https://orcid.org/0000-0002-2043-8566
Abstract
Impaired sleep following trauma such as childhood maltreatment is both a prognostic factor for future mental illness and a feasible intervention point. Yet, associations between childhood maltreatment and objectively measured sleep components are rarely found. New approaches advance the use of multidimensional sleep health scores instead of individual sleep components. However, no such methodology has been used to study the consequences of maltreatment on sleep health in adolescent cohorts so far. We hypothesized that childhood maltreatment will be associated with poorer sleep health in adolescence.
A cross-sectional sample of 494 adolescents at high risk of emotional and behavioral problems (mean age 17.9) completed the Childhood Trauma Questionnaire—Short Form to assess five forms of maltreatment (emotional and physical abuse/neglect and sexual abuse) assessed as continuous sum scores. During nine nights of actigraphy and sleep diary measurements, data on sleep regularity, satisfaction, alertness, timing, efficiency, and duration were collected, which were combined into a sleep health composite score ranging from 0 to 6. Linear regression models were adjusted for age, sex, household income, ethnic origin, educational level, urbanization of living environment, and parental psychopathological problems.
Associations were found between all forms of maltreatment and poorer sleep health (p < .031), except for sexual abuse (p = .224). Partial r effect sizes ranged from −0.12 (95% CI = −0.22 to −0.01) for emotional neglect to −0.18 (−0.28 to −0.08) for total maltreatment.
Maltreatment was associated with impairment in everyday sleep health, reflected in both subjective and objective measurements of sleep.
People with a history of childhood maltreatment often report experiencing sleep problems. In epidemiological research, however, that association is hard to detect using objective methods of measuring sleep. This could be due to researchers focusing on individual parts of sleep rather than measuring sleep health as a whole. We added subjective (sleep diary) and objective (actigraphy) measures of sleep in a combined score to show that maltreatment does associate with worse sleep health in adolescents. The association was detected for both objective and subjective components of sleep.
Introduction
Sleep problems, both transient and persistent, are a hallmark consequence of overwhelming or chronic stress [1, 2]. The link between sleep and stress has been a focus of research for two broad reasons. First, sleep disturbances following stress may serve as early indicators of emerging mental illness [3]. For instance, disrupted sleep patterns are a significant early warning sign of developing post-traumatic stress disorder [1]. Second, beyond its prognostic value, poor sleep quality has been identified as a potential causal factor leading to psychopathology [4, 5]. Consequently, the empirical literature has explored the role of sleep as a mediator in the relationship between psychological stress and subsequent mental health issues, including depression and anxiety [6, 7]. A large body of literature describes childhood maltreatment as a particularly damaging form of stressful experience for both adolescent and adult lifelong health outcomes [8]. Sleep problems in adolescence set up the conditions for continued adverse outcomes in adulthood by affecting academic achievement, somatic and psychological health, and interpersonal relations [9, 10]. Different types of abuse (i.e. emotional, physical, sexual) and neglect (emotional and physical) have been linked with adult sleep disturbances [11]. In adolescents, however, the focus has been much more on interpersonal types of maltreatment, namely physical and sexual abuse [12]. The gap in studying emotional forms of maltreatment and neglect is unfortunate, as sleep in adolescents is reportedly more susceptible to the effect of stress than that of adults [13].
It has previously been reported that all types of maltreatment, including emotional abuse, can result in lower self-reported sleep quality and sleep disturbances in Chinese adolescents [14, 15]. In Canadian adolescents, emotional abuse was also associated with delayed sleep onset and less time asleep [16]. Comparable correlations for all abuse subtypes and self-reported sleep disturbances were also reported in a small longitudinal study of American adolescents as well [17]. There is thus some literature to suggest the associations between childhood maltreatment and sleep are not specific to type of maltreatment. However, it is not clear how that relates to overall sleep health in adolescents, especially if sleep is not self-reported. Following maltreatment, the resulting sleep problems in adolescence can set up the conditions for continued adverse outcomes in adulthood by affecting academic achievement, somatic and psychological health, and interpersonal relations [9, 10]. As such, investigating how childhood maltreatment of wide typology relates to sleep in adolescence is a topic of ongoing importance in developmental research.
The measurement of sleep imposes difficult methodological choices. One approach has found positive associations between impaired sleep and childhood maltreatment by focusing on individual components of sleep in adolescents. Those components include time it takes to fall asleep, termed sleep onset latency [16], waking up during the night [18], susceptibility to nightmares [19], and sleep efficiency (i.e. proportion of time spent asleep in bed) [20] among others which are all reported to be worse following maltreatment. Notably, associations emerge more often and with a bigger effect size when sleep components are studied using subjective self-reports as opposed to objective actigraphy measurements for instance [12]. There are, however, considerable conceptual limitations to this approach. Namely, problems with sleep captured by the individual components are almost never independent but instead co-occur with each other. Not sleeping enough very often overlaps with a late time in bed, for example [21], whereas worse subjectively reported sleep quality coincides with variability of sleep onset [22]. Indeed, in modern sleep research, these co-occurrences are not treated as statistical nuisances to overcome but are instead recognized as meaningful patterns that indicate sleep should be treated as a multidimensional continuum [23]. During the transition from childhood to adolescence, changes in sleep occur rapidly in terms of later time in bed, less sleep time, and more daytime sleepiness among other changes [24, 25]. It is unlikely that maltreatment would affect any of these changing dimensions in isolation, as opposed to acting holistically on overall sleep habits like the onset of adolescence does. Thus, an alternative approach would more appropriately measure sleep in a multidimensional manner that incorporates in one score various domains of sleep health. The RU SATED (Regularity, Satisfaction, Alertness, Timing, Efficiency, and Duration) model promotes this multidimensional approach by proposing a composite score formed by components judged to be most salient for normal sleep health [26]. It has been used to robustly link sleep to physical health [27], depression [28], and mortality in older adults [29] among other health outcomes. An additional benefit of this approach is that both subjective and objective measurements can be combined to get a nuanced understanding of how maltreatment affects sleep health [30]. This method has successfully been implemented in a sample of adults, demonstrating that overall lifetime trauma and, in particular, childhood trauma were associated with sleep health as measured by actigraphy and self-report diaries [31]. As of yet, no such investigation has been conducted in adolescent populations.
The current study investigated the association between childhood maltreatment—categorized as emotional, physical, or sexual abuse, and emotional or physical neglect—and sleep health in a sample of adolescents at high risk of developing emotional and behavioral problems. Sleep health was measured as a multidimensional score composed of self-reported and actigraphy metrics. We hypothesized that more maltreatment of any type will be associated with poorer sleep health.
Methods
The participants of this study were adolescents in the first follow-up measurement of the iBerry (Investigating Behavioral and Emotional Risk in Rotterdam Youth) cohort [32]. The cohort was selected from 16 758 adolescents in their first year of high school (aged 13) in the Greater Rotterdam Area. Screening for participation as part of general preventive youth care was conducted using the self-report Strengths and Difficulties questionnaire [33]. Adolescents with the highest 15% problem scores and a random sample from the lowest 85% problem scores were selected. This selection resulted in a cohort comprising 1022 adolescents, with a 2.5:1 ratio of those at high risk of developing emotional and behavioral problems, who participated in the baseline measurement. This sampling strategy was validated by confirming high-risk adolescents were at higher risk for internalizing and externalizing problems both at age 15 [34] and age 18 [32]. The oversampling of high-risk adolescents confers several advantages. By focusing efforts on including adolescents at high risk of emotional and behavioral problems, more individuals with severe histories of childhood trauma are captured as demonstrated empirically by past work [35]. These are typically the adolescents that are nonresponders in general population studies [36]. As such, the current sample has advantages for ecological validity when estimating associations with childhood maltreatment. Additionally, the larger proportion of adolescents with history of maltreatment allows for higher statistical power and thus better precision of associations. The current cross-sectional analysis uses a subsample from the first follow-up measurement (mean age 18) of 494 adolescents that completed the sleep study of interest for the current research question.
Adolescents were approached to participate in the first follow-up measurement of the iBerry study via a phone call. They were offered the option to also participate in the sleep study starting after their visit to the research center. Every adolescent that consented to participation was included in the sleep study, with no exception criteria based on prior sleep disorders. Information was provided to each adolescent on how to handle the actiwatch device (e.g. not to take it off during showers or sleep) and how to fill out the sleep diary each day. Then, the actiwatch was powered on by the attending research assistant and strapped to the adolescent’s nondominant wrist. A reward of €10 was provided upon return of the actiwatch device. All efforts were made to contact adolescents if they had not returned the device after the sleep study period ended, in order to minimize missing data.
Materials
Childhood maltreatment.
Childhood maltreatment was measured using the Dutch version of the 28-item Childhood Trauma Questionnaire—Short Form (CTQ-SF) [37]. The items inquire about maltreatment experiences during the entire period of childhood and adolescence. Item response categories are scored from 1 to 5, indicating the frequency of maltreatment (ranging from never true to very often true). Five subscales consisting of five items each can be calculated, namely emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect. One item from the sexual abuse scale does not translate well to Dutch (namely “Have you been molested?”). The direct Dutch equivalent carries a different connotation and does not accurately reflect the intended meaning of the original question. It is therefore not used in calculations [38]. Abuse items pertain to abuse perpetrated by anyone, whereas neglect items inquire about experiences in the family. A total maltreatment score can further be calculated by adding all maltreatment subtypes together. There are empirically established cutoffs that indicate severe maltreatment for each subscale. These cutoffs are a score of 16 for emotional abuse, 13 for physical abuse, 13 for sexual abuse (adjusted for the removed item), 12 for emotional neglect, and 18 for physical neglect. McDonald’s omega reliability ranged from 0.85 to 0.96 indicating acceptable reliability, except for the physical neglect subscale with an estimate of 0.60. The CTQ has been validated for use in both Dutch and adolescent populations [38, 39].
Sleep health composite score.
Sleep health was assessed via a combination of six indices derived from objective actigraphy measurements and self-report diary questions. First, actigraphy measurements were collected starting May 2019 and ending February 2022. Of note, some of the nights measured for 36 adolescents coincided with Dutch school holidays. GENEActive accelerometers were worn on the nondominant wrist. The devices can sample activity at a frequency of 50 Hz. Adolescents wore the accelerometer for nine consecutive days and nights, encompassing five weekdays and four weekend days. On average adolescents provided data for 8.6 nights (SD = 0.83). Adolescents who wore the device for less than five nights were excluded from analyses in line with recommendations on sleep research in adolescents [40, 41]. Data from the returned accelerometers were downloaded on a computer and processed using the R package “GGIR” [42]. Before entering the data for processing, naps were manually excluded as indicated by participants in a sleep diary described below. Then, self-reported time in bed and time out of bed in the sleep diaries was further used as a guider in processing the rest intervals. If sleep diary data were missing for a given night, GGIR could instead estimate rest intervals using Heuristic, Data-driven, Cumulative Zero-crossing Analysis. Full specification of the algorithm can be found in the code repository linked in the Statistical analysis section. As a final quality check, visual reports of sleep time were inspected for each participant to ensure the resulting intervals met face validity. This produced metrics on sleep onset, duration, efficiency, and wake-up time. Sleep onset and sleep wake-up time were used to calculate the midpoint of each night. Consequently, the sleep duration, efficiency, and midpoint were used to create four of the six sleep health indices described below.
Alongside the actigraphy accelerometers, adolescents were also supplied with sleep diaries they filled in the day after each night. If diaries were not filled out for more than five nights (weekend or weeknight), the sleep diary data for the participant were coded as missing. The diary inquired about various activities and behaviors throughout the day and, relevant for the current study, about how the adolescent slept the night before. Averages from the sleep diary were used. Namely, three questions inquired directly about the quality of sleep (i.e. “Do you think you slept well last night?”, “Did you feel well rested after getting up?”, and “Do you feel sleep deprived”) with response options “yes” or “no.” To classify a night as “good,” respondents had to indicate that they slept well or felt well rested or did not feel sleep deprived. This formed the fifth sleep index of the composite score. Finally, one question inquired about alertness during the day, namely “Did you feel so sleepy or tired today that it interfered with your activities?” This comprised the sixth and final metric of the composite sleep health score. All sleep measurements were collected following the completion of the childhood maltreatment questionnaires and other self-report instruments. We distinguished between indicators of sleep satisfaction and alertness in this manner based on two considerations. First, we followed prior literature using identical instruments as ours to allow for comparability [43]. Second, we followed recent conceptualizations of sleep satisfaction as expressed feelings about sleep the night before and alertness as daytime dysfunction the following day [44, 45].
The sleep health composite score indices were calculated and combined in a manner analogous to previous empirical literature operationalizing the RU SATED model in adolescents. Indices were dichotomized as “poor” = 0 or “good” = 1 based on previous recommendations or data from healthy adolescents where available [27]. The six indices measured were (1) sleep Satisfaction; (2) Alertness; (3) Timing; (4) Efficiency; (5) Duration; and (6) Regularity. Accordingly, each index was given a numerical value of 1 for good as opposed to 0 for bad in the following manner—for Satisfaction, more than 50% of self-reported sleep quality responses had to be “good” across the nine nights; for Alertness, if more than 50% of the self-reported alertness responses were “good”; for Timing, if the sleep midpoint was on average before 2 am or after 4 am; for Efficiency, the average sleep efficiency was 85% or more; for Duration, the average sleep duration was between 8 and 10 hours; for Regularity, if the standard deviation of the midpoint was less than 1 hour. Summing these indices together produces a sleep health composite score, ranging from 0 to 6. Higher values indicate better overall sleep quality. All calculations were based on both weekend and weeknights, except when stated otherwise in sensitivity analyses.
Confounders.
Demographic data on sex assigned at birth, age, and educational level were collected from the adolescent using self-report questions collected in the same day. Information on ethnic origin was provided by an accompanying parent (i.e. “Dutch” or “non-Dutch”) based on the birth country of the parent. Ethnic groups thus reflected national origin as opposed to shared racial characteristics like skin color or facial features [46]. Likewise, net household monthly income was based on the parent’s income (i.e. “less than €1599,” “€1600–2399,” “€1600–2399,” or “more than €4400”). To obtain the urbanicity of living environment, each adolescent’s home address was categorized as “urban,” “suburban,” or “rural” based on the surrounding number of address per km2 (defined as <1000, 1000–1500, and >1500 addresses/km2, respectively). Parental psychopathological problems were measured using the total score obtained from the self-report Brief Symptoms Inventory [47] filled out by the accompanying parent, of whom 84% were the mothers of the adolescents.
Statistical analysis
A set of linear regression models were used for the current analyses with the six childhood maltreatment scales (i.e. emotional, physical, and sexual abuse; emotional and physical neglect; and total maltreatment) as the main predictors and the sleep health composite score as the main outcome. Each childhood maltreatment predictor was added in a separate regression model resulting in five models. We report on unstandardized beta coefficients and corresponding 95% confidence intervals after adjustments for all confounders, as well as partial r coefficients as a measure of effect size. Additionally, we estimated how much-unmeasured confounding can completely explain the observed associations. To do so, we followed the procedure implemented in the R package “sensemakr” [48]. Briefly, this procedure estimates how many times an unobserved confounder would have to be bigger than a confounder included in the model in order to reduce the association of interest to zero. We chose parental psychopathology as the observed confounder to benchmark unobserved confounding against, as it reflects both genetic and environmental potentially confounding factors. It was also the confounder with strongest correlations with the childhood maltreatment predictors (average Pearson’s r = 0.15) and sleep health outcome (r = 0.10). All analyses were conducted using R version 4.3.1 [49].
Missing cases for covariates amounted to 55 for parental psychopathology, 47 for household income, 27 for ethnic origin, and 18 for education. Actigraphy data were unavailable for 12 adolescents who had less than five valid nights of measurement and were therefore marked as missing. Likewise, 99 adolescents did not contribute sleep diary data. Childhood maltreatment data were missing for 11 further participants. All missing data were handled using multiple imputation (10 imputed data sets with 20 iterations) as implemented in the R package “mice” [50]. Covariates were imputed to reduce the risk of selection bias from estimates, whereas the sleep diary and actigraphy data were imputed to improve precision and thus statistical power [51]. We used psychopathology problem scores from the Youth Self-Report (YSR) scale [52] as auxiliary variables to improve the imputation precision and to compare responders and nonresponders. All other covariates and the childhood maltreatment subscores were also included in the imputation model.
Six sensitivity analyses were conducted. In the first analysis, all five subtypes of childhood maltreatment were entered into the same model to test if any associations are significant over and above the rest. Second, weekend nights were removed and models were refitted, as sleep characteristics are known to differ substantially from weeknights, especially for adolescents [53]. Third, we created two additional sleep health composite scores—one only using the four actigraphy-derived sleep indices (i.e. Regularity, Timing, Efficiency, and Duration) and one using the two sleep diary indices (Satisfaction and Alertness). We then reran the same fully adjusted analyses on the split-up sleep health scores, this time using Poisson regressions to fit distributional assumptions. This sensitivity analysis was conducted to check whether associations were driven by only objective or subjective indices. Fourth, analyses were rerun on the sleep health composite score outcome by dropping one of each six indices. The resulting change in effect size indicates which sleep index was most important for the observed associations with sleep health. In the fifth sensitivity analysis, we used nonimputed sleep diary and actigraphy data only, which resulted in 99 less participants for this final sensitivity analysis consisting of a total 386 participants. In the sixth analysis, we added an interaction term with total emotional and behavioral problems as measured by the YSR, to check if the current sampling strategy moderated the association between maltreatment and sleep, thus limiting generalizability.
All analysis code is available at https://osf.io/hrjz9.
Results
Sample characteristics of the 494 adolescents in the analysis sample are presented in Table 1, stratified by whether any maltreatment was reported. The mean age was 17.9 and 60% of adolescents were female. Most of the sample was of Dutch ethnic origin (81%), of higher household monthly income (52% in the 2400–4399 euro range), and resided in urban living conditions (61%). The average sleep composite score was 2.6 (SD = 1.0), This score suggests that, on average, adolescents met between two and three of the six criteria for good sleep quality. The most often reported category of maltreatment was emotional neglect, where 45% of adolescents were above the “No neglect” cutoff, followed by 33% for emotional abuse. Full maltreatment descriptives are presented in Supplementary Table 1. The sample characteristics of responders in the analysis sample and nonresponders are presented in Supplementary Table 2. Responders were more likely to be younger, female, and in the higher educational level categories. There were no substantial differences in reported maltreatment. Furthermore, two self-report items indicated virtually no differences in sleeping less or more than other people.
Characteristics of the sample, stratified according to whether any maltreatment has been reported
| Total sample n = 494 | No maltreatment* n = 353† | Reported maltreatment* n = 130† | |
|---|---|---|---|
| Age, years | 17.9 (17.4, 18.3) | 17.8 (17.3, 18.2) | 18.0 (17.4, 18.4) |
| Sex | |||
| Male | 199 (40%) | 146 (41%) | 47 (36%) |
| Female | 295 (60%) | 207 (59%) | 83 (64%) |
| Ethnic origin | |||
| Dutch | 376 (81%) | 281 (83%) | 88 (73%) |
| Non-Dutch‡ | 91 (19%) | 56 (17%) | 32 (27%) |
| Household net monthly income, euros | |||
| <€1599 | 40 (8.9%) | 22 (6.8%) | 15 (13%) |
| €1600–2399 | 66 (15%) | 43 (13%) | 20 (18%) |
| €2400–4399 | 234 (52%) | 176 (54%) | 56 (49%) |
| >€4400 | 107 (24%) | 82 (25%) | 23 (20%) |
| Educational level§ | |||
| Special needs secondary education | 9 (1.9%) | 5 (1.5%) | 4 (3.3%) |
| Combined educational level | 30 (6.3%) | 25 (7.3%) | 4 (3.3%) |
| Prevocational secondary education | 115 (24%) | 91 (27%) | 21 (17%) |
| Higher general secondary education | 206 (43%) | 129 (38%) | 72 (59%) |
| Preuniversity education | 116 (24%) | 93 (27%) | 22 (18%) |
| Urbanicity of living environment | |||
| Rural | 100 (20%) | 72 (20%) | 26 (20%) |
| Suburban | 92 (19%) | 65 (18%) | 27 (21%) |
| Urban | 302 (61%) | 216 (61%) | 77 (59%) |
| Parental psychopathology, score | 8.0 (2.0, 11.0) | 6.9 (1.0, 10.0) | 11.3 (4.0, 16.8) |
| Sleep health composite score | 2.6 (2.0, 3.0) | 2.8 (2.0, 3.0) | 2.2 (1.0, 3.0) |
| Total sample n = 494 | No maltreatment* n = 353† | Reported maltreatment* n = 130† | |
|---|---|---|---|
| Age, years | 17.9 (17.4, 18.3) | 17.8 (17.3, 18.2) | 18.0 (17.4, 18.4) |
| Sex | |||
| Male | 199 (40%) | 146 (41%) | 47 (36%) |
| Female | 295 (60%) | 207 (59%) | 83 (64%) |
| Ethnic origin | |||
| Dutch | 376 (81%) | 281 (83%) | 88 (73%) |
| Non-Dutch‡ | 91 (19%) | 56 (17%) | 32 (27%) |
| Household net monthly income, euros | |||
| <€1599 | 40 (8.9%) | 22 (6.8%) | 15 (13%) |
| €1600–2399 | 66 (15%) | 43 (13%) | 20 (18%) |
| €2400–4399 | 234 (52%) | 176 (54%) | 56 (49%) |
| >€4400 | 107 (24%) | 82 (25%) | 23 (20%) |
| Educational level§ | |||
| Special needs secondary education | 9 (1.9%) | 5 (1.5%) | 4 (3.3%) |
| Combined educational level | 30 (6.3%) | 25 (7.3%) | 4 (3.3%) |
| Prevocational secondary education | 115 (24%) | 91 (27%) | 21 (17%) |
| Higher general secondary education | 206 (43%) | 129 (38%) | 72 (59%) |
| Preuniversity education | 116 (24%) | 93 (27%) | 22 (18%) |
| Urbanicity of living environment | |||
| Rural | 100 (20%) | 72 (20%) | 26 (20%) |
| Suburban | 92 (19%) | 65 (18%) | 27 (21%) |
| Urban | 302 (61%) | 216 (61%) | 77 (59%) |
| Parental psychopathology, score | 8.0 (2.0, 11.0) | 6.9 (1.0, 10.0) | 11.3 (4.0, 16.8) |
| Sleep health composite score | 2.6 (2.0, 3.0) | 2.8 (2.0, 3.0) | 2.2 (1.0, 3.0) |
*For descriptive purposes, adolescents were categorized as maltreated if they met the “moderate” cutoff for emotional abuse score (>16), physical abuse (>13), sexual abuse (>13), emotional neglect (>12), or physical neglect (>18). See Supplementary Table 1 for further maltreatment descriptives.
†Mean (IQR); n (%).
‡Category includes Surinamese (n = 19); Asian (12); Dutch Antilles (10); Cape Verdean (7); South America (5); African (3); Moroccan (3); Turkish (2); and Other Western (30).
§Combined educational level refers to participants who have not yet reached the stage of choosing their educational degree. Prevocational, higher general, and preuniversity educational levels in that order increasingly focus away from practical vocational education and toward higher academic education.
Characteristics of the sample, stratified according to whether any maltreatment has been reported
| Total sample n = 494 | No maltreatment* n = 353† | Reported maltreatment* n = 130† | |
|---|---|---|---|
| Age, years | 17.9 (17.4, 18.3) | 17.8 (17.3, 18.2) | 18.0 (17.4, 18.4) |
| Sex | |||
| Male | 199 (40%) | 146 (41%) | 47 (36%) |
| Female | 295 (60%) | 207 (59%) | 83 (64%) |
| Ethnic origin | |||
| Dutch | 376 (81%) | 281 (83%) | 88 (73%) |
| Non-Dutch‡ | 91 (19%) | 56 (17%) | 32 (27%) |
| Household net monthly income, euros | |||
| <€1599 | 40 (8.9%) | 22 (6.8%) | 15 (13%) |
| €1600–2399 | 66 (15%) | 43 (13%) | 20 (18%) |
| €2400–4399 | 234 (52%) | 176 (54%) | 56 (49%) |
| >€4400 | 107 (24%) | 82 (25%) | 23 (20%) |
| Educational level§ | |||
| Special needs secondary education | 9 (1.9%) | 5 (1.5%) | 4 (3.3%) |
| Combined educational level | 30 (6.3%) | 25 (7.3%) | 4 (3.3%) |
| Prevocational secondary education | 115 (24%) | 91 (27%) | 21 (17%) |
| Higher general secondary education | 206 (43%) | 129 (38%) | 72 (59%) |
| Preuniversity education | 116 (24%) | 93 (27%) | 22 (18%) |
| Urbanicity of living environment | |||
| Rural | 100 (20%) | 72 (20%) | 26 (20%) |
| Suburban | 92 (19%) | 65 (18%) | 27 (21%) |
| Urban | 302 (61%) | 216 (61%) | 77 (59%) |
| Parental psychopathology, score | 8.0 (2.0, 11.0) | 6.9 (1.0, 10.0) | 11.3 (4.0, 16.8) |
| Sleep health composite score | 2.6 (2.0, 3.0) | 2.8 (2.0, 3.0) | 2.2 (1.0, 3.0) |
| Total sample n = 494 | No maltreatment* n = 353† | Reported maltreatment* n = 130† | |
|---|---|---|---|
| Age, years | 17.9 (17.4, 18.3) | 17.8 (17.3, 18.2) | 18.0 (17.4, 18.4) |
| Sex | |||
| Male | 199 (40%) | 146 (41%) | 47 (36%) |
| Female | 295 (60%) | 207 (59%) | 83 (64%) |
| Ethnic origin | |||
| Dutch | 376 (81%) | 281 (83%) | 88 (73%) |
| Non-Dutch‡ | 91 (19%) | 56 (17%) | 32 (27%) |
| Household net monthly income, euros | |||
| <€1599 | 40 (8.9%) | 22 (6.8%) | 15 (13%) |
| €1600–2399 | 66 (15%) | 43 (13%) | 20 (18%) |
| €2400–4399 | 234 (52%) | 176 (54%) | 56 (49%) |
| >€4400 | 107 (24%) | 82 (25%) | 23 (20%) |
| Educational level§ | |||
| Special needs secondary education | 9 (1.9%) | 5 (1.5%) | 4 (3.3%) |
| Combined educational level | 30 (6.3%) | 25 (7.3%) | 4 (3.3%) |
| Prevocational secondary education | 115 (24%) | 91 (27%) | 21 (17%) |
| Higher general secondary education | 206 (43%) | 129 (38%) | 72 (59%) |
| Preuniversity education | 116 (24%) | 93 (27%) | 22 (18%) |
| Urbanicity of living environment | |||
| Rural | 100 (20%) | 72 (20%) | 26 (20%) |
| Suburban | 92 (19%) | 65 (18%) | 27 (21%) |
| Urban | 302 (61%) | 216 (61%) | 77 (59%) |
| Parental psychopathology, score | 8.0 (2.0, 11.0) | 6.9 (1.0, 10.0) | 11.3 (4.0, 16.8) |
| Sleep health composite score | 2.6 (2.0, 3.0) | 2.8 (2.0, 3.0) | 2.2 (1.0, 3.0) |
*For descriptive purposes, adolescents were categorized as maltreated if they met the “moderate” cutoff for emotional abuse score (>16), physical abuse (>13), sexual abuse (>13), emotional neglect (>12), or physical neglect (>18). See Supplementary Table 1 for further maltreatment descriptives.
†Mean (IQR); n (%).
‡Category includes Surinamese (n = 19); Asian (12); Dutch Antilles (10); Cape Verdean (7); South America (5); African (3); Moroccan (3); Turkish (2); and Other Western (30).
§Combined educational level refers to participants who have not yet reached the stage of choosing their educational degree. Prevocational, higher general, and preuniversity educational levels in that order increasingly focus away from practical vocational education and toward higher academic education.
Estimates from the main analyses are presented in Table 2. All forms of childhood maltreatment except sexual abuse were significantly negatively associated with the sleep health composite score (all p < .031). The strongest association was with the total maltreatment score combining all subscales of maltreatment (partial r = −0.18 [95% CI −0.28 to −0.08]), followed by emotional abuse (−0.17 [−0.29 to −0.04]), physical abuse (−0.15, [−0.24 to −0.05]), physical neglect (0.14 [−0.24 to −0.04]), emotional neglect (−0.12 [−0.22 to −0.01]), and finally the nonsignificant association with sexual abuse (−0.08 [−0.21 to 0.05]). Of note, the confidence intervals around the association with sexual abuse overlapped considerably with the estimates from all other forms of maltreatment. It was thus not possible to conclude with certainty that sexual abuse had a smaller association with sleep health than other forms of maltreatment. All associations were robust to unmeasured confounding. An unmeasured confounder had to exert 13 times the size of the confounding effect of parental psychopathology in order to reduce the association between physical abuse and sleep health to nonsignificance. It had to be 11 times bigger for physical neglect, 7 times for emotional abuse, and 6 times bigger for emotional neglect. For the total maltreatment score, the unmeasured confounder had to be 12 times bigger than parental psychopathology. None of the associations were significant if the five maltreatment subtypes were entered in the same model (Supplementary Table 3).
Estimates from linear regressions with sleep health composite score as the outcome and six childhood maltreatment scales as the main predictor (n = 494)
| Unstandardized beta [95% CI]* | Partial r [95% CI] | P-value | |
|---|---|---|---|
| Emotional abuse | −0.04 [−0.06 to −0.01] | −0.17 [−0.29 to −0.04] | .009 |
| Physical abuse | −0.08 [−0.14 to −0.03] | −0.15 [−0.24 to −0.05] | .003 |
| Sexual abuse | −0.03 [−0.08 to 0.02] | −0.08 [−0.21 to 0.05] | .224 |
| Emotional neglect | −0.03 [−0.05 to −0.00] | −0.12 [−0.22 to −0.01] | .031 |
| Physical neglect | −0.06 [−0.10 to −0.02] | −0.14 [−0.24 to −0.04] | .005 |
| Maltreatment total | −0.02 [−0.03 to −0.01] | −0.18 [−0.28 to −0.08] | <.001 |
| Unstandardized beta [95% CI]* | Partial r [95% CI] | P-value | |
|---|---|---|---|
| Emotional abuse | −0.04 [−0.06 to −0.01] | −0.17 [−0.29 to −0.04] | .009 |
| Physical abuse | −0.08 [−0.14 to −0.03] | −0.15 [−0.24 to −0.05] | .003 |
| Sexual abuse | −0.03 [−0.08 to 0.02] | −0.08 [−0.21 to 0.05] | .224 |
| Emotional neglect | −0.03 [−0.05 to −0.00] | −0.12 [−0.22 to −0.01] | .031 |
| Physical neglect | −0.06 [−0.10 to −0.02] | −0.14 [−0.24 to −0.04] | .005 |
| Maltreatment total | −0.02 [−0.03 to −0.01] | −0.18 [−0.28 to −0.08] | <.001 |
Higher sleep health scores reflect better sleep, whereas higher maltreatment scores reflect more frequent maltreatment. Each maltreatment estimate was calculated from a separate adjusted regression model.
*Adjusted for age, sex, monthly household income, ethnic origin, educational level, urbanization of living environment, and parental psychopathology.
Estimates from linear regressions with sleep health composite score as the outcome and six childhood maltreatment scales as the main predictor (n = 494)
| Unstandardized beta [95% CI]* | Partial r [95% CI] | P-value | |
|---|---|---|---|
| Emotional abuse | −0.04 [−0.06 to −0.01] | −0.17 [−0.29 to −0.04] | .009 |
| Physical abuse | −0.08 [−0.14 to −0.03] | −0.15 [−0.24 to −0.05] | .003 |
| Sexual abuse | −0.03 [−0.08 to 0.02] | −0.08 [−0.21 to 0.05] | .224 |
| Emotional neglect | −0.03 [−0.05 to −0.00] | −0.12 [−0.22 to −0.01] | .031 |
| Physical neglect | −0.06 [−0.10 to −0.02] | −0.14 [−0.24 to −0.04] | .005 |
| Maltreatment total | −0.02 [−0.03 to −0.01] | −0.18 [−0.28 to −0.08] | <.001 |
| Unstandardized beta [95% CI]* | Partial r [95% CI] | P-value | |
|---|---|---|---|
| Emotional abuse | −0.04 [−0.06 to −0.01] | −0.17 [−0.29 to −0.04] | .009 |
| Physical abuse | −0.08 [−0.14 to −0.03] | −0.15 [−0.24 to −0.05] | .003 |
| Sexual abuse | −0.03 [−0.08 to 0.02] | −0.08 [−0.21 to 0.05] | .224 |
| Emotional neglect | −0.03 [−0.05 to −0.00] | −0.12 [−0.22 to −0.01] | .031 |
| Physical neglect | −0.06 [−0.10 to −0.02] | −0.14 [−0.24 to −0.04] | .005 |
| Maltreatment total | −0.02 [−0.03 to −0.01] | −0.18 [−0.28 to −0.08] | <.001 |
Higher sleep health scores reflect better sleep, whereas higher maltreatment scores reflect more frequent maltreatment. Each maltreatment estimate was calculated from a separate adjusted regression model.
*Adjusted for age, sex, monthly household income, ethnic origin, educational level, urbanization of living environment, and parental psychopathology.
Sensitivity analyses also supported the robustness of the results. Excluding weekend nights from the analysis did not meaningfully change any of the estimates (Supplementary Table 4). Furthermore, all approximations of partial r became smaller, and all associations were nonsignificant after splitting the sleep health score into only objective or only subjective components (Supplementary Table 5). The exceptions were physical abuse and the total maltreatment score, which were significantly associated with the subjective sleep health indices (p = .040 and .039, respectively). However, the observed effect size was smaller. In the next sensitivity analysis, the sleep health composite score was recalculated after dropping one of each of the sleep indices (Supplementary Figure 1). The effect sizes remained similar, indicating no single index completely accounted for the observed association. Excluding the subjective sleep quality index in the sleep health calculation produced the biggest difference in effect sizes, particularly for the association with emotional abuse and physical abuse. In the fifth sensitivity analysis, which did not impute the 99 cases with missing subjective indices score, effect sizes were again closely comparable to the main results (Supplementary Table 6). In the sixth sensitivity analysis, we did not find any evidence for emotional and behavioral problems moderating the association between childhood maltreatment and sleep health (all p < .275, Supplementary Table 7).
Discussion
The current study investigated the association between childhood maltreatment and multidimensional sleep health in a high-risk community sample of adolescents. All types of maltreatment, with the exception of sexual abuse, were associated with worse sleep health. These results were considered robust to unmeasured confounding. Self-report and objective actigraphy components of sleep had to be combined in a single score in order to observe the associations, as childhood maltreatment did not predict just the objective or just the subjective sleep components.
There are key differences in results between the present study and previously published literature on maltreatment and sleep. The most robust associations between childhood maltreatment and sleep have usually been reported in the context of severe sleeping problems. For example, a study on sexually abused adolescents aged 10–18 found that 20%–40% of adolescents suffered from clinical insomnia compared to virtually none in the nonabused control group [54]. A similar 38% difference of severe sleeping problems was reported when comparing children with institutionally confirmed abuse to matched controls at age 5–12 [55]. While these results indicate large effects of maltreatment on sleep disorders, the current study also indicates the consequences of maltreatment could be captured using indicators of everyday sleep health without inquiring about clinical sleep problems. Of relevance, one study used a similar RU SATED outcome to demonstrate that more overall adversity during childhood but not in adulthood is associated with worse sleep health in adults [31]. Likewise, for children, exposure to violence longitudinally predicted later self-reported sleep health measured in early adolescence [56]. Taken together with the current results, the empirical evidence indicates that childhood maltreatment affects overall sleep health in childhood and adolescence, and the impact continues into adulthood.
The vast majority of previous studies on childhood maltreatment and sleep investigated individual sleep components as opposed to overall sleep health [57]. The results emerging from that methodology have been mixed. For instance, some studies in adolescents have reported no association between maltreatment and timing of sleep, but report an association with efficiency [58, 59]. Others have reported the complete opposite in foster children, that is, an association with sleep timing and duration, but not efficiency [60]. Likewise with alertness during the day, some authors report it is associated with maltreatment alongside overall sleep quality [61], yet others report an association with alertness only for the children with the most early life adversities [62]. The evidence base has thus not converged on which components of sleep are most affected by maltreatment, prompting some authors to conclude that sleep duration is not as clinically meaningful as sleep quality, for instance [63]. A more nuanced interpretation, however, might emerge in the context of the present findings. Maltreatment could be considered to affect sleep health in qualitatively different manners across victims of abuse. For some adolescents, impaired sleep might manifest in shortened duration, for instance, for others a shifted sleep rhythm and worse efficiency, and for yet others any possible combination of the sleep metrics studied [64]. By combining the sleep metrics in a multidimensional score and thus allowing different manifestations of impairment to vary between adolescents, between-person differences in sleep health following maltreatment can emerge.
Associations of comparable effect size were found between all types of maltreatment and sleep health, with the exception that the sexual abuse association was nonsignificant. Including all types of abuse in the same model further suggested that none of the maltreatment types individually predicted sleep health. A similar pattern of results has been reported previously for adolescent psychiatric symptoms [65]. This is nevertheless not theoretically accounted for, as different forms of abuse are established to have distinct neurohormonal effects on the developing brain [66, 67]. Psychological accounts also prioritize certain forms of maltreatment as particularly destructive to sleep health. Sexual abuse in particular is thought to compromise safety more so than other forms due to its extreme and taboo nature [68]. Furthermore, childhood sexual abuse is known to often take place in the bedroom of the victim, thus linking time in bed with vulnerability to danger [69]. This line of reasoning is supported by empirical evidence showing childhood sexual abuse can remain associated with impaired sleep up to 50 years later [70]. As a result, many researchers investigating childhood maltreatment and sleep opt to focus exclusively on sexual abuse [12]. It should be noted that while the association with sexual abuse was not significant, the effect size was similar to other significant associations between maltreatment and sleep in the present study. The current results, however, indicate other forms of maltreatment, such as emotional abuse and neglect, can be predictive of adolescent sleep impairments. It could be that by conceptualizing sleep as a multidimensional continuum, the more general consequences of maltreatment could be captured across other domains of abuse. Emotional forms of maltreatment, however, are less often studied in the sleep literature [71, 72] and are sometimes excluded from adversity measurements altogether [73]. Future studies could thus miss important links between maltreatment and general sleep health if their designs are based on published research. It might be justified to expand the prominent study of sexual abuse in the empirical literature and regularly include physical but also emotional forms of abuse and neglect.
Several limitations should be considered in interpreting the results of this study. First, maltreatment was measured using a retrospective self-report instrument. Although this method produces less underreporting of maltreatment than objective records of child abuse, it has nevertheless been described to miss valid cases due to stigma and other psychosocial barriers of disclosing maltreatment [74]. Self-reported maltreatment in general is known to associate more strongly to psychopathology outcomes, indicating that shared informant bias might inflate reported effect sizes [75, 76]. It is unknown whether the current findings generalize to populations with objectively recorded maltreatment. The present study design, on the other hand, subverts the shared informant bias that is prevalent in past empirical literature, namely that the main predictor and outcome are both reported by the same person. By incorporating objective measurements of sleep (effectively an independent informant), we lower the likelihood of spurious associations due to an unmeasured third variable influencing both self-reported maltreatment and self-reported sleep health [77]. This robustness was supported by a sensitivity analysis indicating that the associations were not driven exclusively by the subjective components of the sleep health measure we used. Additionally, robustness analysis on unmeasured confounding revealed that only large unmeasured confounders could completely explain the associations we report on. It should be noted the reliability of the physical neglect measurement was estimated to be low, similar to some previous studies including the original instrument validation [78, 79]. Although the mismeasurement is likely to be random, the lower reliability could have inflated the standard errors and decreased statistical power. As a final point on maltreatment, the prevalences reported in the current sample are somewhat higher than in the Western European general population and more in line with those found in Eastern Europe or Asia [80]. This underscores the fact that the current cohort is strictly not representative of the general population. However, due to the inclusion of lower-risk adolescents as well as high-risk ones, the results can still be generalizable to the general population [81]. As the current cohort includes adolescents of all socioeconomic and ethnic groups, we consider the risk for large unmeasured effect modifiers is low.
In terms of sleep, measurement error remains a concern as both actigraphy and self-report instruments can systematically over- or underestimate sleep [82, 83]. Polysomnography could directly address this source of bias, although that methodology could produce nonrepresentative sleep artifacts as well (e.g. first night effects) [84, 85]. Some of these limitations could be addressed by measuring sleep across two or more periods in an adolescent’s life to obtain a more representative estimate of sleep health [86]. Furthermore, we excluded naps from the current analyses, as they are not directly integrated in the RU SATED model. Naps are nevertheless known to be relevant for adolescent’s sleep and mental health and thus offer an avenue for future research to explore [87]. A last concern to highlight is the possibility of selection bias. No data were collected on diagnoses of sleep disorders, which limits our ability to determine whether the sleep impairments observed are truly subclinical or if they may already be indicative of emerging clinical sleep disturbances. The adolescents who participated in the sleep study differed on some demographic characteristics compared to those who chose not to. They did not, however, differ extensively on their levels of reported maltreatment. Moreover, the adjustment strategy in the current analysis was exhaustive above and beyond past empirical sleep research [12]. Furthermore, in the responder analyses, we found no evidence that those with worse sleep health chose not to take part in the study. On the contrary, the adolescents we report on indicated moderately worse sleep than samples of similar age groups when sleep health was computed in an equivalent manner [27]. Finally, in cross-sectional studies like the current one typically cannot guarantee the temporality of associations under study. This is not the case here, however, as all adolescents first filled out the questionnaires on past childhood maltreatment and only then completed the actigraphy measurements and sleep diaries on current sleep health. Substantial effort has thus been made to reduce bias from multiple sources via statistical and design choices.
Multiple implications can be derived based on the findings of the current study. First, impairments in everyday sleep health following maltreatment can be detected without requiring the diagnosis of clinical sleep disturbances. In children and adolescents, both clinical and subclinical sleep problems can have adverse impact on educational achievements, on physical health (e.g. obesity, immuno-inflammatory response) as well as on psychological problems (e.g. mood, emotional regulation) [88, 89]. This could present an opportunity to prevent deteriorating sleep health and development of psychopathology [90]. Of relevance, there are intervention strategies available for improving sleep health and reducing depressive symptoms [91], which are reportedly more acceptable among adolescents than is seeking help for incipient mental health problems [4]. It should be noted, however, that evidence from a randomized control trial also supports that sleep problems are more resistant to treatment when a history of childhood abuse is present [92]. It should additionally be noted that the effect sizes reported here are rather small in the absolute sense, but typical of the effect sizes reported in behavioral psychiatric research [93]. Therapeutic intervention on factors of similar effect sizes, however, can result in considerable improvements in psychosocial outcomes [94]. It remains for future research to determine how impaired sleep health in maltreated individuals progresses into sleep disturbances across development. This could in turn present new therapeutic challenges in how to address these complex histories of maltreatment [95]. The current study points out that maltreatment of any category can lead to later sleep impairments in multiple dimensions of everyday sleep health.
Supplementary Material
Supplementary material is available at SLEEP online.
Funding
The iBerry Study is funded by the Erasmus University Medical Center and the following institutes of mental health care (GGz): Parnassia Psychiatric Institute Antes, GGz Breburg, GGz Delfland, GGz Westelijk Noord-Brabant, and Yulius. All funding organizations participate in the Epidemiological and Social Psychiatric Research Institute (ESPRi), a consortium of academic and nonacademic research groups.
Disclosure Statement
Financial disclosure: None. Nonfinancial disclosure: None. All procedures were performed in compliance with relevant laws and institutional guidelines and have been approved by the Erasmus MC Medical Ethics Review Committee (MERC) on May 18th, 2021 under number NL76715.078.21. Each participant provided informed consent prior to participating in the current study.
Data Availability
The data underlying this article cannot be shared publicly due to its sensitive nature. The data will be shared on reasonable request to the corresponding author. All code used for the analysis of the data is available at https://osf.io/hrjz9.
Comments