Predicting Adolescent Violence: Impact of Family History, Substance Use, Psychiatric History, and Social Adjustment

Probands were drawn from a pool of adult men (N=121) who either qualified for a lifetime DSM-III-R diagnosis of substance use disorder or had no substance use disorder or any adult axis I or II psychiatric disorder. Psychiatric comorbidity in the probands with substance use disorder was not an exclusionary factor, since this is highly common. The probands with no substance use disorder also had no adult psychiatric disorder so that the children would not be exposed to current paternal psychopathology.

Biological male offspring of the probands were first evaluated when the children were 12–14 years old; they were subsequently tracked to the age of 19 for follow-up evaluation. These boys have been involved in a long-term prospective investigation that began when they were 10–12 years old. However, because alcohol and drug use is infrequent in 10–12-year-old children, this study was conducted when they attained 12–14 years of age, at which time the rate of alcohol and drug use is sufficiently high to enable valid statistical analysis. Only one male child in each family was studied to avoid biased results inasmuch as children comprising a sibship are not independent subjects. The study was confined to boys because their higher rate of violence enabled the detection of predictors in this group of limited size.

Fifteen (12%) of the offspring were lost to follow-up after the baseline evaluation at ages 12–14, thereby reducing the study group size from 121 to 106. Comparisons between the dropouts and remaining subjects did not reveal significant differences in IQ or family socioeconomic status. In addition, the rate of lifetime psychiatric disturbance was similar between the dropouts and remaining subjects for the following diagnostic categories: attention deficit hyperactivity disorder (ADHD), conduct disorder, oppositional defiant disorder, anxiety disorder, and depression. There was no difference in attrition rate between the offspring of probands with and without substance use disorder (14.6% [N=7 of 48] and 11.0% [N=8 of 73], respectively; χ²=0.41, df=1, p=0.52). An additional seven families were excluded because of incomplete data. Thus, the study group was further reduced from 106 subjects to the final study group size of 99 subjects. The longitudinal analyses directed at predicting violence between ages 12–14 and 19 were conducted in this group of 38 boys whose fathers had a substance use disorder and 61 boys whose fathers had no adult psychopathology.

Multiple recruitment sources were used to minimize bias that could potentially occur if all of the probands were recruited from one source. Approximately 20% of the substance use disorder probands were derived from treatment facilities, whereas the remainder were recruited from the community through public service announcements and advertisements as well as by direct telephone contact conducted by a market research firm. The inclusion of men with substance use disorder who recently completed treatment or were in aftercare was based on the rationale that their disorder may be more severe than that of men not in treatment. Hence, combining treatment and nontreatment substance use disorder probands encompasses a broad spectrum of substance use disorder severity. Psychosis, mental retardation, and neurological injury or disease were exclusionary criteria for participation of the family. Recruitment of probands with no adult psychopathology was conducted through advertisements, public service announcements, and the same market research firm used to locate the substance use disorder probands.

After complete description of the study to the subjects, written informed consent was obtained. Parents provided informed consent for their minor children (age 12–14). The boys provided informed consent at the age 19 follow-up evaluation. Each family member was individually administered the research protocol in a private room by a different clinical associate who was not given any information about the subject’s medical, psychiatric, and legal history. The diagnostic interviews were conducted by a staff of four experienced clinical associates. Training the clinical associates involved observation of several interviews and conducting joint interviews in the presence of an experienced interviewer.

Each biological parent was administered the Structured Clinical Interview for DSM-III-R (SCID) (22) to document lifetime abuse/dependence for all categories of compounds that have such liability. The interview was administered by master’s-level clinical associates. The clinical associates underwent extensive training, including ongoing oversight by a board-certified psychiatrist. At the outset of this research program in 1989, the training procedures were found to produce interrater reliabilities exceeding 0.80 for all major diagnostic categories.

Diagnoses were formulated by using the best-estimate procedure (23). The SCID results, in conjunction with all available pertinent medical records and social and legal history, were reviewed in a clinical case conference chaired by a board-certified psychiatrist and consisting of another psychiatrist or psychologist along with the clinical associate who conducted the interview. DSM-III-R criteria were applied to derive lifetime diagnoses, since this research was initiated before the advent of DSM-IV. The aforementioned procedures were also used to characterize lifetime psychopathology other than substance use disorder.

Alcohol, tobacco, and marijuana use frequency were documented in the offspring of the probands at age 12–14 by using section 1A of the Drug Use Screening Inventory (24, 25). The past-month average frequency of consumption for these compounds was the score used to index level of drug involvement. Other drugs were not included because the rate of consumption at age 12–14 was too low to enable statistical analysis.

Offspring psychopathology was assessed with the Schedule for Affective Disorders and Schizophrenia for School-Age Children—Epidemiologic Version (K-SADS-E) (26); both the child and a parent (typically the mother) served as informants. The aforementioned best-estimate procedure was used to formulate lifetime psychiatric disorders that were based on the results of both parent and child K-SADS-E reports in conjunction with available medical, legal, and social history information.

A provisional scale to assess the offspring’s proneness to commit an act of violence was developed in a series of iterative analyses by using items from the revised Drug Use Screening Inventory. Scale construction and subsequent confirmation of its psychometric properties, consisting of six stages, were conducted within a group of 417 boys between 12–14 years of age. First, to narrow the relevant domains, the 10 domain scores of the Drug Use Screening Inventory were correlated with the total score on the Andrew Severity and History of Offenses Scale (27). Only two scale scores, school adjustment (r=0.42, df=415, p<0.001) and peer relations (r=0.45, df=415, p<0.001), demonstrated significant association. Second, to identify the specific items in these two scales that potentially were most predictive of violence, each item was correlated with the Andrew Scale total score by using point biserial correlation. A total of 20 Drug Use Screening Inventory items significantly correlated with violence outcome at age 19. Items having a low correlation (r<0.25) were eliminated, which resulted in one of the items being deleted from the item pool. The third stage involved testing the unidimensionality of the remaining 19 items by using confirmatory factor analysis. This procedure resulted in deletion of one item that had a nonsignificant factor loading. In the fourth stage of scale development, item response theory methods (28) were used to assess item quality by determining the item threshold and discrimination parameters of the remaining 18 items. Item threshold characterizes the severity of the property being assessed; that is, items having a higher threshold are endorsed in more severe cases. Item discrimination refers to the ability of an item to distinguish levels of severity on the trait, which in this study is violence proneness. Items having a discrimination of 0.40 or less were discarded. This analysis yielded a final set of 13 items with an average item discrimination of 0.96 (SD=0.33) and average item threshold of 0.47 (SD=0.53). These 13 items did not overlap with the items from section 1A of the Drug Use Screening Inventory used to document alcohol, tobacco, and marijuana use frequency. The unidimensionality of the Violence Proneness Scale was also established by using confirmatory factor analysis (χ²=26.39, df=21, p<0.20; root mean square error of approximation=0.025; goodness-of-fit index=0.99; adjusted goodness-of-fit index=0.96; nonnormed fit index=0.99). Last, the psychometric properties of the Violence Proneness Scale were documented. The coefficient of internal consistency (Cronbach’s alpha=0.79) and the average correlation of item and total test score (r=0.42) indicate that the Violence Proneness Scale has satisfactory reliability. The items comprising the Violence Proneness Scale and their psychometric characteristics are presented in Table 1.

The dependent variable, violent behavior, was assessed at age 19 with the Andrew Severity and History of Offenses Scale (27). This scale was used because it focuses on discrete behaviors rather than traits such as aggressivity. From a set of 65 items ranked according to severity, 13 items pertain to violent behavior. Although the range of violent behaviors is diverse, the manifest consequence is the same: the potential for intentional physical injury of another person. The questionnaire was completed as a self-report and subsequently reviewed by the research associate for accuracy and completeness. Questions or uncertainties about the purpose of the questionnaire as well as any other issues were addressed by the research associate. Significantly, this questionnaire documents discrete events that have face validity for documenting type and severity of violent behaviors and not dispositional behavior, personality, or psychopathology.

Because the data were binary, multivariate logistic regression analysis was conducted (29). This analysis determined the odds ratio for prediction of presence or absence of violent outcomes between ages 12–14 and 19. The predictor variables were the Violence Proneness Scale score, presence/absence of parental substance use disorder (with the exception of nicotine), offspring’s frequency of past month substance use, and presence of psychiatric disorder in the offspring. IQ and socioeconomic status were controlled in computing the odds ratios.