The Family Stress Study is a prospective cohort study of children aged 2-6 years recruited from 2 Canadian sites: the University of Guelph in Guelph, Ontario, and McMaster University in Hamilton, Ontario. Ethics approval was obtained before the restrictions due to the COVID-19 pandemic were enacted in Ontario in March 2020, but recruitment had not begun. The original protocol, which included an in-person health assessment of adiposity using air displacement (bod pod) and researcher-led measures of child and caregiver height, weight, and child WC, was revised to a caregiver-led at-home health assessment of child and caregiver weight, height, and child WC to adhere to the COVID-19 safety protocols. Measures of COVID-19–related stress were added to the web-based health and behaviors survey. The revised protocol was approved by both research ethics boards. Figure 2 provides an overview of this study’s design.

Between July 2020 and December 2021, participants were recruited from the city of Guelph, the city of Hamilton, and surrounding areas. Recruitment was conducted primarily via the web using paid advertisements on Facebook (Meta) and Instagram (Instagram from Meta), and social media posts, including posts by community organizations that serve families, and through posters in community locations, including libraries and toy stores. All methods of recruitment provided caregivers with study contact information and directed them to this study’s website, where caregivers obtained additional details regarding this study’s process and interested caregivers completed an eligibility screener.

To be eligible to participate, families had to have at least one child aged 2-6 years and have no plans to move from the area within the next 3 years. Participants were excluded if they were (1) enrolled in the Guelph Family Health Study (a longitudinal study led by the investigative team) at the University of Guelph to allow data to be merged across the 2 studies; (2) children born preterm (before 34 weeks gestation); (3) children with any health condition expected to affect cortisol production, obesity, or growth or body composition (eg, Prader Willi syndrome and Cushing disease); and (4) children taking steroid medications, including oral or inhaled corticosteroids, medications for attention-deficit/hyperactivity disorder, or antidepressants or antipsychotic medications.

To participate, caregivers had to live with the child at least 50% of the time and could include parents (biological, related, adoptive, or foster), grandparents, aunts, and uncles. The primary caregiver was the caregiver who first signed up for this study and served as the main contact for the family. Eligible primary caregivers completed a web-based consent to enroll themselves and their oldest eligible child in the target age range (2-6 years). Primary caregivers provided contact information for themselves and, if applicable, an email address for a second caregiver. The secondary caregiver was contacted separately to enroll and complete web-based consent. Study coordinators also connected regularly with families via email, text, or phone to provide them with support throughout the data collection process.

This study was approved on March 11, 2020, and the amended protocol for remote data collection was approved by the University of Guelph Research Ethics Board (11-19-047) on June 12, 2020. The Hamilton Integrated Research Ethics Board approved this study on July 7, 2020, along with approval for the amended protocol on July 20, 2020 (10763). Recruitment and data collection began in July 2020 (Guelph) and August 2020 (Hamilton); primary caregivers enrolled in this study provided informed consent for themselves and their oldest eligible child and secondary caregivers provided informed consent for themselves. All participant data will be deidentified and only group data will be presented in all reports of this study’s results. At each data collection time point, participating families receive a CAD $25 grocery gift card per caregiver for completing a health and behaviors survey (also includes the diet recall for the child), and a CAD $30 grocery gift card per family for completing the at-home health assessment as a thank you for their time.

Chronic stress and children’s weight-related behaviors including diet and physical activity or sleep, cortisol levels, and adiposity are assessed at baseline, 1- and 2-year follow-ups. Further, 1-year intervals for assessment are based on Canadian health practice guidelines that recommend measuring height and weight once per year for children aged 2 years and older [26]. At each time point, primary caregivers complete a web-based health and behaviors survey with questions about family demographics, family stress, child weight-related behaviors, mental health, and their own weight-related behaviors. Secondary caregivers complete an abbreviated version of the health and behaviors survey focused on measures of stress and their own weight-related behaviors. Primary caregivers complete a web-based 1-day dietary assessment for their child. Caregivers complete anthropometric measures, collect a hair sample for cortisol analysis, and fit their child with an activity monitor for the at-home health assessment.

Chronic stress is assessed using 5 different measures of stress: household stress, food insecurity, caregiver depression, children’s negative life events, and COVID-19–related stress. Household stress is quantified using the 15-item Confusion, Hubbub, and Order Scale (CHAOS) [38]. Primary caregivers respond to items including “you can’t hear yourself think in our home” or “there is very little commotion in our home” on a 4-point Likert scale from 1 (very much like your own home) to 4 (not at all like your own home). After reverse scoring relevant items, a total score will be calculated by summing the responses. The full CHAOS score [39-41], as well as the Emotional CHAOS subscale [4], created using 8 of the full-scale items, have been associated with obesity risk and obesogenic eating behaviors among young children. Standardized Cronbach α for the CHAOS survey was found to be 0.78 in the Guelph Family Health Study [42], which is a longitudinal study of families with toddlers and preschoolers [43].

Food security is examined using the Household Food Security Survey Module, an 18-item measure that includes 10 items that assess uncertain, insufficient, or inadequate food access related to adults in the household and 8 items focused on children over the previous 12 months [44]. Households will be categorized based on the total number of positive responses: food secure households (0 responses), marginally food insecure households (1 responses), moderately food insecure households (2 to 5 responses), or severely food insecure households (≥6 responses). While evidence for an association between food insecurity and obesity risk in children has been mixed [45], food insecurity is a prevalent source of stress among families, affecting 1 in 6 Canadian children [46], and has been shown to be associated with poorer diets in children [47].

Caregiver depressive symptoms are assessed using the 20-item short-form Centre for Epidemiologic Studies Depression Scale (CES-D) [48-50]. Both the primary and second caregivers respond to statements about the previous week including “everything I did was an effort” and “I felt lonely.” Responses are to be scored as 0 (less than 1 day), 1 (1-2 days), 2 (3-4 days), or 3 (5-7 days) and, after reverse scoring relevant items, a total score out of 30 will be calculated by summing the responses. The standardized Cronbach α for the CES-D was found to be 0.88 for mothers and 0.80 for fathers of young children in the Guelph Family Health Study [51]. Parental depression assessed using CES-D has been found to be associated with obesity risk and obesity-related behaviors in children [51,52].

Children’s negative life events are reported using an abbreviated List of Threatening Experiences [53], an 8-item measure of negative life events, for example, divorce. Primary caregivers report negative events for their child across their lifetime (or for the past year on the follow-up surveys) by answering “yes” or “no” to whether or not their child has encountered various events, such as “divorce or separation of parents” or “severe disease or accidents.” The total score will be determined by summing the number of negative events. This measure has been found to be associated with obesity risk in children [54].

COVID-19–related stress is assessed by asking both the primary and secondary caregiver about quarantine or illness due to COVID-19, change in employment status due to COVID-19 (eg, loss of job and change in work hours per location), and employment in essential services (eg, first responder, health care worker, critical infrastructure worker, food production and supply, production, supply of medicine, hydro, and gas, and corrections). Given that COVID-19 disrupted home and work life among families with young children, measures assessing work conflict [55], financial insecurity [56], and parenting stress [57] are also included in the caregiver’s web-based survey. Work-to-family conflict is assessed using a 5-item scale adapted from Carlson et al [55]. Caregivers are asked questions including “my work keeps me from family activities more than I would like” and “when I am home from work, I am often too frazzled to participate in family activities or responsibilities.” Responses are on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree) [55]. A total score will be calculated by summing the responses.

Financial insecurity is evaluated using 2 items adapted from the measures by Gundersen and Boushey [56]: “during the past month, was there a time when you were worried you would not be able to pay the mortgage, rent or other bills on time?” and “are you worried about not being able to pay the mortgage, rent or other bills on time over the next 6 months?” to which caregivers respond “yes” or “no.” Families will be coded as financially insecure if they respond positively to either item.

Parenting stress is assessed by asking caregivers 12 items from the parenting distress domain of the short-form parenting index [58]. Items include “I feel trapped by my responsibilities as a parent/caregiver,” “there are quite a few things that bother me about my life,” and “having a child has caused more problems than I expected in my relationship with my spouse/partner (or male/female friend).” Responses are on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree). After reverse scoring relevant items, a total score will be calculated by summing the responses. The standardized Cronbach α in a similar sample was 0.86 among mothers and 0.78 among fathers [59].

Primary caregivers report their child’s sex. Both the primary and secondary caregiver report their education level. Quality of caregiver relationship is assessed for both primary and secondary caregivers using the warmth or affection subscale of the Parenting Acceptance and Rejection Questionnaire [58,71] which measures parents’ feelings of warmth and acceptance (vs rejection) toward their child, a central aspect of relationship quality [72,73] that has been associated with multiple child outcomes [74,75]. Caregivers respond to items including “I say nice things about my children” on a 4-point Likert scale from 1 (almost always true) to 4 (almost never true). After reverse coding relevant items, a total score will be calculated by summing the responses. Both the primary and secondary caregivers completed an abbreviated 30-item version of the Behavior Rating Inventory of Executive Function [76] to assess caregiver self-regulation. A total score will be calculated by summing the responses across the 30 items. Time-varying moderators will be assessed at all 3 time points, whereas time-invariant moderators will only be assessed at baseline.

Primary caregivers report their child’s birth weight, gestational age, duration of breastfeeding, annual and sources of household income, child and caregiver race or ethnicity, and caregiver employment status (full- or part-time). Secondary caregivers report their race or ethnicity and their employment status.

The research team has extensive experience in strategies for retention in longitudinal family studies [77-79]. To maximize retention in the Family Stress Study, many of these same strategies are used including (1) regular emails to maintain connection with participants, (2) mailings with tracking to obtain forwarding address information, and (3) financial incentives for assessments.

To summarize baseline caregiver and child characteristics for this paper, descriptive statistics were used. Percentages are presented for categorical data. Means and SDs are used for continuous data.

This study’s objectives will be tested using an SEM approach for longitudinal models that combines latent growth models (LGMs) with auto-regressive and cross-lagged paths using a special parameterization that separates the time-invariant (between-individuals) and time-varying (within-individuals) components [80,81]. Before constructing the SEM models, distributions of all the variables, their skewness and kurtosis levels, and their presence of extreme or outlier responses above or below 3 times the IQR will be explored and outliers will either be removed or Winsorized.

An incremental model-building approach similar to the recommended approach by Curran et al [80] will be used to test this study’s objectives. The first component in the model-building approach is the LGM of the trajectories across time for the variables’ chronic stressors (explanatory variable), child cortisol and child behavior (mediator variables), and child adiposity (outcome variable). Each variable has its own trajectory and includes an intercept (I), which refers to a baseline value, and a slope (S), which refers to change over time. Of particular importance is that individual child variation in these I and S parameters will be modeled. If any of these components are found to be not statistically significant, they are removed from the model. Typically, the I means and variances are significantly different from 0, but the S means may or may not depict significant linear change. The second component consists of the auto-regressive associations (ie, T2 regressed on T1, and T3 on T2) capturing the rank-order stability of the variables across the 3 time points (eg, the relationship between cortisol at T1 and T2). The third component typically consists of cross-lagged paths or other hypothesized paths between the variables, accounting for growth factors, and stability factors. In this study, this component will consist of the longitudinal mediation paths (ie, from explanatory variable at T1 to mediator at T2, to outcome at T3). The hypothesized model includes 2 partial mediators of the relation between stressors and child adiposity: child cortisol and behaviors. To test mediation, a direct path from stressors at T1 and child adiposity at T3 is also included. A correlational path between the residualized mediators is also included to model any leftover covariance between them. The fourth component will assess the moderation of the mediated paths. For example, to test the hypothesis that the cortisol mediation may differ across the sex of the children, we can use a multiple groups SEM approach that compares the effect size of mediation across male and female children.

As with any longitudinal study, attrition is expected. Patterns of missing data and potential bias at time 2 and time 3 will be investigated by comparing demographics and variable scores of participants who are missing to those who are not missing at each time point. In addition to noting and acknowledging any bias due to missingness, the SEM modelling procedure uses a maximum likelihood estimator which also provides one of the best ways to estimate the model parameter estimates in the presence of missing data. While there is no guarantee that the data are missing completely at random, in longitudinal designs, data from earlier time points compensates for the missingness at later time points in the estimation procedure.

To estimate the power of specific paths in the model, an empirical simulation approach was used [82,83]. To run power analysis for LGM models, Muthén and Muthén [83] proposed a general framework using Monte Carlo simulation. Further, 2 separate a priori power analyses were conducted to determine whether a sample size of 300 is sufficient to detect the cross-lagged effects of interest. The first model assumed data were normally distributed and the second model assumed nonnormality in 10% (n=30) of the cases. For each simulation, a sample size of 300 was specified with 10,000 replications successfully completed. For both the normally distributed and nonnormally distributed data scenarios, all estimated parameters had between 82% and 86% coverage, and the parameter estimate and SE bias were below 5%, suggesting sufficient power.

Despite public health restrictions due to COVID-19, which prevented in-person recruitment, the Family Stress Study was successful in recruiting 359 families into the Family Stress Study cohort, and 301 of those families completed all baseline measures for their child. Web-based recruitment, in particular partnering with various community organizations that serve families with young children, was critical to the recruitment success during this time when in-person engagement was limited. Identifying creative ways to connect and share information with families was also needed to ensure effective data collection. This included having research staff connect regularly with families via email, text, or phone to provide them with support throughout the data collection process and by providing information about using accelerometers to measure physical activity in a family-friendly video that could be shared remotely. These approaches were well received by families and could help inform future family-based research using remote recruitment and data collection.

Although successful in engaging the required sample size, the use of remote recruitment strategies may have impacted the socioeconomic diversity of the sample. While a previous family-based intervention study found that web-based recruitment of families with young children resulted in equivalent participant demographics to traditional recruitment strategies (eg, in-person or via practitioners) [84], other intervention [85] and noninterventional [86] family-based obesity studies have found that in-person recruitment was critical to engaging families, particularly low-income families. Thus, it is possible that had in-person recruitment been possible, more low-income families could have been recruited. Nearly half of the families in the sample reported an annual household income greater than CAD $100,000, which may limit the generalizability of the findings to more socioeconomically disadvantaged families. However, the median income of the sample was CAD $102,800, which is similar to the 2020 median household income (before taxes) for families in Guelph and Hamilton, which was CAD $114,000 and CAD $107,000, respectively [87]. Similarly, in our sample 73.1% (n=263) of the children identified as White, which is similar to the 2021 census data for Guelph and Hamilton which was 73.1% and 72.6%, respectively [87], suggesting that the annual household incomes and race or ethnicity of the children in our sample may be representative of the overall population within these cities.

The need to use remote data collection for the primary outcome, child adiposity, is also a limitation of the Family Stress Study. While the protocols are based on validated methods, the use of caregiver assessments may result in greater error in child adiposity measures than if using researcher-led assessments and this could lead to misclassification for some children. However, the use of caregiver-led assessments allowed the researchers to feasibly implement this study during the COVID-19 pandemic, which represented an unprecedented disruption to family life and associated increase in family stress.

To design effective obesity prevention interventions, there needs to be a clear understanding of key factors influencing excess weight gain in children. By identifying both the direction and the relative contributions of the biological and behavioral pathways linking chronic stress and adiposity, results from the Family Stress Study will help inform interventions to modify these pathways and reduce obesity risk among young children. Specifically, this research will identify the key behavioral drivers linking stress and obesity to inform which weight-related behaviors should be targeted with interventions in families experiencing high stress. This research will also identify whether the quality of caregiver relationships, child sex, and caregiver education and self-regulation moderate the pathways linking stress and obesity; this information is needed to inform what types of families may be best targeted in obesity prevention interventions. Thus, the results of this research will provide a much-needed basis to inform future childhood obesity prevention interventions.