Correlates of Reduced Violent Behavior Among Patients Receiving Intensive Treatment for Posttraumatic Stress Disorder

Violent behavior among veterans returning from service in a war zone has been of concern since the 1960s and 1970s, when accounts of violence among returning Vietnam veterans began appearing in the popular media (1). Research showed that 46% of these veterans had committed at least one violent act in the 12 months prior to assessment and that 37% had committed six or more such acts (2).

Different studies use different outcome variables, making it difficult to compare rates of violence in recent reports. Higher rates of violence are usually reported among veterans with PTSD than in veterans without PTSD. When violence is defined as self-reported kicking, slapping, or getting into fights, the annual rate among U.S. veterans returning from Afghanistan and Iraq with PTSD, 48%, is 2.3 times that for veterans without PTSD (3). Violence is less prevalent when defined as criminal conviction for a violent offence, but the increase in violence risk associated with clinical PTSD (7.3% versus 3.2%) appears to be the same (4). Research on civilian populations similarly shows higher rates of violence among people with PTSD than in population controls (5).

Identifying potential mediators of the link between PTSD and violent behavior is a natural next step toward developing effective interventions, but it has been the subject of only limited research. Veterans with PTSD score higher on measures of anger than do veterans with other psychiatric diagnoses (6–9). Alcohol abuse is common among persons with PTSD, and alcohol use and PTSD together have been shown to be associated with a greater risk of violent behavior compared with either alone (4,10), especially for serious violence (3). Evidence of the contribution of specific PTSD symptom clusters to violent behavior has been inconsistent. Early research implicated numbness (11), but recent studies with larger samples suggest that hyperarousal has a stronger effect (4). Both the overall number and the overall severity of PTSD symptoms are important correlates of violence (4).

We are not aware of any studies that have described reductions in violent behavior following treatment aimed specifically at PTSD and its component symptom clusters. As a consequence, questions relating to which forms or elements of treatment are most likely to reduce violence and which symptom clusters may be the most promising targets for intervention remain unanswered. We sought to identify aspects of patients’ clinical condition, treatment variables, and indicators of PTSD symptom change that were most strongly associated with changes in violent behavior.

Methods

Sample and Data Collection

Data were obtained from an administrative program evaluation of U.S. combat veterans treated in Veterans Health Administration (VHA) specialized intensive PTSD treatment programs throughout the United States between 1993 and 2011. The evaluation was conducted by the VHA’s Northeast Program Evaluation Center (NEPEC). Patients were excluded from the study if they had been admitted to the PTSD program from an inpatient or a residential setting. The intensive treatment programs were provided in day hospitals, short-term acute care admission units, specialized intensive inpatient units, and extended-stay residential treatment facilities. All provided daily treatment. The programs and overall sample characteristics have been described previously (12–14).

Patients received structured baseline assessment at the time of entry to the treatment program and follow-up evaluation four months after discharge. Evaluations were conducted by staff of the treating programs who were not directly involved in the treatment of the particular veterans whom they were assigned to assess. At program entry, data covering sociodemographic and other background information were collected. Data on clinical factors (including PTSD symptoms and substance use), treatment participation, and violent behavior were collected at both time points. Diagnoses and ratings of treatment commitment and treatment participation were based on assessments by program staff. All other data, including clinical measures, relied on patient self-report. The study was approved and given a waiver of informed consent by the U.S. Department of Veterans Affairs Connecticut Health Care System Institutional Review Board.

Patients for whom follow-up data were available did not differ significantly on measures of baseline violence compared with patients with no follow-up data. A stepwise logistic regression showed that the four factors most strongly associated with being lost to follow-up were younger age, clinician ratings of less commitment to treatment at the time of discharge, involuntary discharge for breaking rules, and voluntary discharge against professional advice.

Measures

Violence was assessed at program entry and at follow-up. We used participants’ responses on four items from the National Vietnam Veterans’ Readjustment Study (2). The items asked whether in the past four months the patient had destroyed property, threatened someone with physical violence (without a weapon), had a physical fight with someone, or threatened someone with a weapon. We developed a combined measure of violence by adding together the responses on these four dichotomous questions (yes=1, no=0; possible scores range from 0 to 4). On program entry, the raw and standardized Cronbach alpha coefficients for the items in the combined measure were both .71 (15). For the same measure at follow-up, the raw and standardized alpha coefficients were .68 and .70, respectively.

Background information included age, gender, educational achievement as measured by number of years of full-time education, self-defined race, and marital status. For history of incarceration, a 3-point index was created to reflect whether a patient had been incarcerated for two weeks or more, less than two weeks, or never (score of 2, 1, or 0, respectively). For employment, a 3-point index was generated to indicate whether a veteran was engaged in full-time work, was engaged in part-time work, or was unemployed (score of 2, 1, or 0, respectively). Over the course of the study, the structural emphasis of these treatment programs changed from extended inpatient stays to brief inpatient treatment, nonmedical residential settings, and day hospital programs (16). Year of entry to the PTSD treatment program was recorded as a proxy measure of this changing program emphasis.

At program entry it was recorded whether the patient had been prescribed medication over the past 30 days and any comorbid diagnoses in addition to PTSD. Three features of a patient's participation in the treatment program were recorded at discharge: length of stay in days, the patient’s commitment to treatment during the program (rated by the primary clinician), and treatment outcome (also rated by the primary clinician: successfully completed the program; transferred to a non-PTSD treatment program, typically a hospital; asked to leave for reasons related to rule breaking, most commonly substance use; chose to leave of his or her own accord; or left because of general medical illness).

Alcohol and drug abuse were assessed using composite measures from the Addiction Severity Index (ASI) (17). PTSD symptomatology was assessed at program entry and follow-up with the 14 items of the short form of the Mississippi Scale for PTSD (MISS) (18) and the four items that constitute the NEPEC PTSD assessment tool (13). All items addressed symptoms and behavior during the previous four months. We generated a combined measure of total PTSD severity and six subscales by using all items from the short MISS and the NEPEC scale. Cronbach’s alphas for the 18 items contained in the combined measure were .78 (raw) and .81 (standardized). The six subscales, based on face value interpretation of the items, assessed numbness, arousal, re-experiencing, avoidance, suicidal thoughts, and irritability.

We generated a measure of change in violent behavior by subtracting the results of the baseline assessment from the results of the follow-up assessment. Thus negative values represented reduced violent behavior, and positive values represented increased violent behavior. Change scores were also constructed for the ASI composite alcohol and drug use measures, for the total PTSD scale and each of the six subscales, and for the ASI employment subscale (17), again by subtracting the score at program entry from the score at follow-up.

Statistical Analyses

The analyses were designed to identify correlates of change in violent behavior between program entry and follow-up.

Bivariate correlations were used to identify individual variables that were correlated with change in violence between program entry and follow-up. Because regression to the mean will lead to greater change in violent behavior among veterans with more violent behavior at baseline, we used models that controlled for the values of the violence measures at program entry. We used the same approach in our analysis of clinical change at follow-up. When examining the relationship between change in PTSD symptoms and change in violence, for instance, we controlled for the baseline levels of both PTSD and violence.

All of the variables that were significantly associated with change in violent behavior in these bivariate correlations were then entered into a stepwise regression to identify variables that were independently associated with a change in violent behavior. Baseline measures that were significantly associated with change in violence were forced into the model before the change variables were entered. We further examined a series of four stepwise logistic regression models to evaluate outcomes for each of the four dichotomous violent behavior items constituting the total scale, controlling for the baseline value of each.

We curtailed the addition of variables to the stepwise regression at the point where further variables contributed less than .5% to the cumulative adjusted R². This choice of threshold was based on a calculation of the implications of a .5% increase in R² for the associated change in the dependent variable (violent behavior) when measured in units of standard deviation (SD). A change of .5% in the adjusted R² corresponds to a change in violent behavior of .003 SDs. By conventional criteria, this is below the level of “small” change (19).

All analyses were performed by using SAS, version 9.3, statistical software (20).

Results

Descriptive information about the sample appears in Table 1. Data were available for 35,330 patients at the time of program entry and for 24,099 patients (68.2%) at follow-up, four months after discharge. At program entry, data on violence in the past four months were available for 33,809 patients; 44% (N=14,876) reported that they had destroyed property at some point in the past four months. This figure declined to 23% (N=7,776) at follow-up. At program entry, 58% reported threatening someone without a weapon, 30% reported being involved in a physical fight, and 17% reported threatening someone with a weapon in the past four months. These figures declined to 40%, 15%, and 8%, respectively, at follow-up.

The mean±SD total violence score on program entry was 1.49±1.34. The mean total violence score at follow-up was .91±1.42. The mean change in the violence score from baseline to follow-up (–.58) generated an effect size of –.44. In other words, mean violence at discharge was .44 SDs below the mean violence at program entry. This represents moderate effect size by Cohen’s criteria (19).

After controlling for baseline values of violence and baseline values of all other change measures, we found that many of the background measures were significantly associated with reduction in violence on bivariate analysis (Table 2). The regression coefficients were moderate in size. Thus increased age, being male, educational attainment, being white, being married, and year of program entry were associated with greater reduction in violence (as indicated by a negative coefficient) between treatment entry and follow-up than was the case for the group as a whole, and having a history of incarceration was associated with a smaller reduction in violence (as indicated by a positive coefficient).

Measures of clinical status at the time of program entry, with the exception of substance abuse diagnoses, where the regression coefficients were very small, were not significantly correlated with a change in violence at follow-up. Measures of treatment participation (length of stay, commitment to therapy as rated by program staff, and successfully completing the program) were significantly correlated with a change in violence at follow-up, but the regression coefficients were again small.

Larger, positive, regression coefficients were evident for the measures of clinical change between intake and follow-up. Improved employment (that is, a positive value for the change variable), reduced alcohol use and reduced drug use (a negative value for the change variable) were associated with a reduction in violence beyond what would have otherwise been attributed to baseline status and regression to the mean. Reductions in PTSD symptoms (reflected in negative values for the change variables) were also positively correlated with reductions in violence beyond those for the sample as a whole. This was the case for both individual subscales and total PTSD score.

When expressed as standardized estimates, regression coefficients allow effect size to be calibrated in terms of SDs. Table 2 shows that a reduction in the total PTSD score of one SD (for example, from its mean at program entry of 3.5 to 3.0) was associated with a reduction in the violence score at follow-up of .327 SDs (for example, from its mean at follow-up of .91 to .46).

The results of the stepwise regression are shown in Table 3. The largest contributor to the adjusted R² was violence at program entry, with a large negative coefficient showing that higher baseline levels of violence were associated with greater reductions in violence at four-month follow-up. Measures of alcohol use, drug use, and unemployment at program entry made small contributions to the adjusted R², with a somewhat larger contribution coming from overall PTSD symptoms at program entry. The only change variable contributing significantly to the change in adjusted R² was a change in alcohol use between program entry and follow-up.

Stepwise logistic regression models examining separately the four dichotomous violence items at follow-up confirmed that a reduction in alcohol use remained significantly associated with a reduction in each violent behavior.

Discussion

This study showed that scores on a standard measure of violent behavior declined significantly in a sample of combat veterans treated in specialized intensive PTSD treatment programs between entry into a program and four months postdischarge with a moderate effect size (r=–.44). The correlations between changes in violent behavior and clinical variables, including changes in PTSD symptoms and substance use, have not been demonstrated previously in a sample of veterans receiving intensive treatment. Several aspects of the results warrant further comment.

First, the reduction in violence in this study was somewhat smaller in magnitude for patients who were younger, female, less educated, nonwhite, and not married; who had a history of incarceration; and who been abusing alcohol or drugs when they entered treatment. Female gender has elsewhere been found to be a protective factor against violence, although in mental health samples this effect is often lost as it was here (21,22). The remaining variables that were associated with smaller reductions in violence have consistently been shown to be associated with violent behavior in a range of samples and settings (23,24). We have shown that these variables were also associated with less decline in violence following treatment.

Second, the move away from long-term inpatient treatment in the VHA, as reflected in the year of program entry, was associated with a small but significantly greater reduction in violent behavior. This is consistent with evidence from other studies that the move away from long-term inpatient treatment for PTSD has led to greater clinical improvement, both at discharge and follow-up (15). On the other hand, we found little evidence that treatment participation—as measured by length of stay, commitment to treatment, or successful completion of the program—was associated with substantial reductions in violent behavior. The effect sizes were small and not significant in multivariate analysis. Treatment may have had benefits that were not detected by these therapy-related measures.

Third, on bivariate analysis the reduction in violence was greatest for patients with reductions in substance use and PTSD symptoms. Reduced alcohol use and PTSD symptoms at program entry were further associated with a reduction in violence on multivariate analysis. While the strongest single correlate of reduced violence at follow-up on multivariate analysis was violence at program entry, taken together, these results support targeting PTSD symptoms and substance use when seeking to reduce violence among veterans with PTSD. Intensive treatment may have benefited these patients either because medication and psychotherapy reduced violence directly or, more likely, because treatment reduced the PTSD symptoms and substance use that increased the risk of violence. Some symptom reduction, for instance in re-experiencing and avoidance, may reduce vulnerability to environmental triggers of violence among persons with PTSD. Other changes with treatment, for instance, reduced arousal and irritability, seem likely to improve resilience more generally.

Finally, the decline in violence was most strongly associated with reductions in irritability, suicidality, and re-experiencing and with total PTSD symptom scores. While this bivariate finding was not preserved on multivariate analysis, it lends some support to research linking these symptom clusters to violence and suggesting that PTSD symptoms have a cumulative effect on violence risk (4). Further analysis of the PTSD and substance use change measures showed a high level of intercorrelation (α=.75), suggesting that these change measures reflect a common underlying construct. Patients who drank in part to relieve the distress and arousal of PTSD may have abused alcohol less as these symptoms improved.

Several limitations require acknowledgment. First, because violent behavior is a major reason that individuals with PTSD seek and are admitted for treatment (11), the overall reduction in violence is likely to reflect, at least in part, regression to the mean. Within the overall reduction in violent behavior we have identified variables correlated with greater and lesser reductions. We sought to control for the effect of regression to the mean on these correlations by including violence at program intake as a covariate in the regression analyses. Doing so is not a substitute for an intervention study that randomly assigns patients to treatment and control groups, however. Conclusions regarding causality from these data must be correspondingly cautious.

Second, we have examined the reduction in violence following treatment in a heterogeneous set of programs that used a range of forms of psychotherapy and medication. Although there was also some consistency across sites (for instance, all of these programs provided daily treatment in an inpatient, residential, or day hospital setting to U.S. veterans of recent military service), future research should seek to distinguish the effects of specific therapies and program structures both on symptomatic improvement and on violence risk. Third, we used self-report as the basis of our violence measure. Confirming these reports with official records of arrest and conviction, as well as information from collateral sources, would increase the validity of the outcome measure.

Finally, only a subset of veterans with a diagnosis of PTSD receives mental health services from the VHA, and, of this subset, only a smaller subset receives intensive PTSD treatment. The generalizability of the findings to other kinds of trauma and to women is unknown. Some factors associated with being lost to follow-up, such as younger age, were also related to violence. Although these associations do not demonstrate a bias in our analyses, we cannot exclude the possibility that the results would have been different had follow-up data been collected for all patients.

Previous suggestions that clinical variables are relatively unimportant correlates of violent behavior among samples of persons with a mental disorder (23) are not supported by this study. Not all of the treatment variables that we studied were equally relevant however. Length of stay and perceived commitment to treatment were less strongly associated with a reduction in violent behavior than changes in other clinical phenomena, such as improvements in substance abuse and PTSD symptoms. Most of the significant correlations between violent behavior and clinical variables were no longer significant on multivariate analysis. This suggests that the clinical variables are strongly intercorrelated, a finding consistent with clinical experience.

The mean length of stay in these programs was 44 days. Despite the move away from long-term inpatient psychiatric treatment in the VHA (25) and elsewhere in the U.S. mental health system (26), patients may have experienced benefits not just from treatment but also from nonspecific factors such as shelter, support, and respite from the cumulative effect of life stresses. If so, the benefit of this support for reduced violence remained evident four months after discharge. The role of intensive residential or inpatient treatment has been downplayed in mental health care in recent years, largely for economic reasons. This study suggests that the benefits of specialized intensive PTSD treatment, perhaps in turn reflecting the broader benefits of “asylum,” should not be ignored.

Conclusions

In this study of U.S. veterans receiving intensive treatment for PTSD, reduction in violence during and after treatment was more strongly correlated with reductions in patients’ PTSD symptomatology and substance use than with incarceration history or with sociodemographic and biographical variables traditionally associated with violence. The results suggest that the support, shelter, and asylum that are part of intensive treatment are associated with reduced violent behavior and that such services play an important role in the spectrum of care for war-related PTSD.