Introduction
Childhood maltreatment (CM) and its detrimental consequences are a major public health concern (Chen, Turiano, Mroczek, & Miller, Reference Chen, Turiano, Mroczek and Miller2016; Gilbert et al., Reference Gilbert, Widom, Browne, Fergusson, Webb and Janson2009; Heim, Shugart, Craighead, & Nemeroff, Reference Heim, Shugart, Craighead and Nemeroff2010; Vos et al., Reference Vos, Flaxman, Naghavi, Lozano, Michaud, Ezzati and Murray2012). Across the world, approximately 35% of children have been exposed to emotional abuse, 23% to physical abuse, 18% to neglect (emotional and/or physical), and 13% to sexual abuse, based on self-report studies that suggest a much higher prevalence than informant-based prevalence rates of around 0.3% (Stoltenborgh, Bakermans-Kranenburg, Alink, & van Ijzendoorn, Reference Stoltenborgh, Bakermans-Kranenburg, Alink and van Ijzendoorn2015). The impact of CM is not constrained to a single health outcome but increases the risk for a diversity of psychiatric disorders (Green et al., Reference Green, McLaughlin, Berglund, Gruber, Sampson, Zaslavsky and Kessler2010), worse treatment outcomes (Cakir, Tasdelen Durak, Ozyildirim, Ince, & Sar, Reference Cakir, Tasdelen Durak, Ozyildirim, Ince and Sar2016; Misiak & Frydecka, Reference Misiak and Frydecka2016), decreased social function (Kilian et al., Reference Kilian, Asmal, Chiliza, Olivier, Phahladira, Scheffler and Emsley2018), frequent hospitalizations (Slotema et al., Reference Slotema, Niemantsverdriet, Blom, van der Gaag, Hoek and Sommer2017), and high risk for suicide (Bernegger et al., Reference Bernegger, Kienesberger, Carlberg, Swoboda, Ludwig, Koller and Schosser2015; Hassan, Stuart, & De Luca, Reference Hassan, Stuart and De Luca2016). Such a broad range of adverse consequences points to large individual differences in responses to CM exposure (Edwards, Holden, Felitti, & Anda, Reference Edwards, Holden, Felitti and Anda2003; Howes, McCutcheon, Owen, & Murray, Reference Howes, McCutcheon, Owen and Murray2017; Nemeroff, Reference Nemeroff2004; Vinkers et al., Reference Vinkers, Kalafateli, Rutten, Kas, Kaminsky, Turner and Boks2015; Whitfield, Dube, Felitti, & Anda, Reference Whitfield, Dube, Felitti and Anda2005) and may suggest diversity in pathways of the negative impact of CM.
A likely contributor to the diversity of outcomes of CM exposure is the nature of CM. CM is defined as one or multiple negative life events occurring before the age of 18 years, but the nature of these events diverge largely, and it is not self-evident they all have the same impact. Whereas there are many potential subdivisions of the overall experience of CM, a broad but relevant division is the distinction between abuse and neglect as they comprise fundamentally different psychological experiences. Abuse is defined as any non-accidental act which causes or creates a substantial risk, physical or emotional injury, and covers a highly threatening event. Abused children are more likely to perceive their harmful environment as dependent on their own behavior than neglected children (Humphreys & Zeanah, Reference Humphreys and Zeanah2015). Neglect is defined as the shortcoming, deliberately or through negligence or inability, to take those actions necessary to provide a child with minimally adequate food, clothing, shelter, medical care, supervision, emotional stability, and growth and deprives the child from basic care and stimulating experiences. Differential effects of abuse and neglect have been reported with respect to brain development (Gauthier, Stollak, Messé, & Aronoff, Reference Gauthier, Stollak, Messé and Aronoff1996), recognition of emotional cues (Pollak, Cicchetti, Hornung, & Reed, Reference Pollak, Cicchetti, Hornung and Reed2000), social-emotional adjustment (Scientific Council on the Developing Child, 2012), the hypothalamic–pituitary–adrenocortical axis (Bruce, Fisher, Pears, & Levine, Reference Bruce, Fisher, Pears and Levine2009), and amygdala and hippocampal volumes (Herzog et al., Reference Herzog, Thome, Demirakca, Koppe, Ende, Lis and Schmahl2020; Teicher et al., Reference Teicher, Anderson, Ohashi, Khan, Mcgreenery, Bolger and Vitaliano2018). For understanding the impact of CM on neural development, a conceptual framework distinguishing threat (i.e. abuse) and deprivation (i.e. neglect) has been suggested previously (Sheridan & McLaughlin, Reference Sheridan and McLaughlin2014). As to be expected when growing up in a harmful environment, childhood abuse and neglect often co-occur (Broekhof, Nordahl, Bjørnelv, & Selvik, Reference Broekhof, Nordahl, Bjørnelv and Selvik2022). Experiencing multiple types of CM is related to an accumulation of detrimental consequences later in life, suggesting a dose–response relationship (Hughes et al., Reference Hughes, Bellis, Hardcastle, Sethi, Butchart, Mikton and Dunne2017; Sala, Goldstein, Wang, & Blanco, Reference Sala, Goldstein, Wang and Blanco2014; Steine et al., Reference Steine, Winje, Krystal, Bjorvatn, Milde, Grønli and Pallesen2017). The experience of both abuse and neglect could therefore be seen as a third and more severe type, that of combined CM. Despite these widely recognized psychological and neurodevelopmental differences between the experience of abuse v. neglect in childhood, no previous studies compared their relative differential impact across psychiatric disorders. Existing research on CM has often taken approaches that disregard comparing potential meaningful subdivisions of CM (Cohodes, Kitt, Baskin-Sommers, & Gee, Reference Cohodes, Kitt, Baskin-Sommers and Gee2021; Gee, Reference Gee2021).
Previous studies investigated either one specific CM type, such as sexual abuse, specific disorders, or did not differentiate between CM types (Lewis, McElroy, Harlaar, & Runyan, Reference Lewis, McElroy, Harlaar and Runyan2016; Zhang et al., Reference Zhang, Lin, Yang, Zhang, Pan, Lu and Liu2020). In these studies, childhood abuse has been related to higher risks of schizophrenia (SCZ) spectrum disorder (Croft et al., Reference Croft, Heron, Teufel, Cannon, Wolke, Thompson and Zammit2019; Heins et al., Reference Heins, Simons, Lataster, Pfeifer, Versmissen, Lardinois and Myin-Germeys2011). Childhood neglect, physical, and emotional abuse were associated with risk for major depressive disorder (MDD) (Betz, Rosen, Salokangas, & Kambeitz, Reference Betz, Rosen, Salokangas and Kambeitz2022; Christ et al., Reference Christ, De Waal, Dekker, van Kuijk, Van Schaik, Kikkert and Messman-Moore2019; Humphreys et al., Reference Humphreys, LeMoult, Wear, Piersiak, Lee and Gotlib2020; Infurna, Reichl, Parzer, Schimmenti, & Bifulco, Reference Infurna, Reichl, Parzer, Schimmenti and Bifulco2016; Martins, Von Werne Baes, De Carvalho Tofoli, & Juruena, Reference Martins, Von Werne Baes, De Carvalho Tofoli and Juruena2014). In bipolar disorder (BD), the limited evidence pointed to a stronger association of childhood abuse with symptom severity than childhood neglect (Etain et al., Reference Etain, Aas, Andreassen, Lorentzen, Dieset, Gard and Henry2013). Studies that compare risks of childhood abuse, childhood neglect, and their combination across MDD, BD, and SCZ are absent, despite their relevance from the perspective that if childhood abuse and neglect comprise different etiological pathways in the development of psychopathology (Heim et al., Reference Heim, Shugart, Craighead and Nemeroff2010), this may be reflected in distinct clinical profiles. Even within disorders, two individuals with the same diagnosis can experience different (core) symptoms (Brunoni, Reference Brunoni2017; Cuthbert, Reference Cuthbert2015; Parker, Reference Parker2006), and therefore examining symptom profiles across diagnosis could provide additional insight. Therefore, this study focuses on the relation between the CM types (i.e. abuse, neglect, and combined CM) and psychopathology later in life; both at disorder level (for MDD, BD, and SCZ) and trans-diagnostically, at the symptom levels. As a strategy to inform of causal directions, we additionally make use of Mendelian randomization (MR) (Davies, Holmes, & Davey Smith, Reference Davies, Holmes and Davey Smith2018). MR is a method in epidemiological observational research that leverages genetic variants as instrumental variables (IVs) to explore the likelihood of causal relationships between an exposure (in this case, CM) and an outcome (psychopathology). The method can help mitigate confounding and reverse causation biases present in observational studies by making use of the fact that most genetic variation is at random in large populations, mimicking randomization of exposure in experimental studies. Recent research identified a genetic signal associated with CM exposure, capturing gene–environment correlations (Warrier et al., Reference Warrier, Kwong, Luo, Dalvie, Croft, Sallis and Cecil2021). This allows exploring causal directions of the relationship between abuse and neglect on mental health outcomes, using bidirectional MR (Smith & Hemani, Reference Smith and Hemani2014).
In summary, the primary objective of this study is to investigate the relative impact of different types of CM, including abuse, neglect, and combined maltreatment, on psychiatric disorders, both at the disorder level (MDD, BD, SCZ) and trans-diagnostically at the symptom level. We aim to shed light on potential distinct clinical profiles associated with these maltreatment types and explore causal directions using MR, taking into account recent genetic findings related to CM exposure. This comprehensive approach offers valuable insights into the complex relationship between CM and later-life psychopathology.
Methods
Study participants
Data from three large longitudinal Dutch cohort studies were used for this study (total N = 4037): the genetic risk and outcome in psychosis study, focusing on SCZ spectrum disorders (GROUP, subsample total N = 981, SCZ cases: 577, MDD: 74, controls: 330) (Korver, Piotr, Boos, Simons, & De Haan, Reference Korver, Piotr, Boos, Simons and De Haan2012), the Dutch Bipolar Cohort focusing on bipolar disorder (DBC, total N = 1453, BD: 1255, controls: 198) (Van Bergen et al., Reference Van Bergen, Verkooijen, Vreeker, Abramovic, Hillegers, Spijker and Boks2019), and the Netherlands Study of Depression and Anxiety (NESDA; total N = 1603, MDD: 1166, BD: 84, controls: 353) (Penninx et al., Reference Penninx, Beekman, Smit, Zitman, Nolen, Spinhoven and Van Dyck2008, Reference Penninx, Eikelenboom, Giltay, van Hemert, Riese, Schoevers and Beekman2021), focusing on depressive disorders and anxiety. The distribution of diagnosis categories (MDD, BD, and SCZ) and controls in the total dataset, and for each cohort study, are displayed in online Supplementary Fig. S1. For the MR analyses, genetic data of 143 473 individuals with self-reported white European ancestry were retrieved from the UK Biobank, a large nation-wide cohort study from the United Kingdom (Bycroft et al., Reference Bycroft, Freeman, Petkova, Band, Elliott, Sharp and Marchini2018; Sudlow et al., Reference Sudlow, Gallacher, Allen, Beral, Burton, Danesh and Collins2015). All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Assessments
Baseline assessments in the three cohort studies included history of current psychiatric disorders, standardized interview for DSM diagnosis, or control status. In GROUP, the Comprehensive Assessment of Symptoms and History (CASH) (Andreasen, Flaum, & Arndt, Reference Andreasen, Flaum and Arndt1992) and Schedules for Clinical Assessment in Neuropsychiatry (SCAN) (Wing et al., Reference Wing, Babor, Brugha, Burke, Cooper, Giel and Sartorius1990) were administered. DBC used the Structural Clinical Interview for DSM-IV (SCID-I) (First, Spitzer, Gibbon, & Williams, Reference First, Spitzer, Gibbon and Williams2002) and Mini-International Neuropsychiatric Interview-plus (MINI-plus) (Sheehan et al., Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs, Weiller and Dunbar1998). NESDA used the Composite Interview Diagnostic Instrument (CIDI) (Robins et al., Reference Robins, Wing, Wittchen, Helzer, Babor, Burke and Towle1988). Inclusion criteria for the combined analysis were diagnoses of a major psychiatric disorder at baseline with available data on symptom level, and on CM. Participants diagnosed with solitary anxiety disorder at baseline were excluded due to lack of data on the symptom level in this group. In NESDA, data from multiple waves were collapsed, and MDD was defined as a lifetime diagnosis of MDD. Participants without a history of mental disorder assessed by either the SCID-I, CIDI, CASH, MINI-plus, or SCAN were included as controls. For assessment of CM under the age of 18 years, all three cohorts administered the Childhood Trauma Questionnaire-Short Form (CTQ) (Bernstein et al., Reference Bernstein, Stein, Newcomb, Walker, Pogge, Ahluvalia and Zule2003; Thombs, Bernstein, Lobbestael, & Arntz, Reference Thombs, Bernstein, Lobbestael and Arntz2009). Total CTQ-scores, subscale scores (physical abuse, emotional neglect, emotional abuse, physical neglect, and sexual abuse), and item-scores were reported in all cohorts. Incomplete CTQ scores (0.13% of items missing at random) were imputed at item-level using the multiple imputation algorithm for maximum likelihood estimation. Outcome ranges of pooled outcomes are presented in the result section. Based on the source publication of the CTQ (Bernstein, Reference Bernstein1998; Bernstein et al., Reference Bernstein, Stein, Newcomb, Walker, Pogge, Ahluvalia and Zule2003; Thombs et al., Reference Thombs, Bernstein, Lobbestael and Arntz2009), abuse was defined as moderate or above scores on only CTQ dimensions emotional abuse (score ⩾ 13), physical abuse (score ⩾ 10), or sexual abuse (score ⩾ 8). Neglect was defined as moderate or above scores for only emotional neglect (score ⩾ 15) or physical neglect (score ⩾ 10). Subjects scoring moderate or above on both neglect and abuse subscales were classified as combined CM. Lastly, subjects with CTQ subscale scores below cutoff for both abuse and neglect were classified as ‘no CM’. Comorbid psychiatric and somatic disorders were either unavailable or inconsistently reported across studies and were therefore not taken into account in the analysis.
Dataset harmonization on symptom level
The lifetime presence or absence of symptoms of SCZ, BD, and MDD was reported for all participants of GROUP and DBC. In NESDA, SCZ cases were not included since they fell outside the scope of the study. Therefore, only the presence or absence of bipolar and depressive disorder symptoms was reported for NESDA participants. Patients with completely lacking symptom data were excluded. For the control group and for GROUP participants assessed with the SCAN, symptom data were not available. To harmonize the available data from the SCID-I, CIDI, and CASH, all symptoms were scored dichotomously, with 0 = symptom not present, and 1 = symptom is present. In the DBC study, symptoms coded as 2 = ‘possibly present’ (in <1% of cases) were recoded into 0 = symptom not present. The SCZ symptoms that were scored on a 5-point Likert scale were recoded into dichotomous variables. Symptoms with an original score of 0–2 were reclassified as 0: not/not fully present, not/moderately severe, or not/moderately bizarre. Symptoms with a score of 3–5 were reclassified as 1: present, severe, or bizarre. All symptoms were kept as detailed as possible, for instance hypersomnia and insomnia were both used instead of being merged into a broader term such as sleep disturbances. Some symptoms, however, needed to be combined due to differences in structure of the SCID-I, CIDI, and CASH. Symptoms combined were, for MDD: weight loss + decreased appetite, weight gain + increased appetite, and for BD: expansive mood + irritable mood, increased activity + agitation. For SCZ, all types of delusions were merged into presence or absence of delusions. The same was applied for hallucinations, merging all types into one variable indicating presence or absence of hallucinations.
Statistical analyses
The data were analyzed using Statistical Package for the Social Sciences (SPSS, v26; SPSS Inc., Chicago, Illinois, USA). χ2 tests were performed for analyzing differences in the distribution between groups. In a multivariate logistic regression model, the contribution of CM types abuse, neglect, and combined CM was modeled as three dichotomous indicators (with no CM as reference) with presence/absence of each diagnosis as the outcome. A significant Wald-test statistic for one of the CM types indicated that the experience of this CM type increased the likelihood of a certain diagnosis, as compared to the reference group of no CM. Subsequently, differences between abuse, neglect, and combined CM were analyzed in turn by adapting the reference categories. Similarly, the contribution of CTQ subtypes emotional abuse (EA), physical abuse (PA), sexual abuse (SA), emotional neglect (EN), and physical neglect (PN) for each diagnosis was estimated in a multivariate logistic regression model. At the symptom level, the contribution of abuse, neglect, and combined CM was estimated in a multivariate logistic regression model (with no CM as reference) for the presence or absence of each symptom as a dichotomous outcome. For all these analyses, the assumptions were verified and met. Logistic regression results are presented as odds ratios (OR) in the result section. Confounder analysis was performed by examining the relation of age, gender, and level of education (none, basic, low, intermediate, and high) as determinants, with CM types and diagnosis as the outcome. For education, educational level was coded by using four dummy variables signifying the difference compared to no education. Variables with a significant association with both diagnosis and CM were added as covariates to the multivariate models. Significance of the differences between the OR of a particular CM type between disorders was estimated with the same multivariate logistic regression models but alternating diagnosis categories as reference category in order to obtain head-to-head comparisons (for instance for the contribution of abuse to MDD as compared to SCZ).
The Relative Excess Risk due to Interaction (RERI) (Knol, van der Tweel, Grobbee, Numans, & Geerlings, Reference Knol, van der Tweel, Grobbee, Numans and Geerlings2007, Reference Knol, VanderWeele, Groenwold, Klungel, Rovers and Grobbee2011) was calcluated to measure the deviation from additivity of the exposure effect on an OR scale (Hosmer & Lemeshow, Reference Hosmer and Lemeshow1992). An RERI < 1 indicates a negative interaction and an RERI = 0 means no interaction or exact additivity. If RERI > 0, an interaction on an additive scale is indicated, meaning that the combined effect of two exposures is larger than the sum of the individual effects of the two exposures.
Mendelian randomization
In MR, IVs are genetic variants that are used as proxies of an exposure (e.g. CM), which can be used to estimate the causal effect on an outcome (e.g. psychopathology). In line with the previous genome-wide association studies (GWAS) study of CM in the UK Biobank (Warrier et al., Reference Warrier, Kwong, Luo, Dalvie, Croft, Sallis and Cecil2021), an IV was derived from the UK Biobank. The UK Biobank participants completed the Childhood Trauma Screener (CTS), a retrospectively reported five-item questionnaire that consists of one question per CM subtype, with answers ranging from 0: never true, to 4: very often true (Grabe et al., Reference Grabe, Schulz, Schmidt, Appel, Driessen, Wingenfeld and Freyberger2012; Warrier et al., Reference Warrier, Kwong, Luo, Dalvie, Croft, Sallis and Cecil2021). Abuse was defined when one or more CTS abuse items were scored >0, neglect when one or more neglect items were scored >0, and combined CM when abuse and neglect were both present. In order to ultimately perform bidirectional, two-sample MR, all available genetic data of UK Biobank participants with self-reported white European ancestry were included for GWAS) for traits abuse, neglect, and combined CM. All genotyped and imputed single nucleotide polymorphisms (SNPs) with a minor allele frequency >0.1%, that did not deviate from Hardy–Weinberg equilibrium (p > 1 × 10−6), had a genotyping rate of 95%, or, for imputed SNPs, had an imputation R 2 > 0.4, were used. Participants who had excessive genetic heterozygosity (i.e. who were >5 s.d. from the means of the first two genetic principal components), whose genetic sex did not match their reported sex or who had a genotyping rate <95%, were excluded. GWAS were conducted for over 15 million SNPs using FastGWA-GLMM (using GCTA version 1.93.2) (Jiang et al., Reference Jiang, Zheng, Qi, Kemper, Wray, Visscher and Yang2019). Sex, year of birth, genotyping batch, and the first 10 genetic principal components were included as covariates.
Two-sample MR (Byrne, Yang, & Wray, Reference Byrne, Yang and Wray2017; Slob & Burgess, Reference Slob and Burgess2020) was performed to assess whether genetic predictors of abuse, neglect, and combined CM are associated with SCZ (Trubetskoy et al., Reference Trubetskoy, Pardiñas, Qi, Panagiotaropoulou, Awasthi, Bigdeli, Bryois and van Os2022), BD (Mullins et al., Reference Mullins, Forstner, O'Connell, Coombes, Coleman, Qiao and Andreassen2021), and MDD (Howard et al., Reference Howard, Adams, Clarke, Hafferty, Gibson, Shirali and McIntosh2019). Putative causal links between CM type and diagnosis were investigated bidirectionally, i.e., whether a genetic predictor of CM type enhances the risk of SCZ, BD, and MDD (forward direction) or whether CM type liability is altered because of liability to SCZ, BD, or MDD (backward direction). To avoid bias in MR due to sample overlap (Burgess, Davies, & Thompson, Reference Burgess, Davies and Thompson2016), UK Biobank was excluded from the GWAS data for the three mental health phenotypes of interest. In the first attempt to extract instruments using independent GWAS loci, the standard p value threshold of p < 5 × 10−8 was used. This p value threshold is applied in the MR backward analyses. In the MR forward analyses, no SNPs were selected at a p value threshold of p < 5 × 10−8, therefore the p value was stepwise increased in order to lower the SNP selection threshold, until at least two SNPs were selected. This led to a threshold of p < 1 × 10−6. As a follow-up analysis, designed to interrogate the influence of threshold variation and in order to gauge potential pleiotropy, the SNP selection threshold was further lowered by increasing the p value until the highest number of SNPs was included without significant evidence of horizontal pleiotropy as measured by the Egger's test (corresponding to the threshold p < 3 × 10−6). Then, to ensure independence between IVs, a strict clumping procedure was applied (LD r 2 < 0.001 within 10 Mb, using the 1000 G EUR as the reference panel). Following that, SNP alleles were harmonized between exposure GWAS and outcome GWAS before running the MR analyses. Each MR analysis was conducted using the following methods: inverse variance-weighted (IVW) MR, which assumes that all SNPs are valid instruments; median-weighted, which provides valid estimates even if up to 50% of the instruments are invalid (Bowden, Davey Smith, Haycock, & Burgess, Reference Bowden, Davey Smith, Haycock and Burgess2016); Q statistic, as an assessment of heterogeneity and first indicator of whether there might be pleiotropy; MR-Egger, which accounts for pleiotropy by including an intercept term in the IVW model (Bowden, Smith, & Burgess, Reference Bowden, Smith and Burgess2015); and MR-PRESSO, which accounts for pleiotropy by detecting and removing outliers (Verbanck, Chen, Neale, & Do, Reference Verbanck, Chen, Neale and Do2018). The mean F statistic was used to quantify instrument strength within the univariable IVW analyses, considering a mean F < 20 as indicative of weak instruments. The Steiger test was used to assess the validity of IVs and confirm the direction of causality (Burgess et al., Reference Burgess, Davey Smith, Davies, Dudbridge, Gill, Glymour and Theodoratou2019). All analyses were performed in R (R Studio Team, 2020), using the packages TwoSampleMR (Hemani et al., Reference Hemani, Zheng, Elsworth, Wade, Haberland, Baird and Haycock2018), MendelianRandomization (Yavorska & Burgess, Reference Yavorska and Burgess2017), and MR-PRESSO (Verbanck et al., Reference Verbanck, Chen, Neale and Do2018).
Results
The analyses comprised 3156 cases and 881 controls. Of total cases, 14.3% was diagnosed with SCZ (N = 577), 30.7% with MDD (N = 1240), and 33.2% with BD (N = 1339). Demographics of the total sample are listed in Table 1.
Table 1. Demographic characteristics of the total sample
MDD, major depressive disorder; BD, bipolar disorder; SCZ, schizophrenia; CM, childhood maltreatment; N, number; M, mean; s.d., standard deviation.
χ2 tests showed that the distribution of CM differed significantly across diagnoses including the controls (χ2[9] = 225.17–226.11, p < 0.001) as well as between patient groups only (χ2[6] = 40.21–40.34, p < 0.001). All types of CM were associated with the presence of MDD, BD, and SCZ in separate non-adjusted logistic regression models (online Supplementary Table S1).
Independent t tests for age showed significant differences between CM groups. Neglect and combined CM groups were significantly older (t = 7.44, p < 0.01 and t = 4.31, p < 0.01) and the abuse group was significantly younger (t = 2.77, p = 0.01) than individuals without CM. Compared to the control group, the MDD and BD diagnosis groups were significantly older (t = 8.25, p < 0.01 and t = 5.22, p < 0.01), whereas participants with SCZ were significantly younger (t = 20.87, p < 0.01). χ2 tests showed that CM groups significantly differed by gender (χ2[3] = 22.85–23.58, p < 0.01), with more women in abuse and combined CM groups and relatively more men in the neglect and no CM groups. The CM groups also differed in level of education (χ2[15] = 69.70–70.77, p < 0.01): participants in the no CM group had higher education in contrast to the neglect and combined group, which included more non- and basic educated individuals. To adjust for these potential confounders, age, gender, and education were added as covariates in all the analyses.
ORs and confidence intervals (CIs) of the relation between CM type and major psychiatric disorder after adjusting the model for age, gender, and education are presented in Fig. 1. Variance inflation factors (VIFs) were <1.5 for each predicting variable, indicating absence of multicollinearity.
Figure 1. Forest plot of the adjusted multivariate logistic regression model of the relation between childhood maltreatment (CM) type (abuse, neglect, combined) and diagnosis (MDD, BD, SCZ) with age, gender, and education level added as covariates and healthy controls as the reference group. Dots represent odds ratios (OR), error bars represent 95% confidence intervals (CI). *Significant with α = 0.05.
Childhood abuse, neglect, and combined CM were related to higher odds of MDD, BD, and SCZ compared to healthy controls. Comparing the relation between CM type and disorder shows that the association of both childhood abuse and neglect with MDD and BD was similar, with largely overlapping CIs, as shown in Fig. 1. The association of childhood abuse with SCZ, however, was significantly stronger than the association of childhood abuse with MDD (p = 0.011) and BD (p = 0.048) (see online Supplementary Table S2).
The RERI between CM types shows a significant additive interaction for combined CM in MDD (RERI = 1.4) and BD (RERI = 1.1) compared to the impact of abuse and neglect alone. This indicates that the combined effect of abuse and neglect is larger than the sum of their individual effects. Combined CM showed no additive interaction in SCZ (RERI = 0.6).
Comparing the impact of abuse, neglect, and their combination within diagnostic category (for instance, analysis whether the contribution of abuse was significantly larger that the contribution of neglect to SCZ risk) highlighted a significantly stronger effect of abuse than of neglect in risk for SCZ, and a disproportionately strong impact of combined CM as compared to abuse and neglect alone for MDD and BD, and compared to neglect for SCZ (online Supplementary Table S3), as also indicated by the RERIs.
The relation of abuse and neglect with symptoms of depression, mania, and psychosis across diagnosis
Logistic regression analyses showed that the presence of symptoms of depression, mania, and psychosis differed between CM types, as presented in Table 2. Childhood abuse was the strongest risk factor for feelings of worthlessness/guilt, suicide attempt, delusions, and hallucinations. Childhood neglect showed no association with symptoms of psychosis, and even an opposite relation to the development of delusions. Combined CM was, consequently, not significantly associated with delusions or hallucinations. Both childhood neglect and combined CM stood out as a significant risk factor for reduced need for sleep. Combined CM increased the risks for the same symptoms as abuse and neglect alone, except for delusions. Furthermore, combined CM increased the odds of depressive mood, retardation, and returning thoughts of death the most and showed many more statistically significant associations than abuse and neglect alone.
Table 2. Odds ratio with 95% confidence interval (OR [95%CI]) for presence of symptoms of depression (in MDD, BD and SCZ patients, n = 3156), mania and psychosis (in BD and SCZ patients; n = 1916) after experiencing abuse, neglect, or combined CM
MDD, major depressive disorder; BD, bipolar disorder; SCZ, schizophrenia; CM, childhood maltreatment.
Analyses were adjusted for age, gender, and education.
*Significant with α = 0.05.
Online Supplementary Table S4 summarizes all statistically significant associations between types of CM and symptoms of depression, mania, and psychosis.
Impact of five CM subtypes across psychiatric disorders
χ2 tests showed a significantly different distribution of the five CTQ subscales (emotional and physical abuse, sexual abuse, emotional, and physical neglect) across diagnosis (χ2[15] = 132.84–134.27, p < 0.01). The results of logistic regression of CTQ subscales on psychiatric diagnosis, while adjusting for age, gender, and education (no multicollinearity: VIFs < 1.5) are shown in Table 3. All abuse types were most strongly related to increased odds of SCZ compared to the other diagnoses. Physical abuse stood out as a risk factor specifically for SCZ. Neglect, and especially emotional neglect, was the strongest risk factor for MDD and BD.
Table 3. Odds ratio with 95% confidence interval (OR [CI]) for MDD, BD, or SCZ after experiencing a subtype of childhood maltreatment: emotional abuse or neglect, physical abuse or neglect, or sexual abuse
MDD, major depressive disorder; BD, bipolar disorder; SCZ, schizophrenia.
Analyses were adjusted for age, gender, and education.
*Significant with α = 0.05.
Investigating evidence of causality between CM type and psychiatric disorders using MR
Forward MR analyses were consistent with a causal relationship of childhood abuse with SCZ (IVW = 0.125 [95% CI 0.01–0.24], p = 0.032; online Supplementary Table S5), based on two SNPs. Steiger test indicated a correct direction of causality between the exposure and outcome. Cochran's Q statistic was statistically significant for the analysis of abuse against SCZ, indicating evidence of pleiotropy but may also be the consequence of the limited number of two SNPs. Such limited number of two selected SNPs was also insufficient to perform the weighted median, MR Egger regression, and MR pleiotropy residual sum and outlier (MR-PRESSO) methods. These methods were used as sensitivity analyses and showed no evidence for invalid instruments for the analyses when enough SNPs were selected (online Supplementary Table S5). The F statistic ranged from 24.3 to 45.6, indicating that the estimates were not likely subject to weak instrument bias. The follow-up MR analysis, in which the SNP selection threshold was lowered by increasing the p value to p < 3 × 10−6, confirmed statistical significance with no evidence of horizontal pleiotropy (IVW = 0.112 [95% CI 0.05–0.18], p = 0.001, Q p value 0.186, based on six SNPs; see online Supplementary Table S6). This suggests that the more lenient threshold of 1 × 10−6 is not the main explanation of the found relation, and that the suggestion of pleiotropy disappears with a higher number of SNPs selected. The relation between child abuse and SCZ is the only consistent statistically significant relation in the MR analyses. The MR results for the association between neglect and combined CM and psychiatric disorders showed no consistent patterns of statistically significant causal effect. Backward MR supported a causal relationship of genetic variants linked to SCZ on abuse, neglect, and combined CM (online Supplementary Table S5).
Discussion
This study presents data from three large cohort studies that show that childhood abuse, neglect, and their combination all significantly contribute to an increased risk of SCZ, BD, and MDD. However, childhood abuse and neglect differ in the strength of their relation to diagnosis and clinical symptom profile. Abuse was significantly stronger related to higher odds of developing SCZ compared to its impact on MDD and BD. Neglect was most strongly associated with risk for BD and MDD. Differences between childhood abuse and neglect were also present across diagnosis at the symptom-level. Childhood abuse was associated with a significantly higher risk of suicide attempts, feelings of worthlessness or guilt, delusions, and hallucinations, whereas neglect was significantly related to agitation and reduced need for sleep. Differential effects of child abuse and neglect at the symptom level were most prominent for symptoms of psychosis whereby abuse strongly increased the risk of delusions and hallucinations, in contrast to neglect.
In addition to the distinction between the effects of childhood abuse and neglect alone, this study also shows that experiencing both childhood abuse and neglect (combined CM) is related to disproportionally higher odds of MDD or BD, exceeding the mere additive effect of child abuse plus neglect. Accumulation of stressful life events has been noted before as a risk factor for MDD (Vinkers et al., Reference Vinkers, Joëls, Milaneschi, Kahn, Penninx and Boks2014), and is consistent with the stress resilience model whereby symptoms of psychiatric disorders develop when the impact of adversity exceeds resilience thresholds (Carpenter et al., Reference Carpenter, Carvalho, Tyrka, Wier, Mello, Mello and Price2007; Heim et al., Reference Heim, Jeffrey Newport, Heit, Graham, Wilcox, Bonsall and Nemeroff2013; Houtepen et al., Reference Houtepen, Vinkers, Carrillo-Roa, Hiemstra, Van Lier, Meeus and Boks2016; Tyrka, Price, Marsit, Walters, & Carpenter, Reference Tyrka, Price, Marsit, Walters and Carpenter2012). On the symptom-level, combined maltreatment showed the same pattern, with higher risks for a multitude of symptoms, compared to abuse and neglect alone. Consequently, combined maltreatment exceeds a dose–response relationship and can be seen as a more detrimental type of CM for the risk for BD and MDD.
One of the most consistent findings of this study is the strong relation abuse with SCZ and positive psychotic symptoms. A strong relation between abuse and SCZ fits previous reports of a threefold increase in psychosis risk and hallucinations in patients with a history of child abuse (Croft et al., Reference Croft, Heron, Teufel, Cannon, Wolke, Thompson and Zammit2019; Marchi et al., Reference Marchi, Elkrief, Alkema, van Gastel, Schubart, van Eijk and Boks2022a; Read, Van Os, Morrison, & Ross, Reference Read, Van Os, Morrison and Ross2005; Van Os et al., Reference Van Os, Pries, Delespaul, Kenis, Luykx, Lin and Guloksuz2020; Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer and Bentall2012) and warrants the question about a possible causal relation between child abuse (and not neglect) and SCZ. The findings from the MR analyses lend further support for this hypothesis, even when based on a limited numbers of SNPs due to the modest discovery set of the UK Biobank (N = 143 473). Such a relation would be of great importance to further promote public prevention programs and could provide a personalized treatment perspective for individuals suffering from SCZ with a history of abuse. Considering that a history of CM has a negative influence on prognosis and treatment outcomes (Trotta, Murray, & Fisher, Reference Trotta, Murray and Fisher2015), the question is warranted to what extent individual trauma-focused therapy might decrease their burden (Van Den Berg et al., Reference Van Den Berg, De Bont, Van Der Vleugel, De Roos, De Jongh, Van Minnen and Van Der Gaag2015; Van Den Berg et al., Reference Van Den Berg, De Bont, Van Der Vleugel, De Roos, De Jongh, Van Minnen and Van Der Gaag2018), even in the absence of post-traumatic stress disorder (PTSD). Another prospect of an increased understanding of the CM–SCZ relationship could be the development of predictive models for antipsychotic treatment outcomes, as has previously been done for MDD (Williams, Debattista, Duchemin, Schatzberg, & Nemeroff, Reference Williams, Debattista, Duchemin, Schatzberg and Nemeroff2016). A clinical hypothesis could be that SCZ patients with a history of CM are less likely to respond to antipsychotics and more likely to trauma-focused therapy as compared to patients without such CM history. More fundamentally, these findings could be a starting point for etiology research into a distinct abuse–SCZ pathway. One possible direction for such studies could be the revisiting of molecular pathways linking CM to dopaminergic function specified by CM type (Howes et al., Reference Howes, McCutcheon, Owen and Murray2017). Previous research suggests several other psychological, social, and biological pathways from childhood adversity to SCZ (Alameda et al., Reference Alameda, Rodriguez, Carr, Aas, Trotta, Marino and Murray2020; Sideli et al., Reference Sideli, Murray, Schimmenti, Corso, La Barbera, Trotta and Fisher2020) that could be refined by specifying CM type.
Another consideration on the differential relations between CM type and psychopathology is related to recent evidence of gene–environment correlations (Kendler & Eaves, Reference Kendler and Eaves1986; Knafo & Jaffee, Reference Knafo and Jaffee2013). In the relation between CM and SCZ for instance, genetic liability to SCZ has been associated with CM (Sallis et al., Reference Sallis, Croft, Havdahl, Jones, Dunn, Davey Smith and Munafò2021) and CM has been found to act as a mediator in the relation between genetic risk for SCZ and the occurrence of psychotic-like experiences (Marchi et al., Reference Marchi, Elkrief, Alkema, van Gastel, Schubart, van Eijk and Boks2022), pointing out a role of gene–environment correlation in the emergence of a mental health phenotype. For the current study, the joint effect of both environmental experiences and genetic vulnerability underlying the development of psychopathology (Dalvie et al., Reference Dalvie, Maihofer, Coleman, Bradley, Breen, Brick and Nievergelt2020; Kendler & Eaves, Reference Kendler and Eaves1986) is very relevant and the results therefore should be viewed in light of a broader social context in which CM occurs (Marchi et al., Reference Marchi, Elkrief, Alkema, van Gastel, Schubart, van Eijk and Boks2022; Sideli et al., Reference Sideli, Murray, Schimmenti, Corso, La Barbera, Trotta and Fisher2020; Vinkers et al., Reference Vinkers, Joëls, Milaneschi, Kahn, Penninx and Boks2014). CM subtypes are likely associated with many potential confounders such as life stresses, parental psychopathology, and substance abuse (Doidge, Higgins, Delfabbro, & Segal, Reference Doidge, Higgins, Delfabbro and Segal2017), which occur more frequently in households with lower socio-economic status (Doidge et al., Reference Doidge, Higgins, Delfabbro and Segal2017; Sidebotham & Heron, Reference Sidebotham and Heron2006; Wu et al., Reference Wu, Ma, Carter, Ariet, Feaver, Resnick and Roth2004). Differential influences of abuse and neglect may well be related to specific environmental circumstances, similarly as childhood neglect has been associated with antisocial personality disorder of the care-taker (Mulder, Kuiper, van der Put, Stams, & Assink, Reference Mulder, Kuiper, van der Put, Stams and Assink2018) and neuroticism of the child (Brents, James, Cisler, & Kilts, Reference Brents, James, Cisler and Kilts2018; Hovens, Giltay, Van Hemert, & Penninx, Reference Hovens, Giltay, Van Hemert and Penninx2016).
In addition to pinpointing the relevance of gene–environment effects, the current study underlines the importance of differentiating in the type of adverse childhood experiences with respect to their effects on mental health. Previous research already indicated the relevance of CM even within diagnostic category by reporting clinical and neurobiological differences between maltreated and non-maltreated individuals with the same primary DSM-5 diagnosis (Teicher & Samson, Reference Teicher and Samson2013; Teicher, Gordon, & Nemeroff, Reference Teicher, Gordon and Nemeroff2022). The current study underscores the potential for such refinements. A subdivision within diagnoses based on the effects of childhood adversities may contribute to developing alternative, transdiagnostic approaches in future research. Besides child abuse and neglect, subdividing CM experiences into emotional and physical trauma could also be further investigated (Spinhoven, Elzinga, Van Hemert, De Rooij, & Penninx, Reference Spinhoven, Elzinga, Van Hemert, De Rooij and Penninx2016). Ultimately, further differentiating the effects of various CM types might open doors to targeted therapy and prediction models.
Strengths of this study include the overall sample size with standardized diagnostic assessments, uniform measure of CM, the investigation of effects of CM on both disorder and symptom-level, and the addition of investigating causal inference using MR. However, the results should be interpreted in the context of limitations that are mostly related to the use of data from three separate cohort studies. Consequently, the analyses were restricted to MDD, BD, and SCZ and particularly disorders such as PTSD, anxiety disorders, or personality disorders (Afifi et al., Reference Afifi, Mather, Boman, Fleisher, Enns, MacMillan and Sareen2011; Sistad, Simons, Mojallal, & Simons, Reference Sistad, Simons, Mojallal and Simons2021; Waxman, Fenton, Skodol, Grant, & Hasin, Reference Waxman, Fenton, Skodol, Grant and Hasin2014) could not be taken into account. As another consequence of this approach, symptom-level information on psychosis was not present for the MDD cases. Whereas diagnostic classifications according to the DSM were made using completed validated diagnostic clinical assessments, the presence or absence of a particular symptom was defined based on harmonization of the CIDI, CASH, and SCID items (Andreasen et al., Reference Andreasen, Flaum and Arndt1992; First et al., Reference First, Spitzer, Gibbon and Williams2002; Robins et al., Reference Robins, Wing, Wittchen, Helzer, Babor, Burke and Towle1988). Also, the analysis of multiple symptoms and diagnosis constitute an element of multiple testing that was not adjusted for. Next to that, it is important to acknowledge the potential presence of reverse causality. The most important limitation is that although possible residual confounding or collider bias is minimized by the facts that all cohorts comprised of a mixture of primary and specialty psychiatric care patients, contributed controls recruited in a similar way, and included at least two diagnostic categories, they cannot be ruled out. Also, it should be noted that the MR analyses are preliminary as they are based on a different assessment of CM, and that the forward analysis could only be based on a small selection of SNPs with a sub-genome-wide significant threshold. The lack of association between the CM types and MDD and BD may therefore reflect the limited power of the MR analyses. In the analysis of abuse against SCZ, there is evidence of pleiotropy. This could be explained by increased variance due to the small selection of SNPs in the forward analysis, since the suggestion for pleiotropy disappears in the follow-up analysis including a larger selection of SNPs.
Overall, this study provides evidence that abuse and neglect differ in their impact on risk of major psychiatric disorders and its symptoms, and that their combination is most adverse. The strong relations of abuse with the risk of developing SCZ and hallucinations stand out and are consistent with the possibility of distinct etiological pathway for psychosis. Further understanding of relations between more narrowly defined CM types and psychopathology can increase our etiological understanding and may ultimately guide diagnostic refinements and treatment strategies.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723003471
Acknowledgements
We would like to acknowledge all participants of GROUP, DBC, and NESDA cohort studies. We appreciate and thank Peter Zuithoff for his statistical advice and expertise.
Funding statement
This work is supported by the National Institute of Mental Health (grant number: R01MH 090 553) and the YouthGEMS project, ‘Gene Environment interactions in Mental health trajectories of Youth’ (EU 2020 Health: Youth-GEMs – 101057182 – AMD-101057182-6). C. C. is supported by the European Union's Horizon 2020 Research and Innovation Programme (EarlyCause; grant agreement number: 848158). The GROUP project was supported by a grant from the Netherlands Organization for Health Research and Development (ZonMw), within the Mental Health program (grant number: 10.000.1002). DBC was supported by the National Institute of Mental Health (grant number: R01MH 090 553) and NESDA was funded through the Geestkracht program of the Netherlands Organisation for Health Research and Development (ZonMw, grant number 10-000-1002) and financial contributions by participating universities and mental health care organizations (VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Rob Giel Onderzoekscentrum). The funders had no role in the design and conduct of the studies; data collection, management, analysis, or interpretation; preparation, review, or approval of the manuscript and decision to submit the manuscript for publication.
Competing interests
None.
Open access
