Past-year intimate partner violence perpetration among people with and without depression: an individual participant data (IPD) meta-mediation analysis

Systematic review and IPD meta-mediation analysis. The IPD meta-mediation analysis is registered on PROSPERO (CRD42018082258). The PRISMA-IPD statement has been followed [17].

Potentially eligible studies for the IPD meta-mediation analysis were identified through systematic review searches. Studies were eligible to begin data harmonisation if: study participants were aged 16 and older; they measured depression using either a validated diagnostic tool, validated screening tool, or clinical diagnosis; they measured past-year IPV perpetration; and were representative general population surveys conducted in a high-income country setting. We excluded data from low-income country settings because of likely differences in contextual and societal factors relevant to both IPV perpetration and mental health, such as gender norms, the acceptability of gendered violence [18, 19], income inequality [20], access to healthcare [21], and political instability [22], which would have likely resulted in too much heterogeneity and thus prevented the proposed analyses from being conducted. Individual participant data needed to be available for re-analysis.

The exposure of interest for this review was depression within the past 12 months, defined either in accordance with ICD [23] or DSM [24] criteria (any edition). Unexposed/comparison groups comprised individuals without depression. The outcomes of interest were physical IPV perpetration in the past 12 months, psychological IPV perpetration in the past 12 months, sexual IPV perpetration in the past 12 months, and coercive controlling behaviours in the past 12 months. IPV was defined using as any incident or pattern of incidents of controlling, coercive, threatening behaviour, violence, or abuse between those aged 16 or over who are, or have been, intimate partners regardless of gender or sexuality (UK Home Office, 2012). Mediators of interest were alcohol misuse and drug use within the past 12 months, and IPV victimisation within the past 12 months.

A number of variables were considered putative confounders as they might affect variables in two or more of the following categories: exposures, mediators, or outcomes. These were age, education level, income, relationship status, number of children, childhood abuse or maltreatment (experienced before the age of 18), ethnicity, employment, housing status, and other violent offending [14, 20, 25‐30]. Separate mediation models were fitted for males and females; previous analyses have demonstrated that men are more likely to perpetrate and women are more likely to be the victims of IPV that is injurious, frequent, sexual, and occurring within a context of controlling and coercive behaviour [31].

The search strategy and screening process for identifying studies for inclusion in the IPD meta-mediation analysis was a two-stage process. The first stage involved searching for papers for a systematic review and aggregate data meta-analysis (not reported here), and the second stage involved additional steps to identify studies that fulfilled eligibility criteria for the current IPD meta-mediation analysis. Details are provided in the supplementary information.

Publicly available datasets were downloaded. Access was sought for datasets not in the public domain, either via application to data repositories or principal investigators. All contributing data were cross-sectional. Where studies collected data over multiple waves, one wave was chosen, selected on the basis of participant age (e.g., in which participants were more likely to be or have been in intimate relationships) and/or sample size.

For each eligible study, the variables of interest (exposures, outcome, and hypothesized mediators), as well as variables (covariates) that might act as confounders or stratifiers were extracted from the IPD datasets provided and combined into a master database. To enable pooling, a number of data harmonisation rules were applied. Data harmonisation was only undertaken for outcome, exposure, and mediator variables for which data were available in at least three datasets, and for confounders for which data were available in all datasets being pooled. Details of the variables used for each exposure, outcome, and mediator are provided in the supplementary material, along with how each variable was assessed in each contributing study.

Statistical modelling aimed to assess the effect of mental disorders on past-year IPV perpetration and investigate the presence of indirect effects via various hypothesized mediator variables. All analyses were stratified by sex. Mediators were considered separately. Separate analyses were conducted for each combination of exposure (depression), mediator (alcohol misuse, drug use, any past-year IPV victimisation, and severe IPV victimisation), and outcome variables (any physical IPV perpetration, and severe physical IPV perpetration). Age, education, income, relationship status, and number of children were controlled for as potential confounders in the meta-mediation analyses.

Two-stage IPD meta-mediation analyses were conducted. In the first stage, mediation analyses were run for each individual study, producing both total and direct and indirect effects of exposure on outcome, adjusted for potential confounders. In the second stage, study-level summary statistics for the total, direct, and indirect effects for each exposure, mediator, and outcome combination were combined using fixed-effects meta-analyses, using Stata’s <metan> command [37]. Effect heterogeneity was assessed using the I² statistic. Sensitivity analyses explored the impact of using fixed- versus random-effects meta-analysis; the outputs were minimally different (see supplementary information).

The first stage of the IPD meta-mediation analyses was carried out in Stata using the command <medeff> [38], which uses a simulation approach to conduct causal mediation analysis and can accommodate binary mediators and outcomes [30]. For binary data, the parametric model assumed for the mediator and the outcome is a probit model. As a result, total, direct, and indirect effects are expressed as proportion differences. More specifically, here, the total effect refers to the total effect of depression on IPV perpetration in terms of proportion difference; the indirect effect refers to the part of that effect that is mediated by a given variable, and the direct effect refers to the part of the effect which is unmediated. Each of the total, direct, and indirect effects have been adjusted for confounders. Complete case analysis was used; that is, only participants who provided complete data on all exposure, mediator, and outcome variables were included in the modelling. Due to different levels of missing data across the three mediators, the total effect estimates differ slightly across the mediation analyses. A number of planned mediation analyses could not be run for individual datasets due to very low cell counts and perfect prediction of, for example, (i) IPV victimisation by IPV perpetration among men in the Christchurch Health and Development Study, (ii) depression by IPV victimisation among men in the National Survey for Families and Households and the Adult Psychiatric Morbidity Survey (APMS) 2014, and (iii) depression by IPV victimisation among men in the APMS dataset. This resulted in an inability to analyse drug use as a mediator for both men and women, and IPV victimisation as a mediator for men.

We found that a higher proportion of men and women with depression report physical IPV perpetration in the past year compared to men and women without depression, after controlling for confounders. Across all studies for which data were available, reported prevalence of past-year physical IPV perpetration was higher for women (0.9–28.4%) than it was for men (0.5–8.0%). The same pattern was observed for past-year severe physical IPV perpetration for women (0.5%-10.7%) and men (0.2–3.2%). However, an insufficient number of studies reported data on other types of IPV for men and women, and for physical IPV perpetration, only four studies could contribute to this meta-mediation analysis, highlighting the lack of data available to test our hypotheses. In this meta-mediation analysis, we found no evidence, for either men or women, that alcohol misuse mediated the relationship between depression and physical IPV perpetration in the past year. Past-year IPV victimisation was found to mediate 45% of the total effect of depression and past-year perpetration of physical IPV among women. Also, among women, IPV victimisation was estimated to mediate between 50 and 60% the total effect of depression on past-year perpetration of severe physical IPV.

Depression has previously been shown to be associated with an increased risk of having ever perpetrated physical IPV [3, 5, 39]. This study extends these findings, demonstrating, first, that depression is also associated with having perpetrated physical IPV in the past year, and second, that the association persists after controlling for age, education, income, and number of children. Our finding of higher reported IPV perpetration among women than men has also been reported elsewhere [40‐42]. Our analyses were unable to take account of the frequency, pattern, or context of physically violent acts, and may reflect misclassification bias; elsewhere, analyses have shown that the majority of injurious and high-frequency IPV is experienced by women [31]. Findings may reflect differential under-reporting among men versus women [43], whereby men who perpetrate IPV may seek to underplay their violent behaviour [44], although the mechanisms behind this are poorly understood.

Previous studies have shown that alcohol misuse is causally associated with both IPV perpetration [45] and depression [46, 47]. Whether alcohol misuse mediates the relationship between depression and past-year IPV perpetration has received limited previous attention, but analysis of longitudinal data from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) has shown that substance abuse comorbidity increased the risk of violence perpetration generally in people with mental disorder in a general population sample [48]. These findings contrast with the results of this analysis, which found no evidence that alcohol misuse mediated the relationship between depression and IPV perpetration. This apparent discrepancy may be due to misclassification bias (i.e., under-reporting of alcohol misuse and/or IPV perpetration); insufficient power to detect an association due to small numbers of those reporting all three of IPV perpetration, depression, and alcohol misuse; or high heterogeneity as a result of insufficient variable harmonisation. As there was significant variation in how alcohol misuse was measured across datasets, it is most likely that this result is due to insufficient data harmonisation. It may also be the case that alcohol misuse may confound rather than mediate the association between depression and past-year physical IPV perpetration.

IPV victimisation was found to explain a significant portion of the total effect of depression on IPV perpetration among women. This is in keeping with the findings of a systematic review of longitudinal data, which established that depression is associated with incident IPV victimisation among women [8], and findings that a high proportion of women who report the use of physical violence against intimate partners also report IPV victimisation [10, 49]. Research suggests that many couples experience bidirectional violence [13, 50], and that poor mental health, including depression, is more common where violence is bidirectional versus unidirectional [13, 15, 32, 51]. Self-defence and retaliation are commonly described motivators for women’s use of violence in intimate relationships [52], although this is contentious [42]. Straus (2010), for example, highlights research which has concluded that only a small portion of women’s IPV is explained by self-defence, and notes that most research does not collect the same data for men, and therefore, it cannot be assumed that women’s use of violence differs from men. In this study, we were not able to determine the context of either IPV perpetration or IPV victimisation. We were also not able to investigate whether the association between depression and past-year IPV perpetration was mediated by IPV victimisation for men. Previous analyses of cross-sectional data from Wave 1 of the National Survey of Families and Households (included in this IPD meta-mediation analysis) have shown, however, a stronger link between bidirectional IPV and depression among women than for men [13].

The review used a comprehensive search strategy, and took a systematic approach to data management and harmonisation. Analyses were conducted separately by sex, considered alcohol misuse and IPV victimisation as potential mediators of these relationships, and accounted for a number of key covariates. Eligibility was limited to studies that used samples that were representative of the general population and used validated diagnostic or screening measures of mental disorder, reducing the risk of selection and measurement bias. Datasets were included from several countries, broadening generalisability in a high-income country context.

However, several limitations should be noted. First, due to the use of cross-sectional data, the direction of the observed associations cannot be inferred. Although some longitudinal data were available, we could not conduct analysis of temporal associations between depression and IPV perpetration either, because studies did not measure past-year IPV perpetration and/or mental health at multiple time points, or they did not collect data on the same population in each wave, or because there was significant attrition between time points. Although our aim by analysing associations between mental disorder and IPV perpetration in the past 12 months was to establish that these occurred, if not concurrently, over a short period, findings should be considered preliminary, and tested in future models using longitudinal data. Second, due to limited availability of data, analyses of the association between mental disorder and IPV perpetration could be conducted only for depression, and some mediation analyses could not be carried out for data on men due to small numbers and perfect prediction. Previous research indicates that there is an increased risk of lifetime IPV perpetration across a range of mental disorders [5, 53]; it is not yet clear whether this is also the case for past-year IPV perpetration. When analysing associations between depression and IPV, we were only able to consider associations with physical IPV perpetration and could not investigate the perpetration of psychological and sexual IPV, or of coercive and controlling behaviours. We were also unable to analyse, due to a lack of data, whether IPV victimisation mediated the relationship between depression and past-year physical IPV perpetration among men. Third, heterogeneity in the measurement of potential mediators (e.g., alcohol misuse) meant that data harmonisation was challenging; heterogeneity for each analysis may be attributable to inadequate harmonisation. Several potential mediators of interest were excluded from the analysis, because they could not be adequately harmonised or because they were not consistently available across datasets. Fourth, due to insufficient data across datasets, we could not control for a number of key confounders, such as exposure to IPV during childhood or childhood experiences of abuse, which we know are associated with both mental disorders and IPV perpetration [14, 26, 28]. This may have resulted in an over-estimate of the true effect of depression on past-year IPV perpetration. Finally, data were drawn from high-income countries only to minimise heterogeneity, and thus, findings cannot be generalised to low- and middle-income country settings. Future research should consider replicating the methodology of this study using datasets from low- and middle-income countries.

Mental health services, criminal justice services, and domestic violence perpetrator programmes should be aware that, although most people with mental disorders are not violent, depression is associated with an increased risk of IPV perpetration. Longitudinal research is clearly needed to explore directions of causality, and underlying mechanisms, for men and women, which may differ by sex. To our knowledge, there has also been no research examining whether treatment of depression (e.g., by antidepressants) impacts on the risk of IPV perpetration.

Evidence is lacking on the effectiveness of perpetrator programmes for perpetrators with mental disorders [2, 54] and future research should address this gap as a matter of priority. If effective, these programmes may not only reduce IPV perpetration, but also depression among those who perpetrate IPV and their victims. In clinical settings, professionals working with perpetrators should consider whether there are potentially modifiable risk factors for IPV perpetration that could be addressed. Relevant responses may include psychological therapies for emotional regulation and/or treatment with antidepressants or mood stabilisers; the latter is known to be associated with reduced risk of violent crime [55], but to our knowledge, this has not been investigated in relation to IPV perpetration. Although alcohol misuse was not found to mediate the relationship between depression and IPV perpetration, it is a known risk factor for IPV perpetration that should also be considered an intervention target [56].

Mental health and other professionals working with women with mental disorders who report or are known to be perpetrators of IPV should be aware of the high prevalence of IPV victimisation among this group. Barriers to disclosure of IPV victimisation, and other forms of trauma, should be considered [57] and where elicited, safety of both partners prioritised. Therefore, there may be multiple risks that need to be assessed and considered when managing care. Findings also raise the possibility that interventions aimed at reducing risk of IPV victimisation in women might be helpful in preventing IPV perpetration. Evidence suggests that advocacy (empowerment, safety information, and referrals) interventions, such as those typically provided by specialist violence against women and girls services, can help men and women in terms of both safety and recovery [58, 59].

Our review identified very few studies with the data we needed to investigate associations between mental disorder and IPV perpetration. This study has also highlighted the need to strengthen collection of these data within population-based surveys, and for greater consistency of data collection across these surveys. At a minimum, the collection of data on violence should include physical, sexual, and psychological violence, and, for each type of violence, the number of repetitions in the past year, the sex of the perpetrator and of the victim, and the relationship between the perpetrator and the victim [60]. Studies should additionally collect data on a range of mental disorders, particularly anxiety disorders, personality disorders, and substance abuse disorders, which have been shown elsewhere to be associated with increased risk of IPV perpetration [3, 5], and key covariates, particularly substance misuse, childhood trauma (including physical abuse, sexual abuse, witnessing IPV, and neglect), and IPV victimisation in adulthood.