Transition of Substance-Induced, Brief, and Atypical Psychoses to Schizophrenia

A Systematic Review and Meta-analysis

Benjamin Murrie; Julia Lappin; Matthew Large; Grant Sara


Schizophr Bull. 2020;46(3):505-516. 

In This Article


Search Results

The search strategy identified 6097 potentially relevant publications, of which 5906 were excluded following abstract review, and a further 141 excluded after review of full text (figure 1). Four additional papers were identified through hand searching, resulting in 50 eligible studies included.

Figure 1.

PRISMA flow chart.

The 50 eligible studies[11–13,16,17,19,36,41,42,49–89] (Table 1) provided 79 estimates of transition to a diagnosis of schizophrenia among 40 783 people, including 25 studies of substance-induced psychosis (34 244 people). The mean follow-up period was 4.0 years (range 1–20 y) or 8.4 years when weighted by the number of participants in each study. Study samples included more males than females (mean study proportion male 61%, weighted mean 72%). The mean study age was 28 years (weighted mean 29 y). Studies were from 25 countries including England (5 studies), Denmark (4), United States (4), Ireland, Sweden, Germany, and India (3 each): these were aggregated into regional groupings for subgroup analysis. Diagnoses were most often made by structured interview (22) or by extraction of routine clinical diagnoses from medical records or registers (16). For most studies, the index diagnosis was made in hospital (27) or in mixed hospital and community (13) settings. All but 8 studies[13,17,61,70,72–74,85] examined first-episode cohorts. All eligible studies used a cohort design.

Pooled Rate of Transition to Schizophrenia

Overall one-quarter (25%, 95% CI 18%–35%) of people with substance-induced psychosis had a follow-up diagnosis of schizophrenia (Table 2). This pooled estimate was lower than that for brief, atypical, NOS psychoses, and schizophreniform psychosis (between-group Q 5,830, df 3, P < .0001). There was substantial heterogeneity between studies with non-substance-induced psychosis (Table 2). Amongst the brief, atypical, and NOS group, transition rates were lower in those with brief and atypical psychoses (26 estimates, 30% transition, 95% CI 23%–38%) than in psychosis NOS (18 estimates, 46% transition, 95% CI 40%–52%).

The 25 studies of substance-induced psychosis provided 43 substance-specific estimates (Table 2). Substance-specific estimates differed significantly (Q 137, df 6, P < .0001). Pooled estimates of transition to schizophrenia were highest (34%, 95% CI 25%–46%) for cannabis-induced psychoses, intermediate for amphetamines and hallucinogens, and lowest for alcohol-, sedative- and opioid-induced psychoses. Within-group heterogeneity (I 2) exceeded 90% for all substance types where it could be meaningfully estimated on the basis of more than 3 samples.

Subgroup Analysis

Subgroups of the studies examining substance-induced psychosis were compared (Table 3). After correction for multiple comparisons, study design characteristics did not predict significant between-group differences. Continuous moderators were examined using meta-regression (Table 4). Studies of older cohorts reported lower rates of transition to schizophrenia. There was no association between transition rate and sex, duration of follow-up, proportion of sample followed up or year of publication. There were insufficient studies for meta-regression of PANSS positive, negative, or total scores (reported by 3 studies), percent with comorbid substance use (4 studies) or GAF scores (4 studies).

Study Quality

Study quality did not predict significant between-group differences: studies scoring above and below the median (5 or more on the Newcastle-Ottawa Scale) did not differ in estimates (Table 3). Potential impact of study quality was also examined by subgroup analysis for each integer value of the quality scale, and by meta-regression on quality scores as a continuous variable (supplementary material): no significant effects of study quality were observed using any method.

Publication Bias

There was no apparent impact of publication bias on pooled estimates for psychosis subtypes (Table 5). Egger's test was not significant for any psychosis subgroup, and Duval and Tweedie's trim and fill test was therefore not conducted. Funnel plots for psychosis subgroups are provided as supplementary material.