Background Childhood poverty heightens the risk of obesity in adulthood, but the age at which this risk appears is unclear. We analysed the association between poverty trajectories with body mass index (BMI) Z-scores or the risk of being overweight or obese across four ages (6 years, 8 years, 10 years and 12 years) in childhood.
Methods Data were from the 1998–2010 ‘Quebec Longitudinal Study of Child Development’ cohort (n=698). Poverty was defined using Statistics Canada's thresholds, and trajectories were characterised with a Latent Class Growth Analysis. Multivariable linear and logistic regression models adjusted for sex, whether the mother was an immigrant, maternal education and birth weight.
Results Four income trajectories were identified: a reference group (stable non-poor), and 3 higher exposure categories (increasing likelihood of poverty, decreasing likelihood of poverty or stable poor). Compared with children from stable non-poor households, children from stable poor households had BMI Z-scores that were 0.39 and 0.43 larger than children from stable non-poor households at age 10 years and 12 years, respectively (p<0.05). Compared with children from stable non-poor households, children from stable poor households were 2.22, 2.34, and 3.04 times more likely to be overweight or obese at age 8 years, 10 years and 12 years, respectively (p<0.05).
Conclusion A latency period for the detrimental effects of child poverty on the risk of overweight or obesity was detected. Whether the effects continue to widen with increasing duration of exposure to poverty as the children age should be investigated.
- CHILD HEALTH
Statistics from Altmetric.com
Low socioeconomic status (SES) has been associated with poorer health and health outcomes among children.1 However, the association between SES and obesity among children is not clear, and has been primarily assessed with cross-sectional or retrospective studies. A recent systematic review found that 58% of the 45 cross-sectional studies published since 1990 reported no association, or a mixed association between SES (defined as neighbourhood SES, household income, parental education or parental occupation) and adiposity during childhood.2 The association was weaker in the studies assessing the association with household income. However, because these SES constructs have been shown to be distinctive from one another,3 ,4 and the income fluctuations5 may be more transient than other SES markers, the cross-sectional studies may have mostly undermined these analyses. Because adiposity tracks well from childhood to adulthood,6 ,7 a better understanding of the long-term effects of low SES on adiposity during children's developmental years is needed, but longitudinal studies in children are limited.8–11 Previous studies have primarily focused on late childhood, thus the age at which the association between poverty and adiposity appears during childhood and if the association changes across childhood is unknown. To the best of our knowledge, this is the first study to assess the association in Canadian youth using prospective data of income trajectories. Our objectives were to assess the longitudinal association of poverty and adiposity from birth to ages 6 years, 8 years, 10 years and 12 years. Our secondary objective was to compare our prospective longitudinal associations with measures traditionally conceptualised as cross-sectional (eg, currently living in poverty), or retrospective (eg, any past history of poverty).
The objectives were addressed in the Québec Longitudinal Study of Child Development study, a birth cohort (n=2120) representative of singleton births in the province of Québec, Canada in 1998.12 Full methodological details have been previously published and are briefly mentioned.12 Children were identified through the Québec birth registry and were excluded if they were of unknown gestational age, born before 24 weeks or after 42 weeks, living in Aboriginal territories or remote regions of Québec, or those with special needs (<5% of the target population). The remaining target population was randomly sampled using a multistage cluster sampling design. Children lost to follow-up (n=1046) were excluded, as were children with missing weight or height at ages 6 years, 8 years, 10 years or 12 years, or household income at any of the data collection cycles, resulting in a final analytical sample of 698.
Data collection was conducted annually in the participant's home and has been previously described (http://www.jesuisjeserai.stat.gouv.qc.ca/default.htm).13 The study was approved by the ethics committees of the Institut de la statistique du Québec, CHU Ste-Justine and the Université de Montréal. Parents and youth provided signed informed consent and assent, respectively.
When the child was approximately 5 months, 17 months, 29 months, 41 months and 5 years, 6 years, 7 years, 8 years, 10 years and 12 years of age, parents reported the past year's household income before taxes and deductions. Annual household income was compared with the low-income thresholds, as computed by Statistics Canada which adjusts for household size and geographical region.14 ,15 The low-income thresholds represent the income threshold for which families spend 20% more than the average family for household essentials such as food, shelter and clothing.15
Height and weight were measured by study staff at the four ages of interest. Participants wore light indoor clothing and no shoes. Study staff measured height to the nearest 0.1 cm with a standard measuring tape, and weight to the nearest 0.1 kg with a calibrated spring scale. Weight and height were measured in duplicate, and a third measurement was collected if the first two measurements differed by more than 0.2 kg for weight or 0.5 cm for height. The average of the two closest measurements was used for this analysis.
All data analyses were performed with SAS V.9.2.
Poverty categorisations: longitudinal analysis
We used the Latent Class Growth Analysis to identify different trajectories of poverty exposure. Latent Class Growth Analysis uses maximum likelihood in a semiparametrical mixture model to organise similar observations into groups.16 ,17 The analysis method is informative in describing group-level differences, and has been previously used to assess the association between low SES exposures and blood pressure or the development of asthma among youth.18 ,19
Based on the binary classification of poverty as described above, group-level trajectories from the ten data collection periods were created. Separate Latent Class Growth Analysis for exposure to poverty was conducted from birth to each age of interest. For instance, to develop the poverty trajectories at age 6 years, household income from ages 0–6 years was used; for the trajectories at age 8 years, income from ages 0–8 years was used, and so on. The implications of this are described in our discussion. A series of models varying in three to five trajectories, and complexity (linear, quadratic or cubic) were compared to identify the best fit. Consistent with the literature and the appropriate methods to identify the best fitting model for this analytical approach, the Bayesian Information Criterion from the model fit statistics was used.20 ,21
Poverty categorisations: cross-sectional analysis
Income measures that are traditionally more retrospective or cross-sectional include whether a household had any past history of poverty, or if the household is currently poor. We created these variables based on the binary poverty exposure variables previously described. For instance, for the cross-sectional categorisation, the income only at ages 6 years, 8 years, 10 years and 12 years were of interest, whereas for the retrospective categorisation, all previous exposure of poverty was of interest. Thus, if a child lived in a poor household only before the age of 5 years, the child would not be identified as currently living in a poor household at age 6 years, 8 years, 10 years or 12 years, but would be identified as having a past exposure to poverty at all four of these ages.
Our primary outcomes of interest included body mass index (BMI) Z-scores, and whether the child was overweight or obese. The child's BMI was calculated and converted to age-adjusted and sex-adjusted centiles and Z-scores according to the growth curves of the Centers for Disease Control and Prevention.22 Children's weight status was dichotomised into overweight or obese (BMI >85th centile) or not (BMI <85th centile) based on the Centers for Disease Control and Prevention definitions.22
Sensitivity analyses not excluding children with missing height or weight did not affect our results. Thus we present data for the 698 participants in order to facilitate comparisons and ease of interpretation of the risk estimates over the four ages of interest.
Estimating the longitudinal association between poverty and adiposity
Simple linear and logistic regression assessed the longitudinal association of poverty with BMI Z-scores, and the risk of being overweight or obese, respectively. Multivariable regression models adjusted for a child's birth weight, sex, whether the mother had at least a high school education and whether the mother was an immigrant. Adjusting for current household income did not affect our results and was not included in our final model due to a concern of overadjustment. Parental weight status was only available when the child was 5 months and 10 years of age. Additionally adjusting for parental weight status did not affect our results and was not included in the final model.
Comparison of cross-sectional to longitudinal estimates
The differences between prospective longitudinal data and cross-sectional or retrospective analyses were assessed with κ statistics for descriptive analyses, generalised F-tests for multivariable linear regression models, and changes in the -2 log likelihood for multivariable logistic regression models.
Our study sample (n=698) consisted of slightly more women (n=380) than men (n=318). Compared with our study sample, participants who were excluded or lost to follow-up were more likely to be from single-parent households and from poor households at baseline.
Based on Bayesian Information Criterion fit statistics, a linear four trajectory model best fit our data (figure 1, table 1). These were similar to the trajectories previously reported which were not restricted to those with BMI data, further illustrating the stability of these trajectories in our study sample.23 The trajectories that emerged were identified as consistently above the low-income threshold, consistently below the low-income threshold, and increasing or decreasing exposure to being below the low-income threshold. To simplify the language in our manuscript, we will refer to these groups as ‘stable non-poor’, ‘stable poor’, and ‘increasing likelihood of poverty’ or ‘decreasing likelihood of poverty’. The children in the decreasing likelihood of poverty group had poverty experiences early on in their childhood, the children in the increasing likelihood of poverty had more recent experiences with poverty, and the group of children in the stable poor group had persistent exposure to poverty. The stable poor trajectory group had a fair level of agreement with our cross-sectional measure of currently poor at ages 6 years, 8 years, 10 years and 12 years (κ=0.61, 0.64, 0.60 and 0.54, respectively), and slightly lower agreement with our retrospective measure of any past history of poverty by the ages of 6 years, 8 years, 10 years and 12 years (κ=0.49, 0.31, 0.33 and 0.32, respectively, data not shown).
At all ages, ∼75% of the participants belonged to the stable non-poor group, and the remaining ∼25% were distributed fairly evenly between stable poor, increasing likelihood and decreasing likelihood of poverty (table 2). The majority of respondents were in the same trajectory class at all four ages (87%). The proportion of years spent in poverty was highest among children in our stable poor trajectory.
Association of poverty and BMI Z-scores
Longitudinal exposure to poverty
In multivariable linear regression models after adjusting for covariates, children from stable poor households had BMI Z-scores that were 0.39 (95% CIs (CI 0.1 to 0.7)) and 0.43 (CI 0.1 to 0.7) larger than children from stable non-poor households at ages 10 years and 12 years, respectively (table 3). Children with a decreasing likelihood of poverty had BMI Z-scores at ages 8 years and 12 years that were 0.35 (CI 0.03 to 0.7) and 0.40 (CI 0.1 to 0.7) larger than children from stable non-poor households.
Cross-sectional or retrospective exposure to poverty
After adjusting for covariates, children with any past history of poverty had BMI Z-scores that were 0.22 (CI 0.1 to 0.4), 0.26 (CI 0.1 to 0.4) and 0.34 (CI 0.2 to 0.5) larger than children with no past history at ages 8 years, 10 years and 12 years, respectively. Children currently living in poverty at age 12 years had BMI Z-scores that were 0.33 (CI 0.1 to 0.6) larger than children not currently living in poverty.
Association of poverty and the risk of being overweight or obese
Longitudinal exposure to poverty
In our multivariable logistic regression, children from stable poor households were 2.22 (CI 1.1 to 4.5), 2.34 (CI 1.2 to 4.3) and 3.04 (CI 1.7 to 5.5) times as likely to be overweight or obese at ages 8 years, 10 years and 12 years, respectively, compared with children from stable non-poor households (table 4). Children with an increasing likelihood of poverty were 1.85 (CI 1.1 to 3.2) and 1.71 (CI 1.0 to 2.8) times as likely to be overweight or obese at ages 10 years and 12 years, respectively; children with a decreasing likelihood of poverty were 1.97 (CI 1.0 to 3.7) times as likely to be overweight or obese at age 8 years compared with children from stable non-poor households.
Cross-sectional or retrospective exposure to poverty
After adjusting for covariates, children with any past history of poverty were 1.75 (CI 1.2 to 2.6) and 1.76 (CI 1.2 to 2.5) times as likely to be overweight or obese at ages 10 years and 12 years, respectively, and children that were currently living in poverty at ages 10 years and 12 years were 1.79 (CI 1.0 to 3.2) and 1.84 (CI 1.0 to 3.2) times more likely to be overweight or obese, respectively, compared with children from stable non-poor households.
Comparison of models
Generalised F-tests revealed that the income trajectories did not contribute significantly more information to predicting BMI Z-scores than using the cross-sectional measure of currently poor, or the retrospective measure of any past history of poverty (data not shown). Income trajectories did not contribute significantly more information in predicting the risk of being overweight or obese at age 8 years or age 10 years, but contributed significantly more information at age 12 years (data not shown).
In our Québec birth cohort, children from stable poor households had significantly higher BMI Z-scores by the age of 10 years, and were more likely to be overweight or obese compared with children from stable non-poor households by the age of 8 years. In particular, the association increased as the children aged and underscores the importance of considering the long-term effects of previous exposures to poverty on health, particularly among children. Elevated risk estimates were also noted in our decreasing likelihood of poverty, further underscoring the long-lasting effects of past early-life exposures to poverty.
Several models to explain the pathways through which exposure to poverty during childhood may affect child health have been suggested.24–26 Some evidence suggests the effects of low SES on weight can be cumulative,24 ,25 while others suggest the mobility (such as downward or upward changes in SES over time), or exposure during a critical period is what is detrimental to health.26 Although originally conceptualised as models to explain the effects of childhood exposures on adult health, they are increasingly being used in studies among youth.1 Due to the paucity of longitudinal data, we chose to develop trajectories using data from birth to all four subsequent ages separately, allowing us to investigate the longitudinal association at different ages. This analytical decision resulted in growth trajectories that were not independently constructed across the different ages, but increased the stability of the trajectories, and is consistent with a life course model of cumulative risk.26 Our results suggest that long-term exposure or chronic exposure to poverty is perhaps the most predictive of excess adiposity in childhood, and supports the cumulative model. Our decreasing and increasing trajectories are conceptually similar to the downward/upward mobility model, and results suggest mobility may also have some detrimental effects on adiposity.26 Because the trajectories were stable, considering the exposure only before a certain age in accordance with a critical period perspective did not affect our main results (data not shown).
Although the risk estimates based on our trajectories were larger than those using cross-sectional or retrospective measures, only the risk of being overweight or obese at 12 years of age was significantly larger, likely due to their moderate or fair agreement. However, our longitudinal measures revealed an earlier age of impact on childhood adiposity compared with the cross-sectional measure, and we used 12 years’ worth of prospective longitudinal data to classify our participants regarding their previous exposure to poverty; data that were likely more precise than the measures traditionally collected in retrospective studies. Whether trajectories will continue to significantly contribute to models predicting overweight or obesity among youth as they continue to age should be further investigated.
Strengths and limitations
This study is not without limitations. Subjects were from a representative birth cohort of Québec children, but loss to follow-up and population changes in the demographics of children along the years have resulted in our study no longer being generalisable to Québec children at the studied ages. In particular, children from single parent households, or with lower parental education or household income were more likely to be lost to follow-up. Thus, the risk estimates in our study may be an underestimate of the association between poverty and adiposity in children. Environmental exposures such as poorer walkability, poorer access to fitness centres, and fresh fruits and vegetables have been associated with poverty27–29 but we were unable to investigate these hypothesised mechanisms. In addition, we were unable to adjust for daily caloric intake or metabolic equivalents, and these should be assessed in future studies. Our sample was fairly small; a larger longitudinal sample with more variation in poverty exposures is needed. Lastly, although BMI is a proxy for adiposity, it is inexpensive, non-invasive and strongly correlated with dual energy X-ray absorptiometry, the method of choice in measuring fat mass.30 ,31
Nevertheless, our results extend the current literature in three important ways. First, while past studies have been mostly cross-sectional or retrospective, our study uses a prospective longitudinal design. Second, we assessed the relationship between poverty and adiposity using Latent Class Growth Analysis, a novel method of analysing longitudinal data. This analysis method enabled us to explore group level differences of poverty exposure across childhood. Our descriptive results indicate the trajectories were distinctive measures from the commonly used current or past history to poverty underscoring the complementary information that these trajectories can provide. In particular, the proportion of years spent in poverty was longest for the children identified in the stable poor trajectory class, suggesting this trajectory group was comprised of a more vulnerable and extreme subset of the children with any past history of, or currently living in poverty. The fact that those who are currently living in poverty include children of families who became poor only recently as well as chronically poor ones might explain the mixed results of previous cross-sectional studies on the link between poverty and the risk of being overweight or obese in childhood. To the best of our knowledge, Latent Class Growth Analysis has only been used in one other longitudinal study by Kendzor et al3,2, to assess the association between poverty and adiposity; results are consistent with our study findings. Lastly, we compared single-point measurements with our longitudinal risk estimates and found the estimates from retrospective or cross-sectional methods generally underestimated the risk.
Elevated BMI Z-scores and the increased risk of being overweight or obese was found for children of increasing or decreasing likelihood of poverty, and was significantly elevated for children from stable poor households. Our results suggest that any exposure to poverty may have adverse effects on a child's adiposity, but public health policies to decrease chronic exposure to poverty will help diminish the risk of a child being overweight or obese. Our finding that the association between poverty and adiposity presented a latency effect and increased as children age underscores the necessity of considering lifetime exposures or long-term exposures to poverty on health, particularly among vulnerable populations of youth.
What is already known on this subject
Childhood poverty is associated with adiposity in adulthood. Latent Class Growth Analysis using prospective longitudinal data can be informative in describing group-level differences and identifying the age for which increased risk for adiposity begins, but has not been assessed in the context of adiposity across childhood among Canadian youth.
What this study adds
In this first birth cohort study to assess income trajectories and adiposity among Canadian youth, we found children that were living in stable poor households were significantly more likely to be overweight or obese by the age of 8 years compared with children living in stable non-poor households.
We found that the association between poverty and adiposity presented a latency effect and increased as children aged, underscoring the necessity of considering lifetime exposures or long-term exposures to poverty on health, particularly among vulnerable populations of youth.
Data were collected by the Institut de la Statistique du Québec, Direction Santé Québec. GP and LG hold Canadian Institute of Health Research (CIHR) Applied Public Health Research Chairs. Portions of this manuscript have been presented at the American Heart Association Epidemiology and Prevention, Nutrition, Physical Activity and Metabolism Scientific Sessions in San Diego, California, March 13–16, 2012.
† Dr Marie Lambert passed away on 20th February 2012. Her leadership and devotion to the Québec Adipose and Lifestyle Investigation in Youth (QUALITY) cohort will always be remembered and appreciated.
Contributors LK performed the statistical analysis, interpreted the data and drafted the manuscript. LS, ML, LG, BN and GP contributed to the interpretation of the data, reviewed the manuscript and provided substantive feedback. All authors were involved with the conception of the study and approved the final manuscript.
Funding This study was funded by the CIHR Grant Number MOP-67121 and by the Institut de la Statistique du Québec, Direction Santé Québec. LK is also supported through the CIHR grant (MOP-67121). These funding agencies were not involved in the study design, data analyses, data interpretation or manuscript writing and submission processes.
Competing interests None.
Ethics approval The study was approved by the ethics committees of the Institut de la statistique du Québec, CHU Ste-Justine and the Université de Montréal.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.