Associations between sexual assault and suicidal thoughts and behavior: A meta-analysis

Abstract

Objective:

The goal of the present study was to conduct a quantitative review to determine the degree to which specific forms of suicidality (i.e., lifetime and past-year suicidal ideation and attempts) are associated with sexual assault (SA). It also examined whether the strength of the association between SA and suicidality was moderated by sample characteristics or the type of suicidality assessed.

Method:

A subset of studies (25 samples and 36 effects, reflecting N = 88,376 participants) from a prior meta-analysis assessing associations between SA and psychopathology were examined. Included studies provided the prevalence of suicidality in subsamples that had and had not been exposed to SA and/or an odds ratio comparing the prevalence in these groups. Random effects meta-regression models aggregated prevalence estimates and odds ratios for lifetime and past-year suicidal ideation and suicide attempts in individuals in SA and no-SA groups. Analyses also examined whether sample characteristics (i.e., percent women, college sample) or type of suicidality moderated the magnitude of odds ratios.

Results:

Subsamples exposed to SA reported a substantially higher prevalence of suicidality (27.25%) compared to unassaulted subsamples (9.37%). There were specifically higher rates of lifetime and past-year suicidal ideation, and lifetime suicide attempts in assaulted subsamples than in unassaulted subsamples. No tested moderators had significant associations with the strength of the relationship between SA and suicidality.

Conclusion:

Findings underscore the robust relationship between SA and both suicidal ideation and attempts and suggest that identifying moderators and mediators that explain it is a key directive for future research.

More than one-third of women in the United States experience some type of sexual assault (SA) in their lifetime, and one in five women report experiencing attempted or completed rape (Smith et al., 2017). SA is associated with heightened risk for posttraumatic stress disorder (PTSD) and other forms of psychopathology (Dworkin et al., 2017). Previous studies have found that SA is associated with higher levels of suicidality (i.e., suicidal ideation and attempts) (Ullman & Nadjowski, 2009). Indeed, in a meta-analysis of associations between SA and psychopathology, suicidality was one of the forms of psychopathology most strongly associated with SA (Dworkin et al., 2017). This association is consistent with the broader literature concerning the association between suicidal behavior and trauma exposure in general (Stein et al., 2010), and suicidal behavior and interpersonal violence in particular (DeCou, Wang, Rivara, & Rowhani-Rahbar, 2018). Understanding the strength of this association, and which survivors are at heightened risk of suicidality, is important for informing the assessment, treatment, and prevention of suicidality.

Although meaningful research on the relationship between SA and suicidality exists, the literature remains limited by the inherent difficulty of assessing suicidality given its low base rate. For example, in a study of adulthood and childhood SA and suicidality that utilized the National Comorbidity Survey (Part 1, N=8,098), only 82 survivors reported attempting suicide (Ullman & Brecklin, 2002). Meta-analysis offers a promising strategy to understand patterns across multiple studies that might otherwise be limited by small sample sizes. There is also limited research that aims to understand suicidality as an outcome of adolescent/adult sexual victimization, rather than of childhood SA. This secondary analysis sought to address this gap in the literature and extend findings from a larger meta-analysis focused on lifetime or adolescent/adult SA (Dworkin, Menon, Bystrynski, & Allen, 2017). Although the prior study reported a moderate-sized association between SA and suicidality (g=0.74; Dworkin et al., 2017), it did not examine which forms of suicidality (suicidal ideation versus attempts) SA is associated with, nor did it explore moderators of the association between SA and suicidality. Suicidal ideation and attempts are meaningfully distinct phenomena, and it remains unclear whether SA is associated with increased risk for suicidality generally, or specific forms of suicidality under certain circumstances. In addition, it remains unclear whether the relationship between SA and suicidality is specific to groups with particularly high rates of SA, such as women (Elliott, Mok, & Briere, 2004) or college students (Fedina, Holes, & Backes, 2018), or is common across survivors. Thus, focused investigation of this phenomenon is warranted.

Method

Studies in this analysis reflect the subset of studies from a prior meta-analysis assessing associations between SA and psychopathology (Dworkin et al., 2017) from which suicide-specific prevalence estimates and/or odds ratios (ORs) could be obtained. Detailed methodological information is available in the original manuscript. Literature searches were conducted in PsycINFO, ProQuest Digital Dissertations & Theses, and Academic Search Premier and covered all published and unpublished reports from January 1, 1970 to December 31, 2014. Search terms and procedures are detailed in Appendix A.

See Appendix B for a study inclusion and exclusion diagram. English-language quantitative studies comparing survivors of lifetime or adolescent/adult SA to people who had not experienced SA in terms of psychopathology were eligible. The study must not have exclusively or primarily sampled people seeking psychological treatment or with mental disorders. Studies that assessed SA risk prospectively after the onset of psychopathology were excluded; only baseline data was included for longitudinal studies. For the purposes of this analysis, only studies assessing suicidality were included. One study was excluded that compared a SA group to a group that had experienced some other form of trauma, given that all other studies included in this analysis compared a SA group to a group unselected for trauma exposure.

At least two coders extracted the following information from each source. First, a 2x2 contingency table indicating the number of participants in the SA and no-SA group with and without each form of suicidality was noted. We recorded other types of effect sizes when insufficient information was available to create a contingency table and converted these effect sizes to log odds ratios. Second, we coded the suicidality type (i.e., attempts, ideation, either) and the time frame of suicidality (i.e., lifetime versus past-year). In addition, we coded the following moderators: whether the study used a college sample, and percent of the sample comprised of women. Finally, the type of data source was recorded. Discrepancies were reviewed by the first author and at least one other coder and were resolved by discussion.

Random-effects meta-regression models were tested in R 3.6.1 (R Development Core Team, 2014) using the package metafor (Viechtbauer, 2010). All models were fitted with the rma.uni function using inverse variance weights. Arcsine square root transformations were used for analyses with proportions to normalize and stabilize sampling distributions; proportions were back-transformed for presentation. We conducted sensitivity analyses separately for prevalence estimates and ORs corresponding to lifetime and past-year overall suicidality, ideation, and attempts. Outliers were truncated to the upper bound of the 95% confidence interval obtained from an unconditional model computed using the dataset with outliers excluded.

Results

In total, M = 25 samples and K = 36 effects were included (N = 88,376 participants). Study characteristics, analysis of risk of bias, and forest plots are presented in Online Supplementary Material (Appendices C–E). The pooled prevalence of any suicidal ideation or attempts in the SA groups was 27.25%, compared to 9.37% in the no-SA groups (OR = 3.91, SE = 1.11 p < .001). Prevalence and odds ratios are summarized separately for ideation versus attempts and past year versus lifetime suicidality in Table 1. All odds ratios were statistically significant except for past-year attempts. In an unconditional meta-regression of log ORs, statistically-significant heterogeneity was observed, Q(35) = 604.10, p < .001, suggesting variation that could be accounted for by moderators. However, no moderators evidenced statistically-significant associations with the magnitude of ORs (see Appendix F). Significant heterogeneity remained when moderators were accounted for, QE(30) = 254.16, p < .001.

Table 1.

Pooled estimates of the association between sexual assault and suicidality.

			SA group		No-SA group
	k	OR (SE)	Prevalence	95% CI	Prevalence	95% CI
Any time	36	3.91 (1.11)***	27.25%	21.40%, 33.53%	9.37%	6.61%, 12.55%
Ideation	14	3.05 (1.05)***	32.53%	23.42%, 42.35%	13.28%	8.59%, 18.82%
Attempts	20	4.93 (1.18)***	23.77%	16.23%, 32.26%	5.40%	3.68%, 7.42%
Past year	10	2.56 (1.07)***	21.40%	14.78%, 28.87%	8.96%	5.82%, 12.70%
Ideation	7	2.58 (1.08)***	21.85%	18.48%, 25.43%	12.88%	10.16%, 15.88%
Attempts	2	2.10 (1.78)	12.40%	2.80%, 27.45%	6.14%	4.46%, 8.06%
Lifetime	26	4.50 (1.13)***	29.72%	21.98%, 38.10%	9.60%	5.89%, 14.10%
Ideation	7	3.26 (1.06)***	46.00%	33.85%, 58.42%	15.94%	6.77%, 28.06%
Attempts	18	5.41 (1.17)***	25.36%	17.08%, 34.66%	5.34%	3.43%, 7.63%

Note. k = number of effects

^***p < .001

Discussion

Our findings demonstrate robust associations between SA and both suicidal ideation and attempts across studies that compared SA survivors to those without SA exposure. That the pooled prevalence of SA survivors with suicidal ideation and attempt histories approached one-quarter of survivors was consistent with previous studies (e.g., Davidson et al., 1996). There was not a significant association between SA and odds of past-year attempts, which likely reflects its low base rate, as there was an association between SA and odds of lifetime attempts. It is notable that these findings were derived from pooled estimates based on studies that did not necessarily utilize probability sampling, and thus do not reflect the true prevalence of suicidality among the entire population of SA survivors. It may be that survivors with more severe histories of SA and/or suicidality would be less likely to participate in a research study they knew would include assessment of these issues. This suggests the need for additional research to determine the true prevalence of this disproportionate health outcome via probability sampling. It is also important to note that causal associations between SA and suicidality cannot be inferred from these cross-sectional, observational data, and prospective research is needed to strengthen causal inferences.

There are several limitations to this study. Although a limited set of moderators was tested, none were significant and did not account for the substantial heterogeneity across effects. Moderators were limited to the available data from the larger studies, and thus it was not possible to test theoretically-relevant variables (e.g., thwarted belongingness, perceived burdensomeness, social support, and assault characteristics), which may have accounted for the remaining heterogeneity after moderators were included. Further, our study was a secondary analysis of a previous meta-analysis (Dworkin et al., 2017) and a complete literature review was not undertaken for this analysis specifically. Thus, this analysis did not include more recent studies, and is limited to the search terms and inclusion criteria from the previous meta-analysis. This may introduce bias, which should be investigated in future analyses.

Our findings offer important implications for practice, future research, and policy. They add to a growing mass of literature documenting a robust association between SA and suicidality generally. They also extend them by revealing that SA is associated with both suicidal ideation and attempts specifically and that this effect is present across men and women, and college and non-college samples. This suggests the need for policy makers to consider SA as a public health problem given its adverse health correlates. Clinicians working with SA survivors are encouraged to be vigilant for indicators of suicide risk in their clients, regardless of their sex and college status. However, explanations for the relationship between SA and suicidality remain unclear. Future research should identify moderators and mediators to explain this association, which may cultivate malleable targets for therapeutic intervention.

Clinical Impact Statement.

Study findings suggest that, across studies, there is a strong relationship between sexual assault (SA) and both thinking about (i.e., suicidal ideation) and attempting suicide in one’s lifetime, and suicidal ideation in the past year. Further, they indicate that this relationship is observed regardless of whether studies sample college students or not, and is robust to differences in sample gender. These findings add to a growing mass of research indicating that SA and suicidality are associated, and that clinicians working with SA survivors should be vigilant for signs of suicide risk in their clients.