1 INTRODUCTION

Early childhood is an important period for the development of stress regulation systems (Lupien et al., 2009). Stress regulation refers to an appropriate physiological and psychological response to environmental challenges or threats and the body's ability to return to balance after a stress load (Joseph & Whirledge, 2017). The hypothalamus–pituitary–adrenal (HPA) axis is an essential neuroendocrine system involved in human stress regulation. The HPA axis is activated and produces cortisol in response to psychological or physical stress (Gunnar & Quevedo, 2007; Tsigos & Chrousos, 2002). Cortisol production follows a circadian rhythm in which cortisol levels rise rapidly 30–45 min after waking up in the morning and decline toward the evening being at the lowest before sleep (Gunnar & Quevedo, 2007). Prolonged and repeated exposure to stress may have permanent and adverse effects on the brain and especially on the structures that are developing during the time of a chronic stress load (Lupien et al., 2009). Therefore, repeated exposure to elevated cortisol levels may have particularly harmful influences for young children.

The development of HPA axis system is influenced by the environmental factors such as caregiving and the childcare setting (Gunnar & Quevedo, 2007; Vermeer & Groeneveld, 2017; Vermeer & van IJzendoorn, 2006). Evidence to date suggests that a high quality in early childhood education and care (ECEC) has a positive influence on child cognitive development and later educational outcomes (Loeb et al., 2007). The gains of ECEC are especially beneficial for disadvantaged children and the children from lower socioeconomic families (van Huizen & Plantenga, 2018). Nevertheless, out-of-home childcare has been associated with higher cortisol levels in some children as measured by the diurnal saliva cortisol levels (Vermeer & Groeneveld, 2017). This is shown mostly during the mid-morning and mid-afternoon hours, which may be the most demanding for the youngest children in out-of-home childcare in comparison with the days that they are at home (Drugli et al., 2017; Groeneveld et al., 2010; Talge et al., 2008; Vermeer & van IJzendoorn, 2006; Watamura et al., 2003; Watamura et al., 2009). These results are in line with our recent findings among 2-year-old children indicating that the afternoon cortisol levels were higher in the out-of-home childcare when compared with days spent at home. We also noticed that the total diurnal cortisol production was higher in children cared for at home, who had no experience in out-of-home childcare (Tervahartiala et al., 2019, 2020), which may indicate that different childcare contexts have distinct effects on child stress regulation at various toddlers’ ages.

However, to our knowledge there are no earlier studies that had included a comparison group of children cared for at home by their parents with no experience of out-of-home childcare. The topic is relevant as there are still many children who are not participating in out-of-home childcare. Also, the age at which children start in ECEC varies among countries. The ECEC is usually closely linked to the country's social policy and should be viewed against that. Political decisions usually guide a family's decision making and their possibilities to receive services (OECD, 2020). The ECEC system in Finland is based on the Nordic model that aims to promote democracy and is also an important part of the lifelong learning (Karila, 2012). All the children under school age have a subjective right to participate ECEC regardless of parent's employment status. The ECEC is highly regulated and legislation determines the group sizes and personnel's education qualities in childcare centers. The fees are rather low and free of charge for the lowest income families (Minedu, 2017). In addition, there is rather long home-care allowance that entitles the parent to stay home until the child is 3 years old (OECD, 2020). That is, the social selection for different childcare options in Finland is probably not as high as in countries, where the ECEC fees are higher or the home-care allowance is shorter in duration.

Moreover, only a few studies have followed up the children longitudinally and examined the associations between childcare contexts and stress regulation at multiple age points. The prior findings of Watamura et al. (2003) suggest that child cortisol levels in out-of-home childcare are highest during toddlerhood and then decrease along with child age. The explanation may reside in the peer relations, which emerge more prominently during the toddler period (Watamura et al., 2004). Prior findings indicate that toddlers, who managed to play more frequently and in a more complex manner with other children, had lower diurnal cortisol levels than those who were less involved in play during the out-of-home childcare day (Watamura et al., 2003). The prospective study of Ouellet-Morin et al. (2010) showed that 2-year-old children had a flat diurnal cortisol pattern in out-of-home childcare when compared with their decreasing cortisol levels at home. When the same children were 3 years old, they showed a decreasing pattern both at home and in out-of-home childcare. The results indicate that the differences in the patterns of cortisol secretion at home and in out-of-home childcare in toddlers are transient and diminish as the children develop (Ouellet-Morin et al., 2010). In addition, total diurnal cortisol production during the day typically decreases as the children develop (Simons et al., 2015). Most of the studies have not observed differences between the sex in cortisol production in out-of-home childcare (Vermeer & van IJzendoorn, 2006). The exception is the study of Ouellet-Morin et al. (2010) suggesting higher cortisol levels in boys than in girls attending out-of-home childcare.

In addition to age, a child's individual characteristics, such as temperament, are shown to play a role in early childhood stress regulation (Geoffroy et al., 2006; Phillips et al., 2011; Watamura et al., 2004). Temperament is defined as individual differences in reactivity and self-regulation that have a biological basis but are also influenced by the early childhood social environment, maturation and parenting practices (Rothbart & Bates, 2006; Slagt et al., 2016). Reactivity aspects of temperament characterize one's emotional, motor and attention reactions, intensity and recovery from reactions. Self-regulation, in turn, refers to the ability to regulate this reactivity. According to Mary Rothbart's theory, temperament consists of three main factors: surgency/extroversion, negative affectivity, and effortful control. Surgency/extroversion includes positive anticipation, activation level and sensation seeking, whereas negative affectivity includes fear, anger-frustration, sadness, and discomfort. Effortful control refers to self-regulation, which modulates and regulates this reactivity (Rothbart, 2011).

Prior studies suggest that temperament may be directly related to diurnal cortisol output, but also moderates stress responses in an out-of-home childcare context (Phillips et al., 2011). This is plausible, as both the HPA axis functioning and the child's temperament change dynamically across the early childhood years and are influenced by environmental factors (Badanes et al., 2012; Gunnar & Quevedo, 2007; Vermeer & Groeneveld, 2017). For instance, children lower in effortful control, an aspect of emerging self-regulation, presented with higher total cortisol production during their daily activities when compared with children with better regulatory capacities. This was observed particularly amongst younger toddlers, and the authors hypothesized that the association could result from the daily conflicts and challenges that children with low effortful control encounter (Watamura et al., 2004).

Furthermore, earlier research suggests that young children with fearful temperament, an aspect of negative reactivity, have a larger cortisol increase in stressful situations (O'Connor et al., 2017; Talge et al., 2008). Similarly, a higher level of overall negative reactivity in infants (Albers et al., 2016) as well as in preschoolers (Dettling et al., 2000; Watamura et al., 2002) was associated with higher cortisol levels in out-of-home childcare. Fearfulness and behavioral inhibition are the dimensions of temperament which have been associated with cortisol increases in particular with strange and novel situations (Gunnar & Donzella, 2002).

However, in our previous study on 2-year-old children, we did not find any associations between negative affectivity and cortisol output in the out-of-home childcare context (Tervahartiala et al., 2020). One possible explanation for the distinct results may derive from the different ECEC programs and contexts, which are not fully comparable between different countries (OECD, 2020).

Finally, temperamental surgency/extroversion, an aspect of positive reactivity, may play a role in early childhood stress regulation. Our earlier results indicate that toddlers higher in surgency presented with higher total diurnal cortisol production regardless of the childcare context (Tervahartiala et al., 2020). In addition, higher surgency has been associated with greater cortisol output during a school transition and adaptation to a new school level in preschoolers (Turner-Cobb et al., 2008). These associations could derive from the higher physiological reactivity to environmental stimuli, as the children higher in surgency have also shown higher stress reactivity in an experiment with a competitive challenge (Donzella et al., 2000). Prior research also indicates that children high in surgency with aggressive behavior and low effortful control have presented with higher cortisol levels in preschool environment (Gunnar et al., 2003). Hence, good ability to control emotions may play an important role in stress regulation in particular in out-of-home childcare (Dettling et al., 1999, 2000).

However, very few studies have examined the role of temperament as a contributor to stress regulation in longitudinal settings, where the same children are followed at several age points during early childhood. This would be specifically relevant from the perspective of the findings, where the child age is associated with the cortisol production in out-of-home childcare (Ouellet-Morin et al., 2010; Watamura et al., 2003). Child age is also a plausible moderator in the association between temperament and cortisol production across the childcare contexts. More research is needed to understand the periods of sensitivity and the development in early childhood stress regulation in different childcare settings.

This study builds on our earlier investigations conducted in the same cohort population at the child age of 2 years. In the current study, we aimed to extend our past findings by following up the population up to the age of 3.5 years and investigating whether the associations between childcare group, temperament, and diurnal cortisol change as the children develop.

The assessment of cortisol in saliva is a widely used method in psychoneuroendocrinological research. Due to its noninvasiveness, it is a very useful method particularly for young children (Kirschbaum & Hellhammer, 1994). The diurnal cortisol output can be modeled, for instance, by counting the area under the curve (AUC) of total cortisol production during a day that characterizes the overall activity of the HPA axis, but discards information about the diurnal variation (Saridjan et al., 2014). The diurnal cortisol profiles, in turn, characterize a slope of the cortisol decline over the day and the values of a specific time frame within a day (Adam & Kumari, 2009; Rotenberg et al., 2012).

The first goal of this study was to compare total diurnal cortisol production between children in out-of-home childcare and children in at-home parental care at 3.5 years of age. Second, we aimed at comparing both total diurnal cortisol production and afternoon cortisol levels between the measurement days, Sunday and Monday, within both childcare groups at the child's age of 3.5 years. We expected that there would be no difference in total diurnal cortisol production between the measurement days. Based on an earlier study (Watamura et al., 2003), we also hypothesized that there would be no difference in afternoon cortisol levels between the home day and childcare day in the out-of-home childcare group at the age of 3.5 years. Third, our goal was to examine whether child temperament characteristics would be associated with total diurnal cortisol production in the whole study population. Based on our earlier findings (Tervahartiala et al., 2020), we hypothesized that a higher level of the temperament trait surgency is associated with higher total diurnal cortisol production. This was based on the assumption that children higher in surgency are physiologically more reactive, and surgency would play a general role in early childhood stress regulation (Kabbaj et al., 2000). Child negative affectivity and effortful control were not expected to be associated with cortisol production.

This research was primarily based on the cross-sectional study design at the child age of 3.5 years. Additionally, a post hoc analyses of the age dependency were intended for the findings that would differ from the earlier 2-year-old measurements.

2 METHOD 2.1 Participants

The participants belonged to the larger FinnBrain Birth Cohort Study (N = 3808), which is a population-based pregnancy cohort with aims to identify biomarkers related to prenatal stress and early life stress exposure as well as trajectories for common psychiatric and somatic illnesses. Recruitment took place during the first ultrasound visit during gestational week (gwk) 12 by research nurses in Southwest Finland and the Åland Islands. According to the study inclusion criteria, families with a sufficient knowledge of Finnish or Swedish and with a normal ultrasound screening result were enrolled to the study (Karlsson et al., 2018).

Research recruitment for this sub-study was carried out through personal contact by the research personnel between April 2014 and July 2017. The recruitment process is illustrated in Figure 1. When the current study began, more than half of the FinnBrain Birth Cohort children had reached the age of 2 years and were no longer eligible to participate. Hence, a total of (N = 1881) families, whose child was at an appropriate age, were approached by e-mail and provided preliminary information about the childcare sub-study. Altogether 616 families either responded to the e-mail, or they were personally contacted by the research team in order to assess their eligibility and interest to participate in the study. Out of those who had been contacted, 79 refused to participate and 318 did not meet the inclusion criteria of the current study. Children who attended either out-of-home, center-based childcare, or were cared for at home were eligible to the study. Children attending other forms of childcare (e.g., family-based childcare, which is childcare operated in small groups in the caregiver's own home or 24-h center-based, out-of-home childcare services) were excluded from this study. Family-based childcare was excluded because the context and group setting are different than in center-based out-of-home childcare. The 24-h childcare service was excluded because the daily hours spent in childcare varied, and it would have been difficult to follow the study protocol. Part-time childcare was also excluded, if the child attended childcare only few hours per day or few days per week. Some of the children were just in transition from parental home care to out-of-home childcare at that specific age and were thus not recruited. The child was also not eligible to participate in the study if the family had moved, and no longer lived in the research area.

Flowchart of the recruitment process

Finally, a total of 219 children were eligible and recruited to the study. Altogether, two children quitted the study during the first sample collection. They did not want to participate in the following measurement points either. A total of four children were excluded from the first measurement point because their sample taking failed or they had illness or medication that possibly affected the quality of the cortisol samples. The final sample in the first phase of the study at the age of 2 years (M = 2.13, SD = 0.6) consisted of 213 children of which 106 belonged to the out-of-home childcare group, and 107 children were cared for at home.

2.1.1 Second phase of the study at the age of 3.5 years

When the same children approached the age of 3.5 years, a total of 217 families who participated in the first phase were contacted again and asked to participate in the second phase of the study. Family situations usually change a lot at this age, and the families with a similar childcare arrangement than at the first phase of the study were eligible to participate in the second phase of the study. The final sample in the second phase of the study at the age of 3.5 years (M = 3.59, SD = 0.1) consisted of 111 children.

Specifically, from the out-of-home childcare group, 109 originally recruited families were contacted. The final sample of the children participating in out-of-home childcare consisted of N = 84 children because 13 families could not participate in the second phase of the study, three children were no longer attending out-of-home childcare, and nine families had moved to another place and were no longer able to participate. Children from a total of 32 childcare centers participated. The average group size in the childcare centers was 18.29 (SD = 3.8) children. The participants were not clustered in particular centers, as most of the children participated in different childcare centers or in different groups within the childcare centers. The ECEC is highly regulated in Finland by the government, and children follow a similar schedule and curriculum in each individual childcare center.

From the at-home parental care group, many children, who were cared for at home in the first phase of the study, had started to participate in out-of-home childcare at the age of 3.5 years old and were thus no longer eligible to the study. A total of 108 originally recruited families, who participated in the first phase of the study, were contacted again. Altogether N = 27 children were still cared for at home and were able to participate in the second phase of the study. A total of 70 children had started to attend out-of-home childcare, 10 families were not interested in participating, and one family had moved to another place. One child had attended for a short time in out-of-home childcare between measurement points but returned back to home care and was thus eligible for the study. Most children were cared for at home by a parent (N = 25) and a small minority by another relative (N = 2).

All the study participants gave their written informed consent, and parents gave consent on behalf of their child. This study also meets the ethical guidelines and has been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. The Ethics Committee of the Hospital District of Southwest Finland approved “The FinnBrain Birth Cohort Study” on 14.6.2011 with the protocol number: ETMK: 57/180/2011. This research entitled The Quality of Day Care and the Risk of the Social Exclusion in Early Childhood was approved by The Ethics Committee of the Hospital District of Southwest Finland on 26.11.2013 with the protocol number: ETMK: 137/1801/2013.

2.2 Measures 2.2.1 Diurnal cortisol collection and sample storage

Saliva samples from each child at both age points were collected over 2 days, with four samples during each day being in the morning 30 min after waking, at 10 a.m., between 2 and 3 p.m., and in the evening before sleep. The first day of collection was Sunday, when all the children were at home. The second day of collection was Monday, when the children in the out-of-home childcare group were attending childcare, and the children in at-home parental care spent their weekday at home. For eight children in the first measurement point and for five children in the second measurement point, the samples were not taken on Monday because the children did not attend childcare on Mondays. However, the samples were collected in the childcare center immediately after the day off.

The parents collected saliva samples at home, and childcare personnel collected samples in the childcare center. The research nurse taught the parents and the childcare personnel to take the samples. In addition, parents and childcare personnel were given written information about the sample collection and a teaching video. The saliva samples were collected using Salimetrics© infant swabs (Stratech, Suffolk, UK) by keeping the polymer swab in the child's mouth for 2 min during the collection. Parents and childcare personnel were advised to avoid having the children do physical activity for 30 min and eating for 15 min before sampling.

Saliva samples were placed in the swab storage tubes and kept in a refrigerator from 2 to 5 days between sample taking and delivery to the research center. After delivery, the saliva samples were immediately centrifuged (4°C, 15 min, 1800 × g) and frozen at −70°C. The samples were analyzed by the Finnish Institute of Occupational Health research laboratory in Helsinki, Finland, which regularly participates in the international quality control. The free cortisol in saliva was analyzed using a Cortisol saliva luminescence immunoassay (RE62111; IBL International, Germany). The linear reportable range of the assay was 0.414–88.32 nmol/L. The coefficient of the variation for the intra- and interassay of the method was 5% and 8%, respectively.

2.2.2 Questionnaires

The background data of the mothers (i.e., age, education, and origin) were determined from the cohort research questionnaires during the pregnancy and the Medical Birth Register of the Finnish National Institute for Health and Welfare.

Child temperament was evaluated at the age of 2 years by the mothers using the early childhood behavior questionnaire (Putnam et al., 2006). The questionnaire contains 107 questions with a seven-point Likert-style scale and comprises three main factors of temperament being negative affectivity, surgency/extroversion (reflecting temperamental reactivity), and effortful control (reflecting emerging regulation). Internal consistency scores of the factors in the present sample at the age of 2 years were as follows with negative affectivity having a Cronbach's α = .914, surgency/extroversion with α = .832, and effortful control with α = .876, and at the age of 3.5 years with negative affectivity being α = .889, surgency/extroversion with α = .839, and effortful control with α = .858. For all scales, higher scores reflected higher levels of the particular temperament characteristic in question.

2.3 Data analysis

Area under the curve with respect to ground (AUCG) was used as the measure of total diurnal saliva cortisol (Pruessner et al., 2003). The formulation of AUCG for this study was equal to the method used and described in detail in Tervahartiala et al. (2020). Briefly, the AUCG values were calculated for the time interval 0.5–12 h since awakening, and they were based on the log-transformed cortisol values (see Figure 2(b)). For the children whose first measurement was not made exactly 0.5 h after awakening, the 0.5 h cortisol value was estimated using LOESS regression (Figure 2(a)), and for the children whose last measurement was made before the 12-h limit, the last line in the cortisol curve approximation was linearly continued to the 12-h limit (Figure 2(b)). Missing cortisol values were treated with multiple imputation (100 imputed datasets).

Illustrations of how the 30-min cortisol values were estimated and how the AUCG was defined. (a) Illustration of how the predicted 30-min cortisol values were estimated to achieve a comparable starting point for every individual. The solid black line is the estimated LOESS curve representing the average cortisol curve during the first hours after wake-up. The 30-min cortisol estimation is shown for two examples with the original observations within the red circles, and the corresponding predicted 30-min cortisol values marked by a red star. (b) Definition of AUCG. The red line represents the estimated cortisol curve for a child, whose last saliva sample was taken before the 12-h time period had been reached

As there were two AUCG values (i.e., the Sunday and Monday values) for each child, each study question regarding the total cortisol production was analyzed using a multilevel model with a random intercept for each child. The fixed effects of the models varied by study question and were as follows:  Study question 1: Difference in total diurnal cortisol production between children in out-of-home childcare and children in at-home parental care groups. ∘ Model 1: AUCG = Group + Day + Sex + Age + Education  Study question 2: Difference in total diurnal cortisol production between the measurement days (i.e., Sunday and Monday) separately in each group. ∘ Model 2: AUCG = Day + Sex + Age + Education  Study question 3: Association between each temperament trait and total diurnal cortisol production. ∘ Model 3: AUCG = Temperament + Group + Day + Sex + Age + Education

The predictor variables in the models were: Group (out-of-home childcare or at-home parental care), measurement Day (Sunday or Monday), child Sex (boy or girl), Age (years) and Temperament (surgency/extroversion, negative affectivity, or effortful control measured at 2 years of age), and maternal Education (high school/vocational education, applied university or university degree). To make Models 1 and 3, equal to what was used in our previous study (Tervahartiala et al., 2020), the variances of the random intercepts were not assumed equal between the childcare groups.

All the analyses were first performed on each imputed dataset, and the final results were then obtained by pooling all the results using Rubin's rules (Rubin, 1987). All the results are reported in log(nmol)/l × h units in the Results section.

Description of the analysis of the difference in the afternoon cortisol levels between the days in each group, in Study question 2, is given in the Appendix, as the analysis method differs from the above methods.

Furthermore, description of the post hoc analyses regarding the total diurnal cortisol production, that is, the age dependency of the group difference and the age dependency of the surgency/extroversion association are also given in the Appendix.

All statistical analyses were performed in R 3.6.3 (R Core Team, 2018) with the packages mice (van Buuren & Groothuis-Oudshoorn, 2011) for multiple imputation and nlme (Pinheiro et al., 2018) for fitting the multilevel models.

3 RESULTS 3.1 Demographic characteristics of the participants

Sample characteristics, for both the prior study conducted at the child's age of 2 years and for the present study at the child's age of 3.5 years, are presented in Table 1. Participants were ethnically Caucasian, and the mother's language and origin were primarily Finnish at 97.1% and 98.5%, respectively. The mother's age at childbirth was on average 31.5 (SD = 4.3) years in both childcare groups. Maternal education was higher at the child's age of 2 years in the out-of-home childcare group, while the difference between the groups were no longer observed at the child's age of 3.5 years. Furthermore, the overall educational level was rather high in both groups, as about half of the mothers had a university degree. The proportion of boys and girls did not differ between the childcare groups. The mean age of the children was higher in the first phase of the study in the out-of-home childcare group than in the at-home parental care group, while no age difference between the groups was noted in the second phase of the study. Finally, at either age point, no group differences in the levels of temperament traits were observed.

TABLE 1. Demographic characteristics of the participants Out-of-home childcare At-home parental care Total sample p value 2 yearsa Sample N 106 107 213 Child age (years), mean (SD) 2.26 (0.6) 2.00 (0.5) 2.13 (0.6) .001 Child sex (boys), N (%) 63 (59.4%) 53 (49.5%) 116 (54.5%) .147 Child temperament, mean (SD)b Surgency/extroversion 5.1 (0.6) 5.1 (0.6) 5.1 (0.6) .962 Negative affectivity 2.9 (0.6) 3.0 (0.6) 2.9 (0.6) .280 Effortful control 5.0 (0.6) 5.0 (0.5) 5.0 (0.6) .660 Maternal education, N (%) High school/vocational education 16 (15.1%) 29 (27.1%) 45 (21.1%) .048 Applied university 31 (29.2%) 34 (31.8%) 65 (30.5%) University degree 59 (55.7%) 44 (41.1%) 103 (48.4%) 3.5 years Sample N 84 27 111 Child age (years), mean (SD) 3.60 (0.1) 3.56 (0.1) 3.59 (0.1) .057 Child sex (boys), N (%) 48 (57.1%) 13 (48.1%) 61 (55%) .414 Child temperament, mean (SD)c Surgency/extraversion 5.1 (0.6) 5.0 (0.7) 5.1 (0.6) .521 Negative affectivity 2.8 (0.6) 3.1 (0.5) 2.9 (0.6) .097 Effortful control 5.1 (0.6) 5.0 (0.5) 5.0 (0.5) .550 Maternal education, N (%) High school/Vocational education 12 (14.3%) 9 (33.3%) 21 (18.9%) .073 Applied university 22 (26.2%) 7 (25.9%) 29 (26.1%) University degree 50 (59.5%) 11 (40.7%) 61 (55.0%) p values based on t-test for age, child temperament, and χ2 test for gender and education.

Descriptive statistics of diurnal cortisol values for both the prior study conducted at the child's age of 2 years and for the present study at the child's age of 3.5 years are presented in Table 2.

TABLE 2. Descriptive statistics of diurnal cortisol values (nmol/L) 2 yearsa 3.5 years N Median (interquartile range) Cortisol AUCGM (SD) N Median (interquartile range) Cortisol AUCGM (SD) Out-of-home childcare Sunday 106 5.71 (3.83) 84 5.02 (3.42) Morning 30-min after waking up 93 9.71 (5.61–13.14) 75 8.04 (5.23–13.53) At 10 a.m. 96 3.00 (2.22–4.81) 77 3.03 (2.15–5.34) At 2–3 p.m. 96 2.69 (1.87–4.17) 81 2.53 (1.63–4.61) Evening before sleep 93 1.01 (0.67–1.98) 79 0.82 (0.52–1.53) Monday 106 6.28 (2.36) 84 5.64 (2.53) Morning 30-min after waking up 97 8.76 (5.87–12.84) 79 8.77 (6.00–11.95) At 10 a.m. 99 3.04 (2.28–4.11) 82 2.89 (2.10–3.90) At 2–3 p.m. 91 4.15 (2.43–6.98)