Psychological injury in victims of child sexual abuse: A meta-analytic review

The process for selecting scientific studies began with a search in the meta-search engines (i.e., Google, Yahoo, Google Scholar), which, according to the descriptors, listed between approximately 36,000 to 770,000 results. The vast majority of the search results yielded, with-out compromising the requirements for performing a robust meta-analysis (i.e., sufficient k and N ), a selection of studies par excellence. Thus, the next step was to search for studies in one of the world's leading scientific databases of reference, the Web of Science. All of the databases of scientific papers (Core Collection, Current Contents, Medline, Scielo, KCI-Korean) were searched to include not only scientific literature par excellence, but also cultural concepts (i.e., Latinos, Asiatic), which are referred to in the manuals for the classification of mental disorders (DSM and ICD) and may have differential effects on the symptoms associated to victims of sexual abuse, i.e., depressive and anxiety disorders. Nevertheless, the race or ethnicity of victims of sexual abuse was not related in itself to the manifestation of depressive or anxiety disorders ( Mennen, 1995).

In the initial search for studies in the meta-search engines, both English and Spanish descriptors were used: child maltreatment/maltrato infantil, child sexual abuse/abuso sexual infantil, victimization/ victimización, internalizing disorders/trastornos internalizantes, anxiety/ansiedad, and depression/depresión. In the second search in the Web of Science, the same English language descriptors were used given that all of the descriptors in these databases are in English. Following a method of successive approximations, all of the keywords were reviewed in the selected articles in the search for other potential descriptors. This method identified other descriptors employed by other authors (e.g., child sexual abuse, adolescent sexual abuse, internalizing behaviour disorder, CSA survivors, and sequelae) that were included in the search. In all, the system yielded more than 15,900 (searching by child sexual abuse) and 2,000 (searching by child sexual abuse AND depression OR anxiety) studies that were finally reduced to 78 after applying the following inclusion and exclusion criteria.

Of the studies listed by the system, the following met the inclusion criteria: a) studies assessing the sequelae of CSA/ASA [CSA/ASA understood as the involvement of a child in sexual activity that he or she does not fully comprehend, is unable to give informed consent to, or for which the child is not developmentally prepared and cannot give consent] in terms of depression or anxiety internalizing variables; b) studies reporting the effect sizes of CSA/ASA sequelae, in which the variables and/or statistics enabled the following to be calculated: group size, mean and standard deviation of sequelae measurement variables for each group, prevalence, specificity, and sensitivity; c) studies defining the ground truth for classifying participants as victims of CSA/ASA or the measure (i.e., instrument) of abuse; and d) studies providing descriptive data on the sample employed (e.g., age, sample size).

Studies failing to meet the prescribed requirements were excluded, as were cases where, after contacting the authors, the data required was not facilitated for the computation of the effect sizes. In addition, studies with data errors (e.g., lack of consistency in group size throughout the study not attributable to missing data) were eliminated. Similarly, studies failing to guarantee the mutual exclusion of the victim of sexual abuse condition from other forms of maltreatment were also excluded (e.g., studies undertaking a single comprehensive analysis of victims of sexual and physical abuse or neglect). The inclusion of studies was restricted in time to studies published since 1995 given the profusion of meta-analysis on studies up to 1995 ( Jumper, 1995 ). Thus, by applying these criteria, 78 studies were selected, with a total of 19,360 subjects, from which 149 effect sizes were obtained: 62 for the effects of CSA/ASA on anxiety disorders and 87 for depressive disorders. All of the studies selected had been published in double blind peer-reviewed scientific journals indexed at the Web of Science.

The following data from the studies was coded for the meta-analysis: variables measuring the effects of abuse (i.e., clinical diagnosis, clinical symptoms), measures of abuse, reliability of the measurement instruments, sample characteristics (i.e., size, age, cultural context), and the statistics required for computing the effect sizes.

All of the studies of the sample were examined by two independent researchers, with total agreement in their classifications ( k = 1). Appendix 1 shows the characteristics of the primary studies included in the present meta-analysis.

The weighted effect size was calculated according to the sample size of the variable measuring the effects of CSA/ASA (anxiety or depression). When several measures of anxiety were reported (e.g., generalized, social phobia, specific phobia) or depression (e.g., major depression, dysthymia), the means for the effect sizes of the variables with more than one measure were calculated. Moreover, when the sample was subdivided into subtypes of abuse, the effect sizes were weighted in order to obtain an overall size.

The meta-analytical technique employed was a correlational type procedure ( Hunter & Schmidt, 2004 ), weighting the effect size with the sample size and correcting with the reliability of the predictor and the criterion.

As most of studies fail to provide correlations between sexual abuse and the measures of sequelae of internalizing disorders (i.e., anxiety and depression), an alternative estimator of effect size, i.e., Cohen's d (1988), was computed when studies reported the mean and standard deviations of the experimental group (abused children), but no control group (in these studies the control group taken was the normative population for each instrument). When the results were expressed as student t values (F for one degree of freedom was transformed into t values), Cohen's d was obtained by comparing groups of the same sizes or when homogeneity of variance was observed and Hedges’ g (Hedges & Olkin, 1985 ) were obtained for different sizes or no homogeneity of variance. If the results were expressed in Z values, they were transformed into correlations, phi was obtained from 2 x 2 tables, and when in odds ratio converted to correlations. The formulas for converting the other effect sizes to correlations were taken from Cohen (1988) and Rosenthal (1994).

To estimate the practical utility of the results, the U1 ( Cohen, 1988 ), the odds ratio, and the CLES ( McGraw & Wong, 1992 ) were calculated. Additionally, increases in CSA/ASA symptoms or disorders (injury quantified in relation to population baselines) were directly obtained from r.

For predictor reliability, the measure of sexual abuse, the reliability coefficients obtained from the primary studies were computed. When studies failed to report reliability coefficients or reported concordance, which is not reliability, mean reliability was calculated on the basis of the primary studies. Table 1 shows mean reliability, the standard error, and a 95% confidence interval (indicating the variance of mean reliability) for each context measured. Succinctly, the measure of the predictor, victims of sexual abuse, was reliable (see Table 1 ) with a limit below .77 (that is, high reliability, r > .70; Arce, Velasco, Novo, & Farina, 2014; Nunnally & Bernstein, 1994 ).

Criterion reliability was drawn from the primary studies, the ori-ginal publication of the instrument itself and, in the absence of both, completed with the means for those contingencies where they were unavailable. The mean reliability, standard error, and confidence intervals for each measure are shown in Table 2 . In short, the measure of criteria reliability ranged from .84 to .86, with a lower limit for a 95% confidence interval of .82. Thus, the measures of the criterion were highly reliable ( r > .70).

Outliers in each of the measures (general sequelae, depression, and anxiety) were removed using the criterion ± 2 SD (bilateral) of the mean effect size, so the results were generalizable to 96% of future samples. The results identified 6 outliers (-.12 > r >.64) in sequelae (M = .26, SD = .19), 5 outliers (-.13 > r > .63) in anxiety (M = .25, SD = .19), and 6 outliers (-.14 > r > .66) in depression (M = .26, SD =.20). Thus, a total of 6 studies were eliminated from the meta-analysis as outliers.

The results of the meta-analysis on sequelae of sexual abuse on mental health (internalizing symptoms: depression and anxiety) showed (see Table 3 ) a significant effect (when the confidence interval has no zero, indicating the effect size was significant), positive (between victimization and mental health injury), generalizable (when the credibility interval has no zero, indicating the effect size was generalizable to 90% of other samples), and of a medium size (Cohen's category: r = .30) in general sequelae, depression, and anxiety, explaining 12, 8, and 10% of the variance respectively. In terms of practical utility, victims of CSA/ASA had a 70% higher probability (CLES) of internalizing injury (general sequelae) than non-victims, with 66% for depression and 68% for anxiety. The injury caused was quantified as 34, 28, and 31% in general sequelae, depression, and anxiety, respectively. Finally, the distributions for victims and non-victims (U1) in general sequelae, depression, and anxiety were totally independent (44, 37, and 41% respectively), though they were expected to be similar.

As for the robustness of the significance of the sequelae in CSA/ ASA victims, it is worth noting that 93 studies with no significant results would be required to accept the null hypothesis.

Though results were generalizable in three measures (general sequelae, depression, and anxiety), the literature recommends assessing the potential differential effects of gender, type of measure, and type of abuse (meta-analysis). In short, it is well known that the base rate for symptoms and clinical diagnosis differ from males to females

(Nolen-Hoeksema, 1990, 2002 ). This view is so firmly established that most of the psychometric measurement instruments extensively used in clinical practice score males separately to females. Likewise, according to the leading international organisations classifying mental disorders ( American Psychiatric Association, 2013 ; WHO, 2000), there is a higher prevalence of diagnosed depression and anxiety among females than males, including school-aged children and adolescents. Similarly, the type of abuse suffered (i.e., non-contact, contact, intercourse) has been shown to have effects on sequelae, which relates the severity of the crime to the severity of the injury (mild, moderate, severe) ( Bulik et al., 2001 ; Cutajar et al., 2010; Fergusson et al., 2013; Jonas et al., 2011 ). As for the variables measuring the effects on victims of abuse, i.e., clinical diagnosis and clinical symptoms, the presence of symptoms does not imply that the diagnostic criteria had been met. In other words, they are different measures, given that they measure different constructs, so they may have a differential sensitivity to injury.

Gender effects . The results of the meta-analysis showed (see Table 4) a significant effect, positive, of a small size (Cohen's category: r = .10), and generalizable in depression and anxiety in female CSA/ ASA victims.

In comparison, the meta-analysis revealed for male CSA/ASA victims a significant effect, positive, and of a small size in depression and anxiety, being generalizable in depression, but not so in anxiety (see Table 4 ). Thus, in the latter case, the results exhibited moderators mediated the direction of the effects.

Having contrasted the significance of the differences between the effect sizes, the true correlation between female participants and male participants, sequelae in depression was found to be significantly higher, q s = 0.093, p < .05, in females, but not so for anxiety, q s = 0.073, ns . This translates into quantifying injury in females as suffering from 9% more injury in depression than males. In prevalence rates (odds ratio), injury in depression for female and male victims was 2.26 and 1.60 times greater than for non-victims, and 2.26 and 1.73 times greater in anxiety for female and male victims, respectively, in contrast to non-victims.

Effects of the type of measure . The results of the meta-analysis showed a significant effect, positive, of a medium size, and generalizable in the diagnosis of depressive and anxiety disorders in victims of CSA/ASA (see Table 5).

Likewise, the results of the meta-analysis displayed a significant effect, positive, of a small effect size and generalizable in anxiety and depression symptoms in CSA/ASA victims (see Table 5).

Comparatively, injury was significantly higher, q s = 0.152, p < .01, in the clinical diagnosis of anxiety disorder than reported symptoms of anxiety. Likewise, the diagnosis of a depressive disorder was significantly more sensitive, q s = 0.108, p < .05, for CSA/ASA victims than the report of depressive symptoms. These results, due to the type of measure, explained the differences attributed to sample type ( Rind, Bauserman, & Tromovitch, 1998 ): university population (measure of symptoms) and clinical population (clinical diagnosis).

The meta-analysis of major depressive disorder and dysthymia (persistent depressive disorder) nesting in the diagnosis of depression (see Table 5 ), confirmed a significant and positive effect, of a medium size, and generalizable. Thus, the prevalence of a dysthymic disorder in victims of abuse was 6.59 (odds ratio) times higher than for non-victims, 3.25 higher for major depression. In terms of injury quantification, it was of 46% for dysthymia and 31% for major depression. The difference between effect sizes was significant, q s = 0.176, p < .01, thus the effect size for the diagnosis of dysthymia was significantly larger than for major depressive disorder.

Similarly, the meta-analysis on generalized anxiety disorder, specific phobia, social phobia, and panic disorder were nested in anxiety disorders (see Table 5 ) found a significant and positive effect, of a medium to large, and generalizable for every diagnosis. Indeed, this implied CSA/ASA victims had 5.12, 7.62, 4.85, and 5.60 (odds ratio) greater probability of developing generalized anxiety disorder, specific phobia, social phobia, and panic disorder, respectively, than CSA/ASA non-victims. Injury was quantified as 41, 49, 40, and 43% for generalized anxiety disorder, specific phobia, social phobia, and panic disorder, respectively. The probability of developing these disorders as sequelae was similar except for specific phobia that was significantly higher than social phobia, q s = 0.112, p < .05, and generalized anxiety disorder, q s = 0.100, p < .05.

Effects of the type of abuse . The results of the meta-analysis on the type of abuse suffered (no contact, contact, and penetration) revealed a significant and positive effect, of small size, and generalizable in depression and anxiety (see Table 6 ). The comparison of sizes, showed injury derived from abuse with penetration, both in depression and anxiety, was significantly higher than injury in the no contact abuse condition for depression, q s = 0.093, p < .05, and anxiety, qs = 0.092, p < .05.

Effects of the interaction type of measure and gender . In the diagnosis of depressive disorders (see Table 7 ), the effect sizes were positive and significant for both males and females, of a small size for males and a medium one for females, which were generalizable for females, but not for males (the effects of the moderators could not be assessed in this case due to the very small k ). In comparison, the effect size for females was significantly higher, q s = 0.388, p < .01, than for males, quantifying injury in 42% for female victims and 10% for males. As for prevalence, female CSA/ASA victims had a 5.40 (odds ratio) more probability of meeting the criteria of depressive disorders than female non-victims, whereas males had a 1.44 more probability than male non-victims. Moreover, for the diagnosis of anxiety disorders the effect sizes were positive, significant, and of a small size for both males and females, generalizable for females, but not so for males (once again, moderators could not be found due to the very small k ). Once again, the effect size found in females was significantly higher, q s = 0.104, p < .05, than in males, with injury of 24% for female victims and 14% for males. This reveals that female victims had 2.43 (odds ratio) more probability of developing anxiety disorders than female non-victims, and male victims 1.66 more probability than male non-victims.

The effect sizes in depressive symptoms were significant, positive, of small sizes, generalizable, and similar ( ns ) for both males and females. As for anxiety symptoms, the effect sizes were significant, positive, of small sizes, and generalizable for both males and females. Nonetheless, the effect size was significantly higher, q s = 0.095, p < .05, in males. Thus, the results highlight that male CSA/ASA victims had a 2.76 probability (odds ratio) of developing significantly more anxiety symptoms than male non-victims, and female victims a 1.95 probability than female non-victims.

This meta-analysis entails certain limitations that should be borne in mind when interpreting the data. First, the ground truth of the primary studies for the classification of abuse generally rests on self-reports of a retrospective nature, that relies on individual memory capabilities, and are related to false positives or false alarms ( Amado, Arce, & Fariña, 2015; Schoedl et al., 2010 ). Moreover, victim self-reports of sexual abuse may bias the results towards concealing them (false negatives), in particular for males ( Stoltenborgh et al., 2011 ). Second, primary studies assume that injury to mental health is sequelae to abuse, without appraising other possible causes (cause-effect relationship) ( Jumper, 1995; Vilariño et al., 2013 ). Third, the effect of the variable under analysis in primary studies was not completely isolated as in many studies victims of sexual abuse, physical abuse, neglect, and other categories appear under the same umbrella. Fourth, as some studies had no control group, the normative population was taken as the contrast group, or it was not equivalent to the experimental one with the subsequent potential for distortion in the calculated effect sizes ( Briere, 1992).

Alternatively, the results of the meta-analysis were subject to little variability, that is, Ns > 400 and a large k (Hunter & Schmidt, 2004 ) were highly generalizable (entirely for the female population, and for males with the exception of the diagnosis of a disorder and the general measure of anxiety for the male population), whereas 93 studies with no significant results would be required to annul the evidence supporting the claim that CSA/ASA leads to mental health injuries.

Further research is required to determine which moderators inhibit the generalization of the effects in the general measure of anxiety in the male population and in the diagnosis of depression and anxiety.