Within-community variation in violence and risk of self-harm in California: a population-based case–crossover study

To the Editor:

Self-harm is a leading cause of morbidity and mortality in the United States,¹ and rates are increasing for reasons that are not well understood. Social environments are recognized to be associated with self-harm,² but research to identify features of the social environment that matter most is limited.

Community violence is one potentially modifiable feature of the social environment that may influence self-harm. However, few studies have examined the association of community violence with self-harm^3–10 and, to our knowledge, no research has examined short-term, within-community variation in violence, as opposed to chronic or overall levels of violence.

Within-community variation in violence is directly relevant to the stress-diathesis model of self-harm which posits that incidents of self-harm reflect the confluence of long-term predisposition to self-harm (e.g., due to genetic vulnerability) with exposure to stressful life events that trigger brief periods of elevated risk.¹¹ Thus, increases in community violence (e.g., having neighbors who were recently shot) may trigger self-harm in a vulnerable individual.

Methodologically, chronic community violence is strongly associated with other self-harm risk factors such as economic opportunity, making the effects of these factors difficult or impossible to disentangle, a phenomenon known as structural confounding¹² that has limited past research.^3–10 We address structural confounding by investigating whether within-community variation in violence is associated with self-harm, using a case–crossover design. We compare residents of the same community to themselves at times with relatively high and low levels of violence, thereby controlling for observed and unobserved community- and individual-level factors that are time-invariant over the study period.

We compiled California statewide data on self-harm and community violence 2005–2013 from mortality, emergency department, and inpatient hospitalization discharge records, and conducted a population-based case–crossover study,¹³comparing cases’ exposure at a time relevant to case occurrence to exposure at referent (non-case) times – in this study, exactly 30 days before and after each case.^14–17 Cases were all deaths and hospital visits due to deliberate self-harm (N=396,960). As in previous research,¹⁸ within-community variation in violence was the monthly rate of deaths due to homicide and hospital visits due to assault in the community of residence, with predictable temporal patterning removed using a Kalman smoother.¹⁹ Previous simulation work suggests that the Kalman smoother is superior to a range of other time series methods in the separation of unpredictable versus predictable patterning of violence in California populations.²⁰ Lacking evidence on the critical exposure period (lag time and duration) for the association of within-community variation in violence with self-harm, we selected a reasonable time frame of 30 days prior to injury/referent date to balance capturing short-term, acute effects with pooling enough data to estimate stable rates. We selected bidirectional referent periods with controls drawn as close in time as possible to the case because previous simulation studies of similar settings suggest this approach provides superior control of confounding by trends and seasonality compared to other referent periods.^14–17 This design also helps limit confounding by unmeasured time-varying factors. Although controls drawn from after fatal self-harm are technically no longer at risk, a combination of pre- and post-case exposure is a reasonable approximation for the exposure distribution in the study base (eAppendix 1). We used conditional logistic regression, adjusting for measured time-varying community-level confounders. See eAppendix 1 for further detail on background, methodology, results, and discussion.

After adjustment for confounders, thirty-day periods with higher-than-expected levels of community violence (80^th percentile versus median) were not associated with meaningfully elevated relative odds of self-harm (fatal odds ratio (OR): 1.004 [95% confidence interval [CI]: 0.997, 1.011]; nonfatal OR: 1.005 [CI: 1.003, 1.007]). There was also little variation in associations by demographic subgroup (age, gender, race/ethnicity, or urbanicity). Results were robust in sensitivity analyses using longer and shorter time windows, restriction to communities without residual autocorrelation in exposure, and a case–control design drawing population-based controls from California resident participants of the American Community Survey.

To our knowledge, this is the first study to assess whether increases in violence within communities were associated with greater fatal and nonfatal self-harm in those communities.¹⁰ We found no meaningful associations. This may be because we only assessed self-harm associated with deviations from expected levels, and therefore capture only a small portion of the relationship between community violence and self-harm. The exposure measure in this study may not be the optimal characterization. Previous research focusing on variation across neighborhoods has identified strong associations between long-term community violence and self-harm,¹⁰ but the most salient time frame for elevated risk remains uncertain. This is an area for future research. In addition, other forms of variation (e.g. mass shootings or level shifts caused by interventions) may be important. Future research examining the impacts of violence prevention programs aiming to limit increases in community violence may provide more conclusive evidence.

This study may serve as a model for future research. We leveraged data from comprehensive population-based death, survey, and healthcare utilization data from California to study a potential social ecological driver of self-harm, an outcome for which previous research has been limited by small sample sizes. We combined these data in an efficient way and leveraged the high degree of geographic and temporal precision to study an acute outcome and transient ecological exposure. The case–crossover design enhanced control of unmeasured individual confounders such as genetics and family history, and reduced concerns related to structural confounding and control-selection bias.