Introduction

Child maltreatment is common and causes substantial adverse health, educational and behavioural consequences [1]. Understanding its prevalence and characteristics in nationwide populations is essential to plan and evaluate interventions to reduce maltreatment. However, measurement of child maltreatment is known to be far from universal, and when performed must confront methodological challenges. This study systematically reviews the most comprehensive national studies of the prevalence of child maltreatment, and critically appraises their methodologies to help inform future measurement.

Child maltreatment in its five recognised forms is a major public health issue [2]. Physical and mental diseases are caused through proximal and distal pathways. Immediate physical injuries and conditions include brain injury and failure to thrive, and a panoply of psychological disorders include anxiety, depression, and suicidality. Studies have found serious effects of physical abuse [3,4,5], sexual abuse [6,7], emotional abuse [5,8–10], neglect [5,9,11], and exposure to domestic violence [12–14]. Experiencing multiple forms of maltreatment is common [12], and is associated with more severe outcomes [14,15], including alcohol and drug abuse, mental illness, interpersonal violence, and sexual risk taking [16].

The adoption of coping mechanisms such as smoking, alcohol and drug use can compound the damage by causing diseases in the medium to long term; and chronic stress can cause coronary artery disease, pulmonary fibrosis, and inflammation [17–22]. Potent mediators include prolonged psychological distress, risky behaviours, social withdrawal and dysfunction, impaired cognitive development, low educational and occupational attainment, and interpersonal relationship difficulties. A growing body of evidence is showing child maltreatment affects brain development, shortens telomeres, and produces epigenetic neurobiological changes [23–26]. The disease and economic burdens are substantial: a recent estimate of the cost of disability-adjusted life years (DALYs) lost across East Asia and the Pacific was 1.88% of the region’s GDP, equating to $194 billion in 2012 dollars [27].

As a global policy imperative, the United Nations recognises the gravity of child maltreatment and its consequences. The United Nations Agenda for Sustainable Development includes a target of ending abuse of children [28]. Reliable scientific data on national prevalence is essential to measure progress against this goal, and to inform policy efforts aimed at prevention, early identification and response [29–30].

Nationwide studies of the experience of childhood maltreatment can identify baseline prevalence stratified by maltreatment type, as well as important contextual features including the child’s sex, age, and relationship with the abusive person. Without good measurement techniques and repeated measures over time, we lack understandings of baseline measures, of whether maltreatment is increasing or declining, of changes in maltreatment types over time, and of the efficacy of policy and practice interventions designed to reduce child maltreatment for the whole population and for key sub-populations.

Despite the necessity for good data in public health generally and in child maltreatment in particular, approximately half of all countries have failed to report any kind of robust prevalence estimates [2], and extant studies are often limited to measuring one or few maltreatment types [31]. Accordingly, prevalence estimates are often inadequately specified, and are almost certainly underestimated. In addition, existing studies vary widely in design, sample and methods, and often use non-standardized instruments [5,32]. Where an instrument is non-standardized and untested, the risk may be heightened that the study will fail to capture experiences that constitute maltreatment, and may capture experiences that do not constitute maltreatment, hence producing unreliable results. Importantly, the use of unsound maltreatment constructs and operational definitions also compromises the reliability of recorded measures [33–34]. As an example of this, studies of sexual abuse that do not include non-contact sexual abuse will underestimate prevalence; conversely, studies that include as sexual abuse genuinely consensual acts between peers will overestimate prevalence. Similarly, studies of neglect that do not consider medical neglect will underestimate prevalence. Studies of emotional abuse that include non-abusive yelling will overestimate prevalence.

Optimal methodologies for measuring population characteristics of child maltreatment can ensure adequate detail is captured to yield reliable, detailed, useful data. For best quality estimates, prevalence studies should adopt robust conceptual understandings of maltreatment types and their operational definitions [33]. In addition, prevalence studies need to ask a series of items to obtain accurate data, rather than a single question which will tend to underestimate prevalence [35]. Similarly, to avoid underestimates, items should be behaviourally specific, rather than vague, ambiguous or non-specific [36]. All national prevalence studies face methodological and practical challenges, and studies take different approaches [2,12,14,30]. Ideally, however, all five forms of child maltreatment should be measured simultaneously, since many children experience such poly-victimization and its heightened consequences [1,14,16]. To provide nuanced, useful information, studies should ask about prevalence, and about the specific nature of the acts, their severity, frequency, and timing, and the relationship of the child to the person inflicting the abuse [33]. These factors influence health outcomes and provide evidence about specific risk and protective factors and how these may best be targeted. Rigorous measurement of child maltreatment is complex, but is essential to inform prevention efforts and drive nationwide social change [2,14,29,36,37].

Recent research has reviewed global prevalence estimates [2,31], the nature of population health surveys exploring consequences of child maltreatment [37], and approaches in studies of youth [38]. However, to date, there has not been a systematic review and methodological appraisal of high quality national population prevalence studies of child maltreatment to provide a baseline for future measurement efforts.

This study aimed to investigate three questions. First, what national studies have been conducted of the prevalence and nature of all five, or at least four, major forms of child maltreatment? Second, what methodologies were used in these studies? Third, what does a critical analysis of these studies indicate about the methodological rigour, quality, and practical viability of different approaches? The results of our investigation can inform future efforts to generate baseline prevalence estimates, design policy responses, and chart trends over time, as more societies confront the challenge of childhood maltreatment.

Methods

Screening

As detailed in our search strategy (S1 File), in Phase 1, MS, JD and ED screened records by title. We removed duplicates using electronic software (Endnote), and removed remaining duplicates about the same study, selecting the publication providing the most detailed account. In Phase 2, BM and RP independently screened records by title and abstract. Disagreements were discussed between BM and RP to achieve consensus. To identify any further potential eligible studies at this stage that may not have been captured in the search, all co-authors considered if there were any further known studies requiring inclusion that were not in the Phase 2 shortlist. In Phase 3, BM and RP independently assessed full text of screened in articles. Disagreements were discussed between BM and RP to achieve consensus, with reasons recorded. We screened reference lists of included studies to identify any further potential eligible studies. We used a translator to assist in screening non-English studies. This process resulted in 23 eligible studies (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flow diagram.

https://doi.org/10.1371/journal.pone.0227884.g001

Results

Systematic review

This review identified 23 articles reporting the results of national studies of the prevalence of all five or four of the recognized forms of child maltreatment. One of these articles reported the results of a study conducted simultaneously in nine countries in the Balkan Peninsula, and eight of these national studies met our eligibility criteria [46]. Accordingly, in total, our review identified 30 national studies, conducted in 22 countries. Studies were published between 2005 and 2019. Extracted data revealed study location, scope, participants, data collection method, and instrument. Table 1 presents the extracted information from included studies. The supporting information details the prevalence rates reported by each study (S5 File).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of study characteristics.

https://doi.org/10.1371/journal.pone.0227884.t001

There were four studies in the USA [47–50], three in the UK [51–53], two in Hong Kong [54–55] two in Taiwan [56–57], and two in Germany [58,59]. There was one study in Denmark [60], the Netherlands [61], Switzerland [62], Japan [63], Suriname [64], Saudi Arabia [65], Israel [66], South Africa [67], and Hungary [68]. In the Balkans study [46], eight met eligibility criteria based on the number of types of maltreatment studied: Albania, Bosnia & Herzegovina, Bulgaria, Croatia, the Former Yugoslavian Republic of Macedonia, Greece, Romania, and Serbia; in general for our purposes, we treat these as one study. The Turkish study involved three forms of maltreatment, so was excluded from our analyses.

Fourteen studies measured all five maltreatment types [47–51,53,56–57,61,64–68]. Of nine studies measuring four maltreatment types, seven omitted EDV [46,52,58–60,62–63], and two omitted sexual abuse [54–55]. Eleven studies measured prevalence throughout childhood and in the past year; nine measured prevalence through childhood only, and three measured past year incidence only.

Only nine studies explored all five types of maltreatment across childhood, defined as aged under 18 [48–50,53,64–68]. These studies occurred in seven countries (USA, UK, Suriname, Saudi Arabia, Israel, South Africa and Hungary), and only three involved a sample of adults providing data about experiences over their entire childhood [53,65,68]. Four studies in Germany, the UK and Japan obtained information from adults about all maltreatment across childhood except EDV [52,58–59,63].

Eight studies involved only child participants aged under 18 providing self-report data. Three studies included child and adult participants each providing self-report data. Five studies involved a household’s child participant aged under 18 providing self-report data (four involved children aged 10–17 and one involved children aged 11–17) and the household’s parents providing proxy data about a child aged under the cut-off. Five studies involved only adults providing self-report data (24 year olds; 18–24 year olds; 20–49 year olds; 18 and over). Sample sizes ranged from 1094 to 12,035 participants. Five studies adopted measures to recruit high-risk sub-populations [48,56,60,62,64].

Seven studies were conducted in schools: Taiwan [56–57], the Netherlands [61]. Switzerland [62], Suriname [64], and the Balkans study [46]. Eleven studies were conducted in households by interviews, in Hong Kong [54–55], Hungary [68], the UK [51–53], Germany [58–59], Japan [63], Saudi Arabia [65], and South Africa [67]. Five studies used remote computer assisted telephone interviews (CATI), with four in the USA [47–50], and one in Denmark [60]. Data collection time ranged from 1 month to 2 years.

Methodologies to recruit the sample and accommodate high-risk subpopulations also varied. In most studies, the target population was a close representation of the national population. Studies in schools were done in countries with high school attendance. All studies used random selection. However, few studies used strategies to capture participants from culturally and linguistically diverse groups, or from high-risk groups such as those in out of home care.

Response rates for household studies generally ranged from 56% to 78%, with one reporting a participation rate of 94.8% [67]. Rates in school-based studies showed schools’ participation rate ranging from 49%-79%, and then with almost 100% response rates from children in participating schools. Response rates in CATI studies ranged from 60% to 79.5%, with more recent studies having lower rates [47–49].

Regarding consent to participate, 18 of the studies involved child participants exclusively or with adult participants. Nine studies involved only child participants; in these, two required only the child’s consent [56,62], one required the child’s consent and parental passive consent [64], one required the child’s consent and either passive or active parental consent [46], and five required parental active consent and the child’s consent [54–55,57,61,66–67].

Of the studies involving child participants, seven reported the measures used by research teams when a child was suspected to have been harmed or at risk [46–50,53,67]. Nine studies reported other measures to assist any distressed participants [46,48,52,54,56,60,62,64].

Studies used a range of instruments and approaches to measuring each maltreatment type. Table 2 presents key data extracted from the instrument used in each study. Comprehensive details about the maltreatment items are detailed in the supporting information (S3 File).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Key features of instruments used in prevalence studies.

https://doi.org/10.1371/journal.pone.0227884.t002

Eight studies used the Juvenile Victimization Questionnaire (JVQ). These studies used different versions of the JVQ, either using its original form [72], an enhanced form [48–50], or an adapted version [53,62,66–67]. Two studies used the Conflict Tactics Scale Parent-Child version (measuring physical abuse, emotional abuse, and neglect), and the CTS2 (EDV) [54–55]. Two studies used the ICAST-C, in either its original form [56] or an adapted version [46]. Two studies used the Childhood Trauma Questionnaire [58–59]. Single studies used the Adverse Childhood Experiences International Questionnaire [65], the Adverse Childhood Experiences questionnaire [68], and the Lifestyle and Attitudes Towards Sexual Behavior instrument [63]. Four studies used a blend of instruments [51,57,61,64]. Two studies used self-developed instruments [52,60].

Six studies did not report psychometric data on instrument validity and reliability. Six studies reported psychometric data on the instrument as used [46,54–56,58,72]. Studies using enhanced or adapted versions of instruments generally cited the original instrument’s data but did not report further psychometric tests.

Most studies did not define overarching concepts of each form of maltreatment, instead operationalising these concepts into questions about the participant’s experiences. Approaches to some but not all forms of maltreatment broadly aligned with the nature of maltreatment concepts as established by the scholarly literature. Approaches to physical abuse and sexual abuse were generally sound. Approaches to the construct and operationalisation of emotional abuse were generally sub-optimal, with some exceptions (e.g., [46,52]). Neglect was also rarely well-operationalised, with some exceptions (e.g., [49,52–53,58–59,66].

Studies explored maltreatment experiences in varying depth, reflected by the number and nature of questions asked (Table 2). For sexual abuse, 12 studies asked between five and eight questions. Most studies asked about the relationship with the person inflicting the abuse, and the nature of the acts; more than half asked about frequency; but few asked about severity. Other notable differences included: two studies being limited to sexual abuse by a parent/guardian [51,60]; most studies including contact and non-contact acts, but three studies included contact abuse only [62,65,68]; four studies asking only one question [51,62–63,68].

For physical abuse, eight studies asked only one question, although these included multiple distinct concepts [47–51,62,63,68]. Six studies asked between five and nine questions. Most asked about relationship and nature; more than half asked about frequency; but few asked about severity. A notable difference was in the treatment of spanking on a child’s bottom: seven studies excluded “spanking on your bottom” from the definition of physical abuse [47–50,53,62,66]; four studies included spanking with a bare hand as physical abuse [46,54–56]; and four studies included as physical abuse being hit or spanked on the bottom but only when done with an implement or hard object [51,52,57,64].

For emotional or psychological abuse, eight studies asked between five and eight questions. Most asked about relationship and nature; more than half asked about frequency; but few asked about severity. Other notable differences included: three studies being limited to a single generic question [51,61,64]; seven studies using a single compound question [47–51,62,67]; and only two studies using a detailed scale of items closely aligned with a sound conceptual model [46,52].

For neglect, 12 studies asked between five and 11 questions. Five studies asked one question [47–48,62,63,68]. Most asked about relationship and nature; more than half asked about frequency; but few asked about severity. Six studies asked detailed questions about multiple dimensions of neglect, and their severity [49–50,52,58–59,66]. Other notable differences included: some studies operationalising neglect very broadly, including a parent having low aspirations [51], or not helping with homework [64]; only one study asking about educational neglect [64]; and one study omitting physical and nutritional neglect [46].

For exposure to domestic violence, six studies asked between six and eight questions. Most asked about relationship and nature; more than half asked about frequency; but few asked about severity. Notable differences were: two studies used the comprehensive CTS2 scale of 39 items originally devised for use with adult couples [54–55]; and the original JVQ had two physical assault items [72], and later added six items about threats or property damage by other family members [48–50].

Risk of bias

Table 3 sets out the quality assessment and scoring results for each study. Scores ranged from 6 to 10. Most studies had relatively high internal and external validity. We concluded that studies scoring 9.5 or 10 had minimal bias. Five studies achieved scores of 10: two in Hong Kong [54–55], and one each in Taiwan [56], Israel [66] and South Africa [67]. Five studies achieved scores of 9.5: three in the USA [48–50], one in the UK [53], and the Balkans study [46]. Five other studies achieved scores of 9, from Saudi Arabia [65], the UK [52], Germany [62], Hungary [68], and Taiwan [57]. Four studies scored 7, and two scored 6; here we concluded risk of bias was high, particularly regarding selection bias and non-response bias.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Quality assessment of included studies.

https://doi.org/10.1371/journal.pone.0227884.t003

Discussion

This systematic review identified 30 studies of the prevalence of either four or five forms of child maltreatment, conducted in 22 nations. In addition, many other studies have been conducted of three or fewer maltreatment types, such as studies of sexual, physical and emotional abuse. These have been conducted on a stand-alone basis [76], or as part of a systematic campaign supported by a global public private partnership [77]. By 2019, the Violence Against Children Surveys (VACS), which also measure the prevalence of physical, sexual and emotional abuse, had been conducted in 16 countries and were being planned in a further eight countries in Africa, Asia and the Caribbean [30,77–78]. Other studies have considered the prevalence of a mixture of peer violence and maltreatment by parents or caregivers [79–80]. Accordingly, a good deal of evidence has been generated about the prevalence of child maltreatment in several dozen nations, representing substantial progress in the international understanding of the epidemiology of child maltreatment. However, this review has highlighted the fact that the vast majority of nations lack reliable benchmark national prevalence data on a comprehensive assessment of maltreatment, including measurement of four or five of the recognised five types of maltreatment, and almost all lack follow-up studies to establish trends over time. This study demonstrates the urgent need to conduct more rigorous prevalence studies, particularly those by measuring all relevant types of maltreatment, to generate more accurate understandings of the extent of maltreatment, and to enable progress in reducing child maltreatment against the SDG target.

Our review also shows that there is substantial variation in study participants across the different studies, limiting comparability and introducing certain strengths and limitations which are important to consider in designing future work. Several studies obtained data using parents as proxies for children under 10, and reported reliable responses. This approach may capture data about very young children’s experiences that is otherwise unattainable, although accurate estimates rely on parents being both knowledgeable and truthful in their responses [47]. Yet, the literature reports no evidence of reporter bias in comparisons of adult proxy and youth self-report data [47,48].

Arguably, from a public health perspective, studies provide most comprehensive and reliable estimates when capturing prevalence data over the entire span of childhood up to age 18. Furthermore, where a study’s participants are children and or young adolescents, past year incidence data is useful. Over half of the studies in this review included children as respondents. In these studies, responses benefitted from being direct and proximate to the experience as well as capturing useful stratified data about single year incidence in a closely contemporaneous time period. Developmental evidence suggests children’s and adolescents’ participation is entirely appropriate. While adolescents may generally differ from adults in the attainment of psychosocial capacities to understand long-term consequences, regulate conduct, and withstand social and emotional pressures, their cognitive capacity is not substantially different from that of adults [81–84]. Similarly, apart from those still in early developmental stages, children’s cognition and reliable episodic memory is sufficiently developed to enable participation in survey research [85–86]. This justifies the design of instruments for child and adolescent participants, including the careful approach of the developers of the Juvenile Victimization Questionnaire in designing an instrument suitable for participants as young as eight [72].

Ethically, there is no impediment to involving child and adolescent participants [87]. Adolescents and children are cognitively capable of providing their own consent, and are ethically entitled to do so as autonomous individuals. Moreover, adolescents and children have rights to freedom of expression, and bear the right of participation in matters affecting them. While there remains no consensus on the most justifiable approach to confidentiality and welfare [87–90], we assert that studies can adopt robust measures to balance imperatives of attaining sufficient study participation, while ensuring participant welfare and confidentiality. While confidentiality is a foundational principle in these studies, the exception to this, conveyed to youth participants at the outset, that cases of current or imminent significant risk of danger may be referred to welfare authorities, has been found not to affect response rates [38,53]. Alongside this, studies can adopt stepwise approaches drawing on multiple psychological and legal resources to support participants who disclose severe incidents or who experience distress [87]. However, it is important not overstate the frequency of distress. Several studies have found low rates of distress among youth participants in studies of maltreatment, and the level of youth distress does not differ significantly from that of adults. Furthermore, even distressed participants mostly maintain their involvement was worthwhile [38,91]. A recent US study, for example, found only 0.8% of participants aged 10–17 reported being “pretty or a lot” upset by answering the questions, and even this did not unduly affect their reported willingness to participate [91]. An associated finding is that children in high-risk sub-populations, such as those in out-of-home care, have not been well represented, leading to likely underestimates of prevalence and scarce evidence about specific risk profiles.

Studies that rely on adults’ retrospective accounts offer the substantial benefit of capturing data about experiences across childhood. One limitation of such studies is that they will not obtain recent proximal data of single year incidence. An additional potential limitation, yet to be fully analysed, may be that retrospective accounts are affected by various kinds of recall bias. We acknowledge that some have argued that retrospective studies do not provide data about child abuse experiences that is as accurate as prospective studies [92–93] and have cautioned against sole reliance on retrospective accounts, especially where prevalence estimates are used to draw causal inferences about the effect of maltreatment on biomedical diseases. A recent systematic review and meta-analysis concluded that prospective and retrospective measures of childhood maltreatment identify different groups of individuals [94]. However, it was also recognised that prospective measures may have lower sensitivity than retrospective measures of the experience of maltreatment, and concluded that “the low agreement between prospective and retrospective measures cannot be interpreted to directly indicate poor validity of retrospective measures” and that retrospective measures could have greater ability to identify true cases [94]. The well-known discrepancies between true maltreatment rates and those recorded in many data sources used for prospective studies is attributable to the low correlation between actual experiences and their representation in official data such as crime statistics and child protection service records. Few maltreatment experiences are ever brought to the attention of criminal justice agencies or child protection services. The caution urged regarding retrospective reports appropriately appears more directed towards studies considering causation of disease than estimation of population prevalence. It is also accepted that lack of validity tends to underreport the experience of abuse [95–97], and studies of test-retest reliability regarding retrospective accounts have indicated general stability over time [98]. We acknowledge that retrospective reports may have compromised validity for various reasons, including motivational factors and memory biases, and measurement features including poorly worded questions [92,94]. Overall, however, our view is that retrospective studies of child maltreatment, especially when well-designed with behaviourally-specific questions grounded in sound constructs of maltreatment, with representative samples of the population, offer the opportunity to obtain sufficiently accurate estimates of the prevalence of child maltreatment experiences.

The fourth finding is that while considerable investment is required for all kinds of approach, viable approaches to survey administration are available for diverse geographical settings to accommodate large and small nations, and attain sufficient participation. The implications of this are clear for future study design. School-based studies appeared most often in small nations, which may more readily facilitate centralised educational sector endorsement for the research, or which may have a high commitment to social research. When school leaders agree for their school to participate, children generally participate at a very high rate. Similarly, household studies identified in this review generally occurred in small nations. Both school-based and household studies require substantial numbers of staff, but may be most feasible where labour costs are manageable and where the social ecology is of sufficient strength to support and perhaps even require direct personal involvement in such research. In larger nations, for reasons of practicability and cost, studies used CATI and achieved satisfactory response rates. Perhaps for reasons of cost, and practical difficulty, a challenge remains to capture the experience of culturally and linguistically diverse sub-populations, and hard to reach groups such as children who are not in school, or who are in out of home care. Future research could consider optimal local strategies to respond to this challenge.

Our fifth finding is that selection, design and testing of an appropriate instrument appears an enduring challenge. In this regard, two coexisting needs must be balanced by any study: first, to be practicable in terms of the time and cost required to design, test and administer an instrument and minimise missing data; and second, to achieve sufficient comprehensiveness and ensure construct validity by describing maltreatment types in a way congruent with conceptual understandings [33]. Our review showed that a wide variety of instruments have been used, with psychometric data often not reported. The JVQ was the instrument most often used in either full-form or short-form; moreover, several studies adapted the original JVQ, sometimes adding a considerable number of items. These adapted versions did not appear to have been subjected to testing. While inconsequential modification of a proven instrument obviates the need for re-testing, substantial modification may be further supported by cognitive testing and test-retest reliability. What is relatively clear is that a proven, sound instrument offers both practicable and methodological benefits over a blended tool, or a new unproven instrument.

Our sixth finding is that instruments must soundly operationalise constructs of each maltreatment type by describing them in a way congruent with sound conceptual understandings. This review and critical appraisal found that instruments most often adopted unsound constructs and operationalisation of neglect, and emotional abuse. In particular, many studies did not consider sufficient operational categories of these maltreatment types as required by sound conceptual models, which will lead to under-estimates of prevalence, and will miss the opportunity to capture important information about the nature of specific experiences. Other studies used broad or vague conceptual expressions, which will have the opposite effect of over-estimating prevalence. This finding provides a contextual demonstration of the problem of unsound constructs compromising reliability and validity in general [33,34], and of the ongoing challenge to this field to adopt sound constructs of maltreatment and sound behaviourally-specific examples of these constructs [99]. Additionally in this regard, many studies asked only one question about a maltreatment type, which leads to underestimates of prevalence [36]. Single-item assessment, even through a compound question involving multiple elements of a construct, cannot capture accurate or nuanced data and should be avoided wherever possible. Finally, we found few questions about educational neglect. Arguably, since education is a human right recognised by the United Nations Convention on the Rights of the Child article 28, and is a condition for human flourishing [100] and a protective factor against multiple adversities such as child marriage [101], this is a significant dimension of neglect warranting greater priority. We recommend particularly close attention to how future studies conceptualise and operationalise these forms of maltreatment.

A seventh finding is that few studies asked detailed follow-up questions about the child’s relationship with the person inflicting the acts, and the severity and frequency of the acts. Generally, studies using the JVQ asked the most detailed follow-up questions. Obtaining information about the severity, frequency, timing, and relational setting of abuse and neglect is important, since the closeness of the relationship between the person maltreating the child and the child can have significant effects [102–103], and the timing of maltreatment is also important, with studies finding effects for both sex and age [104]. From a public health perspective, the measurement of maltreatment should ideally move beyond raw prevalence, and yield sufficiently sensitive and nuanced information about these key contextual features of the maltreatment to inform future public health policy and prevention efforts, including the indication of priority areas for responses. The addition of such questions presents challenges for instrument design and implementation, including the time to administer additional questions. However, we recommend such questions wherever possible.

Conclusions

This systematic review and analysis has shown nationwide studies of the prevalence of child maltreatment have been conducted, using methods of administration suited to the setting, and involving child participants, adult participants, or both. However, there are few such nationwide studies of all five or even four maltreatment types, leaving substantial gaps in knowledge about the experience of childhood maltreatment in nearly all countries. Overall, our review and analysis indicates many of the completed studies are generally sound, but some take a more comprehensive and conceptually robust approach to provide nuanced, useful data for researchers and policymakers. To enable measurement of progress against the United Nations Agenda for Sustainable Development Goal 16 of reduction of child abuse, many countries need to invest in robust national prevalence studies. Such studies should measure exposure to domestic violence in addition to physical abuse, sexual abuse, emotional abuse, and neglect. Studies should use an instrument with demonstrated validity and reliability, and must ensure maltreatment types are operationalised appropriately in the questions asked. If participants are children or adolescents under age 18, studies should capture past year incidence, as well as childhood prevalence. Information should be captured about the specific nature, severity and frequency of the maltreatment, and the relationship of the child to the person who inflicted the acts. Such data can best inform the development and monitoring of nationwide prevention efforts.

Supporting information

Acknowledgments

We acknowledge Juliet Davis, Elizabeth Dallaston, and Andrea Boskovic for providing research assistance. We also thank the journal reviewers for their helpful comments.