The Importance of Conduct Disorder in the Treatment of Violence in Schizophrenia: Efficacy of Clozapine Compared With Olanzapine and Haloperidol

Clozapine is the most effective antipsychotic agent for aggressive and violent behavior. Frogley et al. (1) performed a systematic literature review to investigate clozapine’s antiaggressive effects in schizophrenia and found considerable evidence in support of clozapine’s superiority. Citrome et al. (2) and Volavka et al. (3) reported on a large randomized double-blind study of patients with schizophrenia or schizoaffective disorder, but the patients were not selected for aggression. Patients were randomly assigned to receive clozapine, olanzapine, risperidone, or haloperidol. Clozapine was the only agent to reduce hostility, as measured by the Positive and Negative Syndrome Scale (PANSS) (4). This agreement on clozapine’s superior antiaggressive effect in schizophrenia is reflected in published expert consensus guidelines (5).

The evidence for the antiaggressive effect of other second-generation antipsychotic agents is not as strong as it is for clozapine. Some studies report stronger antiaggressive properties for some of these agents compared with first-generation agents (6). A meta-analysis comparing various first- and second-generation medications in the treatment of hostility found a small but significant effect for second-generation antipsychotics compared with first-generation antipsychotics (7). This superiority, however, was present only when higher dosages were used.

Several studies have failed to find stronger antiaggressive properties for these other second-generation antipsychotics. In the two studies mentioned above (2, 3), which found clozapine to have a stronger antiaggressive effect, olanzapine and risperidone were not superior to haloperidol. This was also the case in another study (8), which used data from the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study (9). The CATIE study compared the effectiveness of perphenazine, olanzapine, risperidone, quetiapine, and ziprasidone in the treatment of schizophrenia. The prevalence of aggressive behavior during the first 6 months of the study was determined by a retrospective interview conducted at the end of that period. Overall, the various medications did not differ in their antiaggressive effect (8).

In most of these studies, the authors conducted secondary data analyses from studies that were not primarily designed to identify risk factors for violence, and the patients were not selected for their aggressive behavior (10). In addition, in many of these studies, the measure selected was not actual aggression but the PANSS hostility item, which in these primarily nonviolent populations often does not include actual physical aggression.

An important risk factor for violence in adulthood in persons with schizophrenia is the presence of conduct disorder before age 15 (11, 12). In one study, for example, the number of violent offenses over a 2-year period among male patients with schizophrenia or schizoaffective disorder was increased 2.6-fold for those with a diagnosis of conduct disorder compared with those without the disorder (13).

Because of its importance, the presence of conduct disorder serves as a basis for a typology of violence in persons with schizophrenia proposed by Hodgins (11). In this model, some violent patients, termed “the early starters,” display a pattern of antisocial behavior and violence that emerges in childhood or early adolescence, before the onset of schizophrenia, and that remains stable across the lifespan. Another group of violent offenders with schizophrenia show no antisocial or violent behavior prior to the onset of the illness and become violent afterward.

It would be important, therefore, to know how the presence of conduct disorder affects the treatment of violence in schizophrenia. In the CATIE violence study (8), the effect of conduct disorder on response to treatment was investigated. It was found that adherence to antipsychotic treatment was associated with significantly reduced violence in the group without a history of conduct problems but not in the group with conduct problems. This result was interpreted as indicating that antipsychotic treatment was not effective in reducing violence when patients had conduct disorder. Clozapine was not tested in the study.

There may be several reasons why antipsychotic treatment, and more particularly first-generation antipsychotics, may be less effective in patients with conduct disorder compared with other patients. Violence in patients with conduct disorder is more ingrained and trait-like, as it predates illness onset. Furthermore, these patients exhibit evidence of abnormalities in brain structures (14) and in brain functioning (15) that are not present in patients without conduct disorder. These abnormalities may affect response to treatment.

Although the main effect of first-generation antipsychotics is their antipsychotic action, there is strong evidence that second-generation antipsychotics, and particularly clozapine, have a broader spectrum of effects, which include control of aggression and impulsive behavior. Clozapine has been reported to decrease various forms of aggression in animals (16), including aggression knockout mouse models that are insensitive to its sedative effect (17). In these various animal models, the antiaggressive effect is independent of any antipsychotic or sedative effect.

Clozapine’s broad antiaggressive effect is also evidenced in its superiority over other antipsychotics in the treatment of suicidality (18). In a large international trial, the International Suicide Prevention Trial (InterSePT) (19), clozapine was found to be superior to olanzapine in the treatment of suicidality in schizophrenia. This effect was independent of the reduction in psychotic symptoms. In addition, clozapine has been shown to be effective in the treatment of aggression in children and adolescents with conduct disorder but no other diagnosis (20).

The greater antiaggressive efficacy of the second-generation antipsychotics may be explained by their pharmacological action. They have lower affinity for dopamine D₂ receptors and greater affinities for serotonin (5-HT_1A, 5-HT_2A, 5-HT_2C, 5-HT₃, 5-HT₆, and 5-HT₇) and norepinephrine α₁ and α₂ neuroreceptors than first-generation antipsychotics (21). Norepinephrine and serotonin have been implicated in violent behavior (22). Deficiency or dysregulation of serotonin has been associated with violence (23). Drugs that block the 5-HT_1A and 5-HT₂ receptors appear to have selective antiaggressive effects in animals (22). Clozapine’s antiaggressive effect has been linked to its serotonergic action in animal models (24). The greater limbic selectivity of clozapine (25), shared also to some extent by olanzapine (26), may also play a role in this antiaggressive effect (27). Thus, second-generation antipsychotics, and in particular clozapine, might be effective even in patients with conduct disorder.

Another difference that may exist between patients with and without conduct disorder is the influence of the symptoms of schizophrenia, and more specifically, positive psychotic symptoms, on violence. In the CATIE study (28), positive psychotic symptoms were significantly associated with violence only in patients who did not have conduct problems. In the patients with conduct disorder, there was no relationship between positive symptoms and violence.

Our previous study (29) was a randomized double-blind 12-week clinical study specifically designed to assess the efficacy of clozapine and olanzapine for the treatment of violent behavior in patients diagnosed with schizophrenia or schizoaffective disorder and to compare these two medications with haloperidol, one of the most commonly used first-generation antipsychotics. To be included in the study, the patients had to be physically and repeatedly assaultive.

The overall treatment comparisons of the study (29) indicated that clozapine had a stronger antiaggressive effect than olanzapine and that the latter was superior to haloperidol. The antiaggressive effect was separate from the antipsychotic or sedative action of these medications, as there were no significant differences among the three medication groups in improvement of psychotic symptoms.

The effect of conduct disorder on treatment was not addressed in our original study. Based on the evidence of poorer response to treatment in patients with conduct disorder mentioned above (8), in the present study we examined the effect of conduct disorder on the efficacy of these three medications using the data from our original study.

We hypothesized that because clozapine has strong antiaggressive effects that are independent of its antipsychotic effects, its relative superiority over haloperidol would be even greater in patients with conduct disorder compared with patients without conduct disorder. To a lesser extent, we hypothesized that this would also be true of olanzapine: it would also have a relatively stronger antiaggressive effect than haloperidol in patients with conduct disorder.

Methods

Participants

Our study subjects were 99 patients between ages 18 and 60 years who had diagnoses of schizophrenia or schizoaffective disorder and who were randomly assigned to treatment with clozapine, olanzapine, or haloperidol in a 12-week double-blind study. The Structured Clinical Interviews for DSM-IV Axis I Disorders and for DSM-IV Axis II Disorders (30) were used to confirm diagnosis of schizophrenia and to assess childhood conduct problems, as well as drug and alcohol use and abuse. Besides the diagnostic interview, information was obtained through a systematic review of the patients’ charts, including charts from previous hospitalizations and from other hospitals. These charts included information provided by parents or other relatives but not information from actual school reports or evaluations done in childhood. The diagnosis was based on behaviors up to age 15.

Written consent was required from each patient according to a protocol approved by institutional review boards and compliant with the Declaration of Helsinski. For inclusion in the study, patients were required to have a clearly confirmed episode of physical assault directed at another person which consisted of actual physical contact with the victim, such as striking, kicking, or scratching. There was a further requirement for a separate assaultive incident, whether physical, verbal, or against property, occurring within 4 weeks of this assault.

Patients were excluded if they had been hospitalized for more than a year, if they had a history of nonresponse or intolerance to clozapine, olanzapine, or haloperidol, if they had medical conditions that would be adversely affected by these medications, or if they had received a depot antipsychotic within 30 days before randomization. Patients were also excluded if there was insufficient information to establish a diagnosis of conduct disorder. Eleven patients who participated in our original study were excluded from the present study on this basis.

Treatments

Patients who met the inclusion and exclusion criteria and provided signed informed consent were transferred to the research ward. The entire study was conducted on the research ward to provide a uniform environment for all patients. This setting also allowed for close monitoring of medication administration as all patients were carefully observed when taking their medications, thus ensuring high treatment adherence.

During a baseline screening period of 1 to 2 weeks, patients’ prestudy antipsychotic medications were adjusted so that the dosage at the end of the screening period did not exceed 750 mg/day of chlorpromazine equivalents. The median number of days from the index physical assault to randomization was 20 (interquartile range=14–26 days). After baseline assessments were completed, patients were randomly assigned to one of the three treatment arms: clozapine, olanzapine, or haloperidol. Additional information about treatment, including justification for the dosages selected (31–33), is provided in the Supplemental Methods Section on Treatment in the online supplement.

Measures

Raters who were blind to treatment group assignment performed all clinical research assessments. All study procedures, including blood draws, were identical for all three groups throughout the study in order to preserve the blind assessment. Drug and alcohol misuse were diagnosed using DSM-IV. Records of arrests and prosecutions were obtained from the New York State Division of Criminal Justice Services for all patients. Both violent and nonviolent crimes were recorded. Violent crime included homicide, rape, assault, burglary with a weapon, and armed robbery.

The Modified Overt Aggression Scale (MOAS) (34) was used to rate all incidents during the study period. The PANSS was used to assess psychiatric symptoms. Five PANSS factors, as determined by a factor-analytic study (35), were used to describe symptoms at baseline. Additional information about these scales is provided in the Supplemental Methods Section on Measures in the online supplement.

Statistical Analysis

The principal analyses of efficacy were based on the intent-to-treat principle; thus, all patients who underwent randomized assignment and received at least one dose of study medication were included in the analyses. Differences among the treatment groups in terms of categorical efficacy measures were investigated by generalized linear model analysis, which makes allowance for nonnormally distributed observations. As increasingly higher scores on the MOAS occurred with a decreasing frequency (i.e., a “J curve”), the generalized linear model analysis was based on the Poisson distribution. The MOAS total and physical assault scores were used as the dependent variables in separate analyses. Treatment assignment (medication group) and presence or absence of conduct disorder were used as the independent variables. The interaction between these two variables was also investigated. Post hoc pairwise analyses were conducted when the main effects or interaction effects reached significance.

Odds ratios were calculated to determine the effect size; we computed the odds ratios for a 1-point lower MOAS score to characterize the pairwise medication contrasts and to further contrast the effect of treatment in patients with and without conduct disorder.

The MOAS verbal assault score was not an outcome variable, but it was included to provide additional information about the aggression profiles of the groups with and without conduct disorder as well as the effect of the medications on verbal assaults in these two groups. The above analyses were also performed for this variable.

We used generalized linear model analysis to compare the age distribution for the first arrest for violent crime in both patient groups, as age at first arrest (in years) consisted of nonnormally distributed, left-truncated, and right-skewed count data.

We also investigated the effect of the three medications on psychotic symptoms and examined the role of conduct disorder in this antipsychotic effect. We did so by comparing the change in PANSS total scores and in PANSS positive symptom factor scores among the three medication groups and between the two conduct disorder groups.

In addition, we examined the role of positive psychotic symptoms on violence in the patients with and without conduct disorder. We investigated the effect of change in the PANSS positive symptom factor scores and the interaction between this change and conduct disorder grouping on the MOAS total and physical assault scores. We used least squares mean estimates to provide estimates for a unit improvement or a unit deterioration (i.e., a 1-point decrease or a 1-point increase on the PANSS positive symptom factor scores). A change of 1 scaled point (e.g., “moderate” to “mild” [improvement] or “mild” to “moderate” [worsening]) on the PANSS positive symptom factor scale was used as the unit of analysis.

Age, substance abuse or dependence, and duration of stay in the study were used as covariates in all of the above analyses. In addition, for the PANSS analyses, we used the relevant baseline PANSS value. We also used the baseline PANSS negative symptom factor score, as the patients without conduct disorder had significantly higher scores on this variable than patients with conduct disorder. However, as the inclusion of this variable did not change the results, it was not retained in the final analyses.

SAS, version 9.4 (SAS Institute, Cary, N.C.), was used for all statistical analyses.

Results

Sample Characteristics and Baseline Information

Ninety-nine patients were included in the study. These patients had sufficient historical information about conduct problems prior to age 15 to classify them as having or not having a DSM-IV diagnosis of conduct disorder.

The demographic and clinical characteristics of the patients with and without conduct disorder are summarized in Table 1. Patients in these two groups were similar on most variables but differed in age, duration of illness, substance abuse or dependence, and baseline PANSS negative symptom score, which was higher in patients without conduct disorder.

There were no differences in the number of patients who were arrested for violent crimes; 13 of 46 (28.3%) patients without conduct disorder and 14 of 53 (26.4%) patients with conduct disorder had been arrested for violent crime. There was, however, a significant difference in the age at first arrest for violent crime. In the group without conduct disorder, the mean age was 26.2 years (SD=10.3), while in the group with conduct disorder, the mean age was 21.9 years (SD=6.2) (Wald χ²= 5.37, p=0.02).

The medication groups (i.e., the patients treated with clozapine, olanzapine, or haloperidol) were very similar on all demographic variables, illness characteristics, and baseline psychiatric symptoms. In addition, there were no differences among the three medication groups in the proportion of patients who were taking first- or second-generation antipsychotic agents prior to randomization, and there were no differences in the proportion of patients who were taking other psychotropic medications, including mood stabilizers or antidepressants. There were no differences among the three groups in the number of assaults or the length of hospitalization prior to entry in the study. The median length of hospitalization prior to study entry was 48 days (interquartile range=15–113).

The duration of stay in the study for each group is presented in Table S1 in the online supplement. There were no differences in the duration of stay in the study among the three medication groups (F=1.08, df=2, 98, p=0.34) and no difference between the patients with and without conduct disorder (F=0.84, df=1, 98, p=0.36). A total of 34 of 46 (73.9%) patients without conduct disorder and 34 of 53 (64.2%) patients with conduct disorder completed the full 12-week study. This difference was not significant (χ²=1.09, N=99, p=0.30)

Dosages of each study medication for the two conduct disorder groups are presented in Table S1 in the online supplement. Additional information about dosage is provided in the Supplemental Results Section on Dosages of Medication and Additional Medications in the online supplement.

Assaultive Behaviors

The main goal of this study was to determine the effect of conduct disorder on response to treatment. Medication group had a significant effect in determining both the MOAS total score (F=30.1, df=2, 98, p<0.001) and the physical assault score (F=26.9, df=2, 98, p<0.001). Clozapine was superior to olanzapine (p<0.01), which was superior to haloperidol (p<0.01).

Conduct disorder grouping also had a significant effect in determining the MOAS total score (F=261.3, df=1, 98, p<0.001) and the physical assault score (F=226.0, df=1, 98, p<0.001). Patients with conduct disorder had a significantly higher MOAS total score (N=53; least squares mean=42.6, SE=0.94) than patients without conduct disorder (N=46; least squares mean=22.1, SE=0.74). The odds ratio indicated an approximately twofold increase in assaults when conduct disorder was present compared with when it was absent (odds ratio=1.92, 95% CI=1.78–2.08, z=16.16, p<0.001).

Patients with conduct disorder also had a significantly higher physical assault score (N=53; least squares mean=20.6, SE=0.66) than patients without conduct disorder (N=46; least squares mean=7.4, SE=0.44). These patients were almost three times as likely to have greater physical aggression than those without conduct disorder (odds ratio=2.78, 95% CI=2.44–3.23, z=15.03, p<0.001).

There were significant interactions between medication and conduct disorder groupings in determining the MOAS total (F=14.9, df=2, 98, p<0.001) and physical assault scores (F=17.2, df=2, 98, p<0.001). The means and standard errors for each group, as well as the pairwise group differences, are presented in Table 2 (see also Figure S1 in the online supplement). Overall, clozapine was superior to both haloperidol and olanzapine, and olanzapine was superior to haloperidol, but the relative efficacy of these medications differed between patients with and without conduct disorder.

The odds ratios and confidence intervals for all of these comparisons are provided in Table 3. The data represent the odds of fewer and less severe assaults for the first, as compared with the second, medication in the pair presented in the table. When conduct disorder was not present, patients who received clozapine were almost twice as likely to have a lower MOAS total score (odds ratio=1.92, 95% CI=1.61–2.27) and over three times as likely to have a lower physical assault score (odds ratio=3.09, 95% CI=2.27–3.13) than patients treated with haloperidol. However, when conduct disorder was present, the odds ratios increase: patients who received clozapine were almost three times as likely to have a lower MOAS total score (odds ratio=2.70, 95% CI=2.38–3.03) and four times as likely to have a lower physical assault score (odds ratio=4.12, 95% CI=3.45–4.76)

With respect to the olanzapine-haloperidol comparison, for the MOAS total score, we found that patients without conduct disorder were 1.15 times (95% CI=1.02–1.30) more likely to have a lower score with olanzapine than with haloperidol, while patients with conduct disorder were 1.76 times (95% CI=1.59–1.95) more likely to have a lower score with olanzapine. Similarly, for the physical assault score, patients without conduct disorder were 1.20 times (95% CI=0.99–1.45) more likely to have a lower score with olanzapine than with haloperidol, while patients with conduct disorder were 2.44 times (95% CI=2.12–2.82) more likely to have a lower score.

As mentioned above, the principal analyses of efficacy were based on the intent-to-treat principle; thus, all patients randomly assigned to receive at least one dose of study medication were included in the analyses. In addition, we repeated the main overall analyses including only the patients who completed the full 12 weeks of the study. The results were similar to the results reported above. Medication group had a significant effect in determining both the MOAS total (F=11.0, df=2, 67, p=0.004) and physical assault (F=22.0, df=2, 67, p<0.001) scores. Clozapine was superior to olanzapine, which was superior to haloperidol. Conduct disorder grouping also had a significant effect in determining the MOAS total score (F=159.5, df=1, 67, p<0.001) and the physical assault score (F=61.6, df=1, 67, p<0.001). Patients with conduct disorder had a significantly higher MOAS total score than patients without conduct disorder. There were significant interactions between medication and conduct disorder groupings in determining the MOAS total (F=122.9, df=2, 67, p<0.001) and physical assault scores (F=42.6, df=2, 67, p<0.001). These interactions were in line with the interactions reported above.

Although not used as an outcome variable in this study, verbal assaults were investigated to delineate further the profile of aggressive behaviors. There was a significant main effect for conduct disorder in determining verbal assaults (F=66.8, df=1, 98, p<0.001); patients with conduct disorder had higher scores. The main effect for medication grouping was not significant (p>0.1). There was, however, a significant interaction between medication and conduct disorder groupings (F=27.6, df=2, 98, p<0.001). Among patients without conduct disorder, there were no significant pairwise differences in verbal assaults between the medications (p>0.1 for all three comparisons). On the other hand, among patients with conduct disorder, all three pairwise comparisons between medications were significant. These comparisons pertained to clozapine and haloperidol (z=8.72, p<0.001; odds ratio=0.46, 95% CI=0.39–0.55), clozapine and olanzapine (z=5.45, p<0.001; odds ratio=0.64, 95% CI=0.55–0.75), and olanzapine and haloperidol (z=4.02, p<0.001; odds ratio=0.72, 95% CI=0.61–0.85).

Schizophrenia Symptoms

Differences among the three medication groups and between patients with and without conduct disorder in improvement of symptom severity were investigated. There were no significant differences among the three medication groups and between patients with and without conduct disorder in the change in the total PANSS score or in the change in the PANSS positive symptom factor score; there were no significant interaction effects (p>0.1). Additional information is provided in the Supplemental Results Section on Psychiatric Symptoms in the online supplement.

We also investigated the influence of positive psychotic symptoms on assaultive behavior. There was a significant main effect for change in positive symptoms during treatment in determining the MOAS total score (F=250.0, df=1, 98, p<0.001) and the physical assault score (F=137.0, df=1, 98, p<0.001). With greater improvement in positive symptoms, the MOAS scores were lower. There was also, once again, a significant main effect for the conduct disorder classification in determining the MOAS total (F=223.2, df=1, 98, p<0.001) and physical assault (F=230.2, df=1, 98, p<0.001) scores. Medication grouping and the interaction between conduct disorder and medication grouping, which were used in the principal analyses, were retained in the present analyses, as described in the Methods section. The effect of these variables on the MOAS total and physical assault scores remained highly significant, as they had been in the principal analyses.

Our primary interest in these analyses was to compare the influence of positive symptoms on assaultive behavior in patients with conduct disorder against their influence in patients without conduct disorder, that is, the interaction between conduct disorder classification and change in PANSS positive symptom factor scores. This interaction was significant for both the MOAS total score (F=44.8, df=1, 98, p<0.001) and the physical assault score (F=18.1, df=1, 98, p<0.001).

To clarify the above interactions, the effect of change in PANSS positive symptoms on assaultive behavior among patients with conduct disorder is contrasted to their effect among patients without conduct disorder. As shown in Table 4, in patients with conduct disorder, a worsening of positive symptoms during the study period resulted in a 42% and a 65% increase in the MOAS total and physical assault scores, respectively, compared with these scores when positive symptoms improved. In patients without conduct disorder, worsening of positive symptoms has a stronger effect on assaultive behavior and particularly on physical assaults: the MOAS total score was twice as high and the physical assault score almost two and a half times as high when there was a worsening of symptoms as compared with an improvement in symptoms.

Discussion

Our study is consistent with the literature (11–13) in indicating greater violence among patients with schizophrenia who present with a history of conduct disorder prior to age 15. These patients were more assaultive than those without conduct disorder across the medication groups. This was true for all forms of aggression. This violence was not connected to substance abuse, as patients did not have access to drugs on the unit. Furthermore, past history of substance abuse did not influence the results. Although patients with conduct disorder did not have more arrests for violent crimes than patients without conduct disorder, as determined by official arrest records, they were significantly younger than patients without conduct disorder when they committed their first violent crime. This finding is an important independent confirmation from external records of the earlier onset of violence in this population.

The primary purpose of our study was to determine the effect of conduct disorder on the efficacy of different antipsychotic agents in the treatment of assaultive behavior in patients with schizophrenia. We found that the second-generation antipsychotics had a stronger antiaggressive effect than the first-generation antipsychotic haloperidol. Furthermore, although patients with conduct disorder in all three medication groups presented with more frequent and severe assaults than patients without conduct disorder, the relative pairwise differences among the medications were more prominent when conduct disorder was present. Thus, while clozapine had greater antiaggressive efficacy than haloperidol among patients without conduct disorder (the odds ratio for reduced physical assault scores was three times greater for clozapine than for haloperidol), its superiority was more pronounced among patients with conduct disorder, where the odds ratio was four times greater for clozapine than for haloperidol.

Clozapine was superior to olanzapine in its antiaggressive effect, as was also the case for suicidality in the InterSePT study (19). Although olanzapine was not as good an antiaggressive agent as clozapine, its relative efficacy compared with haloperidol was also greater among patients with conduct disorder compared with patients without conduct disorder.

It should be noted that our study may not have fully assessed the antiaggressive efficacy of clozapine. For clinical reasons, clozapine had to be escalated more slowly to the target dosage than was the case for the other two medications. Therefore, there was less time to assess the efficacy of the full dosage, so its superiority could be even greater than what we found.

The present results differ in two ways from those of the CATIE study (8). In that study, the second-generation antipsychotics did not have a stronger antiaggressive effect than the first-generation agent, and antipsychotic treatment did not have any effect on violence among patients with conduct disorder. There are several reasons why our findings may differ. In the CATIE study, clozapine was not tested; in our study, the antiaggressive effects were strongest for this medication. The patients in the CATIE study, as mentioned above, were not selected on the basis of aggression but were a convenience sample. In our study, on the other hand, the patients had to be physically and repeatedly assaultive to be included. The assessment of violence also differed in the two studies. In the CATIE study, the assessment was based mostly on patients’ self-report, and it was a dichotomous variable (presence or absence of violence). In our study, it was based on observed behavior and was quantified using a carefully psychometrically validated scale, which provided for a reliable and sensitive assessment. In addition, many of the patients in the CATIE study were not adherent to treatment, which decreases assay sensitivity. In our study, there was a careful checking of treatment adherence. Furthermore, it is possible that patients who do not adhere to treatment are more reluctant to admit being violent in their self-report. Thus, the fact that adherence to antipsychotic medication did not result in less violence among patients with a history of conduct disorder in the CATIE study may be due to an underreporting bias in these patients.

We also looked at the effect of antipsychotic treatment on verbal aggression. In the conduct disorder group, the results were similar to the results obtained for the MOAS total and physical assault scores: clozapine was superior to olanzapine, which was superior to haloperidol. In contrast, in the group without conduct disorder, the difference among the groups did not reach significance. It may be that verbal aggression in patients with more ingrained violence (i.e., the conduct disorder group) is similar to physical aggression, while in the group without conduct disorder, it is more strongly linked to psychosis.

We also investigated the effect of antipsychotic treatment on psychiatric symptoms in the three medication groups in patients with and without conduct disorder. There were no significant differences among the medication or conduct disorder groupings and no interaction effect. In our population of repeatedly assaultive patients, the antiaggressive action of the medications does not appear to be mediated through their antipsychotic effect, whether the patients have or do not have conduct disorder.

Although the primary purpose of our study was to determine the influence of conduct disorder on antipsychotic treatment, we investigated as a secondary goal the differential impact of positive psychotic symptoms on violence as a function of conduct disorder. Swanson et al. (28) found that positive symptoms were a determinant of violence only in patients without conduct disorder. In our study, while positive symptoms had a stronger impact in the group without conduct disorder, they still exerted an influence on violence in patients with conduct disorder, as the worsening of positive psychotic symptoms was associated with greater violence.

Swanson et al. (28) have theorized that there may be different pathways to violence in schizophrenia: in patients with conduct disorder, it is the early personality traits and behavioral difficulties that lead to violence, while in patients without conduct disorder, it is the symptoms of the illness that determine the violence. Our results are in partial agreement with this formulation. Patients with conduct disorder present with a violence that is more severe and more resistant to treatment and that appears to be linked to early behavioral problems. Yet even in these patients, coexisting positive symptoms contribute to the violence. Thus, while there are different pathways to violence, they are not mutually exclusive.

Strengths of the study include the fact that the patients were selected on the basis of actual violence, with well-documented episodes of physical assault. In many studies in the literature, the prevalence of actual overt physical assaults is too low to provide useful information. In addition, in our study, the measure of aggression did not depend on self-report but was based on careful monitoring of all assaults.

Weaknesses of the study include the fact that the patients, for the most part, presented with chronic illness; our results may not be generalizable to patients with more acute forms of schizophrenia. In addition, because patients had to provide consent to enter the study, our results may not be applicable to more uncooperative patients.

In summary, this is the first study to examine the efficacy of clozapine in the treatment of violence in patients with schizophrenia and conduct disorder. The findings point to the superiority of clozapine, and to a lesser extent olanzapine, for these patients and have important implications for treatment. It is imperative to obtain a good history of conduct disorder, for if it is present, clinicians should consider clozapine as the primary drug for treatment.