PROTOCOL: Risk factors for femicide

Department of Collective Health, Faculty of Medicinal Sciences, State University of Campinas (UNICAMP), Campinas, São Paulo, Brazil International Research Center Hospital Alemão Oswaldo Cruz, Health Technology Assessment Unit, São Paulo, São Paulo, Brazil Applied Health Research Centre, Li Ka Shing Knowledge Institute, St. Michaelʼs Hospital, Toronto, Ontario, Canada Department of Health Sciences, University of Leicester, Leicester, United Kingdom


| Description of the condition
Femicide is the killing of women, girls and baby girls because of their gender. Femicide is the violent death of women based on gender, whether it occurs within the family, a domestic partnership, or any other interpersonal relationship; in the community, by any person, or when it is perpetrated or tolerated by the state or its agents, by action or omission (OAS 2008). Femicide constitutes a violation of women's rights and can be understood as, the ultimate form of violence against women, which ends in the killing of a woman or women affected. The term femicide was used since 1976 by the sociologist Diana Russell, with the objective of emphasizing differences in the characteristics of the women and men homicides (Russell, 2011).
Femicide became frequently used in academic research, epidemiology, public health, politics, social science, laws, policy making even several conceptual theories that have been expanded through observational epidemiological studies.
The concept of femicide could be too confusing and too broad because of the gender component. International researchers have made efforts to investigate about femicide concept (Sanz-Barbero et al., 2016). Aware of the importance of a clear and operational definition of femicide for data analysis and monitoring systems, an extended definition of femicide taking into account the cultural aspects and the possibility of the women act as aggressor and commit femicide. Femicide could be perpetrated by intimate partners, family members, and in rare occasions the perpetrators can be women either lesbian partners or kin (Weil et al., 2018;WHO, 2012). Then, the most recent definition of femicide is: The killing of a woman because some man or men, although occasionally also some women who accept menʼs values, has or have sentenced her to death adducing whatever reasons, motives or causes, but nonetheless actually and ultimately because he or they believe she has defied (the words they often use are "offended" or "insulted") patriarchal order (in their words "honourable" societies) beyond what her judge (often but not always the same person who kills her) is prepared to tolerate without retaliating in that way (Grzyb et al., 2018;Iranzo, 2015).
Femicide is used in general for conceptualized forms of discrimination and violence against women when it is present an unequal power distribution, sometimes with government complicity and as a result of a cultural constructions. Femicide was translated to Spanish as feminicide and it is used in Latin American countries for the characterization of the cases and in laws. In this review, we will use the terms femicide and feminicide as synonyms.
Killing of women based on this gender is a global issue. There are cases in all countries in the world, they are tolerated, accepted, justified and could remain unpunished (ACUNS, 2014;Sarmiento et al., 2014). The killing of a woman by her partner is often the culmination of long-term violence and can be prevented (UNODC, fathers, brothers, mothers, sisters, and other family members because of their role and status as women (UNODC, 2019).
Several types of gender-related killings of women have been identified: intimate femicide, non-intimate femicide, child femicide, family femicide, femicide because of association/connection, unorganized systematic sexual femicide, organized systematic sexual femicide, femicide because of prostitution or stigmatized occupations, femicide because of trafficking, femicide because of smuggling, transphobic femicide, lesbophobic femicide, racist femicide and femicide because of female genital mutilation (Sarmiento, 2014). Also can be included femicide for accusations of sorcery/witchcraft, dowry deaths and female selective abortions. All of them have a gender component.

| Description of the exposition
Gender-related killings are the extreme manifestation of existing forms of violence against women. Such killings are not isolated incidents that arise suddenly and unexpectedly, but represent the ultimate act of violence which is experienced in a continuum of violence (United Nations, 2012). Femicide is the result of multiple, increasing and continuous manifestations of violence, which are rooted in the historical unequal power relations between men and women and in the systemic gender-based discrimination, supported by pseudo-social values, cultural patterns and practices (ACUNS, 2014).
Several risk factors have been identified related to a woman being victim of femicide: prior domestic violence, gun access, estrangement, threats to kill and threats with a weapon, nonfatal strangulation, and stepchild in the home if a female victim. Other risks included stalking, forced sex, and abuse during pregnancy (Campbell et al., 2003).
Some factors have been identifiable in men for abusing his partner too. The world report of violence has collected them and organized in categories as follows (Krug et al., 2002).
• Individual factors: young age, heavy drinking, depression, personality disorders, low academic achievement, low income, witnessing, or experiencing violence as a child, being abused during childhood, absent, or rejecting father.
• Relationship factors: marital conflict, marital instability, male dominance in the family, economic stress, poor family functioning, isolation of the woman from her family.
• Community factors: weak community sanctions against domestic violence, poverty, unemployed, low social capital, delinquent peer association.
• Societal factors: traditional/rigid gender norms/roles, social norms supportive of violence, sense of ownership over women.

| How the exposition might work
The analysis of violence phenomena must recognize the influence of cultural factors constructed around the roles and behavior of men and women and the diminished power of women explained by lack of access to resources. The pursuit of a single explanatory factor is inadequate. Approximations as intersectionality (Sosa, 2017) and the ecological frameworks/model have been applied to conceptualized violence against women integrating individual, situational/relationship, exosystem/community, and macrosystem/societal factors (Heise, 1998;Krug et al., 2002) or theoretical approaches (Corradi et al., 2016).
There are several risk factors for femicide as mentioned above, the major one intimate partner homicide is prior domestic violence (Campbell et al., 2007). Evidence indicates that the majority of gender motivated killings of women are perpetrated by intimate partners or close family members (UNODC, 2011).
Women are 9 times as likely to be killed by an intimate partner (husband, boyfriend, same-sex partner, or ex) than by a stranger (Campbell et al., 2007). Globally, as many as 38% of all murders of women are committed by intimate partners (WHO 2013). Home is the most likely place for a woman to become a victim of homicide (UNODC, 2011).
Women are often emotionally involved and economically dependent on those who victimize them contributing to the perpetuation and acceptance of violence (Ellsberg et al., 2000;Krug et al., 2002). In general, various types of abuse coexist, for example, sexual, physical, economical, moral, patrimonial, and psychological abuse. Manifestations of violence increase with time and become more severe causing the death of the woman. Not all the femicides occur in this context, but most of them do.
The median time that women spend in a violent relationship is around 5-10 years depending on the womanʼs age. Justifications to continue in a violent relationship include fear of retribution, lack of alternative means of economic support, concern for children, emotional dependence, a lack of support from family or friends, an abiding hope that the abusive man will change, and the stigmatization associated with being unmarried (Ellsberg et al., 2000;Krug et al., 2002).
A womanʼs response to abuse is often limited by the options available to her, having into account the lack of positive response of society. Traditional societies defend menʼs rights of physically punishing their wives based on cultural and religious justifications (Ellsberg et al., 2000;Krug et al., 2002).
Violence against women have several and deep overall consequences. Abusive partner relationships have deep impact on womenʼs health (physical, sexual, reproductive, physiological, behavioral, and fatal health consequences (United Nations, 2015)). Fatal health consequences could be: AIDS-related mortality, maternal mortality, femicide and forced suicide (Krug et al., 2002). The resulting damage also extends its impacts on the health of children. In fact, children may suffer a range of behavioral and emotional disturbances, included psychological, social, physical, and academic consequences e.g., post-traumatic stress, attachment difficulties, weight and appetite changes, and drops in school grades (Alisic et al., 2015). The experience of violence episodes during childhood can be associated with perpetrating or experiencing violence later in life (Renner 2006).

| Why it is important to do this review
Femicide is a research priority because of fatal consequences and protection of the women human rights. Femicide is still remarkably prevalent in many cultures and societies.
What constitutes a "femicide or feminicide" is not entirely clear, and how the concept has been defined within observational research is not universally standardized. This systematic review will explore how femicide has been conceptualized and evaluated in currently available case-control and cohort studies.
Furthermore, we will systematically identify factors associated with the risk of femicide.
A better understanding of its risk factors can help the development of interventions as well as novel preventive strategies to mitigate this problem.

| OBJECTIVES
Our main objective is to systematically identify factors associated with the risk of femicide. Besides, we will investigate how femicide has been defined by researchers.

| Review question
How has femicide been defined in the epidemiological case-control and cohort studies? • Studies that report relative risk (RR) or odds ratio (OR) with the corresponding 95% confidence intervals (CIs) or that offer data to perform the calculation.

Exclusion criteria
• Studies without comparation group or case series.

| Types of participants
We will include studies that studied women victims of homicide. No restriction on age will be imposed.

| Types of expositions
We will include cases of femicide described as: • Female homicide/femicide with firearm

| Types of outcome measures
It is important to notice that femicide is a relatively recent denomination of female homicide. We will include cases of female homicide.
In cases of transsexual population, we will consider gender identity (e.g., if the biological sex of the person is male but gender identity is female, the person will be considered a woman). If it is not possible to establish the gender identity, we will include the particular case as gender identity undefined. We will extract 2 × 2 tables, RR or odds ratio-related estimates-depending on the data availability in each study. When data are available as a point estimate (95% CI) derived from two or more multiple regression models, we will extract the estimates from the most complete model (full model).

Primary outcomes
• Definitions of femicide used in the primary case-control and cohort studies.
• Associated factors to be a victim of femicide estimated by RRs or ORs or available data.
• Associated factors for a person to commit a femicide estimated by RRs or ORs or available data.

Secondary outcomes
Types of exposures to be included, observational studies must describe data on risk factors associated with femicide. Whenever possible, risk factors will be further categorized in victim-related and perpetrator-related factors. Examples of specific risk factors to be systematized: • Age

| Search methods for identification of studies
We will search for all published and unpublished studies in the most common medical databases. We will include Google Scholar searcher to retrieve additional studies and relevant references from related systematic reviews.

| Electronic searches
We will use the following database from their earliest dates to March 2020. The search strategies are presented in the appendices:

| Selection of studies
We will download all titles and abstracts retrieved through the electronic search to EndNote online (Clarivate Analytics, 2015) and remove all duplicated references, we will transfer the data to the covidence platform (Covidence systematic review software) for the process of elaboration of the systematic review. Two reviewers will independently examine the referencesʼ reading titles and abstracts for identified primary studies. Disagreements will be resolved through discussion, or, if necessary, by consulting a third researcher.
We will follow the PRISMA statements for the report selection process (Moher et al., 2009). We will use the Review Manager

for protocol and systematic review final text (The Cochrane
Collaboration, 2014).

| Data extraction and management
We will independently extract study characteristics and outcome data from included studies using covidence; disagreements will be resolved through discussion, or if necessary, by consulting a third researcher. We will contact investigators by email to request further data on methods or results.
With data extracted by a standardized format (Table 1), we will elaborate tables of included, excluded ( Table 2) and ongoing studies (Table 3). We will include studies based on type of studies (case-control and cohort) and we will include all the relevant studies, regardless of the usability of the reported data. When several publications of the same study are found, we will choose the publication with more information and we will exclude the others, register them in the table of excluded studies, justified by the duplicated data presented.
We will compare the magnitude and direction of effects reported by studies using forest plot graphics and evaluating coherence of the data, trying to identify typographical errors in the studies reports.

| Assessment of risk of bias in included studies
For evaluation of the methodological quality of the studies, we will include the Newcastle Ottawa scale (Wells et al., 2019) for casecontrol (Table 4) and cohort studies ( Table 5). Process of assessment of risk bias will be in an independent way by the same authors who extracted data from the studies, and discrepancies resolved as mentioned above. We will provide a risk of bias table with the judgment and the justification of each item, and a summary graph of bias for each study and for all of them.
We will mention quality of information in the results for the readers to be aware of possible bias derived from que quality of the information.

Binary outcomes
Dichotomous data will be analyzed using numbers of events of each study, RRs or ORs, with 95% CIs.
T A B L E 1 Table for data extraction   Supporting Information CAICEDO-RAO ET AL.

Continuous outcomes
Continuous data will be analyzed with means (or mean changes) and standard deviations and will be summarized via standard mean differences.
In cases of missing standard deviations, we will recalculate them from the reported statistics provided in these studies (e.g. CIs, standard errors, p values).
If there are available data for two or more studies, we will combine available data for each outcome. We will combine risk ratios/odds ratios or standardized mean differences using a random-effects model with the restricted maximum-likelihood estimator of between-study variance. If there are more than 10 studies, contour-enhanced plots will be constructed and statistical tests of funnel plot asymmetry will be performed (Eggerʼs test and Harbordʼs test). Heterogeneity will be assessed using both Cochranʼs Q test and the I 2 statistic. We will conduct sensitivity analyses and subgroups analyses in order to investigate possible sources of statistical heterogeneity. If the inconsistency is not explained by sensitive or subgroup analysis, and more than 10 studies are included in the meta-analysis, a meta-regression will be performed (Higgins et al., 2003).

| Unit of analysis issues
The unit of analysis is per dead woman.

| Dealing with missing data
We will attempt to contact study authors to obtain missing data. In cases of missing standard deviations, we will recalculate them from the reported statistics provided in these studies (e.g. CIs, standard errors, p values).

| Assessment of heterogeneity
Heterogeneity will be assessed using both Cochranʼs Q test and the I 2 statistic.

| Assessment of reporting biases
Review authors will aim to minimize the potential impact of reporting bias by ensuring the inclusion of the most important databases and resources to find relevant publications through the comprehensive search for eligible studies and by staying alert for duplication of data.
If we include 10 or more studies in an analysis, we will use a funnel plot to explore publication bias and investigation of the relationship between effect size and study precision.

| Data synthesis
We will provide summary estimates of the strength of association between risk factors and femicide. Summary results will be obtained via a random-effects models using the restricted maximum-likelihood estimator of between-study variance.
If there are more than 10 estimates per risk factor, countourenhanced plot will be constructed and statistical tests of funnel plot asymmetry will be performed (Eggerʼs test and Harbordʼs test).
Heterogeneity will be assessed using both Cochranʼs Q test and the I 2 statistic. We will conduct sensitivity analyses and subgroups analyses in order to investigate possible sources of statistical heterogeneity. If the inconsistency is not explained by sensitive or subgroup analysis, and more than 10 studies are included in the meta-analysis, a meta-regression will be performed.

| Subgroup analysis and investigation of heterogeneity
We plan to perform a sensibility analysis based on the methodological quality of the studies and epidemiological design of the studies and analyses data for victims and aggressors.

| Sensitivity analysis
We will perform sensitivity analysis to determine the effects of the studies judged to be at "high" or "unclear" risk of bias.

| Overall quality of the body of evidence
We will prepare a summary of findings

| Reaching conclusions
We will elaborate conclusions based only on findings from the synthesis (quantitative or narrative) of studies included in the review; we will avoid recommendations, but will recognize the implications of the findings for decision-making, and we will talk about the remaining uncertainties.

ACKNOWLEDGEMENTS
We are grateful to the Social Welfare Campbell Collaboration Group and to the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Capes).

CONTRIBUTIONS OF AUTHORS
Monica Caicedo-Roa: Conceived the review question, developed, completed and advised the protocol.
Tiago Da Vega Pereira: Developed, completed and advised the protocol.
Ricardo Carlos Cordeiro: Edited and advised the protocol.

DECLARATIONS OF INTEREST
The authors declare no conflict of interests