significance regarding the necessity of fulfilment of the research gap which could be acknowledged to be in existence in the reviewing of the accumulated research literature. This literature has been meant to be the framework to build better understanding concerning the investigation of the insecurity scenario related to edible substances and the occurrences of NCDs in the adolescents regarding the behavioural risk factors involved in the LMICs(9). The derived evidence from the following study could be utilised for the purpose of development of the best possible comprehension regarding the policies and interventional elements which could be developed for a specific purpose. This purpose could be understood to be the objective of constituting the most effective health behaviour at the earliest phases of adolescence so that these health behaviours could persist throughout the course of the lives of the subjects under consideration. Another factor which is necessary to be undertaken as a deliberate outcome of such a research could be envisaged by formulation of working architecture of health intervention operations which could effectively contribute to the reduction of the behavioural factors of risk which are incumbent on the deliberations regarding the control of the NCDs. Either complete preclusion or partial prevention of the NCDs from occurring could be contemplated to be the ultimate objective which could be assisted by the achievement of the 2030 Sustainable Development Goal (SDG). This would, in turn, contribute to the curtailment of the premature mortality by one third of the previous volume occurring out of NCDs.

The following research questions were examined in this study:

(1). What is the difference between the prevalence of NCDs behaviouralrisk factors among adolescent from food insecure household and food secure household in LMICs?

(2).Is household food insecurity associated with adolescent behavioural risk behaviours in LMICs after controlling with child, household and community characteristics?

(3). If so, is the association between household food insecurity and adolescents behavioural risk factors after adjusting with covariates differ betweenboys and girls, the area of residence and differ in the four countries respectively?

Aim & objectives

By using the cross-sectionaldata inround 3 older cohortsfrom YL, the aim of this studywas to investigate the predictive role of household food insecurity on adolescent NCDs behavioural risk factors in order to provide evidence on policy implications for NCDs prevention in LMICs. The objectives of this studyincluded:

To determine the prevalence of household food insecurity, adolescent NCDs risk behaviours, and covariates in the pooled sample, and byfour countries.

To examine whether NCDs behavioural risk factors prevalence is different by food security status among the pooled sample and bycountry.

To investigate the linkage between household food insecurity and adolescents NCDs behavioural risk factors after adjusted with covariate among the entire pooled sample.

To investigate the whether the correlation between household food insecurity and adolescent NCDs behavioural risk factors after adjusted with covariatediffer between different gender and area of residence among the pooled sample.

To investigate whether theassociation between household food insecurity and adolescent NCDs behavioural risk factors after adjusted with covariatediffer in the four countriesrespectively.

To identify the area where more research and data collection is needed.

The hypothesis of this study was proposed as the prevalence of NCDs risk behaviours is higher in adolescents from food insecure household than food secure household, and the household food insecurity is positively associated with adolescent NCDs behavioural risk factorsafter controlled with covariate in LMICs. This correlation is different in the subgroup male and female adolescent’s population, adolescents live in the rural and urban area and adolescents from different country. The following sections will present with the methodology, results presentation and interpretation, and discussions of strengths, limitations and further implication for this study.

2. Methods

This section will present the data sources, variable construction and the statistical analysis methods used in this study.

2.1 Data sources

2.1.1 Young Lives Study

YL is a multi-disciplinary, longitudinal study funded by the UK Department for International Development (DFID)since 2002 and run by the University of Oxford(57). This study followed 12,000 children from four LMICs (Ethiopia, the state of Andhra Pradesh and Telangana in India, Peru and Vietnam) for 15 years. There were two cohorts of children involved in this longitudinal study, a younger cohort and an older cohort (See Appendix Figure S1). The younger cohort included the children born in 2001to 2002, with the aim to follow from their period of time as an infantto young adolescent. The older cohort involved the children born in 1994-1994 in order to follow them as they are older children into their adulthood. Four rounds of data have been collected and published every three years (2001-2002, 2006-2007, 2009-2010 and 2013-2014). Individual, household (included household food situation questions) and community level surveys were conducted in each round via face-to-face interview by the fieldworkers with children and adolescents, the adult caregivers of the YL children, and the key representative in the community respectively. Questions about the household food situation were answered by the children’s caregivers. Self-administrated questionnaires answered by adolescents were introduced in round 3 older cohorts to assess theoccurrence of some riskybehaviour. The height and weight were measured and recorded for each child/adolescent by the fieldworker during the interview, and their BMI-for-age z-score were calculated(58). The main purpose of this study was aimed to investigate and improve the understanding of the nature of child poverty in the LMICs(57). The population sampling was choice-based on a multistage strategy(59). First, 20 sentinel sites were selected in each country non-randomly. This was used to make sure the sample can fully represent the different demography, socioeconomic and cultural contexts in each country’s setting. Then, 100 households with children under the age ofyounger cohort and 50 households with children under the age for the older cohort were randomly chosen within each sentinel site in the four countries. More details on the sampling approach had been documented(60–63). The cross-sectional data from YL round 3 older cohorts used in this study included a total of 3722 adolescents and their households, which 1000 from Ethiopia, 1008 from India, 714 from Peru and 1000 from Vietnam.

2.1.2 Dataset construction and data cleaning

A dataset was created and the data cleaning was done before the variablesconstruction. The dataset construction wasmainly done by twosteps as indicated in Figure 3. The first step was to combine the different data from YL round 3 older cohort(included child-level, household-level, and community-level data from four countriesseparately)into one dataset. The merging of child-level, household-level and community-level was first done country by country and combined them as 1:1 by child ID, then a categorical country variable was created toeach country after merged, and coded as 1 for Ethiopia, 2 for India, 3 for Peru and 4 for Vietnam. Then the four newly merged datasets were appended together to create a new overall dataset for this study to use.After that, the dataset was cleaned by only keep the data used for creating exposure, outcome variables and covariates,and then dropped the others in the dataset; then convert any string variables to numeric data to enable analysis. Lastly, checked out and re-coded missing data in the dataset, and re-coded any abnormal data as missing value as well. Some missing values were presented as either number or category code in YL was listed in Appendix Table S3.

2.2 Measures

This studyused the measurement of household food insecurity as the exposure variable, and adolescent NCDsbehavioural risk factors as the outcome variable to test their correlation. The detail of variable construction was listed below.

2.2.1 Exposure variable

The food insecurity variable was created as a binary variable following the approach of Humphries et al. (2015) to determine if the household was food insecure or not(64). The questions used to capture the household’s food security status in the past 12 month contain 9 occurrence questions (answered with ‘yes’ or ‘no’) plus frequency-of-occurrence questionseach (how often this situation happened if answered‘yes’ in the previous question, see Appendix Table S3 for full details)(29).The responses were first classified into different four food security categories, which included food secure, mildly food insecure, moderately food insecure and severely food insecure according to the HFIAS coding algorithm(See Appendix Table S4)(29). After that, a binaryvariable was established if the household was categorised as moderately or severely food insecure previously, and classified them as food-insecure (coded as 1). For the households classified as food secure or mildly food insecure previously, then classified them as food-secure (coded as 0)(64).

2.2.2 Outcome variable

There were three behavioural risk factors with available dataincludedin this study, which arecurrent tobacco use, current alcohol consumption and overweight/obesity. All the factors were demonstrated as a binary indicator to determine whether the adolescent has this behavioural risk factoror not. For current tobacco use, the question used to determine this outcome variablewas ‘how often do you smoke cigarettes now?’, and the answer included ‘every day’, ‘at least once a week’, ‘at least once a month’, and‘hardly ever’. According to the Global Youth Tobacco Survey (GYTS), they identified adolescents as a smoker if they used any tobacco products on one or more days in the one monthperiod(65). Therefore, in this study, adolescents who reported the use of tobacco more than and equal to once a month was considered as the use of tobacco. The adolescents answered in the questionnaire with either ‘every day’, ‘at least once a week’ or ‘at least once a month’ was classified as‘current tobacco use’ (coded as 1). For the adolescent answered with ‘hardly ever’, they are classifiedas ‘non-tobacco use’ (coded as 0). For current alcohol use, the question used to identify this outcome variablewas ‘how often do you usually drink alcohol now?’, the answer included ‘every day’, ‘at least once a week’, ‘at least once a month’, ‘only on special occasions’, ‘hardly ever’, and ‘I never drink alcohol’. Previous literature used the threshold of drinking once in the past one month as alcohol use, and also drinking in the specific occasions may be a part of the social and cultural norm in many LMICs(66). Thus, in this study, for the adolescent with the answer of either‘every day’, ‘at least once a week’ or ‘at least once a month’, they are classified as ‘current alcohol use’ (coded as 1). For the adolescent answered ‘only on special occasions’, ‘hardly ever’ or‘I never drink alcohol’, they are classifiedas ‘non-alcohol use’ (coded as 0). For overweight and obesity, this study chose to use BMI-for-age z scores as a proxy measure to represent this risk factor. A binary variable overweight and obesity was constructed for those adolescents whose BMI-for-age z score was above 1 standard deviation(67). This classification of being overweight/obese is according to WHO growth reference for children aged between 5 and 19. Thus, adolescents with BMI-for-age z score larger than or equal to 1 was classified as being ‘overweight/obese’ (coded as 1), and with z score less than 1 was classified as ‘non-overweight’ (coded as 0). Finally, we created aNCDs behavioural risk factors binary variableby coding adolescentswho hadany of those three risk factorpreviously as 1, and classified them as ‘having NCDs risk factors’, and the adolescents had none of those risk factors was coded as 0, and classified as ‘non-NCDs risk factors’.For adolescents with more than one single risk factor, we did not summarizethem together to create a clustered NCDs risk factors scores index as previous study(68). This may loss some information to show the association between household food insecurity with different level of behavioural risk factors occurrence. However, the use of binary outcome variable was more comply with our research main purpose.

2.2.3 Covariates

The following covariateswere included with the consideration to the prior-known association with both possible cause and possible effect(26,51,52,64,69–74): age in months and gender for YL adolescents, area for residence, household wealth index, household size, household head age and primary caregiver’s education. The adolescent’s age (in months), wealth index (from 0 to 1), household size and household head age (in years) are continuous variable. Area of residence (rural/ urban) and adolescents’ gender (male/ female) were dichotomized as a binary variable. Area for residence is coded as 1 for people living in rural area and coded as 0 for people livingin urban area. The gender ofan adolescent was coded as 1for male, and 0 for female. The primary caregiver’s education was categorised as a three-level categorical variable included ‘no formal education previous’ if caregiver reported with ‘received none education before’, ‘have adult literacy’ or ‘only received religious education’(coded as 0);‘primary-school level education’if caregiver completed ‘Grade 1 to 6’ (coded as 1), and ‘post-secondary and above level education’if caregiver completed ‘Grade 7 to 12’ or ‘university’(coded as 2).

2.3 Data analysis

For descriptive statistics,the frequency and proportionwere used to describe the binary and categorical variables. The mean ± standard deviation (SD) was used to describe the continuous variable. The chi-square test(χ²)with p-value was used to examine whether there was a significantdifference between NCDsbehavioural risk factors prevalence by food insecurity status. Chi-square (χ²) test was chose for this analysis since both exposure and outcome variables were binary. A total of 5 chi-square testswere run both in the pooledsample and by country. Five multivariate logistic regression models with robust standard errorswere constructedto investigate whether household food insecurity is associated with adolescents NCDs behavioural risk factors after adjusted with covariates. The logistic regression was used since the outcome variable – NCDs behavioural risk factors are binary. Model 1 was constructed to investigate their correlation after only controlled with country variable among the pooled sample. Each exposure of interest (covariates)also runs in model 1. The control of the country variable avoided the collinearity with the binary outcome variable. Model 2 was built based on model 1 with fully-adjusted covariates among the pooled sample.Model 3 and model 4 were constructed based on the model 2 with an interaction termto investigate the correlation with modifiable effect among the pooled sample.In model 3, an interaction term between household food insecurity variable and adolescent gender variable added to determine whether the association between household food insecurity and adolescent NCDs risk factors differed in adolescent gender. In model 4, an interaction term between household food insecurity variable and area of residence variable added to examine whether the association between household food insecurity and adolescent NCDs risk factors differed in adolescent lived in rural and urban areas. Model 5 was modified based on model2to determine the association of household food insecurity and adolescent’s risk factors in each country separately. The mathematical formulae for the statistical models in this study are listed in Table 2.

In the entire five multivariate logistic regression models, coefficientsβ1indicated the difference in log odds between food insecure and food secure.After anti-log transformation with, the parameter expressed as Exp(β), which indicated the AOR of having adolescent’s risk factors with household food insecure and food secure after controlling for other variable.The parameter α is an intercept term equal to log odds of having adolescents risk factors when on baseline measurement, which is food secure (x=0). After anti-log transformation, this parameterexpressed as exp(α), and equal to odds of NCDs behavioural risk factorswhenthe household was food secure. The statistical models construction in this study did not involve multiple testing with covariates and robustness check but mainly based on the prior evidence. Also, this empirical strategy can only provide the association but not the causal connection between household food insecurity and adolescent’s behavioural risk factors as thereare some unobserved socioeconomics, household and individual factors may also affect this relation.The 95% confident interval and p-value were calculated for each AOR. The result would be considered as statistically significant if p 0.05. Data was analysed in STATA 14.1.

2.4 Ethical issues

The formal ethical approval for YL study was received fromthe Department of International Development in University of Oxford(75). Informed consent was obtained from every participant involved in this study, which included children, young people, caregivers and all people related to this project in the community. The consent was developed to suitable for the local country setting. Fieldworker explained the research to children in a way they are able to understandfully. Sensitive questions, such as some adolescents’ risk behaviour, areobtained by using self-administered questionnaires to reduce potential social desirability bias and provide respect for personal comfort and confidentiality. Also, data is publicly archived and available to use for free at the UK National Data Repository (http://www.younglives.org.uk/). Thus, no additional ethical approval was needed for this study.

3. Results

This section presents the results from descriptive statistics, chi-square test and multivariatemodel analysis with someinterpretation of the findings.

3.1 Descriptive statistics

3.1.1 Prevalence of household food insecurity among pooled sample and by country

The frequency and proportion of household reporting food insecurity status in the past 12 monthsamong all the pooled samples and by country is shown in Table 3. A total of 1690 households categorised as food insecure in the past 12 months which took upalmost half of thepooled sample(47%). Over half of the householdsreported worry about running out of food in the household (54.5%), not able to eat the food they preferred because of lack of money (72.6%), andhave to eat a limited range of foods due to lack of money (53.5%)in the last 12 months.These situations were related to the mental stress and insufficient quality of food intake due to food insecurity. Households reported with very severe food situation with hunger occurrence were rare, only 6.0% of households had no food to eat, 3.9% of them wentto bed with hungerand 1.7% of them had not eaten for whole day and night in the past 12months. By each country, Ethiopia had the highest proportion of food insecurity household (75.0%) and India had the lowest proportion (28.3%).Ethiopiaalso had the highest proportion of households reported with unable to eat the food they preferred (84.1%),eaten food with a limited food choice (72.4%) and eaten less food than they want (62.5%) in the past 12 months. India had the highest proportion of household who have to eat the food they don't want because of not enough money(28.7%). 73.6% of household in Peru reported worry about running out of food in the past 12 months which was the highest among the other three countries.

1. The value of frequency and proportion (in parentheses) indicated the number of households and the proportion of household were food insecure or experiencing each food situation according to the HFIAS in pooled samples and by country.

2. Households were categorised as food insecure if the household was determined as moderately or severely food insecure according to the HFIAS.

3. The household food situation was assessed according to the 9 occurrence questions in HFIAS.

3.1.2 Prevalence ofNCDsbehavioural risk factorsamong the pooled sample and by country

Table 4 shows the frequency and proportion of adolescents’ overall NCD behavioural risk factors, tobacco use, alcohol consumption and overweight/obesity in pooled samples and by each country. In general, the prevalence of NCD behavioural risk factors among adolescents was relatively low in the sample.A total of 472 adolescents had any form of NCDs behavioural risk factors, which took up13.1% among the pooled sample. 2.8% used tobacco, 5.4% drink alcohol, and 6.2% are overweight/obesity. By country, Peru had the highest proportion of overall NCDs risk behaviours (28.5%) andoverweight/obesity (20.3%).In Ethiopia, 12.8% adolescents had NCDs behavioural risk factors, and it also had the highest prevalence of alcohol consumption by adolescents(11.1%). The other prevalence of NCDs behavioural risk factors were less than 10% in all sample and by country.

1. The value of frequency and proportion (in parentheses) indicated the number of adolescents and the proportion of adolescents had overall NCDs behavioural factors, tobacco use, alcohol use and overweight/obesity in pooled sample and by country.

2. An adolescent who reported with either one of the three NCDs risk factors, which included tobacco use, alcohol use and overweight/obesity was categorisedinto the group of having NCDs risk factors.

3. An adolescent who indicated the frequency of use tobacco as ‘every day’, ‘at least once a week’, and ‘at least once a month’ was categorisedas current tobacco use.

4. Adolescents who indicated the frequency of alcohol consumption as ‘every day’, ‘at least once a week’, and ‘at least once a month’ werecategorisedas current alcohol use.

5. Adolescents who had the BMI for age z-score larger and equal to 1 werecategorisedas overweight/obese.

Prevalence of covariatesamong the pooled sample and by country

The prevalence of covariates included for adjustment in the multivariate model was presented with either mean and SD or frequency and percentage depending on the type of the variable, as shown in Table 5. Among the pooled sample, 50.7% of adolescentswere male and 49.3% were female. Their average age was 180.2 months i.e. 15 years old. More people live in a rural area(62.2%) than in urban areas (37.8%). The average household wealth was 0.5, which was a middle score in the scale range from 0 to 1. The average household size was 5 people per household and the average age of theirhousehold head age was 46 years old. For adolescents’ caregivers’ education levels, 38.5% never had any formal education, 25.4% had primary education level and 36.0% had post-secondary and above education level. The prevalence of covariates by each country also listed in Table 5.

1.Meanand SD(in parentheses) described continuous variables, frequency and proportion(in parentheses)described categorical variables.

2.Household wealth index range from 0 to 1; this index was calculated based on the householdquality index, access to services index, and consumer durable index with equal weights.

3.Caregiver education was categorisedinto 3 group, people who were ‘never received any education’, ‘adult literacy’ and ‘religious education’ was categorised as ‘no formal education’; people with education from ‘Grade 1 to 6’werecategorised as ‘primary level’; people with education from ‘Grade 7to 12’ and ‘university’ was categorised as ‘secondary and post level’.

Association between household food insecurity and adolescent NCDs risk factors

Adolescent NCDs risk factors prevalence by household food security status among the pooled sample and by country

The difference in prevalenceofhaving NCDs risk factors between adolescents from the food-insecure and food-secure household with p-value reported in Figure 4. Among all the pooled sample, there was a significantly higher prevalence of NCDs risk factors among adolescents from food insecure households (15.2%) than from food secure households (11.3%) with a p-value less than 0.01. However, in each country, only Ethiopia showed a significant difference with the adolescentsbehavioural risk factors prevalence by food security status. 14.7% of adolescents who had NCDs risk factors were from food insecure household and only 7.4% were from food secure households in Ethiopia. On the other hand, adolescents in Peru and Vietnam also showed a higher prevalence of NCDs risk factors among whom from food insecure household than food secure household. In Peru, 30.8% adolescents had NCDs risk factors from food insecure household and 26.5% from food secure household. In Vietnam, 7.9% adolescents had NCDs risk factors from food insecure household and7.5% from food secure household. The prevalence of adolescents NCDs risk factors were same by food security status in India (8.3%). However, no significant results are found in those three countries.

1.The p-value (p) was determined by the chi-square (χ²) test forthe difference in the proportion of having NCDs risk factors in adolescents from food-insecure households and food-secure households.

3.2.2 Association between household food insecurity and adolescent NCDs risk factors after adjusted with covariates among thepooled sample

Table 6showed the results from multivariate statistical model 1 and model 2 were used intheassociationbetween household food insecurity and adolescent NCDs risk factors after adjusting with covariates among the pooled sample. In model 1, household food insecurity was a significant determinant of adolescents NCDs risk factors after only controlled with country heterogeneity (AOR: 1.29, 95%CI: 1.05 to 1.58). This association was still significant with a little decrease on theireffect size after fully adjusting with covariates in model 2. Adolescents from food insecure household were 1.26 (95%CI: 1.01 to 1.56) times more likely to have NCDs risk factor comparing to the adolescents from food secure household after fully adjusted with covariates. On the other hand, caregivers’ education level was also a determinant of the occurrence of adolescents NCDs risk factors in both model 1 and model 2. After fully controlled with other covariates, adolescents whom caregivers had primary level of education were 0.72 (95CI%:0.53 to 0.98) times less likely to have NCDs risk factors and those whom caregivers had secondary and post level of education were 0.63 (95CI%: 0.45 to 0.88) times less likely to have NCDs risk factors compared to those whom caregivers had never received any formal education. The country heterogeneity factors determined the occurrence of NCDs risk factors in both model 1 and model 2 as well. In model 1, compared to adolescents from Ethiopia, adolescents in India were 0.62 (95CI%: 0.46 to 0.83) times less likely to have NCDs risk factors while adolescents from Peru were 2.71(95CI%: 2.10 to 3.48) more likely to have NCDs risk factors and adolescents from Vietnam were 0.56 (95CI%: 0.41 to 0.76) times less likely to have NCDs risk factors.Similar significant associations were also found in model 2 with adolescents from India (AOR:0.66, 95CI%: 0.46 to 0.94) and Peru (AOR:3.28, 95CI%: 2.43 to 4.44) comparing to Ethiopia. However, there was no significant association found in adolescent from Vietnam comparing to Ethiopia.

1.model 1: multivariate logistic regression controlled with country variable.

2. model 2: model 1+ full adjusted with covariates listed in the table.

Association between household food insecurity and adolescent NCDs risk factors by adolescents’ gender and by areas of residencein the pooled sample

Table 7 showed the results from model 3 and model 4 to examine whether the association between household food insecurity and adolescents NCDs risk factors differ by adolescents gender and their areas of residence. In model 3, no significant result is found in the interaction term between household food insecurity and adolescent gender, therefore, it cannot tell that the association of household food insecurity and adolescents NCDs risk factorswasdifferent among male and female adolescents. The results also showed that household food insecurity was significantly associated with having NCDs risk factors (AOR: 1.44, 95%CI:1.07 to 1.94)in female adolescents (the referent group of adolescent gender). Therefore, this significant association may apply to the male adolescents group as well. In model 4, no significant result is found in the interaction term between household food insecurity and areas of residence as well. Therefore, it cannot conclude the association between household food insecurity and adolescents NCDs risk factors differ by the areas of residence of adolescents. No significant association found between household food insecurity and NCDs risk factors in adolescents from the urban area (the referent group of areas of residence) from model 4. This insignificant association may apply for the adolescents from the rural area as well.

1. Model 3: model 2 + interaction term, the interaction is between household food insecurity and adolescent gender.

2.Mode 4: model 2+ interaction term, the interaction is between household food insecurity and areas of residence.

Association between household food insecurity and adolescent NCDs risk factors by four countries

Table 8 showed the results from model 5 to investigate the association between household food insecurity and adolescents NCDs risk factors by four countries respectively after adjusting with covariates.The onlysample from Ethiopia showed thathousehold food insecurity was a predictor of the occurrence of adolescents NCDs risk factors (AOR:2.09, 95CI%: 1.23 to 3.56). Also, adolescent lived in the rural area were 2.69 (95CI%: 1.49 to 4.86) times more likely to have NCDs risk factors compared to whom from urban in Ethiopia. However, the sample in India was significantly less likely to have NCDs risk factors if adolescents lived in rural than in urban area (AOR:0.51, 95CI%:0.28 to 0. 93). In Peru, household wealth (AOR:3.68, 95CI%:1.20 to 11.27) and having caregiver who received secondary and post level education rather than no formal education (AOR:0.45, 95CI%:0.25 to 0.81) wereassociated with adolescent NCDs risk factors. In Vietnam, male adolescents were 2.48 (95CI%:1.47 to 4.17) times more likely to have NCDs risk factors than female.

1.Fourmultivariatelogistic regression run based on model 5 among sample in each country with fully adjusted covariates listed in the table.

Discussion

This section will discuss the interpretation of the main findings of this study, its strength and limitations, the potential for policy implementation and area for the future study.

Principle findings and interpretation

This studyprovides new evidences on the association between household food insecurity and having NCDsbehavioural risk factors among adolescents in LMICs. By using the cross-sectional data with adolescents aged 15 from Ethiopia, India, Peru and Vietnam, this study finds a significantly higher prevalence of developing NCDs behavioural risk factors in adolescents from food-insecure household compared to food-secure households. The household food insecurity is also positively correlated with the occurrence of adolescent’sbehavioural risk factors after adjusting with a range of covariates. The research findings are primarily effective regarding the highlighting of the fact that not more than 13% of adolescent participants of the research, who had been selected from the pool of sampled personnel, could put forward the fact that the behavioural risk factors have been rife in them concerning the NCDs. This has been the outcome despite the fact that a relatively small size of sampled personnel has been utilised to determine the research outcome. In this context, it is significant to understand the value of the association since this could lead to the control of the multiplicity of factors which are mostly confounding in nature. Apart from this, the effective size of the correlation which could exist between various scenarios of household based insecurity of edible and consumable substances and the occurrence of NCDs regarding the behavioural risk elements amongst the adolescent populace, could as well be comprehended to be of extensive importance since the levels of risk could only be measured and lowered through the application of the Multivariate Model. However, this is closely incumbent on the fact that this model has to be adjusted closely with the covariate elements as well (AOR:1.26, 95%CI: 1.01 to 1.56). This could be further acknowledged through the performance outcome of the comparison of this model with the actual outcomes which could only be controlled through the heterogeneity of the country (AOR:1.29, 95%CI:1.05 to1.58). The discourse of the preceding study could establish the supposition that this shortcoming could be improved through the strengthening of the research association through selection of a greater number of sample sizes so that greater statistical accuracy could be infused within the research process and more relevant results could be achieved comparatively with the previous studies. Furthermore, it could be further discovered from the discourse of the study that the hypothetical theoretical constructs could be utilised to further reinforce the findings of this study through the establishment of the linkages between two definite research variables. The first one is identifiable to be the insecurity of edible substances at differential households. The second one could be outlined as the development of the necessary understanding regarding the NCDs risk factors in LMICs(46). However, it has been fraught with considerable difficulty to evaluate and examine the effective working mechanism through which the underlying effect on the association could be properly outlined. The primary contention in this regard has been the paucity of data which could provide greater insight regarding the psychological factors which become significant in determining the behaviour of adolescent personnel as well as the diversity associated with the factors of diet. The research also brought forth the effect that the research findings could establish the fact, from the perspective of the association that the adolescents aged not more than that of 15 years acknowledged that food insecurity could be influenced by the status of their households regarding the monetary income and provisioning of proper mechanism to avail edible sustenance. This is directly related to the buffering strategy which many households generally employ to protect their children from the suffering related to food insecurity. However, this buffering strategy may not apply to them as their parents may expect adolescents to handle the food shortage together(52). As mentioned in the previous literature review section, the current study on investigating household food insecurity and NCDs behaviouralrisk factors among adolescents is very limited. The evidence found in this study can complement the findings conducted by Romo(28). As our study found evidence on an overall correlation between household food insecurity and having adolescents NCDs risk factors. Romo’s study identifies a significantassociation with three individual behavioural risk factors, which are inadequate physical activity, unhealthy diet, and tobacco use. Even though no significant association is found with alcohol use, his study identified a significantly higher proportion of alcohol use in adolescents with self-reported hunger status. However, no significant association or higher prevalence of obesity is found in his study.A cross-sectional study in 2009 was based on 2516 children aged between 8 to 17 in the US also indicating no association between food insecurity and obesity(76). Our study considered both adolescent who are overweight and obesity as the NCDs outcomes. There is a possibility that food insecurity is only associated with being overweight but not being obese. Further research needs to clarify this association by using valid measure of both household insecurity and adolescent’s weight status.On the other hand, Romo’s study used hunger as the indicator to measure food insecurity which is a sign of very severe stage of food insecurity in the household(28). Our study categorises household who are moderately and severely food insecurity together as food insecure. Therefore, the developing of any NCDs risk factors can occur in the adolescents from less severe food insecurity household as well. Moreover, our study also shows an advantage in the statistical analysis, as the logistic regression constructed by Romo’s study that is only adjusted with three covariates, which included adolescents’ age, gender and region. However, the occurrence of adolescent NCD risk factors and household food insecurity are associated with many socioeconomic and family factors and the insufficient adjustment of the confounder may influence the reliability and validity of the results. Another important findingfrom the multivariate analysis is that the level of caregivers’ education is negatively associated with the occurrence of NCDs behavioural risk factors of their adolescents. The previous study conducted by Wickrama et al.also indicated a correlation between parental education level with many adolescents risk behaviours(77). There is a potential intergenerational transmission of those risk behaviours, as parents with low education tend to have various risk behaviours, and then those behaviours may pass on to their next generation(77). A study done by Alamian et al. in 2009 stated that caregiver smoking was significant increased the risk of adolescent having behavioural risk factors (AOR: 1.49, 95CI%: 1.09 to 2.03)(68). This can also be explained that caregivers with higher education can predict a better socioeconomic status which will also have the effect on inhibiting the adoption of NCDs behavioural risk factors during adolescence(69). Therefore, the caregivers with a higher education can play a role as a protective factor to prevent NCDs behaviouralrisk factorsoccurrence among adolescents even with the consideration of household food-insecure status.Previous evidence also indicated the interventions for improving adolescents health and reducing their risk behaviours should targeting on their caregivers as well(78).

No difference between the association of household food insecurity and adolescents NCDs behavioural risk factors by adolescent’s gender and by areas of residences found in this study. However, this study found that household food insecurity has a predictor role on NCDs risk factors occurrence among female adolescents. Girls tend to more affected by the food insecurity as the gender discrimination in many LMICs lead to inequality intra-household food allocation(51,52,74). A study conducted by Hadley et al. in 2007 investigated the gender bias towards the food insecurity among adolescents and their intra-household sibling in Ethiopia. More than 40% of adolescents girls reported they arefood insecure compared to their brothers(52).This gender difference in food insecurity also increases their risk of developing behavioural risk factors and deteriorate their health outcome(51).Moreover,adolescents from rural areas may face the issue of health and social inequality than whom from urban areas in LMICs(53).This may lead to the issues with increasing burden of food insecurity and adopting NCDs behavioural risk factors during adolescence(79). However, one clear reason for this inconsistent finding in this study compared to previous evidenceremain unknown, and, this may partly be due to the low sample size of adolescents with NCDs risk factors in this study. This study also identifies the difference in the association between household food insecurity and adolescents NCDs behavioural risk factors by country. A significant association between household food insecurity and adolescent NCDs behavioural risk factorsisonly found in Ethiopia after adjusted with the covariates, however, no significant association found in the other three countries. The potential reason may be explained from two aspects. Moreover, households from Ethiopia have the highest prevalence of food insecurity, and all three aspects of food insecurity issue i.e. anxiety of not enough food, insufficient quality and quantity of food availability were reported. However, the reported food insecurity issue in the other three countries ismore focused on the anxiety with not enough food and quality of food intake but with enough of an amount of food available. This can read as that when compared with India, Peru and Vietnam, household from Ethiopia faced a more severe situation of food insecurity. Adolescents from households with severe food insecurity are more likely to be influenced by and develop NCDsbehaviouralrisk factors(49). This can also be understood that, adolescents from the household with not enough food to eat may suffer from more stress and are more likely to over consume food and eat energy-dense food when there is food available in the household. This possible explanation can also be proved in this study, as among the four countries, only Ethiopia had a significant difference in the proportion of occurring NCDs risk behaviours. As per theresult, they are more likely to accept alcohol and tobacco use, and also gain weight with the copying strategy for food insecurity(49). On the other end, Ethiopia has the lowest proportion of caregivers who has received post-secondary education levels compared with the other three countries, and this may weaken the protective role for reducing the development of NCD risk behaviours of adolescents from food insecurity. Interestingly, we found sample from Peru are more likely to have NCDs risk factors compared with Ethiopia in the multivariate model in pooled sample, however no significant results found in the difference of NCDs risk prevalence and in the multivariate model by Peru. This is also partly due to the relatively small size in Peru compared with other three countries. On the other hand, we also found the determinant of adolescents NCDs behavioural risk factors vary in the four countries. In Ethiopia and India, in the rural area there is a chance of the occurrence with adolescents NCDs risk factors. Household wealth and caregiver’s education level are the determinants of adolescents NCDs risk factors in Peru. Adolescent’s gender is the determinant of adolescents NCDs risk factor in Vietnam. This can be understood as the development of adolescents NCDs behavioural risk factorswhich is a complex processand is influenced by the wider socioeconomic and demographic factors in different country setting(17).

Weakness and strengths

Limitations

Some limitations have been identified in this studythat may have an impact on the outcomes of the results. The limitation of the available data is an important issue. This study has decided to use cross-sectional data from one round only instead of using longitudinal data. This is due to the reason that no data onadolescents risk factorsavailable in all the four countries in the following round. However, the use of cross-sectional data may bring some limitations to the results. We can only identify the prevalence of adolescent risk factorsamong food insecure and food-secure households at the time when the survey was conducted instead of the incidence of adolescent risk factorsduring a longer period of time. Also, the cross-sectional data can only indicate the relationship between household food insecurity and adolescent NCDs risk factorsat this specific time when the survey was conducted.However, it is hard to justify whether the occurrence of NCDs risk factorspreceded the development of those behaviours. In addition, the natures of cross-sectional study are unable to identify the causal inferences between household food insecurity and adolescent NCDs adolescent behaviouralrisk factors. Another limitation is the small sample size of adolescents having NCDsbehavioural risk factors, especially for tobacco and alcohol use. The tobacco use and alcohol consumption areassessed via the self-administrated data.However, this may have the potential of the under-reported by adolescents since in many countries, those aged between 14 and 15 are still under the legal age of using those substances. As a result, this may cause the underestimation of the prevalence of NCDs risk factors among adolescents. In order to overcome the issue of relatively low sample size, this studycombines all the three behaviouralrisk factors together as one composite indicator as the overall NCDs risk factorsoutcome. By doing so, this not only overcomes the issue of underestimating tobacco and alcohol use among the adolescent but also increases the sample size in order to increase the validity and generalisation of the results. The creation of this binary composite outcome indicator also assists to investigate the aim of this study. In addition, physical inactivity and unhealthy diet are also adolescents NCDs behavioural risk factors. However, only Peru hasthe data on adolescents’ physical activity, and also some lack of dietary factors data in the pooled sample. In order to keep the consistent measuring of adolescents riskfactors in all the four countries, we decided to only include NCDs risk factors that are completely available in the pooled sample as the outcome measurement. However, this may also have the potential to underestimate theoverall adolescentsbehavioural risk factors prevalence in the pooled sample without including physical inactivity and an unhealthy diet. Lastly, this studyconstructs a logistic regression model to test the hypothesis of the association between household food insecurity and adolescent risk behaviours, the logistic regression ischosen due to the reason that the outcome of variable food security status is a binary variable. However, this may provide the difficulty to justify the result as only OR can be measured instead of the risk ratio.

Strengths

Even though the existence of limitations in this current study is present, there are still many strengths present in this study. First of all, the study population isfrom the large-scale multi-country cohort study YL, and the study population isselected by a strict design process to make sure they have a good representation of the household and children in LMICs with the consideration of various socioeconomic and geographic statuses. Therefore, the results found, is based on this study population have a good generalizability to represent the populations in LMICs. On the other hand, adolescents’ self-administered survey and BMI measurement are conducted based on the household and not in the school. This can avoid the potential selectin bias for the adolescent sample population due to the high school drop rate or absenteeism, which is particularly high in LMICs(80). This also increases the representation of the results found in this study. The second strength of this study is the way of measuring household food security status by using a validated scale. Household food security has been recognized as being difficult to measure accurately, since it can be identified from many different domains. This studymainly use the HFIAS with 9 occurrence-to-frequency questions, which assess the household food insecurity from three different aspects, which included the anxiety of running out of food in the household, insufficient quality of food, and insufficient food intake and its physical consequence such as hunger. Rather than only measure household food insecurity as not enough food to eat, the inclusion of the questions related to the mental stress and food quality during food-insecure state are more related to the development of NCDs behavioural risk factors examined in this study. For example, the stress due to the uncertainty of food supply may lead to the drinking and smoking behaviours to copeup with mental stress. Also, the stress may lead to overeating when there is enough food available in the household. Low-quality food intake may include the consumption of high-calorie unhealthy food, which may lead to overweight/obesity. Therefore, the use of HFIAS to measure household food security status is appreciated and increases the accuracy of the results in this study. Lastly, both the occurrence of household food insecurity and adolescent NCDbehavioural risk factors are co-associated with many socioeconomic and family factors. The statistical model controlled with various covariates from four domains: adolescents’ characteristic, the area of residence, household characteristic, caregivers’ educationin order to rule out the possible confounding associated with household food insecurity and adolescent NCDs behavioural risk factors.By doing so, this would increase the reliability and validity of the resultsof this study. This study also tested the difference in household food insecurity and adolescentsNCDs behavioural risk factors in the sub-group population (by adolescent’s gender, areas of residence and by different country settings) based on the finding in the available literature in order to increase the results validity.

Policies and interventions implications

The prevention of developing NCD behavioural risk factors among adolescents has been widely accepted as one of the most effective strategies to prevent the incidence of NCDs. Evidence from this study indicated a predictive role of household food insecurity on the adolescents NCDs behavioural risk factors occurrence in LMICs. Therefore, adolescents from food insecure should receive more attention and areconsidered as a vulnerable group with developing NCDs behavioural risk factors. The policy makers and programme organisers should include the measurement of adolescents’ household food security status during implementing the policies and interventions for preventing or reducing adolescents NCDs behavioural risk factors. The data on adolescents’ household food situation should be collected and monitoredregularly with the use of valid food security measuring tools. On the other hand, the current intervention ofreducing the NCD risk behaviours among adolescents showed slow progress may partly due to the reason that the development of adolescents NCDs risk factorsis formed by much wider social and family determinant such as household food insecurity(9). The intervention or policy solely target on preventing NCDs behavioural risk factors may not have a good effectiveness if the household of the adolescents is still food-insecure. Thus, the national policy or interventionin reducing the food insecurity situation in the household should be implemented in addition to the current on-going intervention programs aimed at preventing the development of adolescent NCDs risk factors. A more functioning food system should be constructed in LMICs. Thus, by mitigating the effect of food insecurity, this can increase the effectiveness of the intervention or policy. Even though no difference in household food insecurity and adolescents NCDs behavioural risk factors by adolescents’ gender and area of residence, there is still potential effect on social and health inequality in those sub-groups. Therefore, the policies on reducing household food insecurity and adolescents NCDs risk factors should give particular attention to female and adolescents, lived in the rural area as well. In the local setting, evidence shown in this study isthat the association between household food insecurity and adolescent NCD risk behaviours is differentiated in different countries. Therefore, the policy implementation should be done according to the local socioeconomic, food insecurityand adolescent health status. For countries, with higher prevalence of food-insecure households and adolescent risk factorssuch as Ethiopia, the policy on reducing food-insecurity status in order to prevent the increased burdenof adolescent’srisk factors should be considered as one of the priority factor.

Area for future research

This study provides an overall association of food insecurity and adolescents NCDs behavioural risk factors in LMICs. However, further researches are needed in order to provide the answer to the question that is still not answered by this study. This study does notexplain the underlying mechanism of the association of adolescents NCDs behavioural risk factors by different household food security status. The data on adolescents’psychosocial factors and dietary diversity should be collected to investigatethe potential mechanisms.On the other hand, there is a potential possibility that the occurrence of some NCDsbehavioural risk factors may increase during late adolescence and that people aremore likely to report their use of tobacco and alcohol with increasing age. The mixed-method study with qualitative data collection can be considered as a good strategy to investigate the lived experience of adolescents from food insecure household. Adolescents with increased age tend to be more influenced by the household food insecurity as well(81). Thus, there is a need for relevant data for continuous monitoring of the prevalence of NCDs behavioural risk factors during their late adolescence time and also conduct a study to investigate whether there is an association between household food insecurity and adolescent risk behaviours in late adolescence-hood in LMICs. If a significant association can be determined, the following research can be conducted to investigate if there is any positive association between the increasing level of NCDs risk behaviours and the increasing severity level of household food insecurity On the other hand, thisstudyand most of the currently available researches used cross-sectional data to identify the association between household food insecurity and various adolescent risk behaviours. The future research should use the longitudinal data to determine whether there is any causality relationship between household food insecurity and NCDs risk behaviours among adolescents in LMICs. In addition to these, this studyisnot able to test the impact on adolescents from chronic food-insecure households to the development of adolescent NCD risk behaviours. The longitudinal data should be collected to test the above mentioned research question forfinding an association between adolescents from chronic food-insecure household and the development of NCDs risk behaviours.

Conclusions

In conclusion,by using the cross-sectional data from YL round 3 older cohort, it can be concluded that, there isa significanthigher prevalence of having NCDs behavioural risk factors among the adolescent from food insecure household, and there is an association between household food insecurity and adolescents after adjusted with covariates in the pooled sample among four LMICs, which are Ethiopia, India, Peru and Vietnam. The evidencehas been found in this study that assists to fill the current research as household food insecurity is a predictor of the occurrence of NCDs behavioural risk factors among adolescents in LMICs. Future studies should be conductedby using the longitudinal data to track the trend of the change in the occurrence of NCDs behavioural risk factors when adolescents suffer from chronic food insecurity and to investigate their association.Adolescents from food insecure household should be considered as a vulnerable group for developing NCDs behavioural risk factors.Therefore, a particular attention in this group of adolescents is needed during the policies or intervention implementation. Policy makers and programme organiser on preventing or reducing adolescents NCDs risk factors should collectrelevant data related to the household’s food security situation with valid tool to identify this risky group. Furthermore, based on the current public health intervention programs in preventing the development of adolescent behaviouralrisk factors in LMICs, the policies or interventions about reducing household food insecurity status should also be implemented at the same time to improve theireffectiveness.

References

Omran AR. The Epidemiologic Transition : A Theory of the Epidemiology of Population Change. 2014;83(4):731–57.

Kankeu HT, Saksena P, Xu K, Evans DB. The financial burden from non-communicable diseases in low- and middle-income countries: A literature review. Heal Res Policy Syst. 2013;11(1):1–12.

Alwan A, MacLean DR. A review of non-communicable disease in low- and middle-income countries. Int Health. 2009;1(1):3–9.

Abegunde DO, Mathers CD, Adam T, Ortegon M, Strong K. The burden and costs of chronic diseases in low-income and middle-income countries. Lancet. 2007;370(9603):1929–38.

Vos T, Allen C, Arora M, Barber RM, Brown A, Carter A, et al. Global, regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet. 2016;388(10053):1545–602.

Lim SS, Gaziano TA, Gakidou E, Reddy KS, Farzadfar F, Lozano R, et al. Prevention of cardiovascular disease in high-risk individuals in low-income and middle-income countries: health effects and costs. Lancet. 2007;370(9604):2054–62.

World Health Organization. Health for the World ’ s Adolescents A second chance in the second decade. 2014.

Romo ML. Hunger and Behavioral Risk Factors for Noncommunicable Diseases in School- Going Adolescents in Bolivia , 2012. 2016;1–6.

Eisenmann JC, Gundersen C, Lohman BJ, Garasky S, Stewart SD. Is food insecurity related to overweight and obesity in children and adolescents? A summary of studies, 1995-2009. Obes Rev. 2011;12(501):73–83.

Widome R, Neumark-Sztainer D, Hannan PJ, Haines J, Story M. Eating when there is not enough to eat: Eating behaviors and perceptions of food among food-insecure youths. Am J Public Health. 2009;99(5):822–8.

Alaimo K, Olson CM, Frongillo J. Low family income and food insufficiency in relation to overweight in US children: Is there a Paradox? Arch Pediatr Adolesc Med. 2001;155(10):1161–7.

Lohman BJ, Stewart S, Gundersen C, Garasky S, Eisenmann JC. Adolescent Overweight and Obesity: Links to Food Insecurity and Individual, Maternal, and Family Stressors. J Adolesc Heal. 2009;45(3):230–7.

Baer TE, Scherer EA, Richmond TK, Fleegler EW, Hassan A. Food Insecurity, Weight Status, and Perceived Nutritional and Exercise Barriers in an Urban Youth Population. Clin Pediatr (Phila). 2018;57(2):152–60.

Armour B.S., Pitts M.M. LC-W. Cigarette smoking and food insecurity among low-income families in the United States, 2001. EmbaseAmerican J Heal Promot. 2008;386–92.

Aurino E, Fernandes M, Penny ME. The nutrition transition and adolescents’ diets in low- and middle-income countries: A cross-cohort comparison. Public Health Nutr. 2017;20(1):72–81.

Alemu T, Asgedom G, Liebenberg J, Mekonnen A, Seager J, Tefera B, et al. Young Lives Preliminary Country Report: Ethiopia. University of Oxford, Open Access publications from University of Oxford. 2003.

Humphries DL, Dearden KA, Crookston BT, Fernald LC, Stein AD, Woldehanna T, et al. Cross-Sectional and Longitudinal Associations between Household Food Security and Child Anthropometry at Ages 5 and 8 Years in. J Nutr. 2015;145(8):1924–33.

Hassan F, Asim M, Salim S, Humayun A. House ownership, frequency of illness, fathers’ education: The most significant socio-demographic determinants of poor nutritional status in adolescent girls from low income households of Lahore, Pakistan. Int J Equity Health. 2017;16(1):1–12.

Jung NM, De Bairros FS, Pattussi MP, Pauli S, Neutzling MB. Gender differences in the prevalence of household food insecurity: A systematic review and meta-analysis. Public Health Nutr. 2017;20(5):902–16.

Grantham-McGregor S, Cheung YB, Cueto S, Glewwe P, Richter L, Strupp B. Developmental potential in the first 5 years for children in developing countries. Lancet. 2007;369(9555):60–70.