Methods
Our systematic review was conducted in accordance with PRISMA guidelines6 and registered at PROSPERO (Registration number: CRD42015030125) on 9 December 2015. To obtain representative data on the victim–offender relationship of child homicides, we searched the following databases and electronic resources from first record to the 25 April 2017: Medline, Global Health, Embase, PsycINFO, Social Policy, Popline, Pubmed, Web of Science, LILACS, Medcarib, ADOLEC, Cumulative Index to Nursing and Allied Health Literature, Biosis Citation index, KCI-Korean Journal Database, SciELO citation Index, Western Pacific Region Index Medicus, Index Medicus for the WHO Eastern Mediterranean Region and International Bibliography of Social Sciences IBSS. Controlled vocabulary terms specific to each database were used. For each database, terms included those designed to capture ‘child homicide’ and, where possible, terms to capture perpetrators and to exclude non-relevant article types, such as commentaries. The search strategy is displayed in box 1.
Box 1Detailed search terms
(‘perpetrators’ or ‘perpetrator’ or ‘offender’ or ‘offenders’ or ‘aggressor’ or ‘aggressors’ or ‘father’ or ‘fathers’ or ‘mother’ or ‘mothers’ or ‘parent’ or ‘parents’ or ‘brother’ or ‘brothers’ or ‘sister’ or ‘sisters’ or ‘aunt’ or ‘aunts’ or ‘uncle’ or ‘uncles’ or ‘grandmother’ or ‘grandmothers’ or ‘grandfather’ or ‘grandfathers’ or ‘grandparent’ or ‘grandparents’ or ‘stranger’ or ‘strangers’ or ‘acquaintance’ or ‘acquaintances’ or ‘friends’ or ‘friend’ or ‘partner’ or ‘partners’ or ‘victim-offender relationship’ or ‘victim offender relationship’ or ‘victim-offender relationships’ or ‘victim offender relationships’ or ‘ex-partners’ or ‘ex-partner’ or ‘husband’ or ‘husbands’ or ‘wife’ or ‘wives’ or ‘couple’ or ‘couples’ or ‘boyfriend’ or ‘girlfriend’ or ‘spouse’ or ‘spouses’ or ‘lover’ or ‘spousal’ or ‘boyfriends’ or ‘girlfriends’ or ‘relative’ or ‘relatives’ or ‘family member’ or ‘family members’ or ‘maternal’ or ‘parental’)
AND (‘siblicide’ or ‘neonaticide’ or ‘Infanticide’ or ‘fratricide’ or ‘victim-perpetrator relationship’ or ‘victim perpetrator relationship’ or ‘sororicide’ or ‘family homicide’ or ‘familicide’ or ‘filicide’ or ‘fratricides’ or ‘family murder’ or ‘family homicide suicide’ or ‘familicy’ or ‘family suicide’ or ‘fatal child abuse’ or ‘infant baby dumping’ or ‘child abuse murders’ or ‘child abuse fatalities’ or ‘child abuse fatality’ or ‘parental homicide’ or ‘parental homicides’ or ‘parents who kill’ or ‘mothers who kill’ or ‘fathers who kill’ or ‘grandparents who kill’ or ‘uncles who kill’ or ‘siblings who kill’ or ‘aunts who kill’ or ‘brothers who kill’ or ‘sisters who kill’ or ‘family murder’ or ‘family killing’ or ‘adolescent homicide’ or ‘adolescence homicide’ or ‘adolescent murder’ or ‘adolescence murder’ or ‘adolescent killing’ or ‘adolescence killing’ or ‘child murder’ or ‘child homicide’ or ‘child killing’ or ‘children murder’ or ‘children killing’ or ‘children homicide’ or ‘sibling murder’ or ‘sibling killing’ or ‘sibling homicide’ or ‘infant murder’ or ‘infant killing’ or ‘infant homicide’ or ‘toddler killing’ or ‘toddler murder’ or ‘toddler homicide’ or ‘young adult murder’ or ‘young adult killing’ or ‘young adult homicide’ or ‘sister murder’ or ‘sister killing’ or ‘sister homicide’ or ‘brother murder’ or ‘brother homicide’ or ‘brother killing’ or ‘boy murder’ or ‘boy killing’ or ‘boy homicide’ or ‘girl murder’ or ‘girl killing’ or ‘girl homicide’ or ‘child abandonment’ or ‘infant abandonment’ or ‘toddler abandonment’ or ‘girl abandonment’ or ‘baby abandonment’ or ‘homicide death rates in childhood’ or ‘unnatural sudden infant death’ or ‘unnatural sudden infant deaths’ or ‘childhood homicide’ or ‘childhood homicides’ or ‘child killers’ or ‘fatal maltreatment’)
Studies were eligible for inclusion if they stated a number or proportion of children murdered by distinctly stated perpetrator(s). The definition of homicide followed the definitions used in the individual papers or official statistics. The sample could be based on a country, province or town and be derived from national databases, national representative studies, police, court, mortuary or prison data. Studies were excluded if they only reported data collected from newspaper reports or did not differentiate between attempted and completed homicides. Estimates that combined child homicides with adult homicides were excluded. The citations of included articles were also searched. Two authors (HS and AM-G) screened the 6096 abstracts, and the resulting 563 full texts independently and resolved any disagreements by discussion. One hundred twenty-three studies were finally included (see flow chart in figure 1). HS extracted the data and BD verified each data extraction point for accuracy.
Figure 1Flow chart of the systematic review.
In addition to the systematic review, we surveyed country statistics offices, ministries of justice, home offices or police headquarters of 169 WHO-listed countries to further identify country-level data for child homicide perpetrators and victim–perpetrator relationships. We received responses from 90 countries. Additionally, we made contact with homicide research experts to ask for unpublished data on child homicide perpetrators in their studies. In total, this lead to an additional inclusion of data for 24 countries.
The analysis consisted of two main steps—the selection of one estimate per country year and the calculation of the proportion of child homicides committed by different perpetrator categories for different age groups and by gender separately.
As several countries had more than one estimate available, an algorithm was used to develop a single estimate per country for each of the types of homicide and age group or gender analysed, to avoid potential double counting and to ensure to use the best quality estimates available. Where possible, comparable, non-overlapping data were combined. Otherwise, we chose an estimate according to the following hierarchical order: (1) nationally representative estimates were preferred to provincial estimates, which in turn were prioritised over estimates representing a single town or mortuary, as national studies or larger regional samples would be more generalisable to a country and even out local differences across regions or towns; (2) we chose estimates with more detailed information on the victim–perpetrator relationship, estimates that captured children of varied age groups as this would allow us to group them in a way that would allow comparisons with data from countries that had few perpetrator categories and age groupings; (3) studies that were more recent as it is unknown whether child homicide perpetration trends changed over time across the world; (4) estimates directly from statistics offices were preferred as we assumed that they would have received more checks than data from other sources; and (5) we preferred estimates that covered more years to estimates based on a few years as we assumed they would also be more generalisable and not be influenced by single events.
Approximately two-thirds of the studies only reported combined estimates for up to 20 years on the proportions of child homicide perpetrators, starting with 1970 onwards. The included estimates yielded 51 different victim–perpetrator categories. To facilitate comparisons across countries, we grouped the perpetrators of child homicide into distinct, broad categories: parents, other family members, acquaintances (meaning someone they knew), strangers (capturing people they had no prior contact with) and unknown (not documented perpetrators). As countries used different age categories, countries that only considered children up to age 12 years are also included, if no age category up to 18 years was available.
To calculate the percentage of child homicide perpetrators by country, we divided the total number of homicides in each perpetrator category by the total number of child homicides in that country. To establish the global and regional estimates, the relevant available national percentages were added and divided by the number of countries with available data. For the regional estimates, we grouped countries as per the six WHO regions (Africa, Americas, Eastern Mediterranean, low-income and middle-income Europe, Southeast Asia and Western Pacific). A seventh region was made of the high-income countries from all regions.
After conducting the primary descriptive analysis, secondary analyses investigated parental homicides separately to explore how many children murdered by parents were murdered by their mother or father. Additional analyses investigated the child homicide perpetrators of children below the age of 1 year, neonaticides, adolescents and by gender.
Data were analysed with Stata V.12. Traditional meta-analysis techniques could not be used because nearly all studies were representative of the whole population and not restricted to population samples. Details on individual studies are reported in the supplement (see online supplementary file 1). Because the percentages were skewed, we report the median percentage and the IQR.