Suicidal Ideation and Suicide Attempts in Recent-Onset Schizophrenia

Suicidal Ideation and Suicide Attempts in Recent-Onset Schizophrenia by Alexander S. Young, Keith H. Nuechterlein, Jim Mints, Joseph Ventura, Michael Qitlin, and Robert P. Liberman Abstract The objective of this study was to predict suicidality in people with schizophrenia. Ninety-six patients with recent-onset schizophrenia were rated every 2 weeks for 1 year to examine (1) the temporal course of suicidal ideation and suicide attempts and (2) the extent to which anxiety, depression, and mild suicidal ideation were followed by significant suicidal ideation or a suicide attempt. The severity of suicidality changed rapidly. Low levels of suicidal ideation increased the risk for significant suicidal ideation or a suicide attempt during the subsequent 3 months. Depression was moderately correlated with concurrent suicidality, but not independently associated with future suicidality. Therefore, low levels of suicidal ideation may predict future suicidal ideation or behavior better than depressed mood in individuals with schizophrenia. Key words: Suicidal ideation, depression, anxiety. Schizophrenia Bulletin, 24(4):629-634,1998. Suicide remains a tragically common outcome in people with schizophrenia. In fact, about 10 percent of all people with this disorder die by suicide (Allebeck 1989; Roy 1992). Completed suicides have long-lasting effects on families and clinicians, and suicidal ideation and attempts place stress on the people and systems that provide care for individuals with schizophrenia (Rissmiller et al. 1994). As in other psychiatric disorders, no method exists for predicting suicide in any individual case of schizophrenia (Goldstein et al. 1991; Caldwell and Gottesman 1992). Researchers have, however, characterized individuals who are generally at higher risk of suicide. Retrospective case reviews and large followup studies have identified risk factors for suicide in schizophrenia, including being young and male, being in the first decade of illness, having a high level of premorbid functioning, and having multiple relapses (Caldwell and Gottesman 1990; Rossau and Mortensen 1997). Most studies have not, however, provided a detailed assessment of clinical status before attempted or completed suicide. It is just this clinical information that would be most useful for predicting, intervening, and preventing suicide attempts (Liberman and Eckman 1981; Linehan et al. 1991). Although it is known that individuals who commit suicide often have a previous history of suicide attempts and depressive symptoms, the predictive value of these markers is unclear (Drake et al. 1985). We do not know, for instance, whether these markers identify a state of depression or traits of individuals who are more vulnerable to suicidal behavior (Schotte and Clum 1987; Mann and Arango 1992). The known demographic and clinical risk factors for suicide identify a large population of individuals, many of whom will be at increased risk throughout years of treatment in community settings. More specific clinical indicators might be developed through a better understanding of the time period of increased risk conferred by each clinical risk factor. In this study, we repeatedly evaluated depression, anxiety, and suicidality over a 1-year period to examine (1) the temporal course of suicidal ideation and suicide attempts and (2) the extent to which anxiety, depression, and mild suicidal ideation are followed by significant suicidal ideation or a suicide attempt. Methods Subjects were participants in an intensive longitudinal project examining symptoms, neurocognitive and psychophysiological processes, psychosocial functioning, and environmental stressors in patients with recent-onset schizophrenia. This project and its sample selection crite- Reprint requests should be sent to Dr. A.S. Young, Dept. of Psychiatry, UCLA Neuropsychiatric Institute, 300 UCLA Medical Plaza, Box 956968, Los Angeles, CA 90095-6968. 629

Schizophrenia Bulletin, Vol. 24, No. 4, 1998 A.S. Young et al. ria have been described in detail elsewhere (Nuechterlein et al. 1992). Subjects met the following criteria: (1) first onset of a psychotic disorder within 2 years before project entry; (2) diagnosis by Research Diagnostic Criteria (Spitzer et al. 1978) of schizophrenia or schizoaffective disorder, mainly schizophrenia; (3) age between 18 and 45 years; (4) no evidence of significant and habitual substance abuse in the past 6 months; and (5) no evidence that substance abuse triggered the psychotic episode. Twenty-nine potential subjects were excluded because they had significant and habitual substance abuse during the 6 months before screening or substance abuse that may have triggered their psychosis. Eighty-one percent of subjects were male; 86 percent were Anglo, 6 percent Latino, 3 percent Asian, and 4 percent other. At project entry, the mean age was 23.4 years (range, 18 to 43). Educational level was defined as the highest grade completed at project entry and averaged 12.5 years (range, 8 to 20). Social class for family of origin was assessed using the Hollingshead Two-Factor Index of Social Position (Hollingshead 1971), which is scored from 1 (highest) to 5 (lowest). There were 13 subjects in social class 1, 22 in class 2, 21 in class 3, 34 in class 4, and 5 in class 5 (mean class = 3.0, standard deviation [SD] = 1.2). One of the 96 subjects had missing social class data. Total duration of illness was defined as the number of months since the onset of any prodrome or symptoms, and averaged 15.8 months at project entry. Structured diagnostic assessment determined that 24 subjects had schizophreniform disorder, 62 had schizophrenia, and 10 had schizoaffective disorder according to DSM-HI-R criteria (American Psychiatric Association 1987). All subjects gave written informed consent after the study had been fully explained. Subjects received their mental health treatment through the Aftercare Program, a research clinic at University of California Los Angeles (UCLA). A multidisciplinary treatment team provided case management, socialization activities, and patient and family education. Medication management included a standard starting dosage of fluphenazine decanoate (12.5 mg every 2 weeks) with dosage changes for intolerable side effects and significant symptoms. Adjunctive psychotropic medications (e.g., antidepressants) were used as clinically needed. Subjects were typically seen weekly at the Aftercare Program for the first few weeks after entry, and then every other week on an ongoing basis. A psychiatrist monitored clinical symptoms, prescribed medication, and provided medical supervision for each subject. Social workers and psychologists served as individual case managers and cotherapists for group therapy that emphasized problem-solving strategies and training in social skills. Family education involved a 3- to 4-hour session led by multiple clinical staff for groups of new families. Followup sessions with individual families were provided as clinically indicated. This study examines the first year after subjects were stabilized on outpatient medication in the research clinic. Ratings for this study begin 1 month after the starting outpatient medication dosage was reached, that is, an average of 3 months after subjects entered the research project. Case managers rated anxiety, depression, and suicidal ideation and behavior using the Expanded UCLA version of the Brief Psychiatric Rating Scale (BPRS; Ventura et al. 1993), which is an adaptation of the original BPRS (Overall and Gorham 1962). Mean intraclass correlation coefficients with criterion ratings were between 0.89 and 0.93. The Expanded BPRS includes explicit anchor points for each rating level from "not present" (1) to "extremely severe" (7), and an item developed at our Center to assess suicidality. The suicidality item instructs the interviewer to ask specifically about suicidal ideation and behavior at each interview and uses anchor points derived from prior indicators of severity (Pallis et al. 1982). Suicidality ratings below moderate (< 4) include anchors such as "occasional feelings of being tired of living" that may not reach the threshold for clinical attention or treatment in many clinical settings. Ratings of moderate (4) and moderately severe (5) identify subjects who frequently consider suicide or make a suicide attempt with low lethality. Ratings of severe (6) and extremely severe (7) identify subjects who clearly want to kill themselves or who make a suicide attempt of high lethality. For these analyses, "significant suicidality" was defined as a rating of moderate or higher and "low-level suicidality" as a rating of very mild or mild. Thus, "significant suicidality" included either frequent suicidal ideation or any suicidal behavior. Similarly, "significant depression" and "significant anxiety" indicate a rating of moderate or higher on the relevant item. BPRS ratings were repeated about every 2 weeks for 1 year and were based on symptoms and behaviors since the previous assessment. Associations between ratings of suicidality and depression were evaluated using concurrent and lagged correlations. Analyses were done by performing crosscorrelation analyses within subjects and then pooling results across subjects for significance testing. First, a simple Pearson r was computed within each subject's repeated scores. These correlations were z-transformed, weighted by their squared standard errors, and averaged across subjects. The significance of the resulting average was evaluated with reference to a normal curve test (Fleiss 1981). Because of autocorrelation in these time series (i.e., ratings at a given visit were correlated with those at the next visit), statistical analysis of the significance of these correlations required time series analyses. 630

Suicidal Ideation in Recent-Onset Schizophrenia Schizophrenia Bulletin, Vol. 24, No. 4, 1998 These were performed using conventional time series regression models (Box and Jenkins 1976) computed with SAS (SAS User's Guide 1989) procedure ARIMA. We examined both first-degree autoregressive and moving average models. Results were essentially equivalent, and only the former are presented. Risk analyses were performed using survival regression models (Cox and Oakes 1984) computed with the SAS procedure PHREG. The criterion was a suicidality score of moderate or above. At-risk indicators of low-level suicidality and significant depression were included as time-varying covariates. Results Ninety-six subjects were rated on the Expanded BPRS at approximately 2-week intervals for up to 1 year. Eightythree subjects were rated for at least 40 weeks, with an average of 18.5 ratings per subject. The number of ratings per subject ranged from 2 to 28, with only 11 subjects having fewer than 10 ratings. Subjects who left the project early had relatively few ratings, and ratings were completed more frequently when clinically indicated. Overall, 2 percent of suicidality ratings, 13.6 percent of depressed mood ratings, and 10.2 percent of anxiety ratings were at the moderate level or above. Of the 96 subjects in the sample, 51 rated at "not present" on the suicidality item at every assessment during the year. Twelve subjects were significantly suicidal at one or more times during the year. In 11 subjects, this consisted of suicidal ideation without a suicide attempt. One subject had two suicide attempts of relatively low lethality. There were no deaths from suicide. Thus, "significant suicidality" in this report refers to clinical levels of suicidal ideation and, for one subject, nonlethal suicide attempts. Table 1 presents the rate of significant suicidality by subject gender, age, ethnicity, education, social class, diagnosis, and length of illness. The statistical power to detect differences was limited, and no significant differences were found at the p < 0.05 level. Very few episodes of suicidality had a distinct onset, plateau, and offset. Instead, the level of suicidality changed rapidly. Thirty-two rating periods included ratings of significant suicidality. Of these, 7 were increases from "not present," 6 were increases from "very mild," 9 were increases from "mild," and the remaining 10 were a continuation of previous significant suicidality. Conversely, 0.7 percent of "not present" ratings, 6.8 percent of "very mild" ratings, and 19.1 percent of "mild" ratings were followed by significant suicidality. Most instances of ideation were not followed by a suicide attempt in this sample. Ratings from the 12 subjects with significant suicidality were examined visually. All 12 had suicidality ratings Table 1. Presence of significant suicidality by demographic characteristic Gender Female (n = 18) Male (n = 78) Age 22 years and younger (n = 49) Older than 22 years (n = 47) Ethnicity Minority (n= 13) Anglo-American (n = 83) Education completed High school graduate or less (n = 62) Beyond high school (n = 34) Social class Lower (class 4 or 5; n = 39) Higher (class 1, 2, or 3; n = 56) Diagnosis (DSM-III-R) Schizophreniform (n = 24) Schizophrenia (n = 62) Schizoaffective (n = 10) Months of illness at admisslor to study 2 12 months or fewer (n = 45) Subjects with significant suicidality (%) s l 11.1 12.8 14.3 10.6 7.7 13.3 16.1 5.9 12.8 12.5 8.3 12.9 20.0 8.9 More than 12 months (n = 51) 15.7 1 0.039 0.29 0.32 2.11 0.002 0.90 1.01 Note. DSM-III-R = Diagnostic and Statistical Manual of Mental Disorders, 3rd ed., revised (American Psychiatric Association 1987). 1 No differences are significant at the p < 0.05 level. 2 Lifetime number of months with either prodromal symptoms or characteristic schizophrenic symptoms. of very mild or higher (a 2) on at least one other occasion, and most had numerous occurrences of suicidality over a range of severity levels. Ratings from two subjects are graphed in figure 1. As in subject 1, most subjects had concurrent depressed mood when they were suicidal. However, one individual (subject 2) did not have concurrent depression when suicidal. We used time series analyses to test a conceptual model predicting that current suicidality would be associated with (1) depressed mood at the current assessment, (2) suicidality at the previous assessment, and (3) depressed mood at the previous assessment (mediated through an effect on depressed mood at the current assessment). Thirty-six subjects had variability on both the sui- 631

Schizophrenia Bulletin, Vol. 24, No. 4, 1998 A.S. Young et al. Figure 1. Expanded BPRS ratings for one subject with concurrent suicidaiity and depression and one subject without concurrent suicidaiity and depression S RATINQ m 3 2 1 SUBJECT 1 [ 44, t I SUICIDALITY! DEPRESSED MOOD 20 30 40 WEEK BPRS = Brief Psychiatric Rating Scale-Expanded (Ventura et al. 1993). cidality and depression scales. A moderate concurrent correlation, averaging 0.41 across subjects, was found between suicidaiity and depression. The lagged correlation between one depression rating and the subsequent suicidaiity rating averaged 0.12 across subjects. The time series analysis revealed that the concurrent relationship between depression and suicidaiity was significant (z = 2.55, p = 0.01). However, the lagged relationship was clearly not significant (z = 0.50, p = 0.62). Thus, suicidaiity was significantly associated with depression at the same clinic visit, but not with depression at the previous visit. The above procedures were repeated for anxiety and suicidaiity, and no significant effects were found. We used survival models to test the prognostic significance of depressed mood and low-level suicidal ideation. The criterion was a "significant" suicidaiity rating of moderate or greater. Subjects were defined as at-risk from suicidaiity after they had a low-level suicidaiity rating of very mild or mild. Subjects were defined as at-risk from depression after they had a depressed mood rating of moderate or greater. The ability of each of these two warning flags to predict subsequent significant suicidaiity ratings was assessed for time durations of risk varying from 2 to 52 weeks. s presented in table 2 demonstrate that low-level suicidal ideation was a highly significant predictor. Once a subject displayed low-level ideation, the risk of a subsequent occurrence of a significant suicidaiity rating increased on the order of 20- to 30- fold. When this warning flag was examined as a test, the sensitivity increased from 25 percent at 2 weeks to a maximum of 67 percent at 12 weeks. Sensitivity calculations indicate the percentage of subjects with significant suicidaiity whose first episode of significant suicidaiity was preceded by one or more ratings of low-level suicidaiity. Twelve weeks was the optimal "window" duration for identifying at-risk subjects on the basis of low-level suicidal ideation. Depressed mood did not have a statistically significant effect on subsequent significant suicidaiity (table 2). Any increase appeared to be attributable to the concurrent correlation between depression and suicidaiity. However, low-level suicidaiity increased the odds of a subsequent significant suicidaiity rating even after controlling for depression. Of the 34 subjects who had the onset of very mild or mild suicidal ideation, 24 percent went on to develop significant suicidaiity within 9 weeks. After this Table 2. Time interval (weeks) 2 4 8 12 16 52 1 p<0.01. 2 p < 0.05. 3 p< 0.001. Risk of significant suicidaiity after depression and suicidal ideation Risk from depression 3.6 1.4 1.7 2.5 1.9 0.7 controlling for suicidaiity 0.7 0.1 0.1 0.0 0.1 0.3 Sensitivity (%) 25 50 58 67 67 67 Risk from suicidal ideation 8.2 1 19.6 3 18.3 3 18.9 3 17.8 3 15.3 3 controlling for depression 4.4 2 18.2 3 16.8 3 17.1 3 16.7 3 15.4 3 632 J

Suicidal Ideation in Recent-Onset Schizophrenia Schizophrenia Bulletin, Vol. 24, No. 4, 1998 interval, no further significant suicidality was observed. Of the 12 subjects who developed significant suicidal ideation or made a suicide attempt, 8 had very mild or mild suicidal ideation within the previous 9 weeks. Discussion Our results demonstrate that (1) the severity of suicidality changed rapidly, and suicidality did not occur in apparent episodes; (2) the probability of future significant suicidal ideation or behavior was predicted by low-level suicidal ideation, but not by anxiety or depression; and (3) subjects with low-level suicidal ideation continued to have an increased chance of significant suicidal ideation or a suicide attempt for about 3 months. These results suggest that previous findings of an association between depressed mood and subsequent suicidal behavior may be due to concurrent suicidality. While suicidality and depression often co-occur, it appears that only suicidal ideation predicts future suicidal ideation or behavior. Also, the at-risk window after suicidality may be shorter than the 1-year period that previous retrospective studies have emphasized. This study has several potential limitations. First, completed suicide, suicide attempts, and suicidal ideation are clearly not equivalent. The suicidality in this patient population consisted almost entirely of mild to severe suicidal ideation. Therefore, we cannot assume that our findings apply to individuals who make suicide attempts or die by suicide. However, there is evidence that suicidal ideation and behavior can be viewed along a continuum (Beck et al. 1974; Murphy et al. 1992), suggesting that risk factors for significant suicidality might also identify individuals at risk for completed suicide (Pallis et al. 1984). Second, the subjects in this study were treated in a clinic that provided frequent assessment and more intensive care than that offered by many community clinics (Young et al. 1998). This study also excluded people with a significant recent history of substance abuse at the time of screening for project admission; this group may be at increased risk for suicide. Either of these may have lowered the rate of suicidality and reduced our ability to detect effects that might be seen in another sample. Third, depression and anxiety are just two components of the depressive syndrome. Other components, such as demoralization and hopelessness, may have more predictive power (Beck et al. 1990). However, the study from which these data were drawn was not designed with a specific aim of predicting suicidality and did not include assessment of these items. The effects seen in this sample were strong and consistent with previous studies of suicide in large populations (Caldwell and Gottesman 1990). This has several implications. First, clinicians should be aware that suicidality can change rapidly in people with recent-onset schizophrenia. We observed that severity can change markedly from week to week and could not exclude the possibility that it changes daily. Therefore, clinical interventions that are based on detecting suicidality may need to either identify a subgroup of patients with schizophrenia who are at increased risk for suicidal ideation or behavior and monitor this subgroup on a daily or weekly basis and/or encourage all patients with schizophrenia to report suicidal ideation and behavior immediately and create a system that makes it easy for them to do so. Second, we found that mild suicidal ideation can have prognostic significance. Low levels of ideation may predict future ideation or behavior better than depressed mood in schizophrenia. Also, it is likely that mild suicidality would often have been unknown to us had we not specifically asked about it. Although suicidal patients were often depressed, this was not always the case. Clinicians should consider specifically asking patients with schizophrenia about recent suicidal ideation and behavior and could make this assessment using a structured instrument such as the BPRS. Finally, we found a relatively brief period (9 weeks) during which people with suicidal ideation are at increased risk. If replicated in other samples, this suggests a need for evaluating assessment and treatment interventions that focus on this specific period of increased risk. References Allebeck, P. Schizophrenia: A life-shortening disease. Schizophrenia Bulletin, 15(l):81-89, 1989. American Psychiatric Association. DSM-IH-R: Diagnostic and Statistical Manual of Mental Disorders. 3rd ed. revised. Washington, DC: The Association, 1987. Beck, A.T.; Brown, G.; Berchick, R.J.; Stewart, B.L.; and Steer, R.A. Relationship between hopelessness and ultimate suicide: A replication with psychiatric outpatients. American Journal of Psychiatry, 147:190-195, 1990. Beck, A.T.; Resnik, H.L.P.; and Lettieri, D.J., eds. The Prediction of Suicide. Bowie, MD: Charles Press, 1974. Box, G.E.P., and Jenkins, G.M. Time Series Analysis: Forecasting and Control. San Francisco, CA: Holden- Day, 1976. Caldwell, C.B., and Gottesman, I.I. Schizophrenics kill themselves too: A review of risk factors for suicide. Schizophrenia Bulletin, 16(4):571-589, 1990. Caldwell, C.B., and Gottesman, I.I. Schizophrenia A high-risk factor for suicide: Clues to risk reduction. Suicide and Life-Threatening Behavior, 22:479-493, 1992. 633

Schizophrenia Bulletin, Vol. 24, No. 4, 1998 A.S. Young et al. Cox, D.R., and Oakes, D. Analysis of Survival Data. London, England: Chapman and Hall, 1984. Drake, R.E.; Gates, C; Whitaker, A.; and Cotton, P.G. Suicide among schizophrenics: A review. Comprehensive Psychiatry, 26:90-100, 1985. Fleiss, J.L. Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Wiley & Sons, 1981. Goldstein, R.B.; Black, D.W.; Nasrallah, A.; and Winokur, G. The prediction of suicide: Sensitivity, specificity, and predictive value of a multivariate model applied to suicide among 1906 patients with affective disorders. Archives of General Psychiatry, 48:418-422, 1991. Hollingshead, A.B. Commentary on "the indiscriminate state of social class measurement." Social Forces, 49:563-567, 1971. Liberman, R.P., and Eckman, T. Behavior therapy vs. insight-oriented therapy for repeated suicide attempters. Archives of General Psychiatry, 38:1126-1130, 1981. Linehan, M.M.; Armstrong, H.E.; Suarez, A.; Allmon, D.; and Heard, H.L. Cognitive-behavioral treatment of chronically parasuicidal borderline patients. Archives of General Psychiatry, 48:1060-1064, 1991. Mann, J.J., and Arango, V. Integration of neurobiology and psychopathology in a unified model of suicidal behavior. Journal of Clinical Psychopharmacology, 12:2S-7S, 1992. Murphy, G.E.; Wetzel, R.D.; Robins, E.; and McEvoy, L. Multiple risk factors predict suicide in alcoholism. Archives of General Psychiatry, 49:459-463, 1992. Nuechterlein, K.H.; Dawson, M.E.; Gitlin, M; Ventura, J.; Goldstein, M.J.; Snyder, K.S.; Yee, CM.; and Mintz, J. Developmental processes in schizophrenic disorders: Longitudinal studies of vulnerability and stress. Schizophrenia Bulletin, 18(3):387-425, 1992. Overall, J.E., and Gorham, D.R. The Brief Psychiatric Rating Scale. Psychological Reports, 10:799-812, 1962. Pallis, D.J.; Barraclough, B.M.; Levey, A.B.; Jenkins, J.S.; and Sainsbury, P. Estimating suicide risk among attempted suicides: I. The development of new clinical scales. British Journal of Psychiatry, 141:37^4, 1982. Pallis, D.J.; Gibbons, J.S.; and Pierce, D.W. Estimating suicide risk among attempted suicides: JJ. Efficiency of predictive scales after the attempt. British Journal of Psychiatry, 144:139-148, 1984. Rissmiller, D.J.; Steer, R.; Ranieri, W.F.; Rissmiller, R; and Hogate, P. Factors complicating cost containment in the treatment of suicidal patients. Hospital and Community Psychiatry, 45:782-788, 1994. Rossau, CD., and Mortensen, P.B. Risk factors for suicide in patients with schizophrenia: Nested case-control study. British Journal of Psychiatry, 171:355-359, 1997. Roy, A. Suicide in schizophrenia. International Review of Psychiatry, 4:205-209, 1992. SAS User's Guide: Statistics. 4th ed. Cary, NC: SAS Institute, 1989. Schotte, D.E., and Clum, G.A. Problem-solving skills in suicidal psychiatric patients. Journal of Consulting and Clinical Psychology, 55:49-54, 1987. Spitzer, R.L., Endicott, J.; and Robins, E. Research Diagnostic Criteria: Rationale and reliability. Archives of General Psychiatry, 35:773-782, 1978. Ventura, J.; Lukoff, D.; Nuechterlein, K.H.; Liberman, R.P.; Green, M.F.; and Shaner, A. Brief Psychiatric Rating Scale (BPRS) expanded version (4.0): Scales, anchor points, and administration manual. International Journal of Methods in Psychiatric Research, 3:227-243, 1993. Young, A.S.; Sullivan, G.; Burnam, M.A.; and Brook, R.H. Measuring the quality of outpatient treatment for schizophrenia. Archives of General Psychiatry, 55:611-617, 1998. Acknowledgments Dr. Young's work on this study was supported by a National Alliance for Research on Schizophrenia and Depression Daniel X. Freedman Young Investigator Award. The longitudinal research project was supported by USPHS grant MH-37705 from the National Institute of Mental Health (NIMH) to Dr. Nuechterlein. Diagnostic and symptom assessment training and the data analyses were supported by NIMH Clinical Research Center grant MH-30911 (to Dr. Liberman, Principal Investigator). The authors thank Sun Hwang for her contributions to the data analyses. The Authors Alexander S. Young, M.D., M.S.H.S., is Assistant Clinical Professor, Department of Psychiatry; Keith H. Nuechterlein, Ph.D., is Professor, Departments of Psychiatry and Psychology; Jim Mintz, Ph.D., is Professor, Department of Psychiatry; Joseph Ventura, Ph.D., is Assistant Research Psychologist, Department of Psychiatry; Michael Gitlin, M.D., is Clinical Professor, Department of Psychiatry; and Robert P. Liberman, M.D., is Professor, Department of Psychiatry, all at the University of California, Los Angeles, CA. 634