We used a subset of data from the CHILD Cohort Study, a national population-based birth cohort beginning in 2008 and recruiting from four centers across Canada; Toronto, Manitoba (including participants from Winnipeg, Morden and Winker), Edmonton and Vancouver. Details of the cohort can be found elsewhere (15). The current analysis was limited to the Manitoba site (n = 998). CHILD data were linked with data from the Government of Manitoba (GOM) using Personal Health Identification Numbers. Every other year, the GOM routinely collects Early Development Instrument (EDI) data during kindergarten. The biennial data collection of the EDI limits the sample for the current study to approximately half that of the Manitoba CHILD site. CHILD participants who consented to administrative data linkage and had complete EDI data were included in the study (n = 442; Supplementary Figure 1). Informed written consent was obtained by all participating parents prior to data collection and this study was approved by the Human Research Ethics Boards at McMaster University and University of Manitoba.

The EDI is a 104 item questionnaire that measures children’s readiness for school at kindergarten across five domains of child development; (1) physical health and well-being, (2) social competence, (3) emotional maturity, (4) language and thinking skills, and (5) communication skills and general knowledge (8). Each domain has a score ranging from 0 to 10 with higher scores indicating better development. The EDI categorizes children into one of four categories in each domain based on percentile cutoffs in the population: top (highest 25%); middle (middle 50%), at risk (bottom 25–10%) and vulnerable (bottom 10%). We used percentile data from the entire population of Canadian children and applied the score cut offs to our CHILD cohort sample (Supplementary Table 1). Following this, we derived an outcome variable that classified children as vulnerable or not on one or more EDI domains (i.e., EDI vulnerability) (8).

Latent factors are unobserved variables that are calculated and measured by observed variables (10, 16). In collaboration with early child development experts from the University of Manitoba and the GOM, we developed a hypothesized model of predictors of child development at kindergarten using available CHILD data and grouped the individual variables into latent factors based on previous literature and our study team’s expertise. We evaluated a range of potential predictor variables across 8 original latent factors (n = 35 variables considered in total) and used model fit statistics to select and refine each latent factor by reclassifying variables into new latent factors or omitting variables entirely when they impeded good fit. This process left us with n = 23 total predictor variables classified into 6 latent factors in the final hypothesized model. The latent factors included: (1) prenatal risk behaviours; (2) family stress at 1 year; (3) child health at 1 year; (4) child health and lifestyle at 3 years; (5) family stress at 3 years; and (6) socioeconomic status. Descriptions of the latent factors are below.

“Prenatal risk behaviours” is comprised of data from prenatal questionnaires. Mother’s smoking status during pregnancy was a binary variable (no smoking compared to any smoking during pregnancy). Maternal stress was measured using the Perceived Stress Scale (PSS), a widely used 10-item instrument for measuring perception of stress in the last month (17, 18). The PSS ranges from 0 to 40, with higher scores indicating more stress. Maternal depression was measured using the Centre for Epidemiologic Studies Depression (CES-D) scale, a 20-item measure which asks caregivers to rate their experiences of symptoms associated with depression over the last week (19). CES-D scores range from 0 to 60, with higher scores indicating more depressive symptoms. The stress and depression variables were used as continuous measures. Mother’s diet was collected using the updated Healthy Eating Index (HEI) 2010 total score (range 0 to 130). HEI is a measure of diet quality that meets standards of the United States dietary guidelines using 12 components; higher scores indicate better diet quality (20).

“Family stress at 1 year” is comprised of data from postnatal questionnaires at 6 months and 1 year. These included maternal stress (using the PSS) and depression (using the CES-D) at 6 months, and parenting stress at 1 year. Parenting stress was measured using the Parent–Child Dysfunctional Interaction (P-CDI) sub-scale from the Parenting Stress Index. The P-CDI scale is a 12-items parent-reported measure of parent satisfaction with the interactions with their child (21). The scale ranges from 12 to 60; higher scores indicate more parenting stress between the parent and child. All stress and depression variables were used as continuous measures.

“Child health at 1 year” is comprised of data from hospital birth charts, survey questionnaires at birth and 1 year, and a clinical assessment at 1 year. Weight gain was calculated as the change in weight for age z-scores from birth to 1 year and characterized into a binary variable (weight gain velocity ≤ 0.67, or weight gain velocity > 0.67 (22)). Atopic conditions were characterized into a binary variable (no atopic conditions compared to one or more) using clinical and questionnaire data asking about: child wheezing, atopic dermatitis (physician diagnosis) and atopic conditions to food (using a skin prick test). Smoking in the home was defined as anyone smoking in the home at 1 year of age (yes or no). Number of emergency room (ER) visits were measured in the first year of life and categorized as none, one, two, or three or more.

“Child health and lifestyle at 3 years” is comprised of data from the one and a half to three-year questionnaires. Categories of fruit, vegetables, and sugar-sweetened beverages were created based on questions asked in the child food frequency questionnaire at 3 years. Fruits and vegetables were combined and dichotomized as less than or equal to five servings per day and greater than five servings per day. Sugar sweetened beverages was dichotomized as no servings or any servings per day. Sleep was derived from combining number of hours of night sleep and nap time durations during the weekday from the child three-year questionnaire. Number of ER visits were measured between one and a half and 3 years of age and categorized as none, one, two, or three or more.

“Family stress at 3 years” is comprised of data from the three-year questionnaires. These included maternal stress (using the PSS), maternal depression (using the CES-D) and parenting stress (using the P-CDI). All variables were used as continuous measures.

“Socioeconomic status” is comprised of data from the prenatal questionnaires. Total household income was categorized as < $80,000 or ≥$80,000. Marital status was defined as married/common law or single/never married/divorced/separated. Education was categorized as no post-secondary degree or completed a post-secondary degree. Since most mothers in the CHILD cohort completed some form of university, we were unable to use more granular categories for education. Perceived socioeconomic status was measured using a picture of a ladder with the top of the ladder being defined as people who have the highest standing in their community, and the bottom of the ladder being defined as people who have the lowest standing in their community. Participants scored where they believed they best fit within the community ladder (23).

Males were over 3 times more likely than females to be vulnerable on one or more EDI domains (odds ratio (OR): 3.54, 95% confidence interval (CI): 2.14–6.01, Figure 1). Children were slightly younger in the vulnerable group (9% decrease in being vulnerable per each additional month of age; OR 0.91 95% CI: 0.85–0.98). Having two or more visits to an ER in the first year of life, compared to no visits, was associated with a 2.02-fold increased odds (95% CI: 1.06–3.76) of EDI vulnerability. Daily screen time of greater than 2 hours was associated with a 2.02-fold increased odds (95% CI: 1.14, 3.57) of EDI vulnerability. A one standard deviation increase in maternal stress and depression scores at 6 months or 3 years and parenting stress scores at 3 years were all associated with increased odds of EDI vulnerability (increased odds between 39 and 43% at 6 months and 33 and 45% at 3 years). Protective sociodemographic factors, including having a household income over $80,000 (OR 0.48; 95% CI: 0.28–0.81), a mother who was married or common law (OR 0.16; 95% CI: 0.08–0.32), or a mother who had completed a post-secondary degree (OR 0.47; 95% CI: 0.29–0.75), were associated with lower EDI vulnerability. A one standard deviation increase in maternal prenatal diet quality was associated with a 21% decreased odds (OR 0.79; 95% CI: 0.62–1.00) of EDI vulnerability. Other hypothesized predictors of child development were not statistically different between the vulnerable and not vulnerable groups (ex. older siblings, weekday sleep duration).

All six latent factors passed the thresholds for having good model fit (i.e., CFI or TLI scores ≥0.90 and RMSEA and SRMR scores ≤0.10; Table 2). Most variables (17/23) had standardized loadings that were ≥0.32, which fall under Tabachnick and Fidell’s rule of thumb for minimum loading onto a factor (30). Some variables were < 0.32, but were still included in the model to leverage the richness of the CHILD study data and adhere to our theoretical model of the variables. In addition, variables with low standardized loadings were distributed across the latent factors, rather than grouped onto a single factor, therefore they were not viewed as problematic (31).

Adjusted, standardized coefficients modeling the relationships between each of the latent factors and EDI vulnerability are represented in Figure 2 and Supplementary Table 2. All models maintained good model fit as indicated by the fit indices (Supplementary Table 2). The prenatal risk behaviours latent factor (comprised of prenatal maternal stress, depressing, smoking and diet), was not significantly associated with EDI vulnerability at kindergarten (standardized estimate = 0.09, p = 0.39). Higher family stress at 1 year and 3 years was significantly related to EDI vulnerability: a one standard deviation (SD) increase in family stress at 1 year was associated with a 0.20 standardized increase in EDI vulnerability (p ≤ 0.001); and the association was even stronger for family stress at 3 years (standardized estimate = 0.33, p ≤ 0.001). Higher socioeconomic status (SES, latent class comprised of maternal education and marital status, household income and perceived status) was related to lower EDI vulnerability (standardized estimate = −0.26, p = 0.01). Our latent factor measures of child health at 1 year (standardized estimate = −0.04, p = 0.77) and child health and lifestyle at 3 years (standardized estimate = −0.21, p = 0.18) were not statistically related to lower risk of EDI vulnerability at kindergarten.

To explore whether the observed association between higher SES and decreased EDI vulnerability could be explained by experiencing lower family stress, we performed a mediation analysis using parametric SEM models. After adjustment for covariates, lower family stress at 3 years was a significant mediator in the relationship between SES and EDI vulnerability, accounting for 10.6% of the total path standardized estimate (indirect path standardized estimate = −0.05, p ≤ 0.05; Figure 3 and Supplementary Table 3). Family stress at 1 year did not significantly mediate this relationship (indirect path standardized estimate = −0.03; Figure 3 and Supplementary Table 3); however, it did account for 9.7% of the total path standardized estimate.

Low SES and family stress at 1 and 3 years were significant predictors of EDI vulnerability. The detrimental associations between low SES with child development has been established through decades of previous research (32, 33). Single parenthood, maternal and paternal occupation, living in a more disadvantaged area, and lower household income all have significant relationships with EDI vulnerability (12, 13). While measures of SES are often not directly modifiable, these results demonstrate the need to provide extra support to children living with lower SES.

The family stress latent factor in our analysis, comprised of maternal stress and depression and parenting stress at 1 and 3 years, are modifiable predictors of EDI vulnerability that can be targeted to improve child development. Previous work has established postnatal maternal depression and anxiety as strong predictors of EDI scores (11, 34). It has been suggested that maternal mental health may interfere with the ability for mothers to be engaged in learning with their child, respond sensitively and consistently, and form secure attachments (35–37). Infant attachment insecurity has been associated with poorer child executive functioning at kindergarten (38). Furthermore, in the current study, family stress at 3 years was a significant mediator in the link between SES and early EDI vulnerability at kindergarten. Other studies have examined the interplay between the family environment, SES and child development (10, 33, 34). In line with our study, previous research shows SES and family stress have both independent and combined effects on child development.

Together, our results could support the funding and implementation of early child and parenting support programs among low SES communities in Manitoba. One example is the Abecedarian Program (39), which has been operating in some low SES neighbourhoods in Manitoba since 2012 (40). The Abecedarian approach in Manitoba supports child development through four pillars: language priority, conversational reading, interaction games and enriched caregiving (41). Using a randomized design, the program was shown to improve child language scores, and through qualitative analysis, was shown to strengthen parents’ role in supporting their child’s development in Manitoba (40, 41). The current study could be used as evidence to support the importance of programs, such as Abecedarian, that can bolster child development, particularly in low SES settings. Future work could link participation in the Abecedarian Program with EDI scores, and could use a randomized design to test the effectiveness of other interventions aimed to reduce family and parenting stress on improving child EDI scores.

Child health and lifestyle factors at 1 and 3 years of age did not emerge as significant predictors of EDI vulnerability. However, the standardized estimate of child health and lifestyle at 3 years was of similar magnitude to that of family stress at 1 year (a significant predictor of EDI), indicating that it may still be an important predictor regardless of statistical significance. Contrary to our findings, a previous cross-sectional analysis, found that poorer child health, measured using the Health Utility Index, was significantly related to being vulnerable on one or more EDI domains (12). The Health Utility Index differs from our measure of health as it is a more comprehensive, standardized assessment and was measured at the time of EDI assessment in the aforementioned cross-sectional analysis.

Prenatal risk behaviours were also not related to EDI vulnerability in the current study. Previous studies have shown that prenatal smoking (10, 13) and prenatal maternal depression and anxiety (34) are related to poorer EDI scores. In the current analysis, prenatal maternal depression and stress were not related to EDI vulnerability, however, these same mental health measures at 1 and 3 years were related to EDI vulnerability. Previous work has highlighted the potential deleterious effects of exposure to chronic maternal depression and anxiety (34). While our study did not explicitly measure chronic stress, children exposed to depression or stress at 3 years may also have been exposed to it at earlier time points, indicating a possible chronic exposure. Further research should focus on the associations of chronic stress during the first 3 years of life on child EDI scores. Such analysis was beyond the scope of this paper and would require a larger sample size than what was available in the current study.

This study is strengthened by combining rich longitudinal survey data from the CHILD cohort with a population-based validated measure of child development. The EDI is teacher reported, removing parental bias on reports of child functioning. Compared to previous research, our study expands the number of predictors evaluated and incorporates groups of predictors into statistically and theoretically coherent latent factors, providing a more holistic approach for examining childhood exposures. The current study is limited by a relatively small sample size compared to previous population-based studies, which may result in type two error whereby significant associations exist, but cannot be detected due to lower power. Further, the observational nature of the data limits our ability to draw causal conclusions about relationships between early childhood factors and EDI scores. Future work using randomized designs of potential interventions to address family stress, particularly among those with low SES, could provide stronger evidence for our conclusions. Third, while our models meet the traditional thresholds for good model fit, SEM fit indices may have complicated behaviour as functions of the true underlying model, number of parameters, and sample size (42); the reader may wish to interpret our results with caution. Fourth, we used complete case analysis and found that missing data were not completely at random, potentially biasing our results [i.e., there was more missing data among those who were vulnerable on one or more EDI domains (Supplementary Figure 2)]. Fifth, the CHILD cohort is comprised of a higher SES profile than the general Manitoba population, limiting the generalizability of our results, especially for the most vulnerable. Finally, we lacked information on some key variables that are known to impact child development such as father’s mental health, quality of the home learning environment and participation in early pre-school, daycare or nurseries (43, 44).