Early Poverty and Children’s General Health at 8 Years of Age in the Taiwan Birth Cohort Study

There are growing concerns about the effects of poverty on children’s health, but few studies have examined how the long-term experience of poverty affects the general health of children. We sought to investigate the relationship between children’s poverty experience from birth to school age and their general health status. Data from the Taiwan Birth Cohort Study (TBCS), a nationally representative sample of 17,853 children born in 2005 were used. Mother-rated health of children at age 8 was assessed. We used logistic regression to examine how the duration and timing of child poverty are related to children’s general health, controlling for child and maternal characteristics. Of the 17,853 cohort members, 23.6% experienced early poverty (persistent, 5.9%; occasional, 17.7%) before age 5, and 11.4% experienced concurrent poverty at the age of 8. Children born into poverty and remaining poor were more likely to have poor health than those who were never poor. Early poverty (persistent poverty: odds ratio (OR) = 1.50; 95% confidence interval (CI) 1.27–1.78; occasional poverty: OR = 1.25; 95% CI 1.12–1.39) was more detrimental to children’s general health than concurrent poverty (OR = 1.23; 95% CI 1.09–1.40). A long-lasting gradient effect of early poverty on the general health of children was detected, suggesting an urgent need to prevent children from growing up in persistent poverty.


Introduction
Poverty is a highly significant social determinant of children's health (Commission on Social Determinants of Health 2008). Children who are born and raised in poor families may be exposed to a cluster of specific risk factors or have a greater general 2 Methods

Study Population
The Taiwan Birth Cohort Study (TBCS) is the first large-scale longitudinal study of children in Taiwan. By using two-stage stratified random sampling, a nationally representative cohort of 24,200 babies born in 2005 was initially drawn from the National Birth Report Database, with a sampling rate of approximating 11.7%. Details of the information about the TBCS have been described in a previous publication (Chiang et al. 2010). The first study contact with the cohort children was at age 6 months, which provided information on 21,248 (87.8%) infants. Four waves of follow-up surveys were conducted at 18 months, 3 years, 5.5 years, and 8 years of age, with a response rate of 94.9%, 93.7%, 92.8%, and 91.8%, respectively. All five surveys were conducted in the home via face-toface interviews by trained staff with the mother or primary caregiver (> 93% were with the mother), who gave informed consent.
There were two criteria for inclusion in this study's sample: completion of the five waves of TBCS interview surveys between 2005 and 2013, and that respondents had available data for their poverty histories. Our final analytical sample comprised 17,853 children. The socio-demographic characteristics of the analytical sample were not statistically different from those of the baseline sample. The study was approved by the Institutional Review Board at National Taiwan University Hospital (ID number: 201704076RINB).

Variables
The primary dependent variable in this study was the general health status of children aged 8 years. Based on the mother's rating, we classified children into two groups: fair/ poor (fair, poor, or very poor), and good (very good or good).
The independent variable was the type of child poverty. We used parental income as our indicator of poverty to identify families with less than 50% of the national median income at each wave between 2005 and 2013 of the TBCS survey. If the mother or primary care giver answered '$30,000 New Taiwan Dollars (NTD) or less' to the question 'What is your (parents of child) monthly income in the last year?', then the family was defined as living in poverty (30 NTD≒1 US$; 46 NTD≒£1) (Chiang and Chiang 2018).
Based on each family's history of poverty, we used two features of povertytiming and durationto account for poverty dynamics. The timing of poverty was divided into early poverty (from birth to 5 years of age) and concurrent poverty (at age 8). Early poverty was further grouped into three types: persistent poverty, occasional poverty, and never in poverty (Chiang and Chiang 2018). Persistent poverty was defined as being poor during at least three consecutive waves of the survey. Children occasionally experiencing poverty (occasional poverty) were defined as those meeting the definition of living in poverty at least once but never for longer than two consecutive waves of the survey. Never in poverty was defined as never being in poverty during the five-year window of observation.
The main control variables of this study were child and maternal characteristics. Child characteristics included demographic factors (sex and birth order), neonatal health status (preterm and low birth weight) and early rearing environment (predominant breastfeeding). Predominant breastfeeding after birth was defined as infants who were predominantly breastfed for more than 120 days after birth (Wu et al. 2015). Maternal characteristics included age at birth of child (less than 25 years, 25-29 years, 30-34 years, and more than 35 years), nationality (Taiwan, foreign-born (China, Southeast Asia or others)), and level of education (junior high school or below, senior high school, and college and above). These control variables were selected because they are associated with children's health, and all these covariates were measured when the child was 6-months-old.

Statistical Analysis
We conducted descriptive analyses to provide a socio-demographic profile of the sample and different types of poverty. Bivariate associations of poverty types and each covariate with the health status of children were assessed by χ2 test. After the descriptive analysis, we estimated the unadjusted odds ratio (OR) for children's health by poverty types. Multivariate models of the relation between poverty types and children's health were then presented using logistic regression and controlling for confounding variables. All statistical analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, NC).

Results
Of the 17,853 cohort members, 23.6% of children in the TBCS had experienced poverty before the age of 5 years (5.9% persistent poverty, 17.7% occasional poverty), and 11.4% had experienced poverty at age 8 (Table 1).
Bivariate analysis indicated that children were generally more likely to be in fair/ poor health at 8 years of age if they had ever experienced poverty, particularly those in early poverty and persistent poverty ( Table 2). The results also showed that boys, firstborns, those born before 37 weeks of gestation, those with low birth weight, and children not predominantly breastfed for more than four months had a significantly higher prevalence of fair/poor health at 8 years of age. Mother's characteristics, including delivering the child when younger than 24 years and having a low level of education (less than junior high school), were found to be significantly associated with the fair/poor health of children (Table 3).
Logistic regression analysis demonstrated a relationship between poverty types and children's general health, unadjusted and adjusted for control variables (Table 4). In the unadjusted model, children who lived in persistent poverty (OR = 1.68, 95% confidence interval (CI) 1.46-1.94) and occasional poverty (OR = 1.32, 95% CI 1.20-1.45) during the first five years of life were more likely to have fair/poor health at 8 years than those who had never experienced poverty. We also found that children experiencing concurrent poverty (OR = 1.49; 95% CI 1.34-1.66) were more likely to have poorer health.
The inclusion of control variables in the model attenuated the associations between poverty types and children's general health, but the associations remained significant (Table 4). Early poverty predicted fair/poor health after controlling for concurrent poverty (persistent poverty: OR = 1.50; 95% CI 1.27-1.78; occasional poverty: OR = 1.25; 95% CI 1.12-1.39). Similarly, concurrent poverty was associated with having fair/poor health after controlling for early poverty (OR = 1.23; 95% CI 1.09-1.40). The results also indicate that early poverty has a stronger influence on children's general health than concurrent poverty.

Summary and Discussion
Using the TBCS data, we examined the relationship between the duration and timing of exposure to poverty from birth to school age and children's general health at age 8.
Children born and growing in poor families were more likely to have poor health than those of the never in poverty group. Early poverty before age 5, and particularly persistent poverty, was more detrimental to children's general health than concurrent poverty at age 8. Confirming the accumulation model of life course perspective (Ben-Shlomo and Kuh 2002), we found a gradient effect concerning the duration of poverty on the general health of children in Taiwan. That is, children who experienced persistent poverty had poorer health than children who experienced occasional poverty, and those in occasional poverty had poorer health than children who were never poor. Poverty is highly related to multiple forms of deprivation, including lack of the basic necessities of life, absence of housing facilities, and living in communities with environmental problems. Moreover, poor families are more likely to provide a lower quality of home enrichment, parenting, and child care (Evans 2004). As the duration of exposure to poverty increases, the damaging effect of a cluster of adversities on children's health and development may accumulate gradually over the life course (Ben-Shlomo and Kuh 2002; Black et al. 2017;Lynch and Smith 2005).
Our findings also support the proposition that early poverty has a long-term effect on the general health of children at the age of 8 years (Ben-Shlomo and Kuh 2002). In addition, we found that early poverty had more harmful effects than concurrent poverty on the health of the Taiwanese birth cohort's members. Previous studies have mainly focused on specific health and development conditions before the age of 5 years, showing that the experience of early poverty is highly associated with cognitive development, growth delay, and behavioural problems in children (Duncan et al. 1994;Ehounoux et al. 2009;Kiernan and Mensah 2009;Schoon et al. 2012). Moreover, using the US National Longitudinal Survey of Youth-Children, Chen et al. (2007) found that early poverty had a stronger effect on chronic or limiting conditions of children at the age of 10-11 than current poverty.
Why does early poverty have a long-lasting effect on children's general health? There are two plausible explanations. First, low socioeconomic conditions during the prenatal period and early childhood may affect children's constitution, leading to a permanent change in their general susceptibility to environmental challenges or illnesses (Hertzman  Marmot et al. 1984). For instance, Russ et al. (2014) proposed biological embedding mechanisms-such as the hormonal response system to stress, epigenetic modification, genetic imprinting, learned behaviours, or fetal adaptive response to undernutrition in utero-to explain how social adversity early in life influence longlasting functional and structural changes over the course of human development. Second, while children experience poverty that is accompanied with specific stressors in a critical period, these can cause specific and devastating long-term damage across different pathways (Galobardes et al. 2004;Lynch and Smith 2005;Marmot et al. 1984). For example, less access to cognitively stimulating materials and a negative mother-child relationship in early childhood may impair the structure or function of the brain, causing smaller hippocampal volume, greater or less-regulated amygdala, and higher basal cortisol (Johnson et al. 2016). Thus, future research should be conducted to clarify the mechanisms linking early poverty to children's general health.

Study Strengths and Limitations
The strength of this study lies primarily in its large-scale, nationally representative, longitudinal birth cohort design. We used a longitudinal prospective survey starting at birth, which avoided a recall bias concerning family income, allowed for accurate understanding of children's mobility across poverty status, and confirmed the temporal relationship between poverty and children's health owing to temporal sequence. Moreover, a large sample size provided enough statistical power to test relationships. Our study has some limitations. Self-reported parental income without further checking of actual income may have resulted in bias. However, we found that self-reported income to be negatively correlated with self-perceived financial strain and receipt of social welfare services, and positively correlated with house ownership (Chiang and Chiang 2018). Further, we used the mother's perceptions of the health of their children, which may have a mother's subjective bias, instead of children's self-reports to represent children's health. However, previous study has indicated that maternal ratings of child health are a good indicator of children's health status (Monette et al. 2007). Children who had health problems, such as acute diseases, chronic diseases, asthma attacks, or the experience of hospitalisation, were perceived by their mothers to be in poorer health.

Conclusions
The present findings contribute to understanding in the field of the long-term effect of early-life poverty on the general health status of school-age children. This study demonstrated that persistent poverty during the first five years was an important predictor of children's poor health. Thus, governments should address the issue of early poverty through employment, education, tax and health policies to reduce health inequalities across the life course (Graham and Power 2004;Chiang and Chiang 2018). In addition, paediatricians can assess the long-term economic conditions of children, collaborating with social communities to prevent children from remaining in poverty and improve their health outcomes (Gitterman et al. 2016).