Prevalence of physical abuse by caregivers toward infants aged 0–24 months: a systematic review and meta-analysis

Summary

Background

Although the prevalence of physical abuse in childhood is well studied, the extent among infants–a period of high vulnerability–remains poorly characterized. We aimed to estimate the prevalence of physical abuse of infants (<24 months) by caregivers, using data from anonymous self-report studies.

Methods

A systematic review and meta-analysis were conducted following the Cochrane Handbook. We searched MEDLINE, Embase, PsycINFO, CINAHL, and Web of Science (inception to Sep 3rd, 2025) for studies anonymously reporting the prevalence of physical abuse toward infants (0–24 months) by primary caregivers. Two reviewers independently screened studies and assessed risk of bias using the Joanna Briggs Critical Appraisal tool for prevalence studies. Prevalence estimates were aggregated using Bayesian multilevel logistic regression and heterogeneity quantified with 95% prediction intervals. The protocol was registered with PROSPERO (ID: CRD42023459469).

Findings

We identified 20 relevant studies (>220,000 caregivers) with prevalence estimates spread across 16 types of abuse, spanning North America, Europe, and East Asia. The pooled prevalence of reporting at least one form of abuse in a typical sample was 4·8%, 95% CI [2·6%, 7·5%] or 3·9%, 95% CI [1·9%, 6·4%] excluding “lesser” forms of abuse (e.g., spanking)–but with notable heterogeneity, 95% prediction interval [0·6%, 11·9%]. Aggregate estimates ranged from <3% for severe forms of abuse (e.g., shaking, hitting) to 9·5% for spanking or 20·5%–21·0% for slapping on the wrist (which might be culturally sanctioned in some areas).

Interpretation

Approximately one in twenty infants in a typical sample worldwide experience physical abuse by caregivers, though estimates vary and are likely underreported. Even in the first two years of life, many children face violent caregiving practices, pointing to an urgent need for prevention. The large gap between self-reports and official statistics highlights that most infant abuse remains hidden.

Funding

No funding was received for this research.

Research in context.

Evidence before this study

We conducted the first systematic review and meta-analysis focused specifically on the prevalence of physical abuse of infants (0–24 months). Despite longstanding recognition of the harms of early childhood abuse, no prior synthesis had systematically quantified how often physical abuse occurs in the first two years of life–a critical developmental period when children are uniquely vulnerable and unable to protect themselves or report maltreatment.

Added value of this study

As the first meta-analysis of the prevalence of infant-abuse, this report provides significant novel information which will be of use to clinicians, scientists, policymakers and others. In this study, we synthesized the pooled prevalence of reporting at least one form of abuse, as well as estimates for individual forms of abuse. Our findings highlight the sad reality that a substantial proportion of infants worldwide experience physical abuse by caregivers. The use of rigorous methodology in adherence to PRISMA guidelines and the requirement of anonymous self-report data provide a robust approach to capturing hidden instances of abuse not detected through official registries. Our use of Bayesian logistic regression further accommodated low prevalence estimates and incorporation of reasonable information in the form of priors (e.g., the prevalence in a typical sample would be <50%). Importantly, even so-called “mild” actions (e.g., spanking) are not without consequence, particularly for infants.

Implications of all the available evidence

From a policy perspective, our findings carry significant implications. The fact that approximately one in twenty infants in a typical sample experience caregiver-inflicted physical abuse represents a pressing public health concern. Given the critical nature of infancy as a developmental period for establishing lifelong physical, cognitive, and emotional foundations, our findings underscore the urgency of targeted early interventions and strong policy responses. Countries lacking legislative prohibitions on corporal punishment should consider adopting such laws as part of broader, multifaceted prevention strategies. Beyond legislation, global efforts should prioritize standardized assessments of infant abuse, encourage transparent reporting, and improve parental education and access to support services. Achieving the Sustainable Development Goal of ending violence against children requires prioritizing the youngest and most vulnerable. This will depend on robust public health action, caregiver support, and legislative clarity. Protecting infants from violence is both a moral imperative and foundational to supporting lifelong health and human rights.

Introduction

Child abuse is a pervasive global issue, with physical abuse affecting millions of children and causing both immediate and long-term harm.¹ The World Health Organization defines child physical abuse as: "… the intentional use of physical force against a child that results in, or has a high likelihood of resulting in, harm to the child's health, survival, development or dignity".² This broad definition includes behaviors ranging from culturally sanctioned corporal punishment, such as spanking and slapping a child's wrist, to severe physical assault, such as beating and shaking.³ Such behaviors have been clearly demonstrated to cause lasting harm, including developmental delays, mental illness, increased healthcare utilization, and reduced life expectancy.4, 5, 6 Recent global estimates indicate that up to 1 billion children aged 2–17 years have experienced some form of abuse or neglect in the past year.¹

Physical abuse in infancy (0–24 months) is especially concerning due to infants' heightened vulnerability to the enduring consequences of maltreatment.7, 8, 9, 10 Abuse during this sensitive stage can disrupt critical neural connections and processes, significantly elevating the risk for adverse outcomes, such as cognitive impairment, emotional dysregulation, and behavioral disorders.^9,11, 12, 13 Moreover, infants have the highest rates of death from abuse and neglect compared to any other age group. This makes their maltreatment both a public health emergency and a policy priority.¹⁰ However, prior prevalence studies of child abuse have often focused on broader age ranges (e.g., 0–5 years or 0–17 years) or older children. We failed to identify any prior systematic reviews or meta-analyses of the prevalence of child abuse in which the period of infancy was disaggregated. This may be due to a perception that infants are less likely to be victims of physical abuse than older children, possibly due to their immobility, limited interaction, or the assumption that abuse is more often a response to child misbehavior.^14,15 Cross-national comparisons further underscore the issue: Gilbert and colleagues¹⁰ found that child protection indicators are trending upwards in five of six high-income countries, with the largest increases seen in infants. These findings suggest that risks in this age group are both substantial and often underrecognized, though some increases may also reflect improved detection and early intervention, highlighting the role of surveillance systems in shaping apparent prevalence trends.

Despite these concerns, the true prevalence of caregiver-inflicted physical abuse in infancy remains poorly characterized. Existing estimates vary widely, in large part due to methodological differences, particularly whether abuse is assessed anonymously.16, 17, 18 Assessing child abuse anonymously is critical for maximizing disclosure.^19,20 When parents are aware that their disclosures could result in negative consequences (e.g., reports to child protective agencies) they are less likely to respond honestly. Consequently, estimates of child abuse collected without ensuring anonymity are substantially lower than those from anonymous surveys.¹⁹ Large national surveys using non-anonymous methods, for example, report rates of physical abuse below 1–2%.^10,21,22 Official child welfare reports and registries also likely underestimate the true incidence of infant abuse, since most abuse (especially less severe physical punishment) goes unreported in formal systems.²⁰ Anonymous caregiver self-report surveys reveal otherwise hidden cases, yielding prevalence rates of 1·3% to 9·0% (excluding “mild” forms), and up to nearly 30% (including “milder” forms such as slapping on the wrist).^17,23,24 These discrepancies highlight the critical need for reliable measurement methods and a deeper exploration into the prevalence of infant-specific abuse.

Beyond measurement challenges, definitional disagreement also contributes to inconsistent estimates, particularly regarding whether commonly used disciplinary behaviors such as spanking or slapping should be classified as physical abuse. The inclusion of such behaviors within the definition of physical abuse has been subject to ongoing debate.³ While these actions are often framed as normative forms of discipline in many cultures, an extensive body of research now demonstrates their clear association with more severe forms of maltreatment.^3,25, 26, 27 A large meta-analysis of over 160,000 children found that spanking was strongly and consistently associated with detrimental child outcomes.³ It is also important to note that as parental spanking intensifies, the likelihood of more extreme physical abuse increases correspondingly.25, 26, 27 Conceptually and empirically, this positions physical discipline on a continuum of violence rather than as a distinct category. While not all forms of physical discipline are immediately life-threatening, even “mild” actions such as slapping an infant's wrist carry significant developmental and psychological risks and can escalate to more severe abuse.^3,28,29 Recognizing this continuum provides a more accurate understanding of the full spectrum of caregiver-perpetrated harm during infancy.

With these concerns in mind, the present study aimed to (a) determine the meta-analytic aggregate prevalence of physical abuse of infants (<24 months) reported anonymously by caregivers; and (b) identify potential sources of heterogeneity in current prevalence estimates. To our knowledge, this will be the first systematic review and meta-analysis to focus specifically on the prevalence of physical abuse of infants (0–24 months). This study will clarify existing prevalence data and inform more targeted prevention strategies, contributing to the global agenda to end child violence outlined in the United Nations Sustainable Development Goals (SDGs).¹

Methods

We conducted this systematic review and meta-analysis in accordance with the Cochrane Handbook for Systematic Reviews of Interventions,³⁰ and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.³¹ The protocol was registered with PROSPERO (ID: CRD42023459469). There was no funding source for this research.

Eligibility criteria

Studies were included if they met the following criteria:

• •Participants: Caregivers of infants aged 0–24 months; a caregiver was defined as any individual who assumes the responsibility for providing care for the infant most of the time.
• •Exposure: Caregiver-perpetrated physical abuse defined broadly (e.g., spanking, beating).
• •Outcome: Prevalence of physical abuse, defined broadly, measured via anonymous caregiver self-report.
• •Study Designs: Cross-sectional surveys, cohort studies, baseline data from intervention trials, and descriptive or analytical epidemiological studies were eligible. Clinical or high-risk populations were excluded. Representative, community-based or population-based samples were included.
• •Setting and Timeframe: No geographic, language or publication-date restrictions. While our initial protocol stated that only English-language articles would be included, we revised this criterion during the review process to include studies in any language, provided sufficient methodological and outcome data could be extracted. This decision was made in light of increased accessibility to translation tools and AI-based support, which allowed us to reliably assess non-English studies for eligibility at the title/abstract level. For any studies that passed through the title/abstract level screening, human translators would be accessed to provide validation and ensure accuracy.

Search strategy

The search strategy was developed in MEDLINE alongside a medical research librarian using Medical Subject Headings (MeSH) and keywords for the terms: infant, caregiver, child abuse, violence, prevalence, and self-report. Comprehensive searches were conducted using MEDLINE, Embase, PsycINFO, CINAHL, and Web of Science. The MEDLINE search was translated to four other databases using relevant database-specific subject headings and syntax. MEDLINE's complete search strategy ran in August of 2023 is provided as a representative example (Appendix A). The initial search was conducted September 2023, and updated in August 2023, October 2024 and August 2025.

Selection process

As outlined in Fig. 1, studies were imported into Covidence and duplicates removed. All citations were screened by at least two independent reviewers. After a pilot exercise to ensure consistency across reviewers, title and abstract screening was conducted independently by members of the review team, with full-text screening completed by a subset of reviewers. At every stage, discrepancies were resolved through discussion and consensus between reviewers.

Fig. 1.

PRISMA flow diagram showing the study selection process.

Data extraction process

Data extraction was conducted by two independent reviewers using a standardized Microsoft Excel form. Disagreements were addressed through discussion or, when necessary, adjudicated by a senior team member.

Extracted data included: (1) bibliographic information (authors, year, publication type); (2) study setting (country, income level) and design; (3) sampling strategy and representativeness (national vs. regional, etc.); (4) sample characteristics (sample size, caregiver types/respondents, infant age range, response rate); (5) definitions and measures of physical abuse (specific behaviors queried, time frame of recall, measurement instrument used such as the Conflict Tactics Scale); (6) prevalence estimates—including overall “any physical abuse” prevalence and, when available, prevalence of specific types of abuse (e.g., shaking, spanking, etc.); and (7) risk of bias assessments.

Authors were contacted for any missing data or to request additional clarifications essential for meeting inclusion criteria (e.g., prevalence estimates specific to children aged ≤24 months). In instances where multiple publications analyzed data from the same sample, the study with the most comprehensive prevalence estimates (i.e., reporting a greater number of abuse types) was included to avoid duplication. When a study reported multiple relevant prevalence estimates (e.g., for different types of abuse or different infant age subgroups), each estimate was extracted. For prospective and longitudinal studies including prevalence estimates across more than one time point, the earliest time frame was preferred given it tended to have the largest sample size and facilitated our focus on infancy.^32,33

Moderator analyses

Moderator analyses were considered only if at least 10 studies were available (or 5 per condition for categorical variables).³⁴ Seven exploratory moderators were examined: a) year of study, b) region (Asia, Americas, Europe), c) source (community, routine check-up, survey), d) inclusion age range in months, e) percent employed, f) percent married, and g) number of different types of abuse assessed. Seven additional moderators could not be analyzed given insufficient estimates (e.g., mean age of caregiver or child, gestational age or percent preterm infants) or a restriction of range (child sex: percent male 47–53%; birth order; primiparous/percent first child: primarily 45–54%).

Study risk of bias assessment

Quality assessment was conducted independently by two reviewers using a modified version of the Joanna Briggs Institute (JBI) Critical Appraisal Tool for Prevalence Studies,³⁵ with scores out of 8 (rather than 9) due to the exclusion of the item on anonymity (a requirement for inclusion). Each study was assigned a quality rating from 0 to 8, classifying studies as low (7–8), moderate (4–6), or high risk of bias (0–3; Appendix B). Conflicts were resolved through discussion and joint review of criteria.

Effect size calculation and analysis

No effect size calculations were necessary prior to analysis because statistical models were fit using the number of individuals reporting a given form of abuse as implemented via a Bayesian multilevel logistic regression model using brms 2.22.0³⁶ within R 4.4.0.³⁷ Each model included random effects for both study and the individual estimates, as appropriate. Prediction intervals were calculated reflecting the range of probable prevalence estimates expected in a new study “typical” of the methods and demographics included in the present studies. Prediction intervals are the most meaningful measure of heterogeneity presently available (e.g., IntHout et al., 2016).³⁸ In all cases, prevalence estimates were back-transformed and reported as percentages for ease of interpretation (for priors see Appendix C). We did not evaluate publication bias because it is rarely applicable to epidemiological data (Borenstein, 2019, p. 173).³⁹ For further information pertaining to our modeling approach, please refer to earlier works by our group.40, 41, 42

Role of funding source

No funding support was provided for this research.

Results

Searches identified 21,522 unique records. After screening titles and abstracts, we reviewed 115 full-text articles, of which 20 met our inclusion criteria (Fig. 1). These 20 studies formed the basis of our systematic review and meta-analysis.^{16,17,23,24,32,33,}43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56

Description of studies

The included studies (see Table 1 for study characteristics) were published between 1998 and 2025, and sampled caregivers from seven countries. Studies employed a cross-sectional (n = 8), retrospective cohort (n = 4), prospective cohort (n = 4), longitudinal (n = 2), RCT (n = 1) or quasi-experimental (n = 1) design. All studies relied on anonymous caregiver self-report. Mothers were the respondents in 13 studies, and other caregivers in the remaining 7.^{17,18,24,45,50,52,55} Sample sizes ranged from 73⁵⁴ to over 95,000⁵³ caregivers and in total encompassed reports on over 220,000 infants and caregivers worldwide. Individual participant data were unavailable, so multivariate meta-analysis was not possible.⁴¹

Table 1.

Studies reporting prevalence of infant abuse.

Note. Additional prevalence data provided through author correspondence for Fujiwara et al., 2016a.

Abbreviations: CTS = Conflict Tactic Scale; CTSPC = Parent-Child Conflict Tactic Scale – Parent Child; JVQ = Juvenile Victimization Questionnaire; PBQ = Parenting Behaviors Questionnaire; PRCM = Parental response to child misbehavior questionnaire; RCT = Randomized Controlled Trial.

^aThe respondents were only reported for the entire sample of parents whose children were newborn to 12 years of age (versus parents with children under 2 years as retained for analysis). ^bThere was no referent period established for parents to report spanking (e.g., past month, past year, etc.); child's age was 4–9 months, and 10–18 months. ^cComposition of respondents (mothers, fathers, grandparents, and other guardians) was reported across the entire sample of caregivers of children aged 4–35 months. ^dEight types of abuse measured in total (any corporal punishment): hit or slapped, shook, threw something at the child, pushed, bit, kicked or punched, and tried to or actually hit the child with an object. ^eTwo types of abuse measured in total: shaking and pinching. ^fSix types of abuse measured in total: shaking, hitting, spanking, burning, choking, and slapping. ^gSix types of physical abuse measured in total, with only four verbally described in the article: shook, slapped, kicked, and burned baby. ^hInfant abuse was reported by parents of infants 1, 3, and 6 months old (independent samples, stratified by infant age). ⁱSlap child on the hips was coded as spanking, whereas slap child on the hand was coded as slap minor. ^jFor mothers with 12-month old children, three types of abuse were reported: spanking (with hand), spanking with object (which was coded as hit on bottom with object), and slapping (slap minor [hand], and slap [face]). ^kOnly shaking was included based on the mother's self-report of her own behavior; seven other types of abuse were based on the mother's self-report of both her own and her partner's behavior.

Operational definitions of abuse

The operational definition of physical abuse varied broadly across studies. Twelve focused on severe physical abuse (e.g., shaking or beating), but excluded spanking or other forms of corporal punishment.^23,24,32,43, 44, 45^{,47,49,50,53,55,56} The remaining included any corporal punishment (e.g., spanking, slapping), including severe (e.g., hitting with objects; Table 1). Of the 20 studies, 12 measured the prevalence of caregivers reporting at least one form of physical abuse (“any” abuse),^16,17,23,24,44, 45, 46^{,48,49,52,55,56} whereas the remaining estimates corresponded to individual forms of abuse. Given the substantial heterogeneity in how individual abusive behaviors were defined and measured across studies, we provide an overview of operational definitions and reported frequencies in Appendix D.

Risk of bias

Across the 20 included studies, quality scores ranged from 2 to 8 on the adapted JBI prevalence appraisal tool. Overall, methodological quality was mixed, with most studies falling into the moderate or low risk of bias categories.

• •Low risk of bias (scores 7–8): 9 studies met criteria for low risk.44, 45, 46, 47, 48^,51,55,56 These studies generally used appropriate sampling frames, clear measurement methods, and robust statistical approaches.
• •Moderate risk of bias (scores 4–6): 10 studies fell into the moderate range.^{17,24,32,33,43,49,50,}52, 53, 54 These studies typically had limitations related to sampling methods, response rates, or measurement validity, but nonetheless provided sufficiently reliable prevalence estimates.
• •High risk of bias (scores 0–3): Only one study,¹⁶ was classified as high risk of bias, primarily due to a limited sampling frame, inadequate measurement methods, and insufficient detail on recruitment and response rates.

The prevalence of reporting at least one form of abuse

As depicted in Fig. 2, our analysis of the prevalence of reporting at least one form of abuse revealed an aggregate back-transformed prevalence of 4·8%, 95% CI [2·6%, 7·5%]. The prediction interval ranged from 0·6% to 11·9%, suggesting that if a new study with similar methods were to be conducted, the “true” prevalence of abuse in that study should fall between ∼0.5% and ∼12%. As noted in Fig. 2, there was one apparent outlier—Regalado et al. (2004, 10–18 month sample).¹⁷ This sample included spanking and exhibited a much higher prevalence than other studies (see below for further discussion of spanking). A sensitivity analysis excluding this estimate revealed an aggregate back-transformed prevalence of 4·2%, 95% CI [2·4%, 6·4%] and a reduced prediction interval ranging from 0·6% to 10·5%. A model excluding all estimates containing data pertaining to what might be viewed as “lesser” forms of abuse (e.g., spanking, slapping on the wrist) revealed an aggregate back-transformed prevalence of 3·9%, 95% CI [1·9%, 6·4%] and a prediction interval ranging from 0·4% to 11·0%.

Fig. 2.

Back-transformed prevalence (%) of reported overall abuse, shaking, spanking, smothering, hitting or slapping of a child. Note. Circles and error bars reflect estimated prevalence and 95% confidence intervals as derived from our model with X's reflecting the reported empirical prevalences. Circle size reflects relative sample size. Studies are sorted from lower to highest estimated prevalence. Squares reflect aggregate estimates (i.e., prevalence in a typical study) with thick lines reflecting 95% confidence intervals and thin lines reflecting 95% prediction intervals. Note also that the range of the axes differ to avoid obscure variability.

Recognizing that the implication of physical abuse differs across the development cycle, a further analysis was conducted limited only to samples with a maximum inclusion age of 12 months or lower. This model revealed an aggregate back-transformed prevalence of 4·3%, 95% CI [2·7%, 6·4%], 95% PI [1·7%, 7·8%] for samples based on all forms of abuse (10 effects across 8 studies) or 3·8%, 95% CI [2·0%, 6·1%], 95% PI [1·4%, 6·6%] for samples excluding “lesser” forms of abuse (7 effects across 5 studies). There were insufficient samples here and below to consider children 12 months or older because studies tended to include only infants or a broader range of ages.

Although the intention was to explore potential moderators within these data, all of the prospective variables we sought to explore were either reported too infrequently (e.g., among fewer than 5 studies) or demonstrated too little variability to warrant investigation. The same was true for the individual forms of abuse. Across the moderators with sufficient data, none demonstrated a credible association. Further data are necessary before sources of heterogeneity may be explored.

The prevalence of individual forms of abuse

As depicted in Fig. 2, separate meta-analyses were next conducted for individual forms of abuse for which at least four articles were available. Although in principle a random-effects model can be fit with as few as two estimates, this practice is widely discouraged because the between-study variance and resulting pooled estimate are highly unstable with so little information; as there is no universally accepted minimum, we chose a threshold of four to balance precision with inclusion. All remaining forms of abuse are instead summarized narratively in the following section. Further, for slapping, definitions varied across studies and estimates, with some asking whether caregivers slapped their children (without elaboration as to where) while others specifically asked whether they slapped their children on the wrists (which may be viewed as less severe). These estimates were separated prior to analysis, with the slapping data analyzed meta-analytically and the slapping on the wrist data considered qualitatively.

As noted in Fig. 2, the prevalence of most abusive acts was low and approximately ∼1–3%, including slapping, 1·1%, 95% CI [0·3%, 2·1%], hitting, 1·5%, 95% CI [0·4%, 3·5%], smothering, 1·9%, 95% CI [1·3%, 2·5%], and shaking, 2·3%, 95% CI [1·4%, 3·4%]. The one exception was spanking, 9·5%, 95% CI [3·9%, 17·4%], which was more common than the remaining categories: Infants are 5·6, 95% CI [2·0, 10·4] times more likely to experience spanking compared to the average of slapping, hitting, smothering and shaking. Prediction intervals were similarly consistent, with slapping, hitting, smothering and shaking exhibiting tight prediction intervals ranging no higher than ∼6%. That is to say that the prevalence of these forms of abuse would be expected to be no more than 6% in any sample, similar to those included in our models, and the prevalence could be as low as <1%. However, spanking exhibited a broader range, with a prediction interval ranging from 1·2% to 22·1%. That is to say that within some samples, almost a quarter of caregivers might report spanking their child, whereas others might exhibit prevalence estimates as low as 1·2%.

As earlier, we recognize that the implications of different forms of physical abuse differs across the development cycle and endeavored to estimate the prevalence of abuse amongst children 12 months of age or younger for each individual form. All samples included in our analysis of smothering already fit this definition. Of the remaining forms of abuse, only shaking provided enough data for our model (13 effects across 11 studies). This model revealed an aggregate back-transformed prevalence of 2·7%, 95% CI [1·5%, 4·2%], 95% PI [0·6%, 6·2%]. There were insufficient samples including only children between 12 and 24 months of age to fit comparable models.

Qualitative review of abuses with fewer than four studies

Note that although one study included pinching in its overall estimate of abuse prevalence,⁴⁹ the individual prevalence was unavailable for us to evaluate qualitatively.

Hitting with an object

Both estimates measuring hitting with an object were 0% and derived from the same sample⁴⁵ with one additional estimate of hitting the child on the bottom with an object that was also 0%.³³

Burning, choking, and biting

There was one estimate for burning,⁴⁸ choking,⁴⁸ and biting⁴⁵ each of which were 0%.

Kicking, punching, and throwing

There were two estimates available for kicking,⁴⁵ two for punching,⁴⁵ and two for throwing,⁴⁵ all of which were 0%.

Beating

There was one estimate for beating, which was 7.2%.⁴³ This estimate was higher than expected, but it should be noted that the sample size was quite small (n = 361) and the definition was vague enough that some caregivers may have interpreted it as including spanking.

Pushing

There were two estimates for pushing,⁴⁵ both of which were higher than might be expected, at 9·3% and 17·8%.

Slapping on the wrist

For slapping on the wrist, two estimates^33,54 were available at 20·5% and 21·0%. Notably, these estimates were much higher than the slapping estimates synthesized earlier (which were thought to reflect slapping on the face or other vital area) and instead were more similar in magnitude to spanking.

Discussion

To our knowledge, this is the first systematic review and meta-analysis to focus specifically on the prevalence of physical abuse of infants. In our analysis we focus on infants aged 0–24 months. Synthesizing anonymous caregiver self-reports from international studies, we estimate that approximately one in twenty infants in a typical sample are subjected to physical abuse by caregivers during their first two years.

Our pooled estimate of ∼4.8% of caregivers reporting at least one physically abusive act toward an infant substantially exceeds what is visible in official systems. For example, in U.S. administrative data, infants have the highest recorded maltreatment burden, but even then, the estimate is ∼ 2.5% across all maltreatment types in 2021 (not just physical abuse)–the highest of any age group but still roughly half our estimate.⁵⁷ Cross-national syntheses also show why such a gap might be expected.¹⁰ Child-protection agency indicators (notifications, investigations, substantiations, out-of-home care) vary five-to ten-fold across countries, while “hard” outcomes (maltreatment-related injury admissions and violent deaths) vary far less (generally < two-fold), underscoring how detection and policy shape official counts as much as underlying occurrence.¹⁰

These findings have important implications for abuse prevention. The finding that a substantial proportion of physical abuse of infants is not detected by child-protection agencies underscores the central importance of primary prevention (i.e., preventing abuse before it occurs). While secondary prevention (i.e., intervening to prevent further harm once abuse has been identified) remains critical in cases that come to official attention, it necessarily reaches only a subset of affected infants. Together, these comparisons suggest that anonymous self-report captures forms of caregiver-perpetrated physical aggression in infancy that are systematically under-ascertained in administrative, clinical, and legal data sources, and that different data sources capture different segments of the spectrum, with implications for both primary prevention (reducing incidence) and secondary prevention (improving identification and response when harm is suspected).¹⁰ This discrepancy likely reflects differences in case definitions and thresholds for detection and reporting, as well as the fact that many behaviors—particularly those without immediate injury or external visibility—are more likely to be disclosed anonymously than to come to the attention of services.

Variability in reporting standards and the nature of abusive acts highlights the importance of clear, simple, and unambiguous case definitions that balance sensitivity and specificity.² WHO guidance further emphasizes the value of consistently reporting classification categories (for example, substantiated, suspected, or unsubstantiated) to minimize both false-positive and false-negative case identification.² Recent critical appraisals of the evidence base underpinning abusive head trauma diagnosis have also highlighted challenges related to case classification, including the potential for circular reasoning and incorporation bias when expert opinion or non-independent criteria are used as reference standards.⁵⁸

The use of rigorous methodology and the requirement of anonymous self-report data provide a robust approach to capturing hidden instances of abuse not detected through official registries. Our use of Bayesian logistic regression further accommodated low prevalence estimates and incorporation of reasonable information in the form of priors (e.g., the prevalence in a typical sample would be <50%; see Appendix C). Nevertheless, several limitations must be considered. Measuring child abuse is inherently difficult and even under anonymity, it is often underreported due to fear or social desirability.^52,59 Thus, the figures we report might still underestimate the true rates of abuse. Although there is a theoretical possibility that abuse may be over-reported, we deem this extremely unlikely given the strong social prohibition against these behaviors and the fear of punitive consequences (e.g., child protective services involvement) that most parents have about disclosing these behaviors. Additionally, wide variability across studies underscores both methodological challenges and potential cultural or linguistic influences that must be considered when interpreting these findings.

Observed heterogeneity may stem from how physical abuse is defined across studies, with different behaviors of varying severity reported in different combinations. Prevalence tended to be higher in studies including “milder” disciplinary actions (e.g., spanking, slapping)^17,18,33 and lower in those focusing only on severe maltreatment (e.g., shaking, smothering),^32,46,47,50 however, this trend was not consistent across all studies.^24,53 It is important to distinguish between the developmental impact of mild versus severe forms of physical aggression. Behaviors such as shaking or forcefully striking an infant carry immediate risks of serious injury, including intracranial bleeding, respiratory compromise, and long-term neurological impairment.^60,61 In contrast, behaviors like slapping or spanking may appear less severe but are still associated with elevated risk of escalation to physical abuse and with wide-ranging developmental consequences, particularly for infants.^3,27,28 Research increasingly recognizes that spanking and slapping (even on a non-vital area such as the wrist) can have harmful developmental and psychological effects.^3,28,29 Inconsistent classification of such behaviors across studies highlights the need for standardized definitions to improve surveillance.

A further methodological challenge lies in the variation of recall periods employed across studies, which ranged from the ‘past week'³³ to the ‘past month'^43,46 and up to the ‘past year'^24,62 (Table 1). This heterogeneity likely contributes to a conservative aggregate estimate. Short recall periods (e.g., one week or month) minimize recall bias but fail to capture the cumulative prevalence of abuse across the full 24-month developmental period of infancy. Conversely, studies employing longer retrospective timeframes (e.g., ‘ever’ or ‘past year’) may suffer from recall decay, where caregivers, particularly those reporting on ‘milder’ forms of discipline, may fail to remember isolated incidents from months prior. Consequently, whether due to narrow windows of observation or memory decay over longer periods, the current figures likely underestimate the true lifetime prevalence of physical abuse among infants.

Cultural and legal contexts shape beliefs about child discipline and influence both caregiver behavior and reporting.62, 63, 64, 65 Over the past several decades, developed countries' tolerance of child maltreatment has decreased drastically.^10,66 However, physical actions such as slapping or shaking remain seen as normative rather than abusive in some cultures.⁶⁵ Even in the studies included in this review, hitting, shaking and slapping were often framed as ‘discipline’ measures, rather than abuse,^17,43,47 with some explicitly reporting that many parents saw “mild” physical actions as discipline rather than abuse.¹⁸ The normativity of these behaviors is linked to greater use of physical discipline and a reduced likelihood of identifying these behaviors as harmful.^18,62,67, 68, 69 These cultural dynamics highlight the complexity of accurately measuring physical abuse across diverse settings and the importance of context-sensitive prevention strategies. Efforts to reduce infant abuse must therefore consider both legal reform and cultural change, particularly in societies where physical punishment is widely accepted.

Measurement inconsistency was another limitation. Only four studies used the Conflict Tactics Scale,^24,45,49,56 while others relied on diverse or non-validated tools. The scope of abuse assessed also varied widely, even within studies using the same instrument. Additionally, some surveys were behavior-specific (e.g., slapping in response to crying), while others were general.⁵¹ Given that infant crying is a known trigger for abuse,⁵² its absence from some surveys may have led to underreporting.¹⁸ These inconsistencies suggest that current prevalence estimates may miss specific instances of abuse, particularly when measured under narrow or non-comparable conditions.

Our estimates may also underrepresent abuse in households with multiple young children. Several studies instructed caregivers to report on one focal child,^33,54 while others excluded families with multiple births,⁵³ potentially omitting abusive incidents involving other infants in the same household. This limits the representativeness of data, particularly in cultures with larger families.

From a policy perspective, our findings carry significant implications. The fact that approximately one in twenty infants in a typical sample experience caregiver-inflicted physical abuse represents a pressing public health concern. Given the critical nature of infancy as a developmental period for establishing lifelong physical, cognitive, and emotional foundations, our findings underscore the urgency of targeted early interventions and strong policy responses. Notably, only one of the studies included in our review evaluated an intervention using a randomized controlled trial.⁴⁷ The experimental group watched an educational video about infant crying and the risks of shaking and smothering of infants. Control participants received care as usual. Shaking was reported less often in the intervention group (0.19% compared with 1.69% in the control group). Although self-reported smothering was also less frequent in the intervention group (0.66% vs 1.00%), this difference was not statistically significant. Given the low baseline prevalence of both shaking and smothering among control participants, it is likely that very large samples would be required to detect statistically significant differences of the magnitude observed. Alternatively, the intervention may have been more effective in targeting behaviors explicitly emphasized in the educational content (i.e., shaking), with less impact on smothering-type behaviors. Similar limitations have been noted in broader evaluations of crying-based prevention programmes, which suggest that education alone may be insufficient to influence all forms of severe caregiver response to infant distress.⁷⁰

Taken together, these findings highlight both the promise and the limitations of single-component educational interventions. Countries lacking legislative prohibitions on corporal punishment should consider adopting such laws as part of broader, multifaceted prevention strategies. Beyond legislation, global efforts should prioritize standardized assessments of infant abuse, encourage transparent reporting, and improve parental education and access to support services.

Future research should address the significant knowledge gaps identified in our study. A significant limitation of the current literature is the lack of data from low- and middle-income regions, including Africa, South America, and South Asia. These areas remain critically underrepresented despite the likelihood that cultural norms, legal frameworks, and caregiving practices differ markedly from those in higher-income countries. This gap not only restricts the generalizability of existing prevalence estimates but also reflects a broader, systemic shortfall in global research efforts. Expanding studies in these regions, using consistent methodologies, and evaluating culturally sensitive interventions are urgent priorities for achieving a more representative understanding of infant abuse worldwide.

Similarly, most caregivers investigated were mothers, with no samples focused solely on fathers or gender-diverse caregivers (e.g., male-identified birthing parents, gender queer or gender non-conforming birthing and non-birthing parents). Future research is needed to compare differences in infant abuse prevalence between birthing and non-birthing parents, as well as to explore prevalence estimates among gender-diverse caregivers.

Further, future meta-analyses will benefit from larger and more diverse datasets that allow exploration of moderator variables we were unable to examine in the current review. While we planned to analyze moderators such as child sex, birth order, and gestational age, these could not be evaluated due to insufficient estimates or narrow variability across studies. As more data become available, it will be important to assess whether infant characteristics (e.g., being male, firstborn, or born preterm) or caregiver role are associated with differential risk of physical abuse.

Greater attention is also needed to refine how physical abuse is measured and reported. Future studies may benefit from including questions on the perceived acceptability of various forms of physical aggression (e.g., spanking versus hitting with an object), as behaviors seen as socially acceptable may be more commonly endorsed. Moreover, incorporating items that assess participant comfort and candor in responding to abuse-related questions (e.g., “To what extent did you respond to these questions in a fully honest manner?”) could improve the interpretability of self-reported data and reduce the likelihood of underreporting.

In conclusion, this review highlights a troubling reality: a substantial proportion of infants worldwide experience physical abuse by caregivers. While prevalence estimates vary, the findings underscore clear pathways toward prevention. Consistent with WHO guidance,² prevention should be multi-level, combining upstream public health approaches (e.g., reducing caregiver stressors), strengthened caregiver supports, and clear policy and legislative frameworks to reduce violent caregiving practices in the earliest years of life. Achieving the United Nations Sustainable Development Goal¹ of ending violence against children requires prioritizing the youngest and most vulnerable. This will depend on robust public health action, caregiver support, and legislative clarity. Protecting infants from violence is both a moral imperative and foundational to supporting lifelong health and human rights.

Contributors

CK: Conceptualization, Data Curation, Investigation, Methodology, Project Administration, Supervision, Visualization, Writing—Original Draft, Writing—Review and Editing. NF: Conceptualization, Methodology, Supervision, Investigation, Resources, Project Administration, Writing—Original Draft, Writing—Review and Editing. VK. Conceptualization, Methodology, Investigation, Validation, Writing—Original Draft, Writing—Review and Editing. BS. Methodology, Data Curation, Investigation, Project Administration, Writing—Original Draft, Writing—Review and Editing. EF. Methodology, Supervision, Validation, Data Curation, Writing—Review and Editing. AV. Software, Formal Analysis, Investigation, Data Curation, Writing—Review and Editing. CX. Methodology, Data Curation, Investigation, Writing—Review and Editing. CL. Methodology, Investigation, Data Curation, Writing—Original Draft, Writing—Review and Editing. JF. Conceptualization, Methodology, Software, Investigation, Validation, Visualization, Resources, Data Curation, Supervision, Project Administration, Formal Analysis, Writing—Original Draft, Writing—Review and Editing. All authors had full access to all the data in the study.

All authors had full access to all the data in the study. CK, BS, EF, AV, CX, CL, and JF accessed and verified the underlying data.

Data sharing statement

All data extracted for this study and the analytic code used in the meta-analysis have been posted to a publicly accessible GitHub repository (https://github.com/jmfawcet/infantabuse_public). Additional details are available from the corresponding author upon reasonable request.

Declaration of interests

The authors declare no competing interests.