Participants

The sample is part of a large, school-based, community study of children aged 6 to 12 years (at screening) from 57 schools in two Brazilian cities - Porto Alegre (n = 22) and São Paulo (n = 35). The research protocol was submitted to and approved by the Ethics Committees at the Universidade de São Paulo and the Universidade Federal do Rio Grande do Sul, Brazil, and written consent was obtained from parents of all participants, as well as verbal assent from all children before enrollment. Children who reported child maltreatment were invited to a clinical interview with a psychiatrist to investigate the level of exposure to CM and when appropriate, to be referred to child protection services. Child protection services are the institutions responsible for evaluating each case and, when a suspicion of CM is confirmed, for taking the appropriate legal measures. Additionally, all children with a positive diagnosis of a mental disorder were contacted for psychoeducation about the diagnosis and for referral to family health units for appropriate treatment.

During the screening phase, conducted on their school's registration day, 9,937 informants were interviewed using the Family History Survey (FHS).⁸ From this pool, we recruited two subgroups using a random (n = 958) and a high-risk (n = 1,514) group selection procedure, resulting in a total sample of 2,512 children.⁷ Subjects in the random stratum were selected using a simple randomization procedure based on school rolls, without replacement of non-available subjects. Selection for the high-risk stratum involved a risk-prioritization procedure, focused on individuals with a family history of a disorder and/or ongoing symptoms in one of five targeted domains (attention deficit hyperactivity disorder [ADHD], anxiety, obsessive-compulsive disorder [OCD], psychosis and learning disorders), as detected during screening. Subjects in this second, high-risk stratum were oversampled and, if not available, replaced by the next subject listed in the high-risk sampling frame. The details of the sample selection procedures can be found elsewhere.⁷ Of these 2,512 subjects, all parent reports on child maltreatment were available (94.9% answered by biological mothers and 5.1% answered by biological fathers). A total of 2,213 (88.1%) child reports on CM were also available.

Child maltreatment assessment

The questions about child maltreatment were selected to represent four categories of CM that have been distinguished in the literature: a) physical abuse (infliction of bodily injury by non-accidental means); b) neglect (failure to provide minimum care and/or lack of supervision); c) emotional maltreatment (pervasive and extreme thwarting of a child's basic emotional needs); and d) sexual abuse (sexual contact or attempted contact for purposes of sexual gratification or financial gain).⁹ Similar questions are frequently used in survey research^10-12 and they are close to the constructs evaluated by other instruments that assess child maltreatment more comprehensively, such as the Childhood Trauma Questionnaire.^13-16

Trained lay interviewers administered the research protocol to parents, using the following questions:

The first three questions were also rephrased to make them apply to the children and administered by trained clinical psychologists, as follows:

Responses to these seven questions were rated on a 4-point scale: 0, never; 1, one or two times; 2, sometimes; 3, frequently. The forms used in the study are available on request.

Strengths and Difficulties Questionnaire (SDQ)

The SDQ is a measure of mental health problems in children that can be administered to their parents or teachers or to the children themselves.^17-19 It comprises 20 items related to emotional symptoms, conduct problems, hyperactivity, and peer problems, rated on a 3-point scale, as follows: 0, not true; 1, somewhat true; 2, certainly true. Item scores were summed to create a total difficulty score ranging from 0 to 40. The SDQ also contains five items investigating prosocial behaviors. These questions are then followed by a single item assessing whether the respondent thinks that the child has a problem with "emotions, concentration, behavior, or being able to get along with other people" (response options: no problems, minor problems, definite problems, and severe problems). This last item is used to classify the impact caused by the psychiatric symptoms assessed. The Brazilian version of the SDQ was used in this study. This measure has been translated into Brazilian-Portuguese and has previously been used in large community investigations.²⁰ Parent SDQ reports were available for all of the children in the sample (n = 2,512). Teacher SDQ reports were also available for a subset of 1,189 of the children. Some children had more than one teacher, in which case the teacher who was spending more time with the child was chosen to answer the SDQ.

Psychiatric diagnosis

Psychiatric diagnoses were established using the Development and Well-Being Assessment (DAWBA).²¹ The DAWBA is a structured interview administered by lay interviewers who also record verbatim responses of any reported problems. Verbatim responses and structured questions are then carefully evaluated by psychiatrists, who confirm or refute diagnoses. All questions are closely related to diagnostic criteria from the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV), and focus on current problems causing significant distress or social impairment. The DAWBA has been translated into several languages, and for the present study we used the Brazilian-Portuguese version,²² administered to a biological parent who was legally responsible for the child. Administration was performed in accordance with previously reported procedures.²¹ Nine psychiatrists performed the rating procedures, all trained and supervised by a senior child psychiatrist.

For the purposes of this study we allocated each child to one of five non-overlapping groups: 1) typically developing controls (TDC; n = 1,880): subjects without any psychiatric disorder; 2) ADHD: subjects with any ADHD subtype (n = 143); 3) fear disorders: subjects with separation anxiety disorder, social anxiety disorder, specific phobia, or agoraphobia (n = 108); 4) distress disorders: subjects with generalized anxiety disorder, depression (major or not otherwise specified), or posttraumatic stress disorder (n = 76); or 5) oppositional defiant disorder or conduct disorder (ODD or CD; n = 56). All subjects with co-morbid conditions were excluded from these analyses. These diagnostic groups were chosen on the basis of evidence from twin studies^23,24 and research on symptom structure.^25-29

Data analysis

Since we aimed to combine information from child and parent reports and at the same time to fit a unidimensional measure of child maltreatment, we used confirmatory factor analysis (CFA) to test three theoretical models: a unidimensional model (all items loading into a single child maltreatment estimate); a bifactor model (all items loading into a general maltreatment estimate and residuals loading into a parent group factor and a child group factor); and a higher-order model (a higher-order factor of child maltreatment fully encompassing two lower-order factors - child and parent report). The small number of indicators precluded use of CFA to analyze multitrait-multimethod matrices.

The CFA models were fitted to polychoric correlations among items using the weighted least squares means and variance adjusted (WLSMV) estimator implemented in Mplus 7.0 software.³⁰ Model fit was deemed adequate if the comparative fit index (CFI) and the Tucker-Lewis index (TLI) were ≥ 0.95 and the root square mean error of approximation (RSMEA) was < 0.06. Model fit was deemed acceptable if CFI and TLI were ≥ 0.90 and RMSEA was < 0.08.^31,32

In Mplus, CFA can be used to investigate both factor loadings and thresholds. Factor loadings indicate how well the question reflects the underlying dimension, i.e., the relative contribution that each item makes to the latent trait. Threshold parameters reflect the standardized level of severity of the latent trait at which subsequent response options become more probable than the previous option (i.e. for choosing a response option of frequently or higher vs. sometimes or lower). Confirmatory factor analysis with categorical indicators is equivalent to item response theory (IRT) analyses of these parameters, with factor loadings being analogous to item discrimination parameters, and item thresholds being analogous to item difficulty parameters or item location parameters. Scale reliability was estimated from CFA results.^33,34 Missing data on child reports were assumed to be distributed at random. The Mplus pairwise present method was used to estimate scores for subjects with missing information.

Pearson coefficients were calculated for correlations between the factor scores that emerged from CFA of the model that presented the best fit and the total and subscale scores of parent and teacher SDQ reports. The SDQ scores were used to investigate convergent and divergent validity. We hypothesized that the higher the level of child maltreatment report, the higher the number of psychiatric symptoms, and so such a finding would provide evidence of convergent validity. We also hypothesized that the lower the levels of maltreatment, the higher the number of prosocial behaviors, and so observation of this relationship would provide evidence of divergent validity. Discriminant validity was investigated using one-way analysis of variance (ANOVA) to compare the five groups divided according to their diagnostic categories in the DAWBA, using false discovery rate to control for multiple comparisons.

Confirmatory factor analysis and reliability

A unidimensional model encompassing all seven items did not provide an acceptable fit to the data (χ² = 254.971, degrees of freedom [df] = 14, p < 0.001; RMSEA = 0.083 (90% confidence interval [90%CI] 0.074-0.092); CFI = 0.758; TLI = 0.637). A bifactor model with one general factor encompassing all items and two group factors (parent and child reports) did exhibit an adequate fit (χ² = 26.147, df = 7, p = 0.0005; RMSEA = 0.033 (90%CI 0.020-0.047); CFI = 0.981; TLI = 0.942). However, inspection of the factor loadings revealed that the general factor only captured the variance of the parent report items. Therefore, all child items only loaded onto the child group factor, limiting the utility of the general factor and integration of information from both informants. The higher-order model with two lower-order factors (parent report and child report) and a higher-order factor (child maltreatment) presented the best fit to the data out of the three models tested (χ² = 29.9, df = 13, p = 0.0047; RMSEA = 0.023 (90%CI 0.012-0.034); CFI = 0.983; TLI = 0.972), and the two-factor correlated models also exhibited appropriateness (same fit indices). Therefore, subsequent analyses were only performed for the higher-order model.

Working with this model, inspection of the factor loadings revealed that all factor loadings were above 0.40 (highest: child physical abuse report λ = 0.754; lowest: parent physical neglect report λ = 0.596). Inspection of item thresholds revealed that, as expected, the levels of child maltreatment required for endorsement of the questions followed the following ascending order: emotional abuse questions in both parent and child reports, physical abuse in both reports, physical neglect in both reports and, finally, sexual abuse in parent report (Table 1 and Figure 1).The model-based reliability computed from the model was 0.80 for the higher-order factor, 0.83 for the parent factor and 0.87 for the child factor (Table 2).

Figure 1 - Graphical representation of the higher-order child maltreatment model.  

Discriminant validity

There were significant differences in CM scores between groups selected according to DAWBA results, for the combined score (F_4,2258 = 33.63, p < 0.001, ω2 = 0.055), parent score (F_4,2258 = 33.57, p < 0.001, ω2 = 0.054) and child score (F_4,2258 = 14.57, p < 0.001, ω2 = 0.023). Overall CM levels were higher in all diagnostic category groups than for the typically developing children (except for child scores in the comparison of fear vs. TDC), with moderate to strong effect sizes (Cohen's d ranging from 0.30 to 1.0). Comparisons between different diagnostic groups revealed that CM levels were higher in the distress and ODD/CD groups than in the ADHD and fear groups for combined and parent scores (but not for child scores). Post-hoc comparisons are shown in Table 4 and Figure 2.

Figure 2 - Lower-order and higher-order factor scores for non-overlapping diagnostic groups. ADHD = attention deficit hyperactivity disorder; SE = standard error.  

Supplemental analysis

Table 5 lists the levels of exposure for each type of maltreatment in our sample, in order to provide a clearer report of the level of exposure to CM in the sample. According to both child and parent reports, the most frequent type of CM was emotional abuse. As expected, frequencies of CM were higher in the high-risk subset than in the randomly selected subset (Table 5).

We also used the CFA results to generate a categorical classification of the children's level of exposure. The category thresholds from the CFA results represent the standardized level of exposure to CM at which subsequent response options become more probable than the previous one and these were used to define "high exposure" in each type of maltreatment event, with a value of 1.5 deemed as indicative of high exposure. High exposure in parents' reports was then defined as: physical abuse and physical neglect rated sometimes or frequently, sexual abuse rated rarely, sometimes or frequently, and emotional abuse rated frequently. High exposure in children's reports was defined as: physical abuse rated sometimes or frequently, physical neglect rated rarely, sometimes or frequently, and emotional abuse rated frequently. Based on these criteria, the prevalence of high exposure to CM for any type of trauma in our sample was 12.1% according to parent-report scores and 12.5% according to child-report scores (Table 5).