Community Violence and Internalizing Mental Health Symptoms in Adolescents: Systematic Review.

Purposes: Mental health diseases (MHD) are responsible for 16% of the global burden of disease in adolescents. This review focus on one contextual factor nominated community violence (CV) that can contribute to the development of MHD. Objective: to evaluate the impact of CV on internalizing mental health symptoms (IMHS) in adolescents, to investigate whether different proximity to CV (witness x victim) is associated with different risks and to identify whether gender, age, and race moderate this association. Methods: systematic review of observational studies. Population includes adolescents (10 - 24 years), exposition involves individuals exposed to CV and outcomes consists of IMHS. Selection, extraction and quality assessment were performed independently by two research. Results: 2987 works were identied, after selection and extraction it remained 42. Higher exposure to CV were associated positively with IMHS. Being a witnessing is less harmful for mental health than being a victim. Age and race did not appear in the results as modiers, but masculine gender and family support appear as a protective factor in some studies. Conclusion: This review conrms the positive relationship between CV and IMHS in adolescents and brings relevant information that can direct public efforts to build policies in prevention of both problems. heterogeneity, of internalizing of a small but statiscally signicant greater effect for when to internalizing


Information sources
The search was performed in six allied health research databases: Medline (accessed through PubMed), PsycINFO, Embase, LILACS, Web of Science, and Scopus. Regarding grey literature, only those corresponding to theses and dissertations were included. These were identi ed in the databases above, and "ProQuest Dissertation and Theses" was used to search for full texts. The search was conducted on February 5th, 2019, and no lters for years of publication or language were applied. After the third phase of selection, all studies included in the review had their reference lists analysed by two independent researchers to search for additional works.

Search strategy
Search terms were based on the review question and were constructed with a librarian (APPENDIX I). The main concepts were as follows: "adolescents" OR "youth" OR "teenagers" AND "community violence" OR "urban violence" OR "neighborhood violence" AND "mental health" OR "anxiety" OR "depression" OR "post-traumatic" OR "internalizing" OR "psychological symptoms". A librarian worked on obtaining the full-text works, seeking bibliographic bases, libraries, and contact authors.

Study selection
Data selection was carried out in three stages: title, abstract and full texts. During all phases, two researchers performed critical readings to apply the pre-established inclusion and exclusion criteria. All stages were preceded by a pilot that included 10% of the total number of works in each phase (concordance rate 80-97%). In the rst and second stages of selection, any disagreements were included. At the second stage of selection, we decided to exclude externalizing outcomes. In the third stage, we discussed all the discrepancies. When there were discrepancies, a third researcher was called. All reasons for exclusion were registered. The authors of ve studies were contacted for clari cation.

Data extraction
Data were extracted using Epidata 3.1 with a standardized formulary tested in the pilot. Extracted information included the following: (i) study design, setting, times of measures and recruitment; (ii) demographic population; (iii) exposure characteristics -classi cation subtypes and measurement instrument; (iv) comparison group; (v) outcomes -types and measurement instruments; and (vi) association measures. Again, two review researchers worked independently. All papers included at this phase were discussed. In two studies, a third researcher was consulted to decide about discrepancies. At the end of the extraction phase, 32 studies were divided into two groups: 16 studies with complete information that were included in the meta-analysis and 16 studies with incomplete information included only in the qualitative synthesis.

Assessment of methodological quality
The quality of the studies was also evaluated independently by two researchers. The formulas used were adaptations, also tested in the pilot phase, from a prede ned quality assessment form for cohort/case-control studies and descriptive studies published in the Joanna Briggs Institute Reviewers' Manual [27]. Studies were classi ed into three categories: low, intermediate, and high quality. Researchers de ned cut-off points; all questions had the same weight in the nal punctuation. Discrepancies were discussed, and a consensus was achieved in all cases. Critical appraisal tools are presented in APPENDIX IV.

Synthesis of the results
Results are presented in qualitative synthesis. A subgroup of 21 studies underwent quantitative synthesis. Forest plots were displayed to visualize the results. Heterogeneity was evaluated by the I 2 test, which describes the proportion of variation across the studies due not to chance but rather to heterogeneity [28][29]. The higher the percentage, the higher the level of heterogeneity. Because heterogeneity was still high adopting the random effect model, reasons for these were investigated, and subgroup analysis was conducted -strati cation by proximity of CV (witness and victim) and types of outcomes (PTSD, depression and internalizing symptoms) were performed. Because heterogeneity was still high in almost all forest plots, it was not possible to construct funnel plots to evaluate possible publication bias. We report our ndings in accordance with PRISMA guidelines [30].

Results
After a search in databases, 2987 works were identi ed, and no additional papers were found through other sources. Of these quantities, 1005 duplicates were removed, and the selection phase started with 1982 records. During stages 1 and 2 of selection, 1.119 records were excluded, remaining 863 for the third phase. The eligibility phase started with 42 works. Of these, 21 were included in the quantitative synthesis.
The results are presented in the following manner: 42 studies included in qualitative syntheses had their main characteristics presented in Table 1, and their results were described according to the review objectives. Quality assessments are presented in Table 2, 30 were considered high quality, 11 intermediate and one low quality.
A subgroup of 21 studies could be meta-analysed. The rst forest plots were generated and included all 21 studies. For these, we worked with the concept of general community violence and only one type of outcome, so for the studies that had more than one association measure (for victim and witnessing, for example), a weighted average was calculated, and the same was done for the studies that had more than one outcome. The I 2 value was 53.8%, with a p value of 0,003, thus indicating substantial heterogeneity [31]. Subgroup analysis was conducted with strati cation by proximity of CV (witness and victim) and then with types of outcomes (PTSD, depression and internalizing symptoms). The only graphics presented were the ones with heterogeneity smaller than 60%, wich where the rst forest plot generated with all the 20 studies included and the subgroups of PTSD and internalizing symptoms as outcomes.
The results of the summary measures must be interpreted with caution. Only part od the qualitative synthesis studies presented complete data that would allow inclusion in the quantitative synthesis. The rst graph generated (Fig. 1) shows high heterogeneity, the graphs of soubgroups by PTSD outcome and internalizing symptoms presented do not showed high heterogeneity but represent a small group of studies when compared to the total number included in the review. Still, it was possible to see a small but statiscally signi cant greater effect for PTSD when compared to internalizing symptoms.
High quality Sun et al., Gepty et al, Legend -Answers: Y -Yes, N -No, U -Unde ned. Score: N (1 point); U (2 points); Y (3 points). The studies were ordered according to their quality. Light gray color -low; middle gray -intermediate; dark gray -high quality studies. Obs. Grinshteyn et al. (2018) was avaliated as longitudinal study because it is the study design, but the results presented in Table 1 are in cross-sectional section because the statistical analysis was this kind.
Henry et al.
High quality Chen et al. 2020 Campo-Ríos et al. 2020 Foster et al.
Ho et al.
Plessis et al.
Answers: Y -Yes,  [54] compared their sample perception of violence and objectively measured neighbourhood violence derived from criminal statistics. Perception of violence in the neighbourhood is a different concept when compared to exposure to community violence because the rst one is related to how adolescents see the environment they live in. The authors found that adolescents who perceived their neighbourhood unsafe had a nearly 2.5-fold greater risk of psychological distress than those who believed their neighbourhood was safe. Adolescents who live in areas objectively characterized by high levels of violent crime measured by criminal statistics were no more likely to be distressed than their peers in safer areas.
Aisenberg et al. [44] also did not nd an association between CV and PTSD, and they suggested that other factors, such as one's relationship to the victim and one's physical proximity to the violent event, may in uence this association. It is important to underscore that this is the only study included in this review considered low quality. Donenberg et al. [51] did not nd an association between CV and internalising problems, only for externalising problems in boys, some factors that could have in uenced in this results are a small sample and the fact that in the measure od CV it was considered only witnessing.
The subgroup of 20 studies that were metanalyzed had a summary measure of 1,02 (CI 95% 1,01-1,02) showing that there is a small but statiscally signi cant higher risk of IMHS for adolescents exposed to CV. found an association between all CV measures and mental health outcomes with the same magnitude, and three did not nd an association either for the victim or for witnessing [33-45-71-73].

Differences in
The results indicate that higher proximity to violence (VCV x WVC) was related to a higher risk for internalizing mental health symptoms. Grinshteyn et al. [55], in addition to a gradient risk for the victim, witnessing and knowing about, also found differences between violent events and no violent events, the rst one counting for a higher magnitude. The authors that did not nd an association discuss the possibility of desensitization and other types of violence (school or family violence) as softening the CV's effects on mental health [60].
The metanalysis graphs with victim and witnessing subgroups were not considered because they presented high heterogeneity (61,1 and 67,6% respectively).

Community violence is accessed by criminal statistical and internalizing mental health symptoms.
Six studies measured exposure to CV with criminal rates [32-36-49-54-55-72]. Grinshteyn et al. [55] de ned crime rates using the zip code crime rate per 1000 population. They also collected self-report data and compared. Their results pointed to a gradient risk for victims, witnessing and knowing about, meaning a larger association for the rst ones. When comparing criminal statistics with self-report measures, the results were positively signi cant only for depression and in a smaller magnitude. The authors discussed the importance of these arealevel crime rates to be constructed in smaller geographic units and to be considered a larger variety of crimes. Goldman-Mellor et al. [54] measured perceived neighbourhood safety with self-respondent answers and objectively measured neighbourhood violence using a geospatial index based on FBI Uniform Crime Reports. Their results showed an association for the rst measure but not for the second one, suggesting that perception of neighbourhood violence matters more for mental health than objective levels. Velez-Gomez et al. [72] and Cuartas et al [49] utilized both criminal statistical analysis with homicide rates, the rst group encountered a positive association only for the outcome "ineffectiveness" in early adolescents (10-12 years), the second group a positive relationship for CMD and PTSD.
Gepty et al [36] utilized criminal statistics classifying violent crimes and non-violent crimes and found a positive association with depressive symptoms for the rst violent crime, but not for non-violent crimes. Da Viera et al. [32], worked with criminal statistics related to the adolescents residence and school address, and found that adolescents that lives in areas with low crime and studies in areas with high crime has a larger chance to present anxiety, probably related to feeling of insecurity in the way to school.
3.4. In uence of gender, race, and age on the association between CV and internalizing mental health symptoms.
Thirteen studies analysed gender as a moderator in the relation above, four of them encountered gender as a potential moderator. Bacchini et al [45] and Bone Mccoy et al. [70] found that girls are more affected to negative experiences of CV than boys, reacting high anxiety, depression, sadness and posttraumatic stress symptoms. Haj-Yahia et al [56] found that girls had more internalyzing problems than boys when they are victims of CV, but no witnessing; while Foster et al [53] found positive association between CV and depressive and anxious symptoms only for witnessing but not for victim. The other seven works tested gender as a moderator and did not nd differences between boys and girls in the association [35-38-40-42-46-51-57-58-60-63].
Only two studies, one conducted in Israel [61] with arabs and jewishs and other in Chicago [48] with latins, black and white individuals, tested race as a moderator of the relationship between CV and mental health symptoms. In the rst one, Jewish reported higher levels of WCV, while Arabs higher levels of VCV and posttraumatic symptoms over the last year, but this ethnic a liation did not moderate the relationship between CV exposure and PTSD. Chen et al [48] worked with a large multi-ethnic sample in Chicago and found that latins and black adolescents were more exposure to CV, had higher levels of depression and delinquency, and had more risk factors like low family warmth, peer deviance, school adversity and CV exposure. Besides that, results from regression models showed a higher chance of depression for white adolescents comparing to minorities (black and latins), these are explained in the light of desensitization hypothesis [77][78].
The only study that considered age as a moderator of the relationship above was the once conducted by Gomez et al [72]. Even so, the strati cation occurred with an age group that did not t our inclusion criteria (8-10 years), so the results were presented only for the interval 10-12 years.
3.5. Family support, communication skills, emotional regulation and contextual factors that affects adolescents' mental health when exposed to CV. [64] did not found differences in chances of depression and PTSD for adolescents exposure to VC when family support were present, neither teacher support for the rst one.
Individual characteristics of personality and emotional functioning also appear in some studies as moderators. Le Blanc et al. [26] found that good communication and problem-solving skills protect adolescents exposure to CV from psychological stress. Sun et al. [43] encountered that internal disfunctioning of emotional dysregulation, like self harm for example, potencialize symptoms of PTSD in adolescents exposed to CV; O'Leary [65] found that expressive suppression, that refers to active inhibition of observable verbal and nonverbal emotion expressive behaviour, buffers the effect of CVE on depression; and Gepty et al.
[36] studied ruminative cognitive style, that is the tendency of individual to be caught in cicle of repetitive thoughts, and found that it also increases the chance of depression in adolescents exposure to violent crimes.
Contextual factors were also avaliated as moderators. Cuartas et al [49] studied the effect of living in poor household, having been directly victimized or witnessing a crime, perceived neighborhood as unsafe and social support and found that the rst three of them potencialize the chance for PTSD in adolescent's exposure to VC and that perceived neighborhood as unsafe also worse the chances of CMD. O'Donnell et al.
[64] analysed adolescents from The Republic of Gambia, Africa, and found that positive school climate function as a protective factor between CV exposure and PTSD, it was stronger for witnesses than for victims.
Cultural factors related to ethnicity were also evaluated. Henry et al. [57] studied cultural pride reinforcement and cultural appreciation of legacy as potential moderators between CV and depressive symptoms in a sample exclusively composed of African Americans. Cultural appreciation of legacy was found to be a protective moderator of this relationship, leading to the conclusion that teaching African American youth about their cultural heritage can help them cope with racial discrimination.
3.6. Different risks for different outcomes.
Some studies analysed more than one outcome with the following distribution: depression (20), internalizing symptoms (16), PTSD (15) and anxiety/stress (1). Different outcomes are associated with different magnitudes of CV exposure, as showed in Table 1, and factors analysed as moderators of this association also act differently.
The graphs of metanalyses in subgroups by outcome that showed a heterogeneity below 50% and where therefore presented in this review were the studies with PTSD as outcome and internalizing symptoms. The summary measures for PTSD outcome was greater than 1 (1,12, IC 95% 1,05-1,19), while for the internalizing symptoms outcome was borderline ( 1,02, IC 1,00-1,04).

Discussion
The results of qualitative synthesis reinforced a positive relation encountered in the previous meta-analysis between CV exposure and internalizing mental health symptoms in adolescents [18][19]. Also, the proximity of community violence appeared to be an essential factor contributing to the risk of mental health symptoms. Adolescents who are VCV are at greater risk than younger individuals who are WCV. Only a few studies considered "hearing about community violence" and compared it to witnessing and victim violence; because of the small number of studies, it was not possible to determine the gradient risk for this kind of exposure to violence. The summary measured of the 20 studies in quantitative synthesis showed a small but positive association. The summary measures of the victim and witnessing subgroups could not be considered due to the high heterogeneity. Regarding the outcomes analysed, studies showed different risk magnitudes for different outcomes, and they also varied with CV proximity. The summary measures for PTSD was positive and small, but larger than the one for the subgroup of internalizing symptoms.
Regarding moderators mentioned in objectives (gender, age, and race), although most of the samples found differences in types of CV events mentioned from boys compared to girls, as well as different prevalence in outcomes, only 4 of 8 studies that evaluated gender as moderator con rmed this relationship encountering that girls have a higher chance to present internalising symptoms when exposed to VC compared to boys [45-53-56-70]. These differences between genders are also found in studies that consider externalizing symptoms, however for this outcome, boys have more risk than girls when exposed to VC. A possible explanation for this distinction is the difference in creation between masculine and feminine genders, especially in more traditional societies, where the girl is encouraged to keep her emotions to herself and to have more socially accepted behaviours, while the boys are encouraged to reinforce their masculinity through sometimes violent and deviant behaviours [79].
Age did not appear in any study as a moderator. This result is different from those encountered in Fowler [19]. However, their meta-analysis included children and adolescents, different from ours, which considered only adolescents.
Race was tested as a moderator only in two studies -one conducted in Israel with Jews and Arabs in wich there were no signi cant differences between ethnicities when the effect of community violence was studied concerning posttraumatic stress symptoms, another conducted in Chicago, with white, black and latins in the sample, where race appear as a modi er in the relationship between CV and depression. It is important to highlight the fact that thirteen studies included in this review did not have any information about race of participants, and that in the group of studies that classi ed races' participants, some of them were composed exclusively by African Americans. Previous meta-analysis could not evaluate race as modi ers because of these ethnic's homogeneity of samples [19]. Besides that, literature had showed that racism is linked to poor physical and mental health [80]. Black young boys are the major victims of CV, it is important to study the effects of race on the association of CV and mental health symptoms, as well as possible protective factors and interventions for this population [14]. The study by Henry et al. [57] is an example of how maternal messages of positive reinforcement of black cultural can protect adolescents of this ethnic group exposed to CV from depressive symptoms. Another study evaluated a Security Public Program in the city of Chicago, whose goal was to increase student safety on the way to school by placing more police on strategic streets, its results showed that there was a 14% decrease in crime rates and 2,5% decrease in school absenteeism [81].
Other moderators appeared in some studies as protectors for anxious, depressive, and posttraumatic symptoms. This group's family support and adolescents' ability to talk about problems and positive emotional functioning stands out [26-43-45-56-59-65-66]. The literature previously pointed out this relationship [12]. Studies evaluating polyvictimization have also demonstrated high rates of multiple violence occurring concurrently and an increased risk for internalizing symptoms [62-82-83]. This information is very important to guide programs to prevent mental health symptoms in adolescence. There should be reinforcement of parents' and adolescents' abilities to promote better outcomes, such as evaluation of different types of violence in the past to deal with more speci city in each case.
A possible explanation for studies that found negative effects on mental health only for victims and not for witnessing is the phenomenon of desensitization, since in these samples, they found high rates of WCV. This phenomenon has been described by some authors previously and occurs in areas of high exposure to CV. With chronic and recurrent exposure, individuals do not present as many depressive and anxious symptoms after a certain degree of CV in a process of naturalization of barbarism. [73-77-78] It is important to draw attention to the fact that, in relation to externalizing symptoms, such as aggressive behaviour and delinquency, this does not occur; in addition, there is an increase in these behaviours in a progressive and linear way with an increase in violence, which reinforces the negative aspect of this association (CV and effects on mental health).
Most studies included in this review were conducted in the United States of America (27), followed by South Africa (4) [49] studied the effect of contextual factors such as poverty in the neighborhood and social support as potential moderators of the association of CV and CMD and PTSD, con rming their hypotheses for the former. O'Donnell et al [64] found that a positive school climate was a protective factor for youth who witnessed CVs about posttraumatic stress reactions. The authors highlighted the high levels of self-report hostile school climate that may re ect the school context's structural factors. However, in light of the cultural aspects, any of the studies included in this review compared, for example, urban areas with rural areas. It would be an interesting comparison to examine. Considering these variations attributed to contextual and cultural factors, more studies conducted in different countries and cities would be relevant.
In this review, some studies analysed the difference between the exposure to violence measured by statistical criminalities and CV self-report questionnaires or perceived violence [54][55]. The authors found differences in their results, as described in Sect. 3.3. Four studies that used statistical criminalities showed a positive relationship of CV with at least one mental health outcome analysed, but in a smaller magnitude than CV measured by self-report answers. The rst methodology has relevance because it is less costly and simpler to conduct and therefore has importance, especially in countries where there are few studies in this area. Nevertheless, studies that compare two forms of measuring violence (self-report and criminal statistics) can contribute to a better understanding of the differences between them.
There are strengths and limitations that should be considered in this systematic review. Strengths include an extensive search of databases, contact with authors for clari cation and no lters applied for year or language in the search, all of them contributing to a larger body of literature. Methodologies were constructed with alternate pairs of studies in the selection and extraction phase to avoid selection bias and errors in extraction. Studies included in the review were composed mostly of adolescents from schools or population-based samples and not from mental health services or other types of institutions, leading to a more representative sample. This review utilized a community violence concept that excludes sexual and school interpersonal violence, focusing on and estimating the effect of such violence on adolescents' mental health, which we considered a strength since it brings more speci city to the results. The main limitations were that different tools for exposure and outcome measures were used, leading to heterogeneous results and compromised pooling. Study designs and statistical analysis also differed between studies, which made comparability di cult.

Conclusion
This review con rmed a positive relationship between community violence, excluding sexual assault and school violence, and internalizing mental health symptoms in adolescents. Even though race and age did not appear to be moderator in most of the studies, girls were more sensitive to the effects of the exposure in some studies, showing that gender can be a possible moderator in this relationship. Other factors, such as family constitution, communication skills and emotional functioning, also seem to have an in uence on this association.
This review provides relevant information regarding the health and public safety eld and can serve to direct public efforts to build policies to address the prevention and treatment of both community violence and mental health diseases. This review also contributes to knowledge of these issues among health and education professionals.  Flowchart of the selection and extraction phase Figure 2 forest plot of studies with general community violence as exposure and any type of internalizing mental disorders as outcomes. Figure 3 forest plot of subgroup of studies that considered PTSD as outcome.