The Long-Term Health Consequences of Child Physical Abuse, Emotional Abuse, and Neglect: A Systematic Review and Meta-Analysis

Rosana Norman and colleagues conduct a systematic review and meta-analysis to assess the relationship between child physical abuse, emotional abuse, and neglect, and subsequent mental and physical health outcomes.


Introduction
Child maltreatment is defined as all forms of physical and/or emotional ill-treatment, sexual abuse, neglect or negligent treatment, or commercial or other exploitation of children that results in actual or potential harm to a child's health, survival, development, or dignity in the context of a relationship of responsibility, trust, or power [1]. Four types of maltreatment are commonly recognised: sexual abuse, physical abuse, emotional abuse (also referred to as psychological abuse), and neglect (Table 1).
There is a great deal of uncertainty around estimates of the frequency and severity of child maltreatment worldwide. Furthermore, much violence against children remains largely hidden and unreported because of fear and stigma and the societal acceptance of this type of violence [2]. Globally, prevalence of reported child sexual abuse varies from 2% to 62%, with some of this variation explained by a number of methodological factors including definition of abuse, method of data collection, and type of sample assessed [3]. In high-income countries, the annual prevalence of physical abuse ranges from 4% to 16%, and approximately 10% of children are neglected or emotionally abused [4]. Eighty percent of this maltreatment is perpetrated by parents or parental guardians [4], and poverty, mental health problems, low educational achievement, alcohol and drug misuse, having been maltreated oneself as a child, and family breakdown or violence between other family members are all important risk factors for parents abusing their children [5].
There is growing recognition that different forms of interpersonal violence have a large public health impact [6]. In children, the consequences of violence can vary widely. Physical injuries and, in extreme cases, death are direct consequences. World Health Organization (WHO) estimates of child homicide suggest that infants and very young children are at greatest risk, with rates for the 0-to 4-y age group about double those for 5-to 14-y-olds as a result of their dependency and vulnerability [5]. However, in the majority of non-fatal cases, the direct physical injury causes less morbidity to the child than the long-term impact of the violence on the child's neurological, cognitive, and emotional development and overall health [5].
Child maltreatment is a major public health problem, yet a lack of understanding of its serious lifelong consequences and of the cost and burden on society has hampered investment in prevention policies and programs. In order to effectively respond to the problem, the WHO 2006 report on prevention of child maltreatment [5] recommended expanding the scientific evidence base for the magnitude, consequences, and preventability of child maltreatment.
The relationship between child sexual abuse and adverse psychological consequences in adults is well established [7][8][9], and in the WHO comparative risk assessment study, Andrews and colleagues [3] carried out a systematic review and meta-analysis summarising the evidence of a relationship between child sexual abuse and subsequent mental disorders. This review is currently being updated in the new iteration of the Global Burden of Diseases, Injuries, and Risk Factors Study, aiming to provide global estimates of attributable burden for 1990 to 2010 [10], but other forms of child maltreatment have been omitted.
Exposure to non-sexual child maltreatment, namely, physical abuse, emotional abuse, and neglect, is associated with increased risk of a wide range of psychological and behavioural problems, including depression, alcohol abuse, anxiety, and suicidal behaviour, and with increased risk of HIV and herpes simplex virus type 2 (HSV2) infection [11][12][13][14]. However, the long-term health consequences of these other forms of child maltreatment have not been systematically examined. To address these omissions, clarify the present state of empirical research, and enable the quantification of the health impacts of child neglect, physical abuse, and emotional abuse at the population level using burden of disease and comparative risk assessment methodology, we conducted a systematic review of the scientific literature and quantitative metaanalyses. To the best of our knowledge, this is the first metaanalysis to summarise the evidence for associations between individual types of non-sexual child maltreatment and outcomes related to mental and physical health.

Methods
General recommendations from the PRISMA 2009 revision [15], with regard to processing and reporting of results, were taken Table 1. Definition of child maltreatment.

Type of Maltreatment Description
Physical abuse Physical abuse of a child is defined as the intentional use of physical force against a child that results in-or has a high likelihood of resulting in-harm for the child's health, survival, development, or dignity. This includes hitting, beating, kicking, shaking, biting, strangling, scalding, burning, poisoning, and suffocating. Much physical violence against children in the home is inflicted with the object of punishing.
into account (Text S1). The meta-analysis conforms to the guidelines outlined by the Meta-analysis of Observational Studies in Epidemiology recommendations [16]. Methods and inclusion criteria were specified in advance and documented in a review protocol (Text S2).

Inclusion and Exclusion Criteria
This systematic review and meta-analysis incorporated retrospective and prospective cohort, cross-sectional, and case-control studies meeting the following inclusion criteria: (1) the study reported original, empirical research published in a peer-reviewed journal, (2) the study considered non-sexual child maltreatment as a potential risk factor for loss of health, and (3) the related health outcomes or behavioural risk factors were among those listed in the Global Burden of Diseases, Injuries, and Risk Factors Study [10]. Studies reporting exposure only to combined types of abuse were excluded. Included studies reported odds ratios (ORs) and confidence intervals (CIs) comparing those exposed and not exposed by type of abuse or, alternatively, provided the information from which effect sizes and confidence intervals could be calculated (Text S2).

Search Strategy
Three electronic databases (Medline, EMBASE, and PsycINFO up to 26 June 2012) were searched using full text and Medical Subject Headings (MeSH) terms to identify studies reporting an association between non-sexual child maltreatment and health outcomes (Text S2). Truncation of terms was used to capture variation in terminology. The search was not restricted to the English language, nor restricted by any other means. Searches were conducted using synonyms and combinations of the following search terms: ''maltreatment'', ''physical abuse'', ''psychological abuse'', and ''emotional abuse'', and automatic explosion of the terms ''child abuse'' and ''child neglect''. The search was also not restricted to any particular health outcome. Instead, the broader terms ''risk'', ''adverse effect'', ''consequences'', ''harm'', and ''association'' were used to encompass all studies that investigated any adverse outcome of non-sexual child maltreatment. In addition, reference lists of selected studies were screened for any other relevant study, and additional studies were also identified through contact with study authors. Articles in languages other than English were translated.

Data Collection and Quality Assessment
The full-text article of any study that appeared to meet the inclusion criteria was retrieved for closer examination. Two reviewers (R. E. N. and M. B.) independently assessed articles for eligibility. Disagreements were resolved by consensus. The coders were not masked to the journals or authors of the studies reviewed. A standardised data extraction sheet was developed, and data retrieved included publication details, country where study was conducted, methodological characteristics such as sample size and study design, exposure and outcome measures, type of abuse, and health outcomes (Text S2). The data extraction sheet included a quality assessment tool ( Table 2) to rate the methodological quality of each study based on the Newcastle-Ottawa Scale for assessing the quality of observational studies [17]. Quality assessment was completed independently by two reviewers, and disagreements were resolved by discussion. One author was contacted for further information.

Statistical Analyses
Weighted summary measures were computed using MetaXL, version 1.2 [18], a tool for meta-analysis in Microsoft Excel, with ORs chosen as the principal summary measure. Heterogeneity was quantitatively assessed using the Cochran's Q and I 2 statistics to evaluate whether the pooled studies represent a homogeneous distribution of effect sizes. Evidence of publication bias was investigated by means of funnel plots using the standard error on the y-axis [19].
Meta-analyses were complicated by the presence of significant heterogeneity in the data, likely due to a combination of true variance in these relationships and variability produced by differences in the methodology used to measure exposure and outcomes. We hypothesised that effect size may differ according to the methodological quality of the studies. MetaXL implements a process to explicitly address study heterogeneity caused by differences in study quality. This so-called quality effects (Doi and Thalib) model [20] is a modified version of the fixed-effects inverse variance method that additionally allows giving greater weight to studies of high quality versus studies of lesser quality by using the quality scores assigned to each study to weigh studies not only according to sample size but also by study quality [20,21]. Forest plots were made to visualise individual as well as pooled effects.
To address the effects of important study characteristics and explore heterogeneity, we additionally conducted several prespecified subgroup analyses (depending on data availability) by the following: gender of participants in the sample, geographic location (high income versus low-to-middle income), type of sample (population-based versus non-representative samples), measurement of abuse (self-reported versus official records), assessment of health outcome (structured clinical interview versus self-reported), prospective versus retrospective assessment of abuse and neglect, and appropriate adjustment versus no or inadequate adjustment for confounders.

Mental Disorders
Physically abused (OR = 1.54; 95% CI 1.16-2.04), emotionally abused (OR = 3.06; 95% CI 2.43-3.85), and neglected (OR = 2.11; 95% CI 1.61-2.77) individuals were found to have a higher risk of developing depressive disorders than non-abused individuals (Table 4; Figures S1, S2, S3). The test for heterogeneity was highly significant, with p,0.01 for both abuse types and neglect. Funnel plots indicate the possibility of publication bias for physical abuse, as it appears that some smaller, less precise studies have a greater effect size than the larger studies, and there are no smaller studies to the left (negative) side of the graph, suggesting that some negative studies may never have been published ( Figure S4).
For physical abuse, emotional abuse, and neglect, OR estimates in males were higher than in females, but the difference was not statistically significant (Table S1). The odds of developing depressive disorders with exposure to physical abuse were greatest in prospective studies. Although the OR point estimate was higher in subgroup analyses of studies where exposure to physical abuse was court-substantiated by official records-which would include the more severe cases of abuse (OR = 2.41; 95% CI 1.32-4.41)compared with self-reported physical abuse (OR = 1.56; 95% CI 1. 11-2.19) and physical punishment (OR = 1.20; 95% CI 0.88-1.61), the 95% CIs were overlapping, and these differences were not statistically significant. There was a stronger association between physical abuse and a diagnosis of major depressive disorder using structured interviews (OR = 1.82; 95% CI 1.44-2.30) than when depressive disorders were diagnosed by symptom scales (OR = 1.52; 95% CI 1.03-2.24), but again these differences were not statistically significant (Table S1). Restricting the physical abuse analysis to studies from high-income countries increased the odds of developing depressive disorders to 1.58 (95% CI 1.18-2.12), but the association was not significant in low-to-middleincome countries (Table S1).
However, for neglect in childhood, similar odds of developing depressive disorders were observed in high-and low-to-middle-income countries. Data from two studies suggest a dose-response relationship, with depression more likely with frequent neglect compared with neglect that occurred only sometimes in childhood [13,22]. A dose-response relationship was also reported for emotional abuse and depressive disorders, but not for physical abuse and depressive disorders (Table S1).
Physical abuse (OR = 1.51; 95% CI 1.27-1.79), emotional abuse (OR = 3.21; 95% CI 2.05-5.03), and neglect (OR = 1.82; 95% CI 1.51-2.20) were associated with a significantly increased risk of anxiety disorders ( Figures S5, S6, S7, S8). For physical abuse, significant associations were also observed with posttraumatic stress disorder (PTSD) and panic disorder diagnoses (Table S2). A dose-response relationship was observed with physical abuse but not with emotional abuse and neglect [22], with anxiety disorders more likely with frequent physical abuse than with abuse that occurred only sometimes in childhood (Table S2). Physical abuse, emotional abuse, and neglect were also associated with an almost 3-fold increased risk of developing eating disorders (Figures S9, S10, S11, S12), and physical abuse was associated with a 5-fold increased risk of developing bulimia nervosa meeting Diagnostic and Statistical Manual of Mental Disorders (DSM) diagnostic criteria. Most of the evidence came from retrospective studies, and only one prospective study [23] reported a strong association with neglect in childhood (Table  S3). A dose-response relationship was also observed, with bulimia nervosa more likely with more severe and repeated physical abuse [24] (Table S3). Physical abuse and neglect were also associated with a doubling of the odds of childhood behavioural and conduct disorders (Figures S13, S14, S15). With respect to physical abuse, higher odds of developing conduct and childhood behavioural disorders were observed in prospective than in retrospective studies, but differences were not statistically significant. Studies with non-representative samples had significantly increased effect size for the association between physical abuse and childhood behavioural problems and conduct disorder (OR = 5.98; 95% CI 2.73-13.10) compared with           [189]; CTQ, Childhood Trauma Questionnaire [190]; CTS, Conflict Tactics Scale [191]; CVD, cardiovascular disease; DISC-I, National Institute of Mental Health Diagnostic Interview Schedule for Children I [192]; DISC-II, National Institute of Mental Health Diagnostic Interview Schedule for Children II [193]; DIS-III-R, National Institute of Mental Health Diagnostic Interview Schedule IIIR [194]; EDE, Eating Disorder Examination (a standardised investigatorbased interview that operationalizes DSM-III-R criteria) [195]; ETISR-SF, Early Trauma Inventory Self Report-Short Form [196]; GFES, Global Family Environment Scale [197]; HMO, health maintenance organization; ICI, Incident Classification Interview [198]; K-SADS, Kiddie Schedule for Affective Disorders and Schizophrenia for School-Age Children [199]; MAST, Michigan Alcoholism Screening Test [200]; MINI, Mini International Neuropsychiatric Interview [201]; PBI, Parental Bonding Instrument [202]; PC-PTSD, Primary Care PTSD Screen [203]; PERI Life Events Scale, Psychiatric Epidemiological Research Instrument Life Events Scale [204,205]; PSE, Present State Examination [206]; SSI, Scale for Suicide Ideation [207]; SIBQ, Self-Injurious Behavior Questionnaire [208]; S-L criteria, Silberstein-Lipton criteria [209]; SCID for DSM-III-R, Structured Clinical Interview for DSM-III-R [210]; SCID for DSM-IV, Structured Clinical Interview for DSM-IV [211]; TSS, Traumatic Stress Schedule [212]; YSR, Youth Self-Report [213]. doi:10.1371/journal.pmed.1001349.t003 Table 4. Summary of primary meta-analyses on mental health consequences of child non-sexual maltreatment.  (Table  S4). Physical abuse significantly increased the risk of alcohol problem drinking (risky drinking, alcohol abuse/dependence, binge drinking) (OR = 1.26; 95% CI 1.03-1.55) ( Figure S16) and non-problem drinking (current or ever alcohol use), but the effect did not persist in prospective studies (Table S5). In a subgroup analysis, physical abuse was also significantly associated with a diagnosis of alcohol abuse/dependence meeting DSM criteria (OR = 1.40; 95% CI 1.21-1.64) (Table S5). Alcohol problem drinking was also associated with emotional abuse (OR = 1.27; 95% CI 1.11-1.46) ( Figure S17) but not with neglect in childhood (OR = 1.09; 95% CI 0.87-1.35) ( Figure S18). For alcohol problems, there was no evidence of a dose-response relationship with respect to frequency of abuse and neglect (Table S5) [13]. Gender differences were observed, with the effect of physical abuse on alcohol problems stronger among males, and with females at an increased risk of alcohol problem drinking with exposure to neglect in childhood, but with overlapping confidence intervals (Table S5). Publication bias did not appear to play a role in the association between physical abuse and alcohol problem drinking ( Figure S19).

Sexually Transmitted Infections and Risky Sexual Behaviour
Physically abused (OR = 1.78; 95% CI 1.50-2.10), emotionally abused (OR = 1.75; 95% CI 1.49-2.04), and neglected (OR = 1.57; 95% CI 1.39-1.78) individuals were found to have a significantly higher risk of sexually transmitted infections (STIs) and/or risky sexual behaviour than non-abused individuals (Table 5; Figures S28, S29, S30, S31). For physical abuse and neglect, the association with STIs and risky sexual behaviour was significant in prospective and retrospective studies (Table S8) (Table 5). A dose-response relationship was observed for HIV infection, with a larger effect size reported with more frequent physical and emotional abuse in childhood [13] (Table S8).

Chronic Diseases, Lifestyle Risk Factors, and Other Physical Health Outcomes
With regard to obesity, a significantly increased risk was observed for physical (OR = 1.32; 95% CI 1.06-1.64) and emotional abuse (OR = 1.24; 95% CI 1.13-1.36) but not for neglect (OR = 1.07; 95% CI 0. 97-1.19) in the primary analysis ( Figures S32, S33, S34, S35). Subgroup analysis by assessment of outcome indicated that neglect was associated with a higher risk of developing self-reported obesity, but there was no association with Table 7. Summary of review findings on health consequences of child non-sexual maltreatment for disorders where data were insufficient to include in meta-analyses. obesity defined by waist circumference or body mass index (BMI) measurements (Table S9). In the subgroup analysis by ascertainment of exposure to physical abuse, there was a strong association with obesity in one prospective study, but the magnitude of the effect was reduced in retrospective studies (Table S9). A doseresponse relationship between physical and emotional abuse and obesity has been observed [31] (Table S9). Physical (OR = 1.78; 95% CI 1.26-2.52) ( Figure S36) and emotional abuse (OR = 1.65; 95% CI 1.46-1.87) ( Figure S37) were associated with a significantly increased risk of current smoking, but the association was not significant for neglect in childhood (OR = 1.20; 95% CI 0.98-1.48). One study showed a dose response, with smoking more likely with physical abuse that occurred 3-5 times than with abuse that occurred 1-2 times, but this relationship did not continue into those who had been abused more than six times compared with those who had been abused 3-5 times [32] (Table S10).
Forty-two studies investigated the relationship between nonsexual child maltreatment and lifestyle risk factors, chronic diseases, and other physical health outcomes in adulthood. There is suggestive evidence of a significant association between child physical abuse and arthritis, ulcers, and headache/migraine in adulthood. However, for most other outcomes, including type 2 diabetes (Table S11; Figures S39, S40, S41, S42), hypertension, low exercise, cardiovascular diseases, respiratory diseases, neurological disorders, and cancer, these associations were mostly weak and inconsistent, with little adjustment for lifetime confounders. Pooled estimates were statistically significant in only a limited number of cases (Table 6). A recent prospective investigation of a group of individuals with documented histories of child abuse and neglect followed into middle adulthood provides some evidence that child abuse and neglect may increase the risk of a range of directly measured physical health outcomes after controlling for mental health problems, substance use, smoking, and BMI [33] (Table 7). However, there were insufficient studies examining the association between non-sexual child maltreatment and some of these health outcomes, including anaemia, underweight/malnutrition, hepatitis C, tuberculosis, hearing loss, vision loss, oral health, diarrhoea, allergies, uterine leiomyoma, back pain, breast cancer, and schizophrenia, to undergo meta-analysis (Table 7).

Discussion
To the best of our knowledge, this article presents the first systematic review and meta-analysis of published studies assessing the association between non-sexual child maltreatment and mental and physical health outcomes. We identified 124 studies that examined the association between physical abuse, emotional abuse, and neglect in childhood and various health outcomes.

Does Non-Sexual Child Maltreatment Cause Adverse Health Outcomes?
Evidence for a causal relationship between non-sexual child maltreatment and health outcomes was evaluated within the Bradford Hill framework on the grounds of the following important criteria: strength and consistency of the association, the temporal relationship of the association, evidence of a biological gradient or dose-response relationship, biological plausibility, and consideration of alternate explanations [34] ( Table S12).

Temporality
Both prospective and retrospective studies consistently showed an association between exposure to child physical abuse, emotional abuse, and neglect and adverse health outcomes. The availability of prospective studies provides conclusive evidence of a temporal relationship between exposure to non-sexual child maltreatment and the later development of mental health outcomes, drug use, and STIs and risky sexual behaviour, as in these studies abuse and neglect preceded the onset of health problems in adulthood.
However, only 16 studies were prospective, while the majority of the studies were cross-sectional and relied on adult retrospective report of abuse and neglect in childhood. By definition, these studies cannot prove a temporal relationship between exposure to child maltreatment and the onset of health outcomes. Furthermore, retrospective, self-reported information regarding abuse in childhood may be subject to recall bias, where those with adjustment problems may be more prone to recall or disclose exposure to abuse and neglect. In many cases participants were asked to report on events that would have occurred many years before, and the issue of potentially unreliable recall threatens the validity of the published literature on child maltreatment. At least with respect to child sexual abuse, evidence suggests moderate to good consistency of reports over time [35]. It has also been suggested that biases are probably towards under-reporting rather than over-reporting of abuse [36]. Nevertheless, given that retrospective reports were often the only measure of abuse available, particularly with regard to emotional abuse, we accepted these within the context of the limitations stated.
Although the strength of prospective studies includes the temporal ordering of maltreatment and subsequent health outcomes, with an objective measurement of exposure to abuse, these studies are usually conducted in non-representative samples. Official cases of abuse may only detect those who come to professional attention, and this may alter the strength of the association between non-sexual child maltreatment and adult morbidity. These official cases are also generally skewed towards the lower end of the socioeconomic spectrum and may not be generalisable to child abuse and neglect cases that occur in middleor upper-class children [33]. Those participants who have been identified by child protection agencies as having been exposed to physical abuse or neglect may have received interventions to prevent later pathology. Furthermore, some individuals in the ''never maltreated'' category may actually have experienced maltreatment, given that child maltreatment tends to be underreported. The validity of the various study designs to investigate the long-term health consequences of child maltreatment has been a source of ongoing debate [37,38]. In this meta-analysis we have included prospective and retrospective studies. The subgroup analyses show that with both methodologies there is robust evidence of a significant association between child non-sexual maltreatment and various health outcomes.

Strength of the Association
Associations between child physical abuse, emotional abuse, and neglect and mental disorders, drug use, and suicidal behaviour have been reported in prospective studies and/or large population-based studies. The strength of the relationship between abuse and mental disorders was generally reduced when the effects of important mediating variables were taken into account. Despite some variability, overall, child physical abuse, emotional abuse, and neglect were found to approximately double the likelihood of adverse mental health outcomes when combined in a metaanalysis.

Consistency of the Association
As shown in the forest plots of the effects by study, there was strong consistency and agreement in the estimated effect measures across studies, particularly for neglect and physical abuse, although we suspect publication bias for some of the outcomes. Risk estimates were comparable across different types of samples, for both non-representative and representative populations (Tables S1, S2, S3, S4 and S6, S7, S8). The findings persisted across different study designs, samples, and geographic regions investigated. It can be concluded that there is a highly consistent association between child physical abuse, emotional abuse, and neglect and adverse mental health outcomes, drug use, and STIs and risky sexual behaviour. We did not observe evidence of strong consistent associations for alcohol problems, chronic diseases, or lifestyle risk factors.

Dose-Response Relationship
We found evidence of a dose-response relationship between adverse health outcomes and non-sexual child maltreatment, such that those experiencing more severe abuse or neglect were at greater risk of developing mental disorders than those experiencing less severe maltreatment [39]. In the Chapman et al. [40] study, increasing severity of childhood adversity corresponded with poorer mental health outcomes. Consistent dose-response relationships with repeated, frequent, or severe abuse have been reported for mental disorders and physical abuse [13,24,41] and emotional abuse and neglect [13,22]. Furthermore, there is evidence to suggest that experiencing multiple types of maltreatment may carry more severe consequences, with those exposed to multiple types of abuse at increased odds of developing mental disorders [42,43], and the risk increases with the magnitude of multiple abuse [44]. Dose-response relationships with repeated frequent or severe abuse have also been reported for STIs and physical and emotional abuse [13], obesity and emotional and physical abuse [31], and smoking and physical abuse [32].

Plausibility
With respect to biological plausibility, animal models of mental disorders do not exist, making it particularly difficult to understand the underlying biological mechanisms. Progress in understanding has to be made by association and inference rather than experimental data [3]. There are nevertheless several potential mechanisms that may explain the observed association between abuse and neglect in childhood and increased risk of mental health problems. Neurobiological development can be physiologically altered by maltreatment during a child's early years, which can in turn negatively affect a child's physical, cognitive, emotional, and social growth, leading to psychological, behavioural, and learning problems that persist throughout the life course [45,46]. Moreover, cumulative trauma may further increase risk [47], and some victims of abuse may try to manage the subsequent distress through the use of alcohol, prescription medication, tobacco, or other drugs.
There is emerging evidence that the origins of most adult disease are found among developmental and biological disruptions in childhood. These early life experiences can affect adult mental and physical health either by cumulative damage over time or by the biological embedding of adversities during sensitive developmental periods [48]. There is generally a lag of many years before early adverse experiences are expressed in the form of disease [48].
Andrews and colleagues concluded that despite the lack of a biological link between child sexual abuse and mental disorders, a causal relationship was plausible [3], and that child maltreatment is most likely a contributory cause that acts via other intermediates.

Consideration of Alternate Explanations
It is important to note that the role of genes, environment, and gene-environment interactions in the causation of mental disorders is not well understood. Twin studies provide one of the best ways to examine the interplay between genetic and environmental influences [3], but to the best of our knowledge, these are only available for child sexual abuse. The relationship between abuse and neglect in childhood and subsequent health effects is complex. Although childhood abuse and neglect does result in adverse health outcomes, these outcomes are not independent of broader socioeconomic contexts. Lifestyle factors, access to health care, and neighbourhood characteristics may act as mediators between child abuse and neglect and long-term health consequences [49][50][51]. Exposure to child maltreatment often co-occurs within the context of other family dysfunction, social deprivation, and other environmental stressors that are also associated with mental disorders. Child maltreatment may be a marker of other family problems that together lead to the development of mental disorders. In addition, findings from many studies do not take into account the likely contribution of hereditary influences on the predisposition to mental disorders. Children of depressed parents may be at greater risk of depression through both exposure to maltreatment by their parents and genetic predisposition [43]. Hence, some of the effect of child abuse and neglect on mental disorders may still be explained by confounding. However, the effect of abuse on mental disorders remained significant in the majority of studies included in these meta-analyses after controlling for these co-occurring factors.

Assessment of Causality
In summary, there was robust evidence of significant associations between exposure to non-sexual child maltreatment and increased likelihood of a range of mental disorders, suicide attempts, drug use, STIs, and risky sexual behaviour. An increase in the likelihood of alcohol problem use was not consistently seen. There is weak to limited evidence suggesting a relationship between non-sexual child maltreatment and certain physical disorders and risk factors (Table 8), but more research is required to confirm these relationships.

Study Limitations
Although these findings and conclusions seem to be relatively consistent and robust, they should be interpreted in light of a number of limitations of our analysis.
This meta-analysis may be subject to publication bias because non-significant findings are less likely to be published [52]. This problem is increased when statistical models are employed because often only significant estimates are reported in many studies. This may result in the association between child abuse and neglect and outcomes being overstated, particularly for depressive disorders and anxiety, where publication bias may have played a role. For some of the other conditions there were too few studies to make conclusions with respect to publication bias.
The analysis also suffers from inconsistencies in how child abuse and neglect are defined and measured across the studies, as shown in Table 3. In studies using child protection records, exposure to physical abuse was defined to include injuries such as bruises, welts, burns, abrasions, lacerations, wounds, cuts, and fractures.
Some studies used the Barnett-Cicchetti Maltreatment Classification System [53] which defines physical abuse as a caregiver or responsible adult inflicting physical injury upon a child by other than accidental means. In other studies physical abuse was defined as having been hit, kicked, or punched so hard that the individual had marks or bruising or needed medical attention. Some studies referred to physical punishment [13,54,55] and corporal punishment [56], which may exclude more severe physical abuse, as well as physical assault by caregivers [57]. Emotional abuse definitions also varied considerably and included verbal abuse and being humiliated by a caregiver. Most studies involving neglect referred simply to ''neglect'', while others distinguished between physical and emotional neglect. Similarly, definitions of childhood were not consistent across studies. The complexity of defining and measuring child abuse has been noted in several studies [58][59][60].
Measurement bias with respect to health outcomes and the questionable reliability of self-reported data may also have affected the results. We dealt with this issue in the meta-analysis by adjusting the quality score and performing subgroup analyses. For mental disorders, studies using well-validated and standardised diagnostic instruments were assigned a higher quality score than studies using self-report symptom scales.
Another limitation of meta-analyses of observational studies is that, since individuals cannot be randomly allocated to case groups, the influence of confounding variables cannot be fully evaluated. While most studies presented multivariable adjusted ORs controlling for a range of socio-demographic and study design variables, a few studies presented unadjusted associations between child maltreatment and health outcomes, or adjusted for age and sex only. We again dealt with this issue in our metaanalysis by adjusting the quality score of studies with inadequate control for confounding and by carrying out separate analyses depending on data availability. Some studies also statistically controlled for exposure to other forms of maltreatment by including the different types of abuse in the same model in order to determine the independent contribution of each abuse type. Generally, in studies presenting results from various unadjusted and adjusted models, the association between abuse and physical and mental health outcomes was attenuated when controlling for the effects of mediating variables [61][62][63][64][65][66][67][68][69][70][71][72] and other forms of abuse [73][74][75][76][77][78][79]. However, findings from a recent prospective cohort study indicate that for some physical health outcomes additional control for socioeconomic status, unhealthy behaviour, smoking, and mental health problems seems to play varying roles in attenuating or intensifying these complex relationships [33]. Furthermore, we cannot exclude that residual confounding or unmeasured potential confounders may still remain. Despite evidence of weak associations between non-sexual child maltreatment and chronic diseases, further studies are needed that ensure adequate adjustment for lifetime confounders, because the attributable burden would be appreciable.
Significant heterogeneity exists in the primary analysis of physical and emotional abuse, even after our attempts to control for study quality in quality effects models, and the heterogeneity remained significant in most of the subgroup analyses. Given this situation, combining the effects may not be justified. With respect to neglect, pooled estimates in primary and subgroup analyses did not show significant heterogeneity for many outcomes.

Recommendations
Inconsistencies in the measurement and definition of child maltreatment highlight the importance of international efforts to standardise studies to enhance the comparability of findings. These include defining the cutoff age for childhood (0-18 y, as specified by the United Nations), and breaking this period into smaller age bands that can reflect age-specific patterns [5]. Researchers should select methodologies and instruments with international comparisons in mind. Identical questionnaires, research designs, and interviewing techniques should ideally be used for surveys in different countries [5]. In reality, however, all survey methods will require at least some adaptation to local conditions, and efforts to ensure comparability should involve choosing definitions of abuse and neglect, and questionnaire items, that represent an advanced level of knowledge [80]. To minimise how participants' subjective perceptions and definitions shape the answers, it is recommended that self-report studies clearly specify the behaviours and experiences being investigated, and that each sub-type of abuse and neglect is explored using multiple behaviourally specific questions, instead of a single-item ''label question'' [81].
Examples of international efforts to increase comparability across studies include the WHO's establishment of a global adverse childhood experiences research network, and the International Society for Prevention of Child Abuse and Neglect's Child Abuse Screening Tools (ICAST). The WHO network has developed an international version of the Adverse Childhood Experiences (ACE) questionnaire (the ACE International Questionnaire), for administration to people aged 18 y and older, which is currently being validated through trial implementation as part of broader health surveys in several countries [82]. The ICAST initiative has involved the development of three instruments that ask parents about their use of different behaviours for discipline, young adults (18-24 y) about their exposure to child abuse and neglect in childhood, and older children about their own recent experiences of violence [83].
Child maltreatment deserves increased investment in preventive and treatment strategies. Currently, there is a paucity of evidencebased interventions to reduce child maltreatment. Further research is urgently needed to identify programs that reduce the prevalence of child maltreatment, thereby alleviating an important risk factor for later health problems. Evidence-based systemic interventions that improve parenting strategies and family functioning may be more effective and economical than attempting to treat the wide-ranging deleterious health outcomes in adulthood that arise from maltreatment in the early years of life [48,84].
A broad range of protective factors have been identified that assist in promoting resilience in children exposed to adversity. Self control, problem-solving skills, secure relationships with caregivers, and safe schools and neighbourhoods are known to reduce the risk of adverse consequences in children exposed to trauma [85,86]. There is mounting evidence that exposure to childhood adversity interacting with particular genetic dispositions such as the short allele of the serotonin transporter gene [87] and genes involved in the regulation of the hypothalamic-pituitary axis [88,89] can result in problems with stress regulation and increased risk of anxiety and depression. Epigenetic changes have also been postulated as a mechanism by which transgenerational resilience or vulnerability may occur [90]. In spite of the increased knowledge in this field, it remains a challenge to translate this research into interventions at a population level that can reduce the vulnerability of children exposed to maltreatment [91].

Conclusion
This overview of the evidence suggests a causal relationship between non-sexual child maltreatment and a range of mental disorders, drug use, suicide attempts, sexually transmitted infections, and risky sexual behaviour. There is also emerging evidence that neglect in childhood may be as harmful as physical and emotional abuse. Although these conclusions have been drawn before from single empirical studies, in this article they are demonstrated in aggregate quantitative effects, to our knowledge for the first time.
This review contributes to a better understanding and measurement of the non-injury health impacts of child maltreatment globally and enables quantification of the burden attributable to physical and emotional abuse and neglect at the population level using comparative risk assessment methodology [92]. All forms of child maltreatment should be considered as part of the cluster of interpersonal violence risk factors in future global comparative risk assessments. Attributable burden is likely to be substantial, given the high prevalence of these forms of child maltreatment, the strong associations reported in our analysis, and the fact that related health outcomes are among the leading causes of disease burden globally. Despite the magnitude of the problem and increasing awareness of its high social costs, preventing child maltreatment is not a political priority in most countries. It is imperative that epidemiology and public health approaches find their proper place at the forefront of national and international efforts to understand and prevent child maltreatment [93]. Figure S1 Forest plot for quality-effect meta-analysis of the association between physical abuse and depressive disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S2 Forest plot for quality-effect meta-analysis of the association between emotional abuse and depressive disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S3 Forest plot for quality-effect meta-analysis of the association between neglect and depressive disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S4 Funnel plots to aid assessment of publication bias for depressive disorders and physical abuse. (TIF) Figure S5 Forest plot for quality-effect meta-analysis of the association between physical abuse and anxiety. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S6 Forest plot for quality-effect meta-analysis of the association between emotional abuse and anxiety. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S7 Forest plot for quality-effect meta-analysis of the association between neglect and anxiety. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S8 Funnel plot to aid assessment of publication bias for anxiety and physical abuse. (TIF) Figure S9 Forest plot for quality-effect meta-analysis of the association between physical abuse and eating disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S10 Forest plot for quality-effect meta-analysis of the association between emotional abuse and eating disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S11 Forest plot for quality-effect meta-analysis of the association between neglect and eating disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S12 Funnel plot to aid assessment of publication bias for eating disorders and physical abuse. (TIF) Figure S13 Forest plot for quality-effect meta-analysis of the association between physical abuse and conduct/ childhood behavioural disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S14 Forest plot for quality-effect meta-analysis of the association between neglect and conduct/childhood behavioural disorders. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S15 Funnel plot to aid assessment of publication bias for childhood behavioural/conduct disorders and physical abuse. (TIF) Figure S16 Forest plot for quality-effect meta-analysis of the association between physical abuse and alcohol problem drinking. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S17 Forest plot for quality-effect meta-analysis of the association between emotional abuse and alcohol problem drinking. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S18 Forest plot for quality-effect meta-analysis of the association between neglect and alcohol problem drinking. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S19 Funnel plot to aid assessment of publication bias for alcohol problem drinking and physical abuse. (TIF) Figure S20 Forest plot for quality-effect meta-analysis of the association between physical abuse and drug use. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S21 Forest plot for quality-effect meta-analysis of the association between emotional abuse and drug use. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S22 Forest plot for quality-effect meta-analysis of the association between neglect and drug use. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S23 Funnel plot to aid assessment of publication bias for drug use and physical abuse. (TIF) Figure S24 Forest plot for quality-effect meta-analysis of the association between physical abuse and suicide attempt. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S25 Forest plot for quality-effect meta-analysis of the association between emotional abuse and suicide attempt. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S26 Forest plot for quality-effect meta-analysis of the association between neglect and suicide attempt. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S27 Funnel plot to aid assessment of publication bias for suicide attempt and physical abuse. (TIF) Figure S28 Forest plot for quality-effect meta-analysis of the association between physical abuse and sexually transmitted infections/risky sexual behaviour. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S29 Forest plot for quality-effect meta-analysis of the association between emotional abuse and sexually transmitted infections/risky sexual behaviour. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S30 Forest plot for quality-effect meta-analysis of the association between neglect and sexually transmitted infections/risky sexual behaviour. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S31 Funnel plot to aid assessment of publication bias for sexually transmitted infections/risky sexual behaviour and physical abuse. (TIF) Figure S32 Forest plot for quality-effect meta-analysis of the association between physical abuse and obesity. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S33 Forest plot for quality-effect meta-analysis of the association between emotional abuse and obesity. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S34 Forest plot for quality-effect meta-analysis of the association between neglect and obesity. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S35 Funnel plot to aid assessment of publication bias for obesity and neglect. (TIF) Figure S36 Forest plot for quality-effect meta-analysis of the association between physical abuse and current smoking. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S37 Forest plot for quality-effect meta-analysis of the association between emotional abuse and current smoking. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S38 Funnel plot to aid assessment of publication bias for current smoking and physical abuse. (TIF) Figure S39 Forest plot for quality-effect meta-analysis of the association between physical abuse and type 2 diabetes. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S40 Forest plot for quality-effect meta-analysis of the association between emotional abuse and type 2 diabetes. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S41 Forest plot for quality-effect meta-analysis of the association between neglect and type 2 diabetes. Studies are represented by symbols, the area of which is proportional to the study's weight in the analysis. Output for ORs is set to the (natural) log scale. (TIF) Figure S42 Funnel plot to aid assessment of publication bias for type 2 diabetes and neglect. (TIF)

Editors' Summary
Background. Child maltreatment-the abuse and neglect of children-is a global problem. There are four types of child maltreatment-sexual abuse (the involvement of a child in sexual activity that he or she does not understand, is unable to give consent to, or is not developmentally prepared for), physical abuse (the use of physical force that harms the child's health, survival, development, or dignity), emotional abuse (the failure to provide a supportive environment by, for example, verbally threatening the child), and neglect (the failure to provide for all aspects of the child's well-being). Most child maltreatment is perpetrated by parents or parental guardians, many of whom were maltreated themselves as children. Other risk factors for parents abusing their children include poverty, mental health problems, and alcohol and drug misuse. Although there is considerable uncertainty about the frequency and severity of child maltreatment, according to the World Health Organization (WHO) about 20% of women and 5%-10% of men report being sexually abused as children, and the prevalence of physical abuse in childhood may be 25%-50%.
Why Was This Study Done? Child maltreatment has a large public health impact. Sometimes this impact is immediate and direct (injuries and deaths), but, more often, it is long-term, affecting emotional development and overall health. For child sexual abuse, the relationship between abuse and mental disorders in adult life is well-established. Exposure to other forms of child maltreatment has also been associated with a wide range of psychological and behavioral problems, but the health consequences of physical abuse, emotional abuse, and neglect have not been systematically examined. A better understanding of the long-term health effects of child maltreatment is needed to inform maltreatment prevention strategies and to improve treatment for children who have been abused or neglected. In this systematic review and meta-analysis, the researchers quantify the association between exposure to physical abuse, emotional abuse, and neglect in childhood and mental health and physical health outcomes in later life. A systematic review uses predefined criteria to identify all the research on a given topic; a meta-analysis is a statistical approach that combines the results of several studies.
What Did the Researchers Do and Find? The researchers identified 124 studies that investigated the relationship between child physical abuse, emotional abuse, or neglect and various health outcomes. Their meta-analysis of data from these studies provides suggestive evidence that child physical abuse, emotional abuse, and neglect are causally linked to mental and physical health outcomes. For example, emotionally abused individuals had a three-fold higher risk of developing a depressive disorder than non-abused individuals (an odds ratio [OR] of 3.06). Physically abused and neglected individuals also had a higher risk of developing a depressive disorder than non-abused individuals (ORs of 1.54 and 2.11, respectively). Other mental health disorders associated with child physical abuse, emotional abuse, or neglect included anxiety disorders, drug abuse, and suicidal behavior. Individuals who had been nonsexually maltreated as children also had a higher risk of sexually transmitted diseases and/or risky sexual behavior than non-maltreated individuals. Finally, there was weak and inconsistent evidence that child maltreatment increased the risk of chronic diseases and lifestyle risk factors such as smoking.
What Do These Findings Mean? By providing suggestive evidence of a causal link between non-sexual child maltreatment and mental health disorders, drug use, suicide attempts, and sexually transmitted diseases and risky sexual behavior, these findings contribute to our understanding of the non-injury health impacts of child maltreatment. Although most of the studies included in the meta-analysis were undertaken in high-income countries, the findings suggest that this link occurs in both high-and low-to-middle-income countries. They also suggest that neglect may be as harmful as physical and emotional abuse. However, they need to be interpreted carefully because of the limitations of this meta-analysis, which include the possibility that children who have been abused may share other, unrecognized factors that are actually the cause of their later mental health problems. Importantly, this confirmation that physical abuse, emotional abuse, and neglect in childhood are important risk factors for a range of health problems draws attention to the need to develop evidence-based strategies for preventing child maltreatment both to reduce childhood suffering and to alleviate an important risk factor for later health problems. Children (NSPCC) is a not-for-profit organization that aims to end all cruelty to children in the UK; Childline is a resource provided by the NSPCC that provides help, information, and support to children who are being abused N The Hideout is a UK-based website that helps children and young people understand domestic abuse N Childhelp is a US not-for-profit organization dedicated to helping victims of child abuse and neglect; its website includes a selection of personal stories about child maltreatment