Behind the heterogeneity in the long-term course of first-episode psychosis: Different psychotic symptom trajectories are associated with different patterns of cannabis and stimulant use

Background: Data-driven classification of long-term psychotic symptom trajectories and identification of associated risk factors could assist treatment planning and improve long-term outcomes in psychosis. However, few studies have used this approach, and knowledge about underlying mechanisms is limited. Here, we identify long-term psychotic symptom trajectories and investigate the role of illness-concurrent cannabis and stimulant use. Methods: 192 participants with first-episode psychosis were followed up after 10 years. Psychotic symptom trajectories were estimated using growth mixture modeling and tested for associations with baseline characteristics and cannabis and stimulant use during the follow-up (FU) period. Results: Four trajectories emerged: (1) Stable Psychotic Remission (54.2 %), (2) Delayed Psychotic Remission (15.6 %), (3) Psychotic Relapse (7.8 %), (4) Persistent Psychotic Symptoms (22.4 %). At baseline, all unfavorable trajectories (2 – 4) were characterized by more schizophrenia diagnoses, higher symptom severity, and longer duration of untreated psychosis. Compared to the Stable Psychotic Remission trajectory, unstable trajectories (2,3) showed distinct associations with cannabis/stimulant use during the FU-period, with dose-dependent effects for cannabis but not stimulants (Delayed Psychotic Remission: higher rates of frequent cannabis and stimulant use during the first 5 FU-years; Psychotic Relapse: higher rates of sporadic stimulant use throughout the entire FU-period). The Persistent Psychosis trajectory was less clearly linked to substance use during the FU-period. Conclusions: The risk for an adverse long-term course could be mitigated by treatment of substance use, where particular attention should be devoted to preventing the use of stimulants while the use reduction of cannabis may already yield positive effects.


Introduction
Psychotic disorders are long-term relapsing-remitting disorders.There are, however, remarkably few studies of first-episode psychosis (FEP) cohorts with an observation period of more than five years from the first treatment.The existing studies agree on a substantial heterogeneity in long-term outcomes of FEP, including in the remission and recovery from positive psychotic symptoms (Canal-Rivero et al., 2023;Chan et al., 2020a;Hansen et al., 2023;Kotov et al., 2017;Morgan et al., 2017;O'Keeffe et al., 2022;Peralta et al., 2023;ten Velden Hegelstad et al., 2013).Importantly, heterogeneity in the course of positive psychotic symptoms also exceeds the binary differentiation between those who remit/recover and those who do not.Instead, there are considerable inter-individual differences in how psychotic symptoms wax and wane over time, with some displaying more episodic and others more stable course characteristics (Morgan et al., 2021;Morgan et al., 2014).Identifying these course types and their underlying mechanisms is essential to improve outcome prediction and clinical interventions.The complex variations are best captured by data-driven approaches such as growth mixture modeling (Morgan et al., 2021;Ram and Grimm, 2009).However, to our knowledge, only three FEP studies have tried to identify the complex long-term trajectories of positive psychotic symptoms in this manner (Austin et al., 2015;Cuesta et al., 2024;Morgan et al., 2021).Of these, only one has analyzed psychotic symptomatology per follow-up year (Morgan et al., 2021).Furthermore, the mechanisms underlying the heterogeneity in the observed psychotic symptom trajectories remain poorly understood.Typically, baseline clinical and demographic characteristics are tested for their association with the identified course types, but findings have been inconsistent and unspecific (Austin et al., 2015;Cuesta et al., 2024;Morgan et al., 2021).This may be because baseline characteristics fail to capture mechanisms influencing clinical changes later in the course of illness.
A candidate mechanism for late changes is illness-concurrent substance use, particularly of cannabis and stimulants (cocaine, amphetamine).Cannabis and stimulants are the most frequently used substances in FEP besides alcohol (Abdel-Baki et al., 2017).Both are assumed to influence the dopaminergic system, the core neurotransmitter system implicated in psychotic disorders and psychotic relapse (Howes and Kapur, 2009;Maia and Frank, 2017;Murray et al., 2017;Remington et al., 2014), can cause psychotic-like symptoms in healthy individuals, are associated with conversion to schizophrenia following an episode of substance-induced psychosis (Rognli et al., 2023;Tikka and D'Souza, 2019), and have been linked to exacerbations of psychotic symptoms in persons with psychotic disorders (Brown et al., 2020;Hasan et al., 2020;San et al., 2013;Schoeler et al., 2016a), including treatment-compliant individuals (Levi et al., 2023).Individual differences in the long-term use of cannabis and/or stimulants may thus contribute to the observed heterogeneity in long-term symptom trajectories.Yet, how long-term use patterns of cannabis and stimulants are associated with long-term psychotic symptom trajectories in FEP has not yet been investigated.Multiple short- (Sara et al., 2014;Schoeler et al., 2016b;Schoeler et al., 2017;Schoeler et al., 2016c) and one long-term follow-up study (Weibell et al., 2017) have found that continued substance use during the first two years after an index episode is associated with poorer outcomes.However, the role of use beyond these first two years has not been explored.Importantly, the effects may be dosedependent.This has been observed in short-term follow-up studies where more frequent cannabis use was associated with higher relapse risk (Schoeler et al., 2016b;Schoeler et al., 2016c).Similar studies for stimulants are lacking.
In sum, investigations of substance use patterns as a possible mechanism behind different long-term trajectories of psychotic symptoms need to consider the type, frequency, and time-point of use.The current study thus aims to address this knowledge gap by a) identifying the longterm trajectories of psychotic symptoms in a cohort of individuals with FEP recruited in their first year of treatment and then followed up after 10 years and b) examining how these trajectories are associated with patterns of cannabis and stimulant use during the follow-up period.

Participants
In total, 513 individuals with FEP were recruited during their first year of treatment with consecutive inclusion of in-and outpatients in the Norwegian regions of Oslo and Innlandet from 2004 to 2012 as part of the Thematically Organized Psychosis (TOP) project.All participants met the diagnostic criteria for non-affective or affective psychotic disorders, had experienced at least one psychotic episode, and not previously received adequate treatment for psychosis (Kreis et al., 2024).Organic and substance induced psychotic disorders were not included.Further inclusion criteria were: 18-65 years old, estimated IQ ≥ 70, sufficient knowledge of a Scandinavian language to participate in clinical interviews, ability to give informed consent.
The study was conducted in line with the Helsinki Declaration and approved by the Regional Committee for Medical Research Ethics South East Norway.All participants gave written informed consent before participation, including to the linkage of study data to the national patient registry.After ten years (M = 9.70, SD = 1.47), 192 were successfully followed up with personal interviews (Fig. 1).Follow-up time differed slightly between sites (Oslo: M = 10.29,SD = 0.84, n = 156; Innlandet: M = 7.10, SD = 0.41, n = 36).

Clinical assessments
All assessments were conducted by psychologists or medical doctors trained in the comprehensive semi-structured protocol.Medical records were used to validate and supplement information.

Baseline assessments
The Structured Clinical Interview for Mental Disorders (SCID-I; First et al., 1995) was used to confirm diagnoses according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV).Demographic and clinical information included age, gender, years of education, age at onset (first psychotic episode), and duration of untreated psychosis (DUP; number of weeks from the start of the first psychotic episode to the start of the first adequate treatment, log-transformed for analyses).Overall symptom severity was assessed with the Positive and Negative Syndrome Scale (PANSS total score; Kay et al., 1987), and clinical insight with PANSS item g12.Current antipsychotic treatment was recorded, and medication adherence was measured with the MARS-5, an abbreviated version of the Medication Adherence Report Scale validated across different patient groups (Chan et al., 2020b).Only MARS scores of participants who were currently prescribed antipsychotic medication and/or lithium were included in the analyses.The prescribed daily dose of the primary antipsychotic (PDD) was divided by the Defined Daily Dose (DDD; WHO Collaborating Centre for Drug Statistics Methodology Norwegian Institute of Public Health, 2019), as a proxy for antipsychotic medication load (PDD/DDD ratio).Daily nicotine use (use of cigarettes and/or snuff) was registered, as well as recent (past 6 months) use of cannabis, stimulants, alcohol, and other substances (opiates, hallucinogens, benzodiazepines).

Follow-up assessments
At follow-up, the number of weeks with manifest psychotic symptoms during each follow-up year (starting with the first year of treatment) was calculated based on a semi-structured interview using a visual analog timeline, supplemented with information from clinical records ('psychosis timeline').A level of manifest psychotic symptoms was defined as having positive psychotic symptoms corresponding to a score ≥ 4 on PANSS items assessing delusions, hallucinatory behavior, or unusual thought content.
The frequency of cannabis and stimulant use during the first (year 1-5) and the second (year 6-10) half of the follow-up period was assessed using a rating scale ranging from 0 (no use) to 4 (daily use; with 1 = sporadic, 2 = monthly, and 3 = weekly use).For analyses, these ratings were categorized into use frequency categories with the levels 'never' (0), 'sporadic' (1,2), and 'frequent' (3,4).Information on the use of cocaine, amphetamine, and 3,4-methylenedioxymethamphetamine (MDMA) was assessed separately but summarized as one variable to reflect stimulant use in the analyses.If a participant used several stimulants, the one with highest use frequency was used to represent stimulant use frequency.The frequency of alcohol intoxication (consumption of alcohol to become intoxicated or drunk) and use of other substances (opiates, hallucinogens, solvents) were assessed in the same manner.

Statistical analyses
To capture subgroup-dependent patterns of variability in the course of manifest psychotic symptoms over the 10-year follow-up period and to identify the optimal number of subgroups (latent classes) to represent these patterns, we applied growth mixture modeling (Muthén et al., 2002;Ram and Grimm, 2009), using the R package 'lcmm' (Proust-Lima et al., 2017;version 2.0.2, applied in R version 4.2.2).Time (number of weeks) with psychotic symptoms was treated as the dependent and year (range 1-10, in integer steps) as the independent variable.The effect of year can be moderated by latent class membership, with different effects for different subgroups, yielding different psychosis trajectories and returning most likely class membership for each participant.Analyses were conducted and presented following the Guidelines for Reporting on Latent Trajectory Studies (GRoLTS; van de Schoot et al., 2017).Prior to modeling, the variable capturing time with psychotic symptoms was normalized, following Proust-Lima and colleagues (Philipps et al., 2014;Proust-Lima and Philipps, 2023).On the normalized scale, a value of 0 reflects the minimum sample value of weeks with psychotic symptoms (0) and a value of 100 the maximum (52; see Supplementary Information).
To determine the most appropriate trajectory representation (i.e., the growth pattern of psychotic symptoms over time), single-class mixed models were fitted with the normalized psychotic symptom variable as the outcome and differing polynomial degrees in the fixed and random effects of year (Ram and Grimm, 2007).Models including no growth, linear, quadratic, and cubic growth were tested.Model fit was assessed and compared with the Bayesian Information Criterion (BIC), with lower values indicating better fit (Raftery, 1995;Schwarz, 1978).The bestfitting model included linear, quadratic, and cubic fixed and random effects of year (Supplementary Information, Table S3).
Then, the growth mixture model was defined, where intercept and slopes for all polynomial degrees of the year effect were allowed to differ between latent classes.Random effects were simplified to intercept and slope of the linear year effect only (Supplementary Information).The model was fitted for 1 to 6 classes.Again, the BIC was used to determine which class solution best fit the data (van de Schoot et al., 2017).Additionally, entropy values were extracted as a measure of class partition.Here, values closer to 1 indicate higher and values closer to 0 lower accuracy of participant classification into the different latent classes (van de Schoot et al., 2017).Further, posterior class membership probabilities were inspected, reflecting the discrimination power of the model.Here, mean probabilities >0.8 of belonging to the assigned class reflect good discrimination (Proust-Lima et al., 2017; Supplementary Information, Tables S4-6).In addition to the statistical fit criteria, class size and distinctiveness of classes were examined when selecting the best and most reasonable model regarding the number of classes (Supplementary Information, Table S4, Fig. S1).Only solutions where each class contained ≥5 % of the sample and where class trajectories were dissimilar enough to suggest real differences were considered (Morgan et al., 2021;Ram and Grimm, 2009;van de Schoot et al., 2017).
Data on the psychosis timeline was assumed to be missing at random, allowing for unbiased estimation of the statistical model (Proust-Lima et al., 2017).Missingness was primarily explained by inclusion site (Innlandet > Oslo) and none of the clinical or demographic baseline variables were independently associated with missingness (Supplementary Information, Table S2).
Associations between latent trajectory classes, baseline characteristics, and substance use frequency during follow-up were investigated with multinomial logistic regression.Significant bivariate analyses were followed up with multivariable analyses to assess independent associations.Variance inflation factors (VIFs) were inspected to rule out multicollinearity (VIF > 5; James et al., 2021).

Study sample
At baseline, the study sample (n = 192) had slightly better clinical insight (mean difference = 0.27, range 1-7) and included a lower proportion of men (52 % vs. 62 %) and of participants reporting recent stimulant use (8 % vs. 15 %) compared to participants who were lost to follow-up or excluded from analyses due to a missing psychosis timeline (n = 321; see Supplementary Information, Table S1).However, data from the national patient registry indicated no statistically significant group differences in clinical severity, as approximated by the use of specialized health services over the 10-year period (Supplementary Information).

Identification of four distinct psychotic symptom trajectories
A four-class growth mixture model proved to be the best fit, with good discrimination power, satisfactory class partition, sufficiently large class sizes, the second-best model fit according to statistical selection criteria alone, and the single best fit when taking all selection criteria into account (Supplementary Information, Tables S4-6, Fig. S1).

Baseline characteristics associated with psychotic symptom trajectories
The majority of participants with Persistent Psychotic Symptoms-or Psychotic Relapse trajectories had a schizophrenia spectrum disorder (72 % and 80 %, respectively).The Delayed Psychotic Remission trajectory consisted mostly of participants with a schizophrenia spectrum disorder (57 %) or 'other psychosis' (40 %), and the Stable Psychotic Remission trajectory included the highest percentage of participants with a psychotic bipolar disorder (44 %; Table 1).
Bivariate multinomial logistic regressions revealed no significant associations between trajectories and demographic characteristics (Table 2).However, participants with less favorable trajectories (Delayed Psychotic Remission, Psychotic Relapse, Persistent Psychotic Symptoms) were more likely to have a 'core' schizophrenia diagnosis (i.e., as opposed to all other diagnoses, including schizophreniform and schizoaffective disorder), greater symptom severity (PANSS total), and a longer DUP than the Stable Psychotic Remission trajectory (Table 2).The Persistent Psychotic Symptoms and the Delayed Psychotic Remission trajectories, in particular, were characterized by higher rates of recent cannabis and stimulant use as well as daily nicotine intake, and participants in both the Persistent Psychotic Symptoms and the Psychotic Relapse trajectory had significantly larger clinical insight impairments.There was no association between trajectories and medication adherence.
In multivariable analyses (Supplementary Information, Table S8), mainly associations for the Delayed Psychotic Remission and the Persistent Psychotic Symptoms trajectories remained significant, with a statistically significant effect of DUP and overall symptom severity when other factors were controlled for.For the Persistent Psychotic Symptoms trajectory, this was also true for recent cannabis use.The effect of a schizophrenia diagnosis remained significant only for the Psychotic Relapse trajectory.There were no signs of multicollinearity, with VIFs of all included variables < 2.

Associations between cannabis and stimulant use during follow-up and psychotic symptom trajectories
The psychotic symptom trajectories were significantly and distinctly associated with specific patterns of cannabis and stimulant use during the 10-year follow-up period.
Compared to the Stable Psychotic Remission trajectory, the Delayed Psychotic Remission trajectory was characterized by higher rates of frequent, but not sporadic use of cannabis and stimulants during the first (year 1-5) but not the second half of the follow-up period (year 6-10; Fig. 3, Table 3).In contrast, the Psychotic Relapse trajectory was characterized by higher rates of sporadic stimulant use both during the first and the second half, with no significant links to cannabis use (Table 3).The Persistent Psychotic Symptoms trajectory showed the least prevalent associations with substance use, with higher rates of frequent but not sporadic use of cannabis during the first half of the follow-up period only (Fig. 3, Table 3).Frequent use of stimulants during the second half of the follow-up period was too rare to yield reliable estimates (Fig. 3, Tables 3, S9) and stimulant use was therefore dichotomized for the respective multivariable model.Counts per trajectory and for each level of cannabis/stimulant use frequency are presented in the Supplementary Information (Tables S9-10).The frequency of alcohol intoxication was not associated with trajectory class membership (Table 3), and the use of other substances was rare and thus not controlled for in statistical analyses (Supplementary Information).
When controlling the use of cannabis for stimulant use (and vice versa) in multivariable analyses conducted separately for the first and the second half of the follow-up period (Supplementary Information, Tables S11-12), only the associations between stimulant use and the Psychotic Relapse trajectory remained significant (sporadic use during the first half of the follow-up period, and any stimulant use during the second half).There were no signs of multicollinearity, with VIFs of all included variables < 2.

Discussion
The findings of this study demonstrate high variability in the longterm trajectories of psychotic symptoms in FEP.Most participants (~54 %) achieved stable psychotic remission, and a smaller but substantial proportion (~22 %) showed persistent psychotic symptoms.In contrast to these comparatively stable symptom trajectories, symptoms were more unstable and fluctuating in a minority of the study participants (Delayed Psychotic Remission trajectory, ~16 %, Psychotic Relapse trajectory, ~8 %).Course types, specifically the unstable ones, were associated with particular patterns of cannabis and stimulant use during the follow-up period.Hence, differences in substance use could be part of the mechanisms underlying the observed heterogeneity of the long-term course of psychotic symptoms.
The different psychotic symptom trajectories are similar to those identified by a recent study using similar methods (Morgan et al., 2021).This strengthens their validity and relevance, as these course types seem to exist across FEP samples.All unfavorable trajectories were associated with a baseline schizophrenia diagnosis, higher overall symptom severity, and a longer DUP.These unspecific associations demonstrate the difficulty of predicting the specific course of psychosis based on baseline attributes (Morgan et al., 2021;Suvisaari et al., 2018).Furthermore, there was a lack of alignment between the associations of trajectories with cannabis/stimulant use at baseline compared to cannabis/stimulant use during follow-up.This illustrates that baseline assessments of substance use do not reliably capture future patterns of use and their effects on psychosis.
Analyses of substance use during the follow-up period revealed distinct associations with psychosis trajectories, with converging temporal patterns between trajectories and substance use patterns.Specifically, the Delayed Psychotic Remission trajectory was associated with frequent use of cannabis and stimulants only in the first 5 years of the 1 Class names were chosen to best reflect the observed patterns of symptom development.Note that the term 'remission' here is used purely descriptively, to capture the decrease of psychotic symptom frequency to very low levels (around 0), but does not refer to any established consensus criteria.More detailed descriptions of the patterns in each trajectory are provided in the Fig. 2 notes.
follow-up, i.e., a period when the frequency of psychotic episodes was still comparatively high (Fig. 2).In contrast, the Psychotic Relapse trajectory, characterized by an increase in psychotic symptom frequency between year 3 and year 8 (Fig. 2), was associated with sporadic stimulant use during the entire follow-up period.This suggests that the illness-concurrent use of cannabis and stimulants played a role in shaping the course of psychosis over time, possibly eliciting or exacerbating psychotic symptoms.The design of the current study does not allow for causal inference, and substance use could follow the manifestation of psychotic symptoms instead of preceding it, in an attempt to self-medicate.However, when cannabis use and positive psychotic symptoms co-occur, cannabis is rarely used to alleviate positive symptoms (Dekker et al., 2009;Hanna et al., 2017;Mané et al., 2015).Furthermore, recent studies clearly point toward a causal effect of cannabis use on subsequent psychotic relapses (Levi et al., 2023;Schoeler et al., 2016c).
Associations between cannabis use and psychotic symptom trajectories were dose-dependent, mirroring findings from short-term followup studies on relapse risk (Schoeler et al., 2016b;Schoeler et al., 2016c).In contrast, the effect of stimulants was less dose-dependent, with both sporadic and frequent use linked to different trajectories.Neurobiologically, both cannabis and stimulants are assumed to affect the dopaminergic system, the dysfunction of which is assumed to underlie psychotic disorders and psychotic relapse (Howes and Kapur, 2009;Maia and Frank, 2017;Murray et al., 2017;Remington et al., 2014).The effects of stimulants on dopaminergic transmission seem well established (Maia and Frank, 2017;Remington et al., 2014), and the intake of stimulants can trigger and exacerbate psychotic symptoms in individuals with psychosis at lower doses than in individuals without (Curran et al., 2004;Lieberman et al., 1990;Remington et al., 2014).In contrast, the neurobiological effects of cannabis and how they relate to psychosis are less clear.They appear to involve dopaminergic transmission to a lesser degree than stimulants (Murray et al., 2017), as well as glutamatergic and GABAergic transmission (Fischer et al., 2023;Mason et al., 2022;Remington et al., 2014;Sami and Bhattacharyya, 2018).Sporadic use of stimulants may be sufficient to exacerbate psychotic symptoms and accelerate psychotic relapse due to their more direct effect on the dopaminergic system, whereas only frequent use of cannabis will achieve the same effect.An alternative explanation for the diverging findings on dose-dependency is that frequent cannabis users are more likely also to use other drugs, including stimulants (Millar et al., 2021).Hence, the effect of frequent cannabis use on psychotic symptom trajectories could be driven by the concurrent use of stimulants.This interpretation is partially supported by the multivariable model results.Here, the effects of cannabis were non-significant once stimulant use was controlled for, whereas the effects of stimulant use remained significant for the Psychotic Relapse trajectory after controlling for cannabis.While these results must be interpreted with caution due to low counts of frequent substance use, they highlight the critical role of stimulant use for the course of psychosis, with detrimental effects occurring already at sporadic use frequency and independently of cannabis use.
The substance use patterns in the Persistent Psychotic Symptoms trajectory showed less convergence with the course of psychotic symptoms, limited to frequent cannabis use during the first half of the followup period.Therefore, it is likely that other factors contributed to the persistent psychotic symptoms throughout the entire follow-up period, including treatment resistance.Factors other than substance use may also have contributed to the course of psychotic symptoms in the remaining trajectories.These include, but are not limited to, stressful life events, psychosocial support, pharmacological and psychological treatments, and treatment adherence during the follow-up period.In the context of substance use, medication adherence seems particularly relevant, as the effect of substance use on psychotic relapse risk may be partially mediated by lower medication adherence (Colizzi et al., 2016;Schoeler et al., 2017).However, substance use is associated with increased relapse risk even in treatment compliant individuals (Levi     et al., 2023;Rubio et al., 2020), and medication adherence at baseline had no effect on trajectories in the current study.While fluctuations in adherence patterns over time cannot be ruled out, it thus seems unlikely for medication adherence to explain the observed findings in their entirety.
The current study focused on the course of positive psychotic symptoms.Other symptoms relevant to long-term outcomes, such as negative, cognitive, or affective symptoms, may evolve differently (Austin et al., 2015;Cuesta et al., 2024) and their association with illness-concurrent substance use patterns may also differ.Limitations of the current study include the lack of information on treatment adherence during the follow-up period, the potential for  retrospective bias when reconstructing past psychotic episodes and the study attrition.Attrition is a common problem in prospective studies of first-episode cohorts.Nevertheless, utilizing registry data on specialized psychiatric health service use during the 10-year follow-up period available for all participants, we found no evidence for differences in illness severity between the study sample and those lost to follow-up.The low sample sizes in the unstable trajectories and some substance use frequency categories, however, led to large confidence intervals in some of the analyses, and the effect of substance use frequency was possibly underestimated in multivariable analyses.The findings should therefore be replicated in larger samples.

Conclusions
These findings demonstrate that some of the variability in the longterm course of psychotic symptoms, especially when symptoms are fluctuating, may be driven by patterns of concurrent substance use.They further highlight the critical role of stimulant use, which increases the risk for a more unfavorable course already when used at low frequency.In contrast, the detrimental effect of cannabis appears to be dosedependent.Based on this, cannabis using individuals with psychosis should be encouraged to reduce their frequency of use if complete cessation seems unrealistic, in line with harm-reduction strategies.These findings provide hope by suggesting that the course of psychotic symptoms may be modulated by interventions targeting substance use.Substance use patterns should thus routinely be monitored, addressed, and incorporated into treatment plans.The factors underlying a persistent course of psychotic symptoms and ways to mitigate their effect remain to be established.

Fig. 2 .
Fig. 2. Estimated trajectories of the 4-class model.Note.Average time with psychotic symptoms per follow-up year is displayed for each trajectory class.The Stable Psychotic Remission class (54.2 %) is characterized by remission of psychotic symptoms within the first two years and persistently low frequency throughout the remainder of the 10-year follow-up period.The Delayed Psychotic Remission class (15.6 %) is characterized by high, though steadily decreasing frequency of psychotic symptoms within the first half of the follow-up period [year 1-5] and remission during the second half [year 6-10]).The Psychotic Relapse class (7.8 %) is characterized by medium frequency of psychotic symptoms that improved slightly within the first 3 years and worsened during the second half of the follow-up period before improving during the last 2 years.The Persistent Psychotic Symptoms class (22.4 %) is characterized by high frequency of psychotic symptoms that persisted throughout the entire 10-year follow-up period.

Fig. 3 .
Fig. 3. Cannabis and stimulant use during follow-up by trajectory.Note.Percentage of participants within each use frequency category (transparency) per substance type (color) and trajectory class (x-axis).Numbers have been rounded to the closest integer.

Table 1
Descriptive statistics of baseline variables by trajectory (N = 192).

Table 2
Bivariate multinomial logistic regression results of baseline variables.
Note.OR = Odds ratio (supplied with lower and upper boundaries of the 95 % confidence interval), Core SCZ = core diagnosis of schizophrenia, DUP (log) = logtransformed duration of untreated psychosis in weeks, PANSS = PANSS total score, Insight = clinical insight impairment assessed with PANSS item g12, PDD/DDD = ratio of prescribed daily dose to Defined Daily dose of the primary antipsychotic (antipsychotic medication load), MARS = medication adherence score, Other subst.= substances other than nicotine, alcohol, cannabis, or stimulants.Reference levels of categorical predictors are 'female' for gender, and 'no' for core SCZ and substance use variables (incl.nicotine and alcohol).Values are rounded to 2 and 3 (p-values only) decimal places.

Table 3
Bivariate multinomial logistic regression results of substance use variables.