Declining Clinical Course of Psychotic Disorders Over the Two Decades Following First Hospitalization: Evidence From the Suffolk County Mental Health Project

The cohort was assembled by the Suffolk County Mental Health Project, an epidemiologic study of first-admission psychosis (11, 16, 17). Participants were recruited from the 12 psychiatric inpatient units of Suffolk County, New York, between 1990 and 1995. Inclusion criteria were first admission either current or within 6 months, clinical evidence of psychosis, ages 15–60, IQ >70, proficiency with English, residency in Suffolk County, and no apparent medical etiology for the psychosis. The study was approved annually by the institutional review boards of Stony Brook University and the participating hospitals.

We initially interviewed 675 participants (72% of referrals); 628 of them met the eligibility criteria. Follow-ups were conducted at 6 months, 24 months, 48 months, 10 years, and 20 years. Seventy-nine participants died during the 20 years. Of the 549 survivors, 373 were successfully contacted at year 20 and constitute the analysis sample. Of these, 68.9% lived in Suffolk County, 6.4% moved to another county in New York, and 24.7% moved to another state. Surviving nonparticipants—patients who completed baseline assessment but did not participate in the 20-year follow-up assessments and were not known to be deceased (N=176)—were similar to the analysis sample on baseline study variables (demographic characteristics, diagnosis, and symptoms) (Table 1), but they were less likely to be Caucasian (67.6% compared with 77.7%) and on average had more severe reality distortion symptoms (Cohen’s d=0.23). For the analysis sample (N=373), we had 2,046 observations across six waves (i.e., the data were 91.4% complete). Follow-ups that were done over the telephone (277 across the six waves) did not allow the behavioral ratings necessary for scoring inexpressivity, resulting in 1,769 observations available for inexpressivity. The primary analyses employed maximum likelihood estimation and thus used all available data.

Interviews were conducted by master’s-level mental health professionals. Medical records and interviews with significant others were solicited at every assessment. This multisource information was used to complete the following rating scales about past-month symptoms: the Scale for the Assessment of Negative Symptoms (SANS) (18), the Scale for the Assessment of Positive Symptoms (SAPS) (19), the Brief Psychiatric Rating Scale (BPRS) (20), and the Structured Clinical Interview for DSM-III-R (SCID) (21). For the present sample, we scored four reliable factor-analytically derived subscales from the SANS and SAPS: inexpressivity (Cronbach’s alpha≥0.88, nine items), apathy-asociality (alpha≥0.81, six items), reality distortion (alpha≥0.80, 14 items), and disorganization (alpha≥0.72, 11 items) (11). Mania was operationalized with the excitement rating of the BPRS. Depressive symptoms were assessed with the current depression module of the SCID, administered without skip-outs. We constructed a nine-symptom depression composite (score range, 9–27) that has excellent reliability (alpha≥0.81) and validity (11, 17). All ratings were highly reliable (see the Supplemental Methods section in the data supplement that accompanies the online edition of this article).

Primary DSM-IV diagnoses were formulated at the 10-year point by consensus of study psychiatrists using all available information (17). The same process was performed in previous waves, including the 6-month assessment. Diagnoses were grouped into five categories: schizophrenia/schizophreniform/schizoaffective disorder, bipolar disorder with psychosis, depression with psychosis, substance-induced psychosis, and other/undetermined psychosis (e.g., psychotic disorder not otherwise specified). Psychiatrists made consensus ratings of the Global Assessment of Functioning (GAF) for the best month of the year before interview, an index that captures both symptom burden and functional impairment. At year 20, psychiatrists also rated the overall pattern of clinical course, using the DOSMeD criteria (see the Supplemental Methods section in the data supplement) (22). For interpretability, we grouped eight course categories into three: single episode (i.e., baseline episode resolved, no recurrence), multiple episodes, and continuous illness (i.e., no remission).

We investigated trajectories of each disorder on seven outcome measures: the GAF (primary measure) and the six symptom dimensions. At baseline all participants were highly symptomatic and hospitalized; therefore, baseline could not be included in the model, and we started charting trajectories from month 6. We focus here on mean disorder trajectories. Within-group heterogeneity was reported previously (11).

First, we examined clinical course in bivariate analyses, using paired t tests to compare outcomes at follow-ups subsequent to month 6. We used independent-samples t tests to compare outcomes between disorders. Next, we charted trajectories of disorders across all waves by fitting multilevel spline regression models with a random intercept (see the Supplemental Methods section in the data supplement) (23–25). Models were fitted for each disorder separately; they estimated trajectories of individual participants and then calculated the mean trajectory for the group. These analyses took advantage of variation in follow-ups around target dates (i.e., some were done late and others early), which allowed us to chart trajectories through year 23. However, data were limited for years 5–8 and 13–16 (< 20 observations/year), and these portions of the trajectories were estimated less precisely. Spline regression is a piecewise regression that allows different slopes in different segments of the predictor variable. We considered up to three segments (the largest number suggested by descriptive analyses). Transition points between segments were determined empirically by testing the full range of possible transition points and selecting the model with the best Bayesian information criterion (26). The number of segments was determined similarly. Finally, we added age and antipsychotic medication as time-varying covariates to the resulting models to determine whether the observed changes were independent from variation in these covariates.

Table 2 summarizes the outcomes and antipsychotic medication use of the five diagnostic groups across the two decades. The pattern is notable for worse outcomes in schizophrenia. Differences among the other groups were less pronounced, although the bipolar disorder with psychosis group and the psychotic depression group often had better outcomes than the substance-induced and other/undetermined psychoses groups. Notably, within-group variability was substantial and often dwarfed between-group differences. The overall pattern of clinical course over 20 years indicated that schizophrenia typically followed a chronic course (74.1% of patients continuously ill), whereas an episodic course was common in bipolar disorder with psychosis (79.5%) and psychotic depression (66.7%), and the other two groups fell between them. The psychotic depression, substance-induced psychosis, and other/undetermined groups were too small (Ns <50) for planned analyses; therefore, we combined groups based on similarity of course, resulting in three larger categories: schizophrenia (N=175), mood disorders with psychosis (N=137), and other psychoses (N=61).

Next, we tested the significance of changes within the three groups from month 6 to each follow-up wave (Figure 1). GAF scores remained stable or improved from month 6 through month 48 for each group, but thereafter they declined by 13 points in the schizophrenia group and 9 points in the other groups. With regard to specific symptom dimensions, apathy-asociality also remained stable or improved through month 48 but then worsened (Cohen’s d values, 0.35–0.73, comparing year 20 to month 6). Inexpressivity improved through year 10, but by year 20 it returned to initial levels. Reality distortion symptoms were at stable low levels throughout the follow-up period in the psychotic mood disorders and other psychoses groups. In the schizophrenia group, reality distortion ratings were stable through month 48 but then increased substantially (d=0.45), whereas disorganization worsened even more (d=0.61). These increases are particularly notable given that rates of antipsychotic medication use remained largely stable in the schizophrenia group, while they declined dramatically in the other groups (see Figure S1 in the data supplement). In contrast, depression ratings decreased and mania/excitement ratings remained stable across the interval.

We also compared clinical course among disorders, focusing on the initial outcome (month 6), the long-term outcome (year 20), and the change between these waves (see Table S1 in the data supplement). Compared with the mood disorders with psychosis group, the schizophrenia group had consistently worse outcomes on the GAF and in ratings for apathy-asociality, inexpressivity, reality distortion, and disorganization. Moreover, worsening was greater in the schizophrenia group than in the psychotic mood disorders group on these outcomes, except for inexpressivity. No differences were observed between disorder groups on depression and mania/excitement. The only difference between the other psychoses group and the psychotic mood disorders group was higher ratings in the former on reality distortion symptoms at year 20.

Although highly accurate, 10-year diagnosis may be confounded by illness course during the first decade. To consider the impact of this potential confounding, we repeated the analyses using 6-month diagnosis (see Figure S2 in the data supplement). Schizophrenia trajectories were virtually unchanged on the GAF and on ratings for apathy-asociality, inexpressivity, reality distortion, and disorganization. Trajectories of the psychotic mood disorders and other psychoses groups were more severe using 6-month diagnoses compared with 10-year diagnoses. This pattern is consistent with misclassification, that is, that some patients who followed a schizophrenia trajectory were assigned other diagnoses at month 6. Indeed, we previously found that many 6-month nonschizophrenia cases in this cohort were later reclassified as schizophrenia, whereas few patients were reclassified from schizophrenia to other diagnoses (17). The reanalysis with 6-month diagnosis had little impact on trajectories of mood symptoms.

Next, we used multilevel spline regression to estimate trajectories for the three groups across the entire follow-up period for each outcome. The number of segments was determined empirically (see Table S2 in the data supplement). Selected models either were linear (i.e., had only one segment) or were allowed one change in the trajectory’s slope (i.e., had two segments). Estimated trajectories (Figure 2) closely resembled longitudinal patterns obtained by smoothing raw data (see Figure S3 in the data supplement), which suggests that the models represented the data well. Table 3 lists magnitude of change in symptom scores per year (unadjusted columns).

GAF score declined significantly in the schizophrenia group; it improved initially in the psychotic mood disorders group and the other psychoses group but declined significantly after approximately year 7 (Figure 2). Apathy-asociality ratings worsened in all groups, although in the psychotic mood disorders group it improved through about year 7 and then deteriorated well beyond the initial level. Inexpressivity lessened until approximately year 7 but increased thereafter, except in the other psychoses group, where the trajectory was flat throughout. Reality distortion ratings increased in the schizophrenia group but remained stable in the other groups. Disorganization ratings worsened in all groups. Depression ratings improved in all groups, but in the psychotic mood disorders group, improvement plateaued at year 2. Mania/excitement ratings did not change significantly across the interval. Changes in GAF score were primarily driven by changes in apathy-asociality and reality distortion ratings (see Table S3 in the data supplement).

To test whether the observed patterns reflect effects of aging or changes in treatment rather than illness evolution, we repeated the analyses controlling for age and antipsychotic use at each assessment point (Table 3). Age had no effect on psychopathology after accounting for time since baseline. Antipsychotic use was associated with worse GAF scores, inexpressivity ratings, and apathy-asociality ratings overall, but with lower disorganization and mania/excitement ratings in the schizophrenia group. Also, antipsychotic use was associated with lower reality distortion ratings in the schizophrenia group, but greater symptom ratings in the other psychoses group. This pattern may indicate medication side effects or selection effects (for example, sicker participants are more likely to receive long-term treatment with antipsychotics). Adjustment for these two variables did not change the findings for illness trajectories except that three slopes became nonsignificant in the other psychoses group and two became nonsignificant in the psychotic mood disorders group. Adjustments for various other potential confounders had little impact on the pattern of results (see the Supplementary Results section and Table S4 in the data supplement).